add index_stats to python api

Added a small bit of documentation for the `dim` feature which is
provided by the new `text-embedding-3` model series that allows users to
shorten an embedding.

Happy to discuss a bit on the phrasing but I struggled quite a bit with
getting it to work so wanted to help others who might want to use the
newer model too

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.4.3
+current_version = 0.4.12
 commit = True
 message = Bump version: {current_version} → {new_version}
 tag = True
@@ -9,4 +9,4 @@ tag_name = v{new_version}
 
 [bumpversion:file:rust/ffi/node/Cargo.toml]
 
-[bumpversion:file:rust/vectordb/Cargo.toml]
+[bumpversion:file:rust/lancedb/Cargo.toml]

.cargo/config.toml

@@ -0,0 +1,40 @@
+[profile.release]
+lto = "fat"
+codegen-units = 1
+
+[profile.release-with-debug]
+inherits = "release"
+debug = true
+# Prioritize compile time over runtime performance
+codegen-units = 16
+lto = "thin"
+
+[target.'cfg(all())']
+rustflags = [
+    "-Wclippy::all",
+    "-Wclippy::style",
+    "-Wclippy::fallible_impl_from",
+    "-Wclippy::manual_let_else",
+    "-Wclippy::redundant_pub_crate",
+    "-Wclippy::string_add_assign",
+    "-Wclippy::string_add",
+    "-Wclippy::string_lit_as_bytes",
+    "-Wclippy::string_to_string",
+    "-Wclippy::use_self",
+    "-Dclippy::cargo",
+    "-Dclippy::dbg_macro",
+    # not too much we can do to avoid multiple crate versions
+    "-Aclippy::multiple-crate-versions",
+    "-Aclippy::wildcard_dependencies",
+]
+
+[target.x86_64-unknown-linux-gnu]
+rustflags = ["-C", "target-cpu=haswell", "-C", "target-feature=+avx2,+fma,+f16c"]
+
+[target.aarch64-apple-darwin]
+rustflags = ["-C", "target-cpu=apple-m1", "-C", "target-feature=+neon,+fp16,+fhm,+dotprod"]
+
+# Not all Windows systems have the C runtime installed, so this avoids library
+# not found errors on systems that are missing it.
+[target.x86_64-pc-windows-msvc]
+rustflags = ["-Ctarget-feature=+crt-static"]

.github/workflows/build_linux_wheel/action.yml

@@ -0,0 +1,58 @@
+# We create a composite action to be re-used both for testing and for releasing
+name: build-linux-wheel
+description: "Build a manylinux wheel for lance"
+inputs:
+  python-minor-version:
+    description: "8, 9, 10, 11, 12"
+    required: true
+  args:
+    description: "--release"
+    required: false
+    default: ""
+  arm-build:
+    description: "Build for arm64 instead of x86_64"
+    # Note: this does *not* mean the host is arm64, since we might be cross-compiling.
+    required: false
+    default: "false"
+runs:
+  using: "composite"
+  steps:
+    - name: CONFIRM ARM BUILD
+      shell: bash
+      run: |
+        echo "ARM BUILD: ${{ inputs.arm-build }}"
+    - name: Build x86_64 Manylinux wheel
+      if: ${{ inputs.arm-build == 'false' }}
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        working-directory: python
+        target: x86_64-unknown-linux-gnu
+        manylinux: "2_17"
+        args: ${{ inputs.args }}
+        before-script-linux: |
+          set -e
+          yum install -y openssl-devel \
+            && curl -L https://github.com/protocolbuffers/protobuf/releases/download/v24.4/protoc-24.4-linux-$(uname -m).zip > /tmp/protoc.zip \
+            && unzip /tmp/protoc.zip -d /usr/local \
+            && rm /tmp/protoc.zip
+    - name: Build Arm Manylinux Wheel
+      if: ${{ inputs.arm-build == 'true' }}
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        working-directory: python
+        target: aarch64-unknown-linux-gnu
+        manylinux: "2_24"
+        args: ${{ inputs.args }}
+        before-script-linux: |
+          set -e
+          apt install -y unzip
+          if [ $(uname -m) = "x86_64" ]; then
+            PROTOC_ARCH="x86_64"
+          else
+            PROTOC_ARCH="aarch_64"
+          fi
+          curl -L https://github.com/protocolbuffers/protobuf/releases/download/v24.4/protoc-24.4-linux-$PROTOC_ARCH.zip > /tmp/protoc.zip \
+            && unzip /tmp/protoc.zip -d /usr/local \
+            && rm /tmp/protoc.zip

.github/workflows/build_mac_wheel/action.yml

@@ -0,0 +1,25 @@
+# We create a composite action to be re-used both for testing and for releasing
+name: build_wheel
+description: "Build a lance wheel"
+inputs:
+  python-minor-version:
+    description: "8, 9, 10, 11"
+    required: true
+  args:
+    description: "--release"
+    required: false
+    default: ""
+runs:
+  using: "composite"
+  steps:
+    - name: Install macos dependency
+      shell: bash
+      run: |
+        brew install protobuf
+    - name: Build wheel
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        args: ${{ inputs.args }}
+        working-directory: python
+        interpreter: 3.${{ inputs.python-minor-version }}

.github/workflows/build_windows_wheel/action.yml

@@ -0,0 +1,33 @@
+# We create a composite action to be re-used both for testing and for releasing
+name: build_wheel
+description: "Build a lance wheel"
+inputs:
+  python-minor-version:
+    description: "8, 9, 10, 11"
+    required: true
+  args:
+    description: "--release"
+    required: false
+    default: ""
+runs:
+  using: "composite"
+  steps:
+    - name: Install Protoc v21.12
+      working-directory: C:\
+      run: |
+        New-Item -Path 'C:\protoc' -ItemType Directory
+        Set-Location C:\protoc
+        Invoke-WebRequest https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win64.zip -OutFile C:\protoc\protoc.zip
+        7z x protoc.zip
+        Add-Content $env:GITHUB_PATH "C:\protoc\bin"
+      shell: powershell
+    - name: Build wheel
+      uses: PyO3/maturin-action@v1
+      with:
+        command: build
+        args: ${{ inputs.args }}
+        working-directory: python
+    - uses: actions/upload-artifact@v3
+      with:
+        name: windows-wheels
+        path: python\target\wheels

.github/workflows/cargo-publish.yml

@@ -16,7 +16,7 @@ jobs:
     # Only runs on tags that matches the make-release action
     if: startsWith(github.ref, 'refs/tags/v')
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
       - uses: Swatinem/rust-cache@v2
         with:
           workspaces: rust
@@ -26,4 +26,4 @@ jobs:
           sudo apt install -y protobuf-compiler libssl-dev
       - name: Publish the package
         run: |
-          cargo publish -p vectordb --all-features --token ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          cargo publish -p lancedb --all-features --token ${{ secrets.CARGO_REGISTRY_TOKEN }}

.github/workflows/docs.yml

@@ -24,12 +24,16 @@ jobs:
     environment:
       name: github-pages
       url: ${{ steps.deployment.outputs.page_url }}
-    runs-on: ubuntu-22.04
+    runs-on: buildjet-8vcpu-ubuntu-2204
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
+      - name: Install dependecies needed for ubuntu
+        run: |
+          sudo apt install -y protobuf-compiler libssl-dev
+          rustup update && rustup default        
       - name: Set up Python
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v5
         with:
           python-version: "3.10"
           cache: "pip"
@@ -42,7 +46,7 @@ jobs:
       - name: Set up node
         uses: actions/setup-node@v3
         with:
-          node-version: ${{ matrix.node-version }}
+          node-version: 20
           cache: 'npm'
           cache-dependency-path: node/package-lock.json
       - uses: Swatinem/rust-cache@v2
@@ -62,8 +66,9 @@ jobs:
         run: |
           npx typedoc --plugin typedoc-plugin-markdown --out ../docs/src/javascript src/index.ts
       - name: Build docs
+        working-directory: docs
         run: |
-          PYTHONPATH=. mkdocs build -f docs/mkdocs.yml
+          PYTHONPATH=. mkdocs build
       - name: Setup Pages
         uses: actions/configure-pages@v2
       - name: Upload artifact
@@ -72,4 +77,4 @@ jobs:
           path: "docs/site"
       - name: Deploy to GitHub Pages
         id: deployment
-        uses: actions/deploy-pages@v1
+        uses: actions/deploy-pages@v1

.github/workflows/docs_test.yml

@@ -18,26 +18,28 @@ on:
 env:
   # Disable full debug symbol generation to speed up CI build and keep memory down
   # "1" means line tables only, which is useful for panic tracebacks.
-  RUSTFLAGS: "-C debuginfo=1"
+  RUSTFLAGS: "-C debuginfo=1 -C target-cpu=native -C target-feature=+f16c,+avx2,+fma"
   RUST_BACKTRACE: "1"
 
 jobs:
   test-python:
     name: Test doc python code
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        python-minor-version: [ "11" ]
-        os: ["ubuntu-22.04"]
+    runs-on: "buildjet-8vcpu-ubuntu-2204"
     steps:
     - name: Checkout
-      uses: actions/checkout@v3
+      uses: actions/checkout@v4
+    - name: Install dependecies needed for ubuntu
+      run: |
+        sudo apt install -y protobuf-compiler libssl-dev
+        rustup update && rustup default
     - name: Set up Python
-      uses: actions/setup-python@v4
+      uses: actions/setup-python@v5
       with:
-        python-version: 3.${{ matrix.python-minor-version }}
+        python-version: 3.11
         cache: "pip"
         cache-dependency-path: "docs/test/requirements.txt"
+    - name: Rust cache
+      uses: swatinem/rust-cache@v2        
     - name: Build Python
       working-directory: docs/test
       run:
@@ -52,45 +54,43 @@ jobs:
         for d in *; do cd "$d"; echo "$d".py; python "$d".py; cd ..; done
   test-node:
     name: Test doc nodejs code
-    runs-on: ${{ matrix.os }}
+    runs-on: "buildjet-8vcpu-ubuntu-2204"
+    timeout-minutes: 60
     strategy:
-      matrix:
-        node-version: [ "18" ]
-        os: ["ubuntu-22.04"]
+      fail-fast: false
     steps:
     - name: Checkout
-      uses: actions/checkout@v3
+      uses: actions/checkout@v4
       with:
         fetch-depth: 0
         lfs: true
     - name: Set up Node
-      uses: actions/setup-node@v3
+      uses: actions/setup-node@v4
       with:
-        node-version: ${{ matrix.node-version }}
+        node-version: 20
     - name: Install dependecies needed for ubuntu
-      if: ${{ matrix.os == 'ubuntu-22.04' }}
       run: |
         sudo apt install -y protobuf-compiler libssl-dev
-    - name: Install node dependencies
-      run: |
-        cd docs/test
-        npm install 
+        rustup update && rustup default
     - name: Rust cache
       uses: swatinem/rust-cache@v2
-    - name: Install LanceDB
+    - name: Install node dependencies
       run: |
-        cd docs/test/node_modules/vectordb 
+        sudo swapoff -a
+        sudo fallocate -l 8G /swapfile
+        sudo chmod 600 /swapfile
+        sudo mkswap /swapfile
+        sudo swapon /swapfile
+        sudo swapon --show
+        cd node
         npm ci
         npm run build-release
-        npm run tsc
-    - name: Create test files
-      run: |
-        cd docs/test
-        node md_testing.js
+        cd ../docs
+        npm install
     - name: Test
       env:
         LANCEDB_URI: ${{ secrets.LANCEDB_URI }}
         LANCEDB_DEV_API_KEY: ${{ secrets.LANCEDB_DEV_API_KEY }}
       run: |
-        cd docs/test/node
-        for d in *; do cd "$d"; echo "$d".js; node "$d".js; cd ..; done
+        cd docs
+        npm t

.github/workflows/make-release-commit.yml

@@ -26,7 +26,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out main
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           ref: main
           persist-credentials: false
@@ -37,10 +37,10 @@ jobs:
         run: |
           git config user.name 'Lance Release'
           git config user.email 'lance-dev@lancedb.com'
-      - name: Set up Python 3.10
-        uses: actions/setup-python@v4
+      - name: Set up Python 3.11
+        uses: actions/setup-python@v5
         with:
-          python-version: "3.10"
+          python-version: "3.11"
       - name: Bump version, create tag and commit
         run: |
           pip install bump2version

.github/workflows/node.yml

@@ -24,27 +24,6 @@ env:
   RUST_BACKTRACE: "1"
 
 jobs:
-  lint:
-    name: Lint
-    runs-on: ubuntu-22.04
-    defaults:
-      run:
-        shell: bash
-        working-directory: node
-    steps:
-    - uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-        lfs: true
-    - uses: actions/setup-node@v3
-      with:
-        node-version: 20
-        cache: 'npm'
-        cache-dependency-path: node/package-lock.json
-    - name: Lint
-      run: |
-        npm ci
-        npm run lint
   linux:
     name: Linux (Node ${{ matrix.node-version }})
     timeout-minutes: 30
@@ -57,7 +36,7 @@ jobs:
         shell: bash
         working-directory: node
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v4
       with:
         fetch-depth: 0
         lfs: true
@@ -89,7 +68,7 @@ jobs:
         shell: bash
         working-directory: node
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v4
       with:
         fetch-depth: 0
         lfs: true
@@ -128,7 +107,7 @@ jobs:
       # this one is for dynamodb
       DYNAMODB_ENDPOINT: http://localhost:4566
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v4
       with:
         fetch-depth: 0
         lfs: true

.github/workflows/nodejs.yml

@@ -0,0 +1,114 @@
+name: NodeJS (NAPI)
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    paths:
+      - nodejs/**
+      - .github/workflows/nodejs.yml
+      - docker-compose.yml
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+env:
+  # Disable full debug symbol generation to speed up CI build and keep memory down
+  # "1" means line tables only, which is useful for panic tracebacks.
+  RUSTFLAGS: "-C debuginfo=1"
+  RUST_BACKTRACE: "1"
+
+jobs:
+  lint:
+    name: Lint
+    runs-on: ubuntu-22.04
+    defaults:
+      run:
+        shell: bash
+        working-directory: nodejs
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+        lfs: true
+    - uses: actions/setup-node@v3
+      with:
+        node-version: 20
+        cache: 'npm'
+        cache-dependency-path: nodejs/package-lock.json
+    - uses: Swatinem/rust-cache@v2
+    - name: Install dependencies
+      run: |
+        sudo apt update
+        sudo apt install -y protobuf-compiler libssl-dev
+    - name: Lint
+      run: |
+        cargo fmt --all -- --check
+        cargo clippy --all --all-features -- -D warnings
+        npm ci
+        npm run lint
+        npm run chkformat
+  linux:
+    name: Linux (NodeJS ${{ matrix.node-version }})
+    timeout-minutes: 30
+    strategy:
+      matrix:
+        node-version: [ "18", "20" ]
+    runs-on: "ubuntu-22.04"
+    defaults:
+      run:
+        shell: bash
+        working-directory: nodejs
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+        lfs: true
+    - uses: actions/setup-node@v3
+      with:
+        node-version: ${{ matrix.node-version }}
+        cache: 'npm'
+        cache-dependency-path: node/package-lock.json
+    - uses: Swatinem/rust-cache@v2
+    - name: Install dependencies
+      run: |
+        sudo apt update
+        sudo apt install -y protobuf-compiler libssl-dev
+        npm install -g @napi-rs/cli
+    - name: Build
+      run: |
+        npm ci
+        npm run build
+    - name: Test
+      run: npm run test
+  macos:
+    timeout-minutes: 30
+    runs-on: "macos-14"
+    defaults:
+      run:
+        shell: bash
+        working-directory: nodejs
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+        lfs: true
+    - uses: actions/setup-node@v3
+      with:
+        node-version: 20
+        cache: 'npm'
+        cache-dependency-path: node/package-lock.json
+    - uses: Swatinem/rust-cache@v2
+    - name: Install dependencies
+      run: |
+        brew install protobuf
+        npm install -g @napi-rs/cli
+    - name: Build
+      run: |
+        npm ci
+        npm run build
+    - name: Test
+      run: |
+        npm run test

.github/workflows/npm-publish.yml

@@ -15,7 +15,7 @@ jobs:
         working-directory: node
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - uses: actions/setup-node@v3
         with:
           node-version: 20
@@ -45,13 +45,13 @@ jobs:
             runner: macos-13
           - arch: aarch64-apple-darwin
             # xlarge is implicitly arm64.
-            runner: macos-13-xlarge
+            runner: macos-14
     runs-on: ${{ matrix.config.runner }}
     # Only runs on tags that matches the make-release action
     if: startsWith(github.ref, 'refs/tags/v')
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Install system dependencies
         run: brew install protobuf
       - name: Install npm dependencies
@@ -66,7 +66,7 @@ jobs:
           name: native-darwin
           path: |
             node/dist/lancedb-vectordb-darwin*.tgz
-         
+
 
   node-linux:
     name: node-linux (${{ matrix.config.arch}}-unknown-linux-gnu
@@ -80,10 +80,25 @@ jobs:
           - arch: x86_64
             runner: ubuntu-latest
           - arch: aarch64
-            runner: buildjet-4vcpu-ubuntu-2204-arm
+            # For successful fat LTO builds, we need a large runner to avoid OOM errors.
+            runner: buildjet-16vcpu-ubuntu-2204-arm
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
+      # Buildjet aarch64 runners have only 1.5 GB RAM per core, vs 3.5 GB per core for
+      # x86_64 runners. To avoid OOM errors on ARM, we create a swap file.
+      - name: Configure aarch64 build
+        if: ${{ matrix.config.arch == 'aarch64' }}
+        run: |
+          free -h
+          sudo fallocate -l 16G /swapfile
+          sudo chmod 600 /swapfile
+          sudo mkswap /swapfile
+          sudo swapon /swapfile
+          echo "/swapfile swap swap defaults 0 0" >> sudo /etc/fstab
+          # print info
+          swapon --show
+          free -h
       - name: Build Linux Artifacts
         run: |
           bash ci/build_linux_artifacts.sh ${{ matrix.config.arch }}
@@ -104,7 +119,7 @@ jobs:
         target: [x86_64-pc-windows-msvc]
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
       - name: Install Protoc v21.12
         working-directory: C:\
         run: |
@@ -154,7 +169,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           ref: main
           persist-credentials: false

.github/workflows/pypi-publish.yml

@@ -2,30 +2,91 @@ name: PyPI Publish
 
 on:
   release:
-    types: [ published ]
+    types: [published]
 
 jobs:
-  publish:
-    runs-on: ubuntu-latest
-    # Only runs on tags that matches the python-make-release action
-    if: startsWith(github.ref, 'refs/tags/python-v')
-    defaults:
-      run:
-        shell: bash
-        working-directory: python
+  linux:
+    timeout-minutes: 60
+    strategy:
+      matrix:
+        python-minor-version: ["8"]
+        platform:
+          - x86_64
+          - aarch64
+    runs-on: "ubuntu-22.04"
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
       - name: Set up Python
         uses: actions/setup-python@v4
         with:
-          python-version: "3.8"
-      - name: Build distribution
-        run: |
-          ls -la
-          pip install wheel setuptools --upgrade
-          python setup.py sdist bdist_wheel
-      - name: Publish
-        uses: pypa/gh-action-pypi-publish@v1.8.5
+          python-version: 3.${{ matrix.python-minor-version }}
+      - uses: ./.github/workflows/build_linux_wheel
         with:
-          password: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
-          packages-dir: python/dist
+          python-minor-version: ${{ matrix.python-minor-version }}
+          args: "--release --strip"
+          arm-build: ${{ matrix.platform == 'aarch64' }}
+      - uses: ./.github/workflows/upload_wheel
+        with:
+          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          repo: "pypi"
+  mac:
+    timeout-minutes: 60
+    runs-on: ${{ matrix.config.runner }}
+    strategy:
+      matrix:
+        python-minor-version: ["8"]
+        config:
+          - target: x86_64-apple-darwin
+            runner: macos-13
+          - target: aarch64-apple-darwin
+            runner: macos-14
+    env:
+      MACOSX_DEPLOYMENT_TARGET: 10.15
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.12
+      - uses: ./.github/workflows/build_mac_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          args: "--release --strip --target ${{ matrix.config.target }}"
+      - uses: ./.github/workflows/upload_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          repo: "pypi"
+  windows:
+    timeout-minutes: 60
+    runs-on: windows-latest
+    strategy:
+      matrix:
+        python-minor-version: ["8"]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ inputs.ref }}
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: 3.${{ matrix.python-minor-version }}
+      - uses: ./.github/workflows/build_windows_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          args: "--release --strip"
+          vcpkg_token: ${{ secrets.VCPKG_GITHUB_PACKAGES }}
+      - uses: ./.github/workflows/upload_wheel
+        with:
+          python-minor-version: ${{ matrix.python-minor-version }}
+          token: ${{ secrets.LANCEDB_PYPI_API_TOKEN }}
+          repo: "pypi"

.github/workflows/python-make-release-commit.yml

@@ -26,7 +26,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
     - name: Check out main
-      uses: actions/checkout@v3
+      uses: actions/checkout@v4
       with:
         ref: main
         persist-credentials: false
@@ -37,10 +37,10 @@ jobs:
       run: |
         git config user.name 'Lance Release'
         git config user.email 'lance-dev@lancedb.com'
-    - name: Set up Python 3.10
-      uses: actions/setup-python@v4
+    - name: Set up Python
+      uses: actions/setup-python@v5
       with:
-        python-version: "3.10"
+        python-version: "3.11"
     - name: Bump version, create tag and commit
       working-directory: python
       run: |

.github/workflows/python.yml

@@ -14,49 +14,133 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  linux:
+  lint:
+    name: "Lint"
     timeout-minutes: 30
-    strategy:
-      matrix:
-        python-minor-version: [ "8", "9", "10", "11" ]
     runs-on: "ubuntu-22.04"
     defaults:
       run:
         shell: bash
         working-directory: python
     steps:
-    - uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-        lfs: true
-    - name: Set up Python
-      uses: actions/setup-python@v4
-      with:
-        python-version: 3.${{ matrix.python-minor-version }}
-    - name: Install lancedb
-      run: |
-        pip install -e .[tests]
-        pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
-        pip install pytest pytest-mock ruff
-    - name: Format check
-      run: ruff format --check .
-    - name: Lint
-      run: ruff .
-    - name: Run tests
-      run: pytest -m "not slow" -x -v --durations=30 tests
-    - name: doctest
-      run: pytest --doctest-modules lancedb
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - name: Install ruff
+        run: |
+          pip install ruff==0.2.2
+      - name: Format check
+        run: ruff format --check .
+      - name: Lint
+        run: ruff .
+  doctest:
+    name: "Doctest"
+    timeout-minutes: 30
+    runs-on: "ubuntu-22.04"
+    defaults:
+      run:
+        shell: bash
+        working-directory: python
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: "pip"
+      - name: Install protobuf
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - name: Install
+        run: |
+          pip install -e .[tests,dev,embeddings]
+          pip install tantivy
+          pip install mlx
+      - name: Doctest
+        run: pytest --doctest-modules python/lancedb
+  linux:
+    name: "Linux: python-3.${{ matrix.python-minor-version }}"
+    timeout-minutes: 30
+    strategy:
+      matrix:
+        python-minor-version: ["8", "11"]
+    runs-on: "ubuntu-22.04"
+    defaults:
+      run:
+        shell: bash
+        working-directory: python
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Install protobuf
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.${{ matrix.python-minor-version }}
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - uses: ./.github/workflows/build_linux_wheel
+      - uses: ./.github/workflows/run_tests
+      # Make sure wheels are not included in the Rust cache
+      - name: Delete wheels
+        run: rm -rf target/wheels
   platform:
-    name: "Platform: ${{ matrix.config.name }}"
+    name: "Mac: ${{ matrix.config.name }}"
     timeout-minutes: 30
     strategy:
       matrix:
         config:
-          - name: x86 Mac
+          - name: x86
             runner: macos-13
-          - name: Arm Mac
-            runner: macos-13-xlarge
-          - name: x86 Windows
+          - name: Arm
+            runner: macos-14
+    runs-on: "${{ matrix.config.runner }}"
+    defaults:
+      run:
+        shell: bash
+        working-directory: python
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - uses: ./.github/workflows/build_mac_wheel
+      - uses: ./.github/workflows/run_tests
+      # Make sure wheels are not included in the Rust cache
+      - name: Delete wheels
+        run: rm -rf target/wheels
+  windows:
+    name: "Windows: ${{ matrix.config.name }}"
+    timeout-minutes: 30
+    strategy:
+      matrix:
+        config:
+          - name: x86
             runner: windows-latest
     runs-on: "${{ matrix.config.runner }}"
     defaults:
@@ -64,21 +148,22 @@ jobs:
         shell: bash
         working-directory: python
     steps:
-    - uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-        lfs: true
-    - name: Set up Python
-      uses: actions/setup-python@v4
-      with:
-        python-version: "3.11"
-    - name: Install lancedb
-      run: |
-        pip install -e .[tests]
-        pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
-        pip install pytest pytest-mock
-    - name: Run tests
-      run: pytest -m "not slow" -x -v --durations=30 tests
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - uses: Swatinem/rust-cache@v2
+        with:
+          workspaces: python
+      - uses: ./.github/workflows/build_windows_wheel
+      - uses: ./.github/workflows/run_tests
+      # Make sure wheels are not included in the Rust cache
+      - name: Delete wheels
+        run: rm -rf target/wheels
   pydantic1x:
     timeout-minutes: 30
     runs-on: "ubuntu-22.04"
@@ -87,21 +172,22 @@ jobs:
         shell: bash
         working-directory: python
     steps:
-    - uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-        lfs: true
-    - name: Set up Python
-      uses: actions/setup-python@v4
-      with:
-        python-version: 3.9
-    - name: Install lancedb
-      run: |
-        pip install "pydantic<2"
-        pip install -e .[tests]
-        pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
-        pip install pytest pytest-mock
-    - name: Run tests
-      run: pytest -m "not slow" -x -v --durations=30 tests
-    - name: doctest
-      run: pytest --doctest-modules lancedb
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          lfs: true
+      - name: Install dependencies
+        run: |
+          sudo apt update
+          sudo apt install -y protobuf-compiler
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: 3.9
+      - name: Install lancedb
+        run: |
+          pip install "pydantic<2"
+          pip install -e .[tests]
+          pip install tantivy
+      - name: Run tests
+        run: pytest -m "not slow" -x -v --durations=30 python/tests

.github/workflows/run_tests/action.yml

@@ -0,0 +1,17 @@
+name: run-tests
+
+description: "Install lance wheel and run unit tests"
+inputs:
+  python-minor-version:
+    required: true
+    description: "8 9 10 11 12"
+runs:
+  using: "composite"
+  steps:
+    - name: Install lancedb
+      shell: bash
+      run: |
+        pip3 install $(ls target/wheels/lancedb-*.whl)[tests,dev]
+    - name: pytest
+      shell: bash
+      run: pytest -m "not slow" -x -v --durations=30 python/python/tests

.github/workflows/rust.yml

@@ -32,7 +32,7 @@ jobs:
         shell: bash
         working-directory: rust
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v4
       with:
           fetch-depth: 0
           lfs: true
@@ -55,7 +55,7 @@ jobs:
         shell: bash
         working-directory: rust
     steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v4
       with:
           fetch-depth: 0
           lfs: true
@@ -70,18 +70,20 @@ jobs:
       run: cargo build --all-features
     - name: Run tests
       run: cargo test --all-features
+    - name: Run examples
+      run: cargo run --example simple
   macos:
     timeout-minutes: 30
     strategy:
       matrix:
-        mac-runner: [ "macos-13", "macos-13-xlarge" ]
+        mac-runner: [ "macos-13", "macos-14" ]
     runs-on: "${{ matrix.mac-runner }}"
     defaults:
       run:
         shell: bash
         working-directory: rust
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
         with:
             fetch-depth: 0
             lfs: true
@@ -99,7 +101,7 @@ jobs:
   windows:
     runs-on: windows-2022
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
       - uses: Swatinem/rust-cache@v2
         with:
           workspaces: rust
@@ -117,3 +119,4 @@ jobs:
           $env:VCPKG_ROOT = $env:VCPKG_INSTALLATION_ROOT
           cargo build
           cargo test
+  

.github/workflows/update_package_lock_run.yml

@@ -8,7 +8,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
-        uses: actions/checkout@v3
+        uses: actions/checkout@v4
         with:
           ref: main
           persist-credentials: false

.github/workflows/upload_wheel/action.yml

@@ -0,0 +1,29 @@
+name: upload-wheel
+
+description: "Upload wheels to Pypi"
+inputs:
+  os:
+    required: true
+    description: "ubuntu-22.04 or macos-13"
+  repo:
+    required: false
+    description: "pypi or testpypi"
+    default: "pypi"
+  token:
+    required: true
+    description: "release token for the repo"
+
+runs:
+  using: "composite"
+  steps:
+    - name: Install dependencies
+      shell: bash
+      run: |
+        python -m pip install --upgrade pip
+        pip install twine
+    - name: Publish wheel
+      env:
+        TWINE_USERNAME: __token__
+        TWINE_PASSWORD: ${{ inputs.token }}
+      shell: bash
+      run: twine upload --repository ${{ inputs.repo }} target/wheels/lancedb-*.whl

.gitignore

@@ -22,6 +22,11 @@ python/dist
 
 **/.hypothesis
 
+# Compiled Dynamic libraries
+*.so
+*.dylib
+*.dll
+
 ## Javascript
 *.node
 **/node_modules
@@ -29,8 +34,11 @@ python/dist
 node/dist
 node/examples/**/package-lock.json
 node/examples/**/dist
+dist
 
 ## Rust
 target
 
-Cargo.lock
+**/sccache.log
+
+Cargo.lock

.pre-commit-config.yaml

@@ -5,17 +5,14 @@ repos:
     -   id: check-yaml
     -   id: end-of-file-fixer
     -   id: trailing-whitespace
--   repo: https://github.com/psf/black
-    rev: 22.12.0
-    hooks:
-    -   id: black
 -   repo: https://github.com/astral-sh/ruff-pre-commit
     # Ruff version.
-    rev: v0.0.277
+    rev: v0.2.2
     hooks:
     - id: ruff
--   repo: https://github.com/pycqa/isort
-    rev: 5.12.0
-    hooks:
-    - id: isort
-      name: isort (python)
+- repo: https://github.com/pre-commit/mirrors-prettier
+  rev: v3.1.0
+  hooks:
+    - id: prettier
+      files: "nodejs/.*"
+      exclude: nodejs/lancedb/native.d.ts|nodejs/dist/.*

Cargo.toml

@@ -1,28 +1,40 @@
 [workspace]
-members = ["rust/ffi/node", "rust/vectordb"]
+members = ["rust/ffi/node", "rust/lancedb", "nodejs", "python"]
 # Python package needs to be built by maturin.
 exclude = ["python"]
 resolver = "2"
 
+[workspace.package]
+edition = "2021"
+authors = ["LanceDB Devs <dev@lancedb.com>"]
+license = "Apache-2.0"
+repository = "https://github.com/lancedb/lancedb"
+description = "Serverless, low-latency vector database for AI applications"
+keywords = ["lancedb", "lance", "database", "vector", "search"]
+categories = ["database-implementations"]
+
 [workspace.dependencies]
-lance = { "version" = "=0.9.7", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.9.7" }
-lance-linalg = { "version" = "=0.9.7" }
-lance-testing = { "version" = "=0.9.7" }
+lance = { "version" = "=0.10.2", "features" = ["dynamodb"] }
+lance-index = { "version" = "=0.10.2" }
+lance-linalg = { "version" = "=0.10.2" }
+lance-testing = { "version" = "=0.10.2" }
 # Note that this one does not include pyarrow
-arrow = { version = "49.0.0", optional = false }
-arrow-array = "49.0"
-arrow-data = "49.0"
-arrow-ipc = "49.0"
-arrow-ord = "49.0"
-arrow-schema = "49.0"
-arrow-arith = "49.0"
-arrow-cast = "49.0"
+arrow = { version = "50.0", optional = false }
+arrow-array = "50.0"
+arrow-data = "50.0"
+arrow-ipc = "50.0"
+arrow-ord = "50.0"
+arrow-schema = "50.0"
+arrow-arith = "50.0"
+arrow-cast = "50.0"
+async-trait = "0"
 chrono = "0.4.23"
 half = { "version" = "=2.3.1", default-features = false, features = [
     "num-traits",
 ] }
+futures = "0"
 log = "0.4"
 object_store = "0.9.0"
 snafu = "0.7.4"
 url = "2"
+num-traits = "0.2"

README.md

@@ -51,12 +51,19 @@ npm install vectordb
 const lancedb = require('vectordb');
 const db = await lancedb.connect('data/sample-lancedb');
 
-const table = await db.createTable('vectors',
-      [{ id: 1, vector: [0.1, 0.2], item: "foo", price: 10 },
-       { id: 2, vector: [1.1, 1.2], item: "bar", price: 50 }])
+const table = await db.createTable({
+  name: 'vectors',
+  data:  [
+    { id: 1, vector: [0.1, 0.2], item: "foo", price: 10 },
+    { id: 2, vector: [1.1, 1.2], item: "bar", price: 50 }
+  ]
+})
 
 const query = table.search([0.1, 0.3]).limit(2);
 const results = await query.execute();
+
+// You can also search for rows by specific criteria without involving a vector search.
+const rowsByCriteria = await table.search(undefined).where("price >= 10").execute();
 ```
 
 **Python**

ci/build_linux_artifacts.sh

@@ -13,7 +13,9 @@ docker build \
     .
 popd
 
+# We turn on memory swap to avoid OOM killer
 docker run \
     -v $(pwd):/io -w /io \
+    --memory-swap=-1 \
     lancedb-node-manylinux \
     bash ci/manylinux_node/build.sh $ARCH

dockerfiles/Dockerfile

@@ -0,0 +1,27 @@
+#Simple base dockerfile that supports basic dependencies required to run lance with FTS and Hybrid Search
+#Usage docker build -t lancedb:latest -f Dockerfile .
+FROM python:3.10-slim-buster
+
+# Install Rust
+RUN apt-get update && apt-get install -y curl build-essential && \
+  curl https://sh.rustup.rs -sSf | sh -s -- -y
+
+# Set the environment variable for Rust
+ENV PATH="/root/.cargo/bin:${PATH}"
+
+# Install protobuf compiler
+RUN apt-get install -y protobuf-compiler && \
+  apt-get clean && \
+  rm -rf /var/lib/apt/lists/*
+
+RUN apt-get -y update &&\
+  apt-get -y upgrade && \
+  apt-get -y install git
+
+
+# Verify installations
+RUN python --version && \
+  rustc --version && \
+  protoc --version
+
+RUN pip install tantivy lancedb

docs/README.md

@@ -33,3 +33,12 @@ You can run a local server to test the docs prior to deployment by navigating to
 cd docs
 mkdocs serve
 ```
+
+### Run doctest for typescript example
+
+```bash
+cd lancedb/docs
+npm i
+npm run build
+npm run all
+```

docs/mkdocs.yml

@@ -57,6 +57,16 @@ plugins:
           - https://arrow.apache.org/docs/objects.inv
           - https://pandas.pydata.org/docs/objects.inv
 - mkdocs-jupyter
+- ultralytics:
+    verbose: True
+    enabled: True
+    default_image: "assets/lancedb_and_lance.png" # Default image for all pages
+    add_image: True # Automatically add meta image
+    add_keywords: True # Add page keywords in the header tag
+    add_share_buttons: True # Add social share buttons
+    add_authors: False # Display page authors 
+    add_desc: False
+    add_dates: False
 
 markdown_extensions:
 - admonition
@@ -67,7 +77,9 @@ markdown_extensions:
     line_spans: __span
     pygments_lang_class: true
 - pymdownx.inlinehilite
-- pymdownx.snippets
+- pymdownx.snippets:
+    base_path: ..
+    dedent_subsections: true
 - pymdownx.superfences
 - pymdownx.tabbed:
     alternate_style: true
@@ -88,14 +100,18 @@ nav:
     - Building an ANN index: ann_indexes.md
     - Vector Search: search.md
     - Full-text search: fts.md
+    - Hybrid search:
+      - Overview: hybrid_search/hybrid_search.md
+      - Comparing Rerankers: hybrid_search/eval.md
+      - Airbnb financial data example: notebooks/hybrid_search.ipynb
     - Filtering: sql.md
     - Versioning & Reproducibility: notebooks/reproducibility.ipynb
+    - Configuring Storage: guides/storage.md
   - 🧬 Managing embeddings:
     - Overview: embeddings/index.md
-    - Explicit management: embeddings/embedding_explicit.md
-    - Implicit management: embeddings/embedding_functions.md
-    - Available Functions: embeddings/default_embedding_functions.md
-    - Custom Embedding Functions: embeddings/api.md
+    - Embedding functions: embeddings/embedding_functions.md
+    - Available models: embeddings/default_embedding_functions.md
+    - User-defined embedding functions: embeddings/custom_embedding_function.md
     - "Example: Multi-lingual semantic search": notebooks/multi_lingual_example.ipynb
     - "Example: MultiModal CLIP Embeddings": notebooks/DisappearingEmbeddingFunction.ipynb
   - 🔌 Integrations:
@@ -129,6 +145,7 @@ nav:
   - ⚙️ API reference:
     - 🐍 Python: python/python.md
     - 👾 JavaScript: javascript/modules.md
+    - 🦀 Rust: https://docs.rs/vectordb/latest/vectordb/
   - ☁️ LanceDB Cloud:
     - Overview: cloud/index.md
     - API reference:
@@ -147,14 +164,18 @@ nav:
     - Building an ANN index: ann_indexes.md
     - Vector Search: search.md
     - Full-text search: fts.md
+    - Hybrid search:
+      - Overview: hybrid_search/hybrid_search.md
+      - Comparing Rerankers: hybrid_search/eval.md
+      - Airbnb financial data example: notebooks/hybrid_search.ipynb
     - Filtering: sql.md
     - Versioning & Reproducibility: notebooks/reproducibility.ipynb
+    - Configuring Storage: guides/storage.md
 - Managing Embeddings:
   - Overview: embeddings/index.md
-  - Explicit management: embeddings/embedding_explicit.md
-  - Implicit management: embeddings/embedding_functions.md
-  - Available Functions: embeddings/default_embedding_functions.md
-  - Custom Embedding Functions: embeddings/api.md
+  - Embedding functions: embeddings/embedding_functions.md
+  - Available models: embeddings/default_embedding_functions.md
+  - User-defined embedding functions: embeddings/custom_embedding_function.md
   - "Example: Multi-lingual semantic search": notebooks/multi_lingual_example.ipynb
   - "Example: MultiModal CLIP Embeddings": notebooks/DisappearingEmbeddingFunction.ipynb
 - Integrations:
@@ -193,6 +214,9 @@ extra_css:
   - styles/global.css
   - styles/extra.css
 
+extra_javascript:
+  - "extra_js/init_ask_ai_widget.js"
+
 extra:
   analytics:
     provider: google

docs/package-lock.json

@@ -0,0 +1,132 @@
+{
+  "name": "lancedb-docs-test",
+  "version": "1.0.0",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {
+    "": {
+      "name": "lancedb-docs-test",
+      "version": "1.0.0",
+      "license": "Apache 2",
+      "dependencies": {
+        "apache-arrow": "file:../node/node_modules/apache-arrow",
+        "vectordb": "file:../node"
+      },
+      "devDependencies": {
+        "@types/node": "^20.11.8",
+        "typescript": "^5.3.3"
+      }
+    },
+    "../node": {
+      "name": "vectordb",
+      "version": "0.4.6",
+      "cpu": [
+        "x64",
+        "arm64"
+      ],
+      "license": "Apache-2.0",
+      "os": [
+        "darwin",
+        "linux",
+        "win32"
+      ],
+      "dependencies": {
+        "@apache-arrow/ts": "^14.0.2",
+        "@neon-rs/load": "^0.0.74",
+        "apache-arrow": "^14.0.2",
+        "axios": "^1.4.0"
+      },
+      "devDependencies": {
+        "@neon-rs/cli": "^0.0.160",
+        "@types/chai": "^4.3.4",
+        "@types/chai-as-promised": "^7.1.5",
+        "@types/mocha": "^10.0.1",
+        "@types/node": "^18.16.2",
+        "@types/sinon": "^10.0.15",
+        "@types/temp": "^0.9.1",
+        "@types/uuid": "^9.0.3",
+        "@typescript-eslint/eslint-plugin": "^5.59.1",
+        "cargo-cp-artifact": "^0.1",
+        "chai": "^4.3.7",
+        "chai-as-promised": "^7.1.1",
+        "eslint": "^8.39.0",
+        "eslint-config-standard-with-typescript": "^34.0.1",
+        "eslint-plugin-import": "^2.26.0",
+        "eslint-plugin-n": "^15.7.0",
+        "eslint-plugin-promise": "^6.1.1",
+        "mocha": "^10.2.0",
+        "openai": "^4.24.1",
+        "sinon": "^15.1.0",
+        "temp": "^0.9.4",
+        "ts-node": "^10.9.1",
+        "ts-node-dev": "^2.0.0",
+        "typedoc": "^0.24.7",
+        "typedoc-plugin-markdown": "^3.15.3",
+        "typescript": "*",
+        "uuid": "^9.0.0"
+      },
+      "optionalDependencies": {
+        "@lancedb/vectordb-darwin-arm64": "0.4.6",
+        "@lancedb/vectordb-darwin-x64": "0.4.6",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.4.6",
+        "@lancedb/vectordb-linux-x64-gnu": "0.4.6",
+        "@lancedb/vectordb-win32-x64-msvc": "0.4.6"
+      }
+    },
+    "../node/node_modules/apache-arrow": {
+      "version": "14.0.2",
+      "license": "Apache-2.0",
+      "dependencies": {
+        "@types/command-line-args": "5.2.0",
+        "@types/command-line-usage": "5.0.2",
+        "@types/node": "20.3.0",
+        "@types/pad-left": "2.1.1",
+        "command-line-args": "5.2.1",
+        "command-line-usage": "7.0.1",
+        "flatbuffers": "23.5.26",
+        "json-bignum": "^0.0.3",
+        "pad-left": "^2.1.0",
+        "tslib": "^2.5.3"
+      },
+      "bin": {
+        "arrow2csv": "bin/arrow2csv.js"
+      }
+    },
+    "node_modules/@types/node": {
+      "version": "20.11.8",
+      "resolved": "https://registry.npmjs.org/@types/node/-/node-20.11.8.tgz",
+      "integrity": "sha512-i7omyekpPTNdv4Jb/Rgqg0RU8YqLcNsI12quKSDkRXNfx7Wxdm6HhK1awT3xTgEkgxPn3bvnSpiEAc7a7Lpyow==",
+      "dev": true,
+      "dependencies": {
+        "undici-types": "~5.26.4"
+      }
+    },
+    "node_modules/apache-arrow": {
+      "resolved": "../node/node_modules/apache-arrow",
+      "link": true
+    },
+    "node_modules/typescript": {
+      "version": "5.3.3",
+      "resolved": "https://registry.npmjs.org/typescript/-/typescript-5.3.3.tgz",
+      "integrity": "sha512-pXWcraxM0uxAS+tN0AG/BF2TyqmHO014Z070UsJ+pFvYuRSq8KH8DmWpnbXe0pEPDHXZV3FcAbJkijJ5oNEnWw==",
+      "dev": true,
+      "bin": {
+        "tsc": "bin/tsc",
+        "tsserver": "bin/tsserver"
+      },
+      "engines": {
+        "node": ">=14.17"
+      }
+    },
+    "node_modules/undici-types": {
+      "version": "5.26.5",
+      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-5.26.5.tgz",
+      "integrity": "sha512-JlCMO+ehdEIKqlFxk6IfVoAUVmgz7cU7zD/h9XZ0qzeosSHmUJVOzSQvvYSYWXkFXC+IfLKSIffhv0sVZup6pA==",
+      "dev": true
+    },
+    "node_modules/vectordb": {
+      "resolved": "../node",
+      "link": true
+    }
+  }
+}

docs/package.json

@@ -0,0 +1,20 @@
+{
+  "name": "lancedb-docs-test",
+  "version": "1.0.0",
+  "description": "auto-generated tests from doc",
+  "author": "dev@lancedb.com",
+  "license": "Apache 2",
+  "dependencies": {
+    "apache-arrow": "file:../node/node_modules/apache-arrow",
+    "vectordb": "file:../node"
+  },
+  "scripts": {
+    "build": "tsc -b && cd ../node && npm run build-release",
+    "example": "npm run build && node",
+    "test": "npm run build && ls dist/*.js | xargs -n 1 node"
+  },
+  "devDependencies": {
+    "@types/node": "^20.11.8",
+    "typescript": "^5.3.3"
+  }
+}

docs/requirements.txt

@@ -2,4 +2,5 @@ mkdocs==1.5.3
 mkdocs-jupyter==0.24.1
 mkdocs-material==9.5.3
 mkdocstrings[python]==0.20.0
-pydantic
+pydantic
+mkdocs-ultralytics-plugin==0.0.44

docs/src/ann_indexes.md

@@ -7,26 +7,18 @@ for brute-force scanning of the entire vector space.
 A vector index is faster but less accurate than exhaustive search (kNN or flat search).
 LanceDB provides many parameters to fine-tune the index's size, the speed of queries, and the accuracy of results.
 
-Currently, LanceDB does *not* automatically create the ANN index.
-LanceDB has optimized code for kNN as well. For many use-cases, datasets under 100K vectors won't require index creation at all.
-If you can live with <100ms latency, skipping index creation is a simpler workflow while guaranteeing 100% recall.
+## Disk-based Index
 
-In the future we will look to automatically create and configure the ANN index as data comes in.
-
-## Types of Index
-
-Lance can support multiple index types, the most widely used one is `IVF_PQ`.
-
-* `IVF_PQ`: use **Inverted File Index (IVF)** to first divide the dataset into `N` partitions,
-   and then use **Product Quantization** to compress vectors in each partition.
-* `DiskANN` (**Experimental**): organize the vector as a on-disk graph, where the vertices approximately
-   represent the nearest neighbors of each vector.
+Lance provides an `IVF_PQ` disk-based index. It uses **Inverted File Index (IVF)** to first divide
+the dataset into `N` partitions, and then applies **Product Quantization** to compress vectors in each partition.
+See the [indexing](concepts/index_ivfpq.md) concepts guide for more information on how this works.
 
 ## Creating an IVF_PQ Index
 
 Lance supports `IVF_PQ` index type by default.
 
 === "Python"
+
      Creating indexes is done via the [create_index](https://lancedb.github.io/lancedb/python/#lancedb.table.LanceTable.create_index) method.
 
      ```python
@@ -46,25 +38,20 @@ Lance supports `IVF_PQ` index type by default.
      tbl.create_index(num_partitions=256, num_sub_vectors=96)
      ```
 
-=== "Javascript"
-     ```javascript
-     const vectordb = require('vectordb')
-     const db = await vectordb.connect('data/sample-lancedb')
+=== "Typescript"
 
-     let data = []
-     for (let i = 0; i < 10_000; i++) {
-         data.push({vector: Array(1536).fill(i), id: `${i}`, content: "", longId: `${i}`},)
-     }
-     const table = await db.createTable('my_vectors', data)
-     await table.createIndex({ type: 'ivf_pq', column: 'vector', num_partitions: 256, num_sub_vectors: 96 })
+     ```typescript
+     --8<--- "docs/src/ann_indexes.ts:import"
+
+     --8<-- "docs/src/ann_indexes.ts:ingest"
      ```
 
 - **metric** (default: "L2"): The distance metric to use. By default it uses euclidean distance "`L2`".
-We also support "cosine" and "dot" distance as well.
+  We also support "cosine" and "dot" distance as well.
 - **num_partitions** (default: 256): The number of partitions of the index.
 - **num_sub_vectors** (default: 96): The number of sub-vectors (M) that will be created during Product Quantization (PQ).
-For D dimensional vector, it will be divided into `M` of `D/M` sub-vectors, each of which is presented by
-a single PQ code.
+  For D dimensional vector, it will be divided into `M` of `D/M` sub-vectors, each of which is presented by
+  a single PQ code.
 
 <figure markdown>
   ![IVF PQ](./assets/ivf_pq.png)
@@ -78,7 +65,7 @@ Using GPU for index creation requires [PyTorch>2.0](https://pytorch.org/) being
 
 You can specify the GPU device to train IVF partitions via
 
-- **accelerator**: Specify to ``cuda`` or ``mps`` (on Apple Silicon) to enable GPU training.
+- **accelerator**: Specify to `cuda` or `mps` (on Apple Silicon) to enable GPU training.
 
 === "Linux"
 
@@ -92,7 +79,7 @@ You can specify the GPU device to train IVF partitions via
      )
      ```
 
-=== "Macos"
+=== "MacOS"
 
      <!-- skip-test -->
      ```python
@@ -104,12 +91,11 @@ You can specify the GPU device to train IVF partitions via
      )
      ```
 
-Trouble shootings:
+Troubleshooting:
 
-If you see ``AssertionError: Torch not compiled with CUDA enabled``, you need to [install
+If you see `AssertionError: Torch not compiled with CUDA enabled`, you need to [install
 PyTorch with CUDA support](https://pytorch.org/get-started/locally/).
 
-
 ## Querying an ANN Index
 
 Querying vector indexes is done via the [search](https://lancedb.github.io/lancedb/python/#lancedb.table.LanceTable.search) function.
@@ -127,6 +113,7 @@ There are a couple of parameters that can be used to fine-tune the search:
   Note: refine_factor is only applicable if an ANN index is present. If specified on a table without an ANN index, it is ignored.
 
 === "Python"
+
      ```python
      tbl.search(np.random.random((1536))) \
          .limit(2) \
@@ -134,41 +121,35 @@ There are a couple of parameters that can be used to fine-tune the search:
          .refine_factor(10) \
          .to_pandas()
      ```
-     ```
+
+     ```text
                                               vector       item       _distance
      0  [0.44949695, 0.8444449, 0.06281311, 0.23338133...  item 1141  103.575333
      1  [0.48587373, 0.269207, 0.15095535, 0.65531915,...  item 3953  108.393867
      ```
 
-=== "Javascript"
-     ```javascript
-     const results_1 = await table
-         .search(Array(1536).fill(1.2))
-         .limit(2)
-         .nprobes(20)
-         .refineFactor(10)
-         .execute()
+=== "Typescript"
+
+     ```typescript
+     --8<-- "docs/src/ann_indexes.ts:search1"
      ```
 
 The search will return the data requested in addition to the distance of each item.
 
-
 ### Filtering (where clause)
 
 You can further filter the elements returned by a search using a where clause.
 
 === "Python"
+
      ```python
      tbl.search(np.random.random((1536))).where("item != 'item 1141'").to_pandas()
      ```
 
-=== "Javascript"
+=== "Typescript"
+
      ```javascript
-     const results_2 = await table
-         .search(Array(1536).fill(1.2))
-         .where("id != '1141'")
-         .limit(2)
-         .execute()
+     --8<-- "docs/src/ann_indexes.ts:search2"
      ```
 
 ### Projections (select clause)
@@ -176,34 +157,42 @@ You can further filter the elements returned by a search using a where clause.
 You can select the columns returned by the query using a select clause.
 
 === "Python"
+
      ```python
      tbl.search(np.random.random((1536))).select(["vector"]).to_pandas()
      ```
-     ```
-        vector                                             _distance
+
+
+     ```text
+                                                   vector _distance
      0  [0.30928212, 0.022668175, 0.1756372, 0.4911822...  93.971092
      1  [0.2525465, 0.01723831, 0.261568, 0.002007689,...  95.173485
      ...
      ```
 
-=== "Javascript"
-     ```javascript
-     const results_3 = await table
-         .search(Array(1536).fill(1.2))
-         .select(["id"])
-         .limit(2)
-         .execute()
+=== "Typescript"
+
+     ```typescript
+     --8<-- "docs/src/ann_indexes.ts:search3"
      ```
 
 ## FAQ
 
+### Why do I need to manually create an index?
+
+Currently, LanceDB does _not_ automatically create the ANN index.
+LanceDB is well-optimized for kNN (exhaustive search) via a disk-based  index. For many use-cases,
+datasets of the order of ~100K vectors don't require index creation. If you can live with up to
+100ms latency, skipping index creation is a simpler workflow while guaranteeing 100% recall.
+
 ### When is it necessary to create an ANN vector index?
 
-`LanceDB` has manually-tuned SIMD code for computing vector distances.
-In our benchmarks, computing 100K pairs of 1K dimension vectors takes **less than 20ms**.
-For small datasets (< 100K rows) or applications that can accept 100ms latency, vector indices are usually not necessary.
+`LanceDB` comes out-of-the-box with highly optimized SIMD code for computing vector similarity.
+In our benchmarks, computing distances for 100K pairs of 1K dimension vectors takes **less than 20ms**.
+We observe that for small datasets (~100K rows) or for applications that can accept 100ms latency,
+vector indices are usually not necessary.
 
-For large-scale or higher dimension vectors, it is beneficial to create vector index.
+For large-scale or higher dimension vectors, it can beneficial to create vector index for performance.
 
 ### How big is my index, and how many memory will it take?
 
@@ -221,4 +210,4 @@ On `SIFT-1M` dataset, our benchmark shows that keeping each partition 1K-4K rows
 `num_sub_vectors` specifies how many Product Quantization (PQ) short codes to generate on each vector. Because
 PQ is a lossy compression of the original vector, a higher `num_sub_vectors` usually results in
 less space distortion, and thus yields better accuracy. However, a higher `num_sub_vectors` also causes heavier I/O and
-more PQ computation, and thus, higher latency. `dimension / num_sub_vectors` should be a multiple of 8 for optimum SIMD efficiency.
+more PQ computation, and thus, higher latency. `dimension / num_sub_vectors` should be a multiple of 8 for optimum SIMD efficiency.

docs/src/ann_indexes.ts

@@ -0,0 +1,53 @@
+// --8<-- [start:import]
+import * as vectordb from "vectordb";
+// --8<-- [end:import]
+
+(async () => {
+  // --8<-- [start:ingest]
+  const db = await vectordb.connect("data/sample-lancedb");
+
+  let data = [];
+  for (let i = 0; i < 10_000; i++) {
+    data.push({
+      vector: Array(1536).fill(i),
+      id: `${i}`,
+      content: "",
+      longId: `${i}`,
+    });
+  }
+  const table = await db.createTable("my_vectors", data);
+  await table.createIndex({
+    type: "ivf_pq",
+    column: "vector",
+    num_partitions: 16,
+    num_sub_vectors: 48,
+  });
+  // --8<-- [end:ingest]
+
+  // --8<-- [start:search1]
+  const results_1 = await table
+    .search(Array(1536).fill(1.2))
+    .limit(2)
+    .nprobes(20)
+    .refineFactor(10)
+    .execute();
+  // --8<-- [end:search1]
+
+  // --8<-- [start:search2]
+  const results_2 = await table
+    .search(Array(1536).fill(1.2))
+    .where("id != '1141'")
+    .limit(2)
+    .execute();
+  // --8<-- [end:search2]
+
+  // --8<-- [start:search3]
+  const results_3 = await table
+    .search(Array(1536).fill(1.2))
+    .select(["id"])
+    .limit(2)
+    .execute();
+  // --8<-- [end:search3]
+
+  console.log("Ann indexes: done");
+})();

docs/src/basic.md

@@ -11,43 +11,83 @@
 ## Installation
 
 === "Python"
+
       ```shell
       pip install lancedb
       ```
 
-=== "Javascript"
+=== "Typescript"
+
       ```shell
       npm install vectordb
       ```
 
-## How to connect to a database
+=== "Rust"
+
+    !!! warning "Rust SDK is experimental, might introduce breaking changes in the near future"
+
+    ```shell
+    cargo add vectordb
+    ```
+
+    !!! info "To use the vectordb create, you first need to install protobuf."
+
+    === "macOS"
+
+        ```shell
+        brew install protobuf
+        ```
+
+    === "Ubuntu/Debian"
+
+        ```shell
+        sudo apt install -y protobuf-compiler libssl-dev
+        ```
+
+    !!! info "Please also make sure you're using the same version of Arrow as in the [vectordb crate](https://github.com/lancedb/lancedb/blob/main/Cargo.toml)"
+
+## Connect to a database
 
 === "Python"
+
       ```python
       import lancedb
       uri = "data/sample-lancedb"
       db = lancedb.connect(uri)
       ```
 
-      LanceDB will create the directory if it doesn't exist (including parent directories).
+=== "Typescript"
 
-      If you need a reminder of the uri, use the `db.uri` property.
+    ```typescript
+    --8<-- "docs/src/basic_legacy.ts:import"
 
-=== "Javascript"
-      ```javascript
-      const lancedb = require("vectordb");
+    --8<-- "docs/src/basic_legacy.ts:open_db"
+    ```
 
-      const uri = "data/sample-lancedb";
-      const db = await lancedb.connect(uri);
-      ```
-      
-      LanceDB will create the directory if it doesn't exist (including parent directories).
+=== "Rust"
 
-      If you need a reminder of the uri, you can call `db.uri()`.
+    ```rust
+    #[tokio::main]
+    async fn main() -> Result<()> {
+        --8<-- "rust/lancedb/examples/simple.rs:connect"
+    }
+    ```
 
-## How to create a table
+    !!! info "See [examples/simple.rs](https://github.com/lancedb/lancedb/tree/main/rust/lancedb/examples/simple.rs) for a full working example."
+
+LanceDB will create the directory if it doesn't exist (including parent directories).
+
+If you need a reminder of the uri, you can call `db.uri()`.
+
+## Create a table
+
+### Directly insert data to a new table
+
+If you have data to insert into the table at creation time, you can simultaneously create a 
+table and insert the data to it.
 
 === "Python"
+
     ```python
     tbl = db.create_table("my_table",
                     data=[{"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
@@ -59,6 +99,7 @@
     to the `create_table` method.
 
     You can also pass in a pandas DataFrame directly:
+
     ```python
     import pandas as pd
     df = pd.DataFrame([{"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
@@ -66,120 +107,210 @@
     tbl = db.create_table("table_from_df", data=df)
     ```
 
-=== "Javascript"
-    ```javascript
-    const tb = await db.createTable(
-        "myTable",
-        [{"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
-         {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}]
-    )
+=== "Typescript"
+
+    ```typescript
+    --8<-- "docs/src/basic_legacy.ts:create_table"
     ```
 
     If the table already exists, LanceDB will raise an error by default.
     If you want to overwrite the table, you can pass in `mode="overwrite"`
     to the `createTable` function.
 
+=== "Rust"
 
-!!! info "Under the hood, LanceDB is converting the input data into an Apache Arrow table and persisting it to disk in [Lance format](https://www.github.com/lancedb/lance)."
+    ```rust
+    use arrow_schema::{DataType, Schema, Field};
+    use arrow_array::{RecordBatch, RecordBatchIterator};
 
-### Creating an empty table
+    --8<-- "rust/lancedb/examples/simple.rs:create_table"
+    ```
+
+    If the table already exists, LanceDB will raise an error by default.
+
+!!! info "Under the hood, LanceDB converts the input data into an Apache Arrow table and persists it to disk using the [Lance format](https://www.github.com/lancedb/lance)."
+
+### Create an empty table
 
 Sometimes you may not have the data to insert into the table at creation time.
-In this case, you can create an empty table and specify the schema.
+In this case, you can create an empty table and specify the schema, so that you can add
+data to the table at a later time (such that it conforms to the schema).
 
 === "Python"
+
       ```python
       import pyarrow as pa
       schema = pa.schema([pa.field("vector", pa.list_(pa.float32(), list_size=2))])
       tbl = db.create_table("empty_table", schema=schema)
       ```
 
-## How to open an existing table
+=== "Typescript"
 
-Once created, you can open a table using the following code:
+    ```typescript
+    --8<-- "docs/src/basic_legacy.ts:create_empty_table"
+    ```
+
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/simple.rs:create_empty_table"
+    ```
+
+## Open an existing table
+
+Once created, you can open a table as follows:
 
 === "Python"
-      ```python
-      tbl = db.open_table("my_table")
-      ```
 
-      If you forget the name of your table, you can always get a listing of all table names:
+    ```python
+    tbl = db.open_table("my_table")
+    ```
 
-      ```python
-      print(db.table_names())
-      ```
+=== "Typescript"
 
-=== "Javascript"
-      ```javascript
-      const tbl = await db.openTable("myTable");
-      ```
+    ```typescript
+    const tbl = await db.openTable("myTable");
+    ```
 
-      If you forget the name of your table, you can always get a listing of all table names:
+=== "Rust"
 
-      ```javascript
-      console.log(await db.tableNames());
-      ```
+    ```rust
+    --8<-- "rust/lancedb/examples/simple.rs:open_with_existing_file"
+    ```
 
-## How to add data to a table
-
-After a table has been created, you can always add more data to it using
+If you forget the name of your table, you can always get a listing of all table names:
 
 === "Python"
-      ```python
 
-      # Option 1: Add a list of dicts to a table
-      data = [{"vector": [1.3, 1.4], "item": "fizz", "price": 100.0},
-              {"vector": [9.5, 56.2], "item": "buzz", "price": 200.0}]
-      tbl.add(data)
-
-      # Option 2: Add a pandas DataFrame to a table
-      df = pd.DataFrame(data)
-      tbl.add(data)
-      ```
+    ```python
+    print(db.table_names())
+    ```
 
 === "Javascript"
-      ```javascript
-      await tbl.add([{vector: [1.3, 1.4], item: "fizz", price: 100.0},
-                     {vector: [9.5, 56.2], item: "buzz", price: 200.0}])
-      ```
 
-## How to search for (approximate) nearest neighbors
+    ```javascript
+    console.log(await db.tableNames());
+    ```
 
-Once you've embedded the query, you can find its nearest neighbors using the following code:
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/simple.rs:list_names"
+    ```
+
+## Add data to a table
+
+After a table has been created, you can always add more data to it as follows:
 
 === "Python"
-      ```python
-      tbl.search([100, 100]).limit(2).to_pandas()
-      ```
 
-      This returns a pandas DataFrame with the results.
+    ```python
 
-=== "Javascript"
-      ```javascript
-      const query = await tbl.search([100, 100]).limit(2).execute();
-      ```
+    # Option 1: Add a list of dicts to a table
+    data = [{"vector": [1.3, 1.4], "item": "fizz", "price": 100.0},
+            {"vector": [9.5, 56.2], "item": "buzz", "price": 200.0}]
+    tbl.add(data)
 
-## How to delete rows from a table
+    # Option 2: Add a pandas DataFrame to a table
+    df = pd.DataFrame(data)
+    tbl.add(data)
+    ```
+
+=== "Typescript"
+
+    ```typescript
+    --8<-- "docs/src/basic_legacy.ts:add"
+    ```
+
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/simple.rs:add"
+    ```
+
+## Search for nearest neighbors
+
+Once you've embedded the query, you can find its nearest neighbors as follows:
+
+=== "Python"
+
+    ```python
+    tbl.search([100, 100]).limit(2).to_pandas()
+    ```
+
+    This returns a pandas DataFrame with the results.
+
+=== "Typescript"
+
+    ```typescript
+    --8<-- "docs/src/basic_legacy.ts:search"
+    ```
+
+=== "Rust"
+
+    ```rust
+    use futures::TryStreamExt;
+
+    --8<-- "rust/lancedb/examples/simple.rs:search"
+    ```
+
+By default, LanceDB runs a brute-force scan over dataset to find the K nearest neighbours (KNN).
+For tables with more than 50K vectors, creating an ANN index is recommended to speed up search performance.
+LanceDB allows you to create an ANN index on a table as follows:
+
+=== "Python"
+
+    ```py
+    tbl.create_index()
+    ```
+
+=== "Typescript"
+
+    ```{.typescript .ignore}
+    --8<-- "docs/src/basic_legacy.ts:create_index"
+    ```
+
+=== "Rust"
+
+    ```rust
+     --8<-- "rust/lancedb/examples/simple.rs:create_index"
+    ```
+
+!!! note "Why do I need to create an index manually?"
+    LanceDB does not automatically create the ANN index, for two reasons. The first is that it's optimized
+    for really fast retrievals via a disk-based index, and the second is that data and query workloads can
+    be very diverse, so there's no one-size-fits-all index configuration. LanceDB provides many parameters
+    to fine-tune index size, query latency and accuracy. See the section on
+    [ANN indexes](ann_indexes.md) for more details.
+
+## Delete rows from a table
 
 Use the `delete()` method on tables to delete rows from a table. To choose
 which rows to delete, provide a filter that matches on the metadata columns.
 This can delete any number of rows that match the filter.
 
 === "Python"
-      ```python
-      tbl.delete('item = "fizz"')
-      ```
 
-=== "Javascript"
-      ```javascript
-      await tbl.delete('item = "fizz"')
-      ```
+    ```python
+    tbl.delete('item = "fizz"')
+    ```
+
+=== "Typescript"
+
+    ```typescript
+    --8<-- "docs/src/basic_legacy.ts:delete"
+    ```
+
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/simple.rs:delete"
+    ```
 
 The deletion predicate is a SQL expression that supports the same expressions
 as the `where()` clause on a search. They can be as simple or complex as needed.
 To see what expressions are supported, see the [SQL filters](sql.md) section.
 
-
 === "Python"
 
       Read more: [lancedb.table.Table.delete][]
@@ -188,11 +319,12 @@ To see what expressions are supported, see the [SQL filters](sql.md) section.
 
       Read more: [vectordb.Table.delete](javascript/interfaces/Table.md#delete)
 
-## How to remove a table
+## Drop a table
 
 Use the `drop_table()` method on the database to remove a table.
 
 === "Python"
+
       ```python
       db.drop_table("my_table")
       ```
@@ -201,13 +333,20 @@ Use the `drop_table()` method on the database to remove a table.
       By default, if the table does not exist an exception is raised. To suppress this,
       you can pass in `ignore_missing=True`.
 
-=== "JavaScript"
-      ```javascript
-      await db.dropTable('myTable')
+=== "Typescript"
+
+      ```typescript
+      --8<-- "docs/src/basic_legacy.ts:drop_table"
       ```
 
       This permanently removes the table and is not recoverable, unlike deleting rows.
-      If the table does not exist an exception is raised. 
+      If the table does not exist an exception is raised.
+
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/simple.rs:drop_table"
+    ```
 
 !!! note "Bundling `vectordb` apps with Webpack"
 

docs/src/basic_legacy.ts

@@ -0,0 +1,92 @@
+// --8<-- [start:import]
+import * as lancedb from "vectordb";
+import { Schema, Field, Float32, FixedSizeList, Int32, Float16 } from "apache-arrow";
+// --8<-- [end:import]
+import * as fs from "fs";
+import { Table as ArrowTable, Utf8 } from "apache-arrow";
+
+const example = async () => {
+  fs.rmSync("data/sample-lancedb", { recursive: true, force: true });
+  // --8<-- [start:open_db]
+  const lancedb = require("vectordb");
+  const uri = "data/sample-lancedb";
+  const db = await lancedb.connect(uri);
+  // --8<-- [end:open_db]
+
+  // --8<-- [start:create_table]
+  const tbl = await db.createTable(
+    "myTable",
+    [
+      { vector: [3.1, 4.1], item: "foo", price: 10.0 },
+      { vector: [5.9, 26.5], item: "bar", price: 20.0 },
+    ],
+    { writeMode: lancedb.WriteMode.Overwrite }
+  );
+  // --8<-- [end:create_table]
+
+  // --8<-- [start:add]
+  const newData = Array.from({ length: 500 }, (_, i) => ({
+    vector: [i, i + 1],
+    item: "fizz",
+    price: i * 0.1,
+  }));
+  await tbl.add(newData);
+  // --8<-- [end:add]
+
+  // --8<-- [start:create_index]
+  await tbl.createIndex({
+    type: "ivf_pq",
+    num_partitions: 2,
+    num_sub_vectors: 2,
+  });
+  // --8<-- [end:create_index]
+
+  // --8<-- [start:create_empty_table]
+  const schema = new Schema([
+    new Field("id", new Int32()),
+    new Field("name", new Utf8()),
+  ]);
+  const empty_tbl = await db.createTable({ name: "empty_table", schema });
+  // --8<-- [end:create_empty_table]
+
+  // --8<-- [start:create_f16_table]
+  const dim = 16
+  const total = 10
+  const f16_schema = new Schema([
+      new Field('id', new Int32()),
+      new Field(
+        'vector',
+        new FixedSizeList(dim, new Field('item', new Float16(), true)),
+        false
+      )
+    ])
+  const data = lancedb.makeArrowTable(
+      Array.from(Array(total), (_, i) => ({
+        id: i,
+        vector: Array.from(Array(dim), Math.random)
+      })),
+      { f16_schema }
+    )
+  const table = await db.createTable('f16_tbl', data)
+  // --8<-- [end:create_f16_table]
+
+  // --8<-- [start:search]
+  const query = await tbl.search([100, 100]).limit(2).execute();
+  // --8<-- [end:search]
+  console.log(query);
+
+  // --8<-- [start:delete]
+  await tbl.delete('item = "fizz"');
+  // --8<-- [end:delete]
+
+  // --8<-- [start:drop_table]
+  await db.dropTable("myTable");
+  // --8<-- [end:drop_table]
+};
+
+async function main() {
+  await example();
+  console.log("Basic example: done");
+}
+
+main();

docs/src/concepts/index_ivfpq.md

@@ -81,24 +81,4 @@ The above query will perform a search on the table `tbl` using the given query v
 * `to_pandas()`: Convert the results to a pandas DataFrame
 
 And there you have it! You now understand what an IVF-PQ index is, and how to create and query it in LanceDB.
-
-
-## FAQ
-
-### When is it necessary to create a vector index?
-
-LanceDB has manually-tuned SIMD code for computing vector distances. In our benchmarks, computing 100K pairs of 1K dimension vectors takes **<20ms**. For small datasets (<100K rows) or applications that can accept up to 100ms latency, vector indices are usually not necessary.
-
-For large-scale or higher dimension vectors, it is beneficial to create vector index.
-
-### How big is my index, and how much memory will it take?
-
-In LanceDB, all vector indices are disk-based, meaning that when responding to a vector query, only the relevant pages from the index file are loaded from disk and cached in memory. Additionally, each sub-vector is usually encoded into 1 byte PQ code.
-
-For example, with 1024-dimension vectors, if we choose `num_sub_vectors = 64`, each sub-vector has `1024 / 64 = 16` float32 numbers. Product quantization can lead to approximately `16 * sizeof(float32) / 1 = 64` times of space reduction.
-
-### How to choose `num_partitions` and `num_sub_vectors` for IVF_PQ index?
-
-`num_partitions` is used to decide how many partitions the first level IVF index uses. Higher number of partitions could lead to more efficient I/O during queries and better accuracy, but it takes much more time to train. On SIFT-1M dataset, our benchmark shows that keeping each partition 1K-4K rows lead to a good latency/recall.
-
-`num_sub_vectors` specifies how many PQ short codes to generate on each vector. Because PQ is a lossy compression of the original vector, a higher `num_sub_vectors` usually results in less space distortion, and thus yields better accuracy. However, a higher `num_sub_vectors` also causes heavier I/O and more PQ computation, and thus, higher latency. `dimension / num_sub_vectors` should be a multiple of 8 for optimum SIMD efficiency.
+To see how to create an IVF-PQ index in LanceDB, take a look at the [ANN indexes](../ann_indexes.md) section.

docs/src/embeddings/custom_embedding_function.md

@@ -17,6 +17,7 @@ Let's implement `SentenceTransformerEmbeddings` class. All you need to do is imp
 
 ```python
 from lancedb.embeddings import register
+from lancedb.util import attempt_import_or_raise
 
 @register("sentence-transformers")
 class SentenceTransformerEmbeddings(TextEmbeddingFunction):
@@ -81,7 +82,7 @@ class OpenClipEmbeddings(EmbeddingFunction):
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
-        open_clip = self.safe_import("open_clip", "open-clip") # EmbeddingFunction util to import external libs and raise if not found
+        open_clip = attempt_import_or_raise("open_clip", "open-clip") # EmbeddingFunction util to import external libs and raise if not found
         model, _, preprocess = open_clip.create_model_and_transforms(
             self.name, pretrained=self.pretrained
         )
@@ -109,14 +110,14 @@ class OpenClipEmbeddings(EmbeddingFunction):
         if isinstance(query, str):
             return [self.generate_text_embeddings(query)]
         else:
-            PIL = self.safe_import("PIL", "pillow")
+            PIL = attempt_import_or_raise("PIL", "pillow")
             if isinstance(query, PIL.Image.Image):
                 return [self.generate_image_embedding(query)]
             else:
                 raise TypeError("OpenClip supports str or PIL Image as query")
 
     def generate_text_embeddings(self, text: str) -> np.ndarray:
-        torch = self.safe_import("torch")
+        torch = attempt_import_or_raise("torch")
         text = self.sanitize_input(text)
         text = self._tokenizer(text)
         text.to(self.device)
@@ -175,7 +176,7 @@ class OpenClipEmbeddings(EmbeddingFunction):
             The image to embed. If the image is a str, it is treated as a uri.
             If the image is bytes, it is treated as the raw image bytes.
         """
-        torch = self.safe_import("torch")
+        torch = attempt_import_or_raise("torch")
         # TODO handle retry and errors for https
         image = self._to_pil(image)
         image = self._preprocess(image).unsqueeze(0)
@@ -183,7 +184,7 @@ class OpenClipEmbeddings(EmbeddingFunction):
             return self._encode_and_normalize_image(image)
 
     def _to_pil(self, image: Union[str, bytes]):
-        PIL = self.safe_import("PIL", "pillow")
+        PIL = attempt_import_or_raise("PIL", "pillow")
         if isinstance(image, bytes):
             return PIL.Image.open(io.BytesIO(image))
         if isinstance(image, PIL.Image.Image):

docs/src/embeddings/default_embedding_functions.md

@@ -9,6 +9,9 @@ Contains the text embedding functions registered by default.
 ### Sentence transformers
 Allows you to set parameters when registering a `sentence-transformers` object.
 
+!!! info
+    Sentence transformer embeddings are normalized by default. It is recommended to use normalized embeddings for similarity search.
+
 | Parameter | Type | Default Value | Description |
 |---|---|---|---|
 | `name` | `str` | `all-MiniLM-L6-v2` | The name of the model |
@@ -44,6 +47,7 @@ LanceDB registers the OpenAI embeddings function in the registry by default, as
 | Parameter | Type | Default Value | Description |
 |---|---|---|---|
 | `name` | `str` | `"text-embedding-ada-002"` | The name of the model. |
+| `dim` | `int` |  Model default   | For OpenAI's newer text-embedding-3 model, we can specify a dimensionality that is smaller than the 1536 size. This feature supports it |
 
 
 ```python
@@ -119,7 +123,7 @@ texts = [{"text": "Capitalism has been dominant in the Western world since the e
 tbl.add(texts)
 ```
 
-## Gemini Embedding Function
+### Gemini Embeddings
 With Google's Gemini, you can represent text (words, sentences, and blocks of text) in a vectorized form, making it easier to compare and contrast embeddings. For example, two texts that share a similar subject matter or sentiment should have similar embeddings, which can be identified through mathematical comparison techniques such as cosine similarity. For more on how and why you should use embeddings, refer to the Embeddings guide.
 The Gemini Embedding Model API supports various task types:
 
@@ -155,6 +159,52 @@ tbl.add(df)
 rs = tbl.search("hello").limit(1).to_pandas()
 ```
 
+### AWS Bedrock Text Embedding Functions
+AWS Bedrock supports multiple base models for generating text embeddings. You need to setup the AWS credentials to use this embedding function.
+You can do so by using `awscli` and also add your session_token:
+```shell
+aws configure
+aws configure set aws_session_token "<your_session_token>"
+```
+to ensure that the credentials are set up correctly, you can run the following command:
+```shell
+aws sts get-caller-identity
+```
+
+Supported Embedding modelIDs are:
+* `amazon.titan-embed-text-v1`
+* `cohere.embed-english-v3`
+* `cohere.embed-multilingual-v3`
+
+Supported parameters (to be passed in `create` method) are:
+
+| Parameter | Type | Default Value | Description |
+|---|---|---|---|
+| **name** | str | "amazon.titan-embed-text-v1" | The model ID of the bedrock model to use. Supported base models for Text Embeddings: amazon.titan-embed-text-v1, cohere.embed-english-v3, cohere.embed-multilingual-v3 |
+| **region** | str | "us-east-1" | Optional name of the AWS Region in which the service should be called (e.g., "us-east-1"). |
+| **profile_name** | str | None | Optional name of the AWS profile to use for calling the Bedrock service. If not specified, the default profile will be used. |
+| **assumed_role** | str | None | Optional ARN of an AWS IAM role to assume for calling the Bedrock service. If not specified, the current active credentials will be used. |
+| **role_session_name** | str | "lancedb-embeddings" | Optional name of the AWS IAM role session to use for calling the Bedrock service. If not specified, a "lancedb-embeddings" name will be used. |
+| **runtime** | bool | True | Optional choice of getting different client to perform operations with the Amazon Bedrock service. |
+| **max_retries** | int | 7 | Optional number of retries to perform when a request fails. |
+
+Usage Example:
+
+```python
+model = get_registry().get("bedrock-text").create()
+
+class TextModel(LanceModel):
+    text: str = model.SourceField()
+    vector: Vector(model.ndims()) = model.VectorField()
+
+df = pd.DataFrame({"text": ["hello world", "goodbye world"]})
+db = lancedb.connect("tmp_path")
+tbl = db.create_table("test", schema=TextModel, mode="overwrite")
+
+tbl.add(df)
+rs = tbl.search("hello").limit(1).to_pandas()
+```
+
 ## Multi-modal embedding functions
 Multi-modal embedding functions allow you to query your table using both images and text.
 

docs/src/embeddings/embedding_explicit.md

@@ -1,141 +0,0 @@
-In this workflow, you define your own embedding function and pass it as a callable to LanceDB, invoking it in your code to generate the embeddings. Let's look at some examples.
-
-### Hugging Face
-
-!!! note
-    Currently, the Hugging Face method is only supported in the Python SDK.
-
-=== "Python"
-    The most popular open source option is to use the [sentence-transformers](https://www.sbert.net/) 
-    library, which can be installed via pip.
-
-    ```bash
-    pip install sentence-transformers
-    ```
-
-    The example below shows how to use the `paraphrase-albert-small-v2` model to generate embeddings 
-    for a given document.
-
-    ```python
-    from sentence_transformers import SentenceTransformer
-
-    name="paraphrase-albert-small-v2"
-    model = SentenceTransformer(name)
-
-    # used for both training and querying
-    def embed_func(batch):
-        return [model.encode(sentence) for sentence in batch]
-    ```
-
-### OpenAI
-
-Another popular alternative is to use an external API like OpenAI's [embeddings API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
-
-=== "Python"
-      ```python
-        import openai
-        import os
-
-        # Configuring the environment variable OPENAI_API_KEY
-        if "OPENAI_API_KEY" not in os.environ:
-        # OR set the key here as a variable
-        openai.api_key = "sk-..."
-
-        # verify that the API key is working
-        assert len(openai.Model.list()["data"]) > 0
-
-        def embed_func(c):
-            rs = openai.Embedding.create(input=c, engine="text-embedding-ada-002")
-            return [record["embedding"] for record in rs["data"]]
-      ```
-
-=== "JavaScript"
-      ```javascript
-        const lancedb = require("vectordb");
-
-        // You need to provide an OpenAI API key
-        const apiKey = "sk-..."
-        // The embedding function will create embeddings for the 'text' column
-        const embedding = new lancedb.OpenAIEmbeddingFunction('text', apiKey)
-      ```
-
-## Applying an embedding function to data
-
-=== "Python"
-    Using an embedding function, you can apply it to raw data
-    to generate embeddings for each record.
-
-    Say you have a pandas DataFrame with a `text` column that you want embedded,
-    you can use the `with_embeddings` function to generate embeddings and add them to 
-    an existing table.
-
-    ```python
-     import pandas as pd
-     from lancedb.embeddings import with_embeddings
-
-     df = pd.DataFrame(
-        [
-            {"text": "pepperoni"},
-            {"text": "pineapple"}
-        ]
-    )
-     data = with_embeddings(embed_func, df)
-
-     # The output is used to create / append to a table
-     # db.create_table("my_table", data=data)
-    ```
-
-    If your data is in a different column, you can specify the `column` kwarg to `with_embeddings`.
-
-    By default, LanceDB calls the function with batches of 1000 rows. This can be configured
-    using the `batch_size` parameter to `with_embeddings`.
-
-    LanceDB automatically wraps the function with retry and rate-limit logic to ensure the OpenAI
-    API call is reliable.
-
-=== "JavaScript"
-    Using an embedding function, you can apply it to raw data
-    to generate embeddings for each record.
-
-    Simply pass the embedding function created above and LanceDB will use it to generate
-    embeddings for your data.
-
-    ```javascript
-    const db = await lancedb.connect("data/sample-lancedb");
-    const data = [
-    { text: "pepperoni"},
-    { text: "pineapple"}
-    ]
-
-    const table = await db.createTable("vectors", data, embedding)
-    ```
-
-## Querying using an embedding function
-
-!!! warning
-    At query time, you **must** use the same embedding function you used to vectorize your data.
-    If you use a different embedding function, the embeddings will not reside in the same vector
-    space and the results will be nonsensical.
-
-=== "Python"
-     ```python
-     query = "What's the best pizza topping?"
-     query_vector = embed_func([query])[0]
-     results = (
-        tbl.search(query_vector)
-        .limit(10)
-        .to_pandas()
-     )
-     ```
-
-     The above snippet returns a pandas DataFrame with the 10 closest vectors to the query.
-
-=== "JavaScript"
-     ```javascript
-      const results = await table
-        .search("What's the best pizza topping?")
-        .limit(10)
-        .execute()
-     ```
-
-     The above snippet returns an array of records with the top 10 nearest neighbors to the query.

docs/src/embeddings/embedding_functions.md

@@ -3,61 +3,126 @@ Representing multi-modal data as vector embeddings is becoming a standard practi
 For this purpose, LanceDB introduces an **embedding functions API**, that allow you simply set up once, during the configuration stage of your project. After this, the table remembers it, effectively making the embedding functions *disappear in the background* so you don't have to worry about manually passing callables, and instead, simply focus on the rest of your data engineering pipeline.
 
 !!! warning
-    Using the implicit embeddings management approach means that you can forget about the manually passing around embedding
-    functions in your code, as long as you don't intend to change it at a later time. If your embedding function changes,
-    you'll have to re-configure your table with the new embedding function and regenerate the embeddings.
+    Using the embedding function registry means that you don't have to explicitly generate the embeddings yourself. 
+    However, if your embedding function changes, you'll have to re-configure your table with the new embedding function 
+    and regenerate the embeddings. In the future, we plan to support the ability to change the embedding function via
+    table metadata and have LanceDB automatically take care of regenerating the embeddings.
+
 
 ## 1. Define the embedding function
-We have some pre-defined embedding functions in the global registry, with more coming soon. Here's let's an implementation of CLIP as example.
-```
-registry = EmbeddingFunctionRegistry.get_instance()
-clip = registry.get("open-clip").create()
 
-```
-You can also define your own embedding function by implementing the `EmbeddingFunction` abstract base interface. It subclasses Pydantic Model which can be utilized to write complex schemas simply as we'll see next!
+=== "Python"
+    In the LanceDB python SDK, we define a global embedding function registry with
+    many different embedding models and even more coming soon. 
+    Here's let's an implementation of CLIP as example.
+
+    ```python
+    from lancedb.embeddings import get_registry
+
+    registry = get_registry()
+    clip = registry.get("open-clip").create()
+    ```
+
+    You can also define your own embedding function by implementing the `EmbeddingFunction` 
+    abstract base interface. It subclasses Pydantic Model which can be utilized to write complex schemas simply as we'll see next!
+
+=== "JavaScript""
+    In the TypeScript SDK, the choices are more limited. For now, only the OpenAI
+    embedding function is available.
+
+    ```javascript
+    const lancedb = require("vectordb");
+
+    // You need to provide an OpenAI API key
+    const apiKey = "sk-..."
+    // The embedding function will create embeddings for the 'text' column
+    const embedding = new lancedb.OpenAIEmbeddingFunction('text', apiKey)
+    ```
 
 ## 2. Define the data model or schema
-The embedding function defined above abstracts away all the details about the models and dimensions required to define the schema. You can simply set a field as **source** or **vector** column. Here's how:
 
-```python
-class Pets(LanceModel):
-    vector: Vector(clip.ndims) = clip.VectorField()
-    image_uri: str = clip.SourceField()
-```
+=== "Python"
+    The embedding function defined above abstracts away all the details about the models and dimensions required to define the schema. You can simply set a field as **source** or **vector** column. Here's how:
 
-`VectorField` tells LanceDB to use the clip embedding function to generate query embeddings for the `vector` column and `SourceField` ensures that when adding data, we automatically use the specified embedding function to encode `image_uri`.
+    ```python
+    class Pets(LanceModel):
+        vector: Vector(clip.ndims) = clip.VectorField()
+        image_uri: str = clip.SourceField()
+    ```
 
-## 3. Create LanceDB table
-Now that we have chosen/defined our embedding function and the schema, we can create the table:
+    `VectorField` tells LanceDB to use the clip embedding function to generate query embeddings for the `vector` column and `SourceField` ensures that when adding data, we automatically use the specified embedding function to encode `image_uri`.
 
-```python
-db = lancedb.connect("~/lancedb")
-table = db.create_table("pets", schema=Pets)
+=== "JavaScript"
 
-```
+    For the TypeScript SDK, a schema can be inferred from input data, or an explicit
+    Arrow schema can be provided.
 
-That's it! We've provided all the information needed to embed the source and query inputs. We can now forget about the model and dimension details and start to build our VectorDB pipeline.
+## 3. Create table and add data
 
-## 4. Ingest lots of data and query your table
-Any new or incoming data can just be added and it'll be vectorized automatically.
+Now that we have chosen/defined our embedding function and the schema, 
+we can create the table and ingest data without needing to explicitly generate
+the embeddings at all:
 
-```python
-table.add([{"image_uri": u} for u in uris])
-```
+=== "Python"
+    ```python
+    db = lancedb.connect("~/lancedb")
+    table = db.create_table("pets", schema=Pets)
 
-Our OpenCLIP query embedding function supports querying via both text and images:
+    table.add([{"image_uri": u} for u in uris])
+    ```
 
-```python
-result = table.search("dog")
-```
+=== "JavaScript"
 
-Let's query an image:
+    ```javascript
+    const db = await lancedb.connect("data/sample-lancedb");
+    const data = [
+    { text: "pepperoni"},
+    { text: "pineapple"}
+    ]
 
-```python
-p = Path("path/to/images/samoyed_100.jpg")
-query_image = Image.open(p)
-table.search(query_image)
-```
+    const table = await db.createTable("vectors", data, embedding)
+    ```
+
+## 4. Querying your table
+Not only can you forget about the embeddings during ingestion, you also don't
+need to worry about it when you query the table:
+
+=== "Python"
+
+    Our OpenCLIP query embedding function supports querying via both text and images:
+
+    ```python
+    results = (
+        table.search("dog")
+        .limit(10)
+        .to_pandas()
+    )
+    ```
+
+    Or we can search using an image:
+
+    ```python
+    p = Path("path/to/images/samoyed_100.jpg")
+    query_image = Image.open(p)
+    results = (
+        table.search(query_image)
+        .limit(10)
+        .to_pandas()
+    )
+    ```
+
+    Both of the above snippet returns a pandas DataFrame with the 10 closest vectors to the query.
+
+=== "JavaScript"
+
+    ```javascript
+    const results = await table
+    .search("What's the best pizza topping?")
+    .limit(10)
+    .execute()
+    ```    
+    
+    The above snippet returns an array of records with the top 10 nearest neighbors to the query.
 
 ---
 
@@ -100,4 +165,5 @@ rs[2].image
 
 ![](../assets/dog_clip_output.png)
 
-Now that you have the basic idea about implicit management via embedding functions, let's dive deeper into a [custom API](./api.md) that you can use to implement your own embedding functions.
+Now that you have the basic idea about LanceDB embedding functions and the embedding function registry,
+let's dive deeper into defining your own [custom functions](./custom_embedding_function.md).

docs/src/embeddings/index.md

@@ -1,8 +1,14 @@
-Due to the nature of vector embeddings, they can be used to represent any kind of data, from text to images to audio. This makes them a very powerful tool for machine learning practitioners. However, there's no one-size-fits-all solution for generating embeddings - there are many different libraries and APIs (both commercial and open source) that can be used to generate embeddings from structured/unstructured data.
+Due to the nature of vector embeddings, they can be used to represent any kind of data, from text to images to audio. 
+This makes them a very powerful tool for machine learning practitioners. 
+However, there's no one-size-fits-all solution for generating embeddings - there are many different libraries and APIs 
+(both commercial and open source) that can be used to generate embeddings from structured/unstructured data.
 
-LanceDB supports 2 methods of vectorizing your raw data into embeddings.
+LanceDB supports 3 methods of working with embeddings.
 
-1. **Explicit**: By manually calling LanceDB's `with_embedding` function to vectorize your data via an `embed_func` of your choice
-2. **Implicit**: Allow LanceDB to embed the data and queries in the background as they come in, by using the table's `EmbeddingRegistry` information
+1. You can manually generate embeddings for the data and queries. This is done outside of LanceDB.
+2. You can use the built-in [embedding functions](./embedding_functions.md) to embed the data and queries in the background.
+3. For python users, you can define your own [custom embedding function](./custom_embedding_function.md)
+   that extends the default embedding functions.
 
-See the [explicit](embedding_explicit.md) and [implicit](embedding_functions.md) embedding sections for more details.
+For python users, there is also a legacy [with_embeddings API](./legacy.md).
+It is retained for compatibility and will be removed in a future version.

docs/src/embeddings/legacy.md

@@ -0,0 +1,99 @@
+The legacy `with_embeddings` API is for Python only and is deprecated.
+
+### Hugging Face
+
+The most popular open source option is to use the [sentence-transformers](https://www.sbert.net/) 
+library, which can be installed via pip.
+
+```bash
+pip install sentence-transformers
+```
+
+The example below shows how to use the `paraphrase-albert-small-v2` model to generate embeddings 
+for a given document.
+
+```python
+from sentence_transformers import SentenceTransformer
+
+name="paraphrase-albert-small-v2"
+model = SentenceTransformer(name)
+
+# used for both training and querying
+def embed_func(batch):
+    return [model.encode(sentence) for sentence in batch]
+```
+
+
+### OpenAI
+
+Another popular alternative is to use an external API like OpenAI's [embeddings API](https://platform.openai.com/docs/guides/embeddings/what-are-embeddings).
+
+```python
+import openai
+import os
+
+# Configuring the environment variable OPENAI_API_KEY
+if "OPENAI_API_KEY" not in os.environ:
+# OR set the key here as a variable
+openai.api_key = "sk-..."
+
+client = openai.OpenAI()
+
+def embed_func(c):    
+    rs = client.embeddings.create(input=c, model="text-embedding-ada-002")
+    return [record.embedding for record in rs["data"]]
+```
+
+
+## Applying an embedding function to data
+
+Using an embedding function, you can apply it to raw data
+to generate embeddings for each record.
+
+Say you have a pandas DataFrame with a `text` column that you want embedded,
+you can use the `with_embeddings` function to generate embeddings and add them to 
+an existing table.
+
+```python
+    import pandas as pd
+    from lancedb.embeddings import with_embeddings
+
+    df = pd.DataFrame(
+        [
+            {"text": "pepperoni"},
+            {"text": "pineapple"}
+        ]
+    )
+    data = with_embeddings(embed_func, df)
+
+    # The output is used to create / append to a table
+    tbl = db.create_table("my_table", data=data)
+```
+
+If your data is in a different column, you can specify the `column` kwarg to `with_embeddings`.
+
+By default, LanceDB calls the function with batches of 1000 rows. This can be configured
+using the `batch_size` parameter to `with_embeddings`.
+
+LanceDB automatically wraps the function with retry and rate-limit logic to ensure the OpenAI
+API call is reliable.
+
+## Querying using an embedding function
+
+!!! warning
+    At query time, you **must** use the same embedding function you used to vectorize your data.
+    If you use a different embedding function, the embeddings will not reside in the same vector
+    space and the results will be nonsensical.
+
+=== "Python"
+     ```python
+     query = "What's the best pizza topping?"
+     query_vector = embed_func([query])[0]
+     results = (
+        tbl.search(query_vector)
+        .limit(10)
+        .to_pandas()
+     )
+     ```
+
+     The above snippet returns a pandas DataFrame with the 10 closest vectors to the query.

docs/src/examples/image_embeddings_roboflow.md

@@ -43,7 +43,7 @@ pip install lancedb
 We also need to install a specific commit of `tantivy`, a dependency of the LanceDB full text search engine we will use later in this guide:
 
 ```
-pip install tantivy@git+https://github.com/quickwit-oss/tantivy-py#164adc87e1a033117001cf70e38c82a53014d985
+pip install tantivy
 ```
 
 Create a new Python file and add the following code:

docs/src/examples/modal_langchain.py

@@ -1,6 +1,5 @@
 import pickle
 import re
-import sys
 import zipfile
 from pathlib import Path
 
@@ -79,7 +78,10 @@ def qanda_langchain(query):
     download_docs()
     docs = store_docs()
 
-    text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200,)
+    text_splitter = RecursiveCharacterTextSplitter(
+        chunk_size=1000,
+        chunk_overlap=200,
+    )
     documents = text_splitter.split_documents(docs)
     embeddings = OpenAIEmbeddings()
 

docs/src/extra_js/init_ask_ai_widget.js

@@ -0,0 +1,11 @@
+document.addEventListener("DOMContentLoaded", function () {
+    var script = document.createElement("script");
+    script.src = "https://widget.kapa.ai/kapa-widget.bundle.js";
+    script.setAttribute("data-website-id", "c5881fae-cec0-490b-b45e-d83d131d4f25");
+    script.setAttribute("data-project-name", "LanceDB");
+    script.setAttribute("data-project-color", "#000000");
+    script.setAttribute("data-project-logo", "https://avatars.githubusercontent.com/u/108903835?s=200&v=4");
+  script.setAttribute("data-modal-example-questions","Help me create an IVF_PQ index,How do I do an exhaustive search?,How do I create a LanceDB table?,Can I use my own embedding function?");
+    script.async = true;
+    document.head.appendChild(script);
+  });

docs/src/faq.md

@@ -40,7 +40,7 @@ LanceDB and its underlying data format, Lance, are built to scale to really larg
 
 No. LanceDB is blazing fast (due to its disk-based index) for even brute force kNN search, within reason. In our benchmarks, computing 100K pairs of 1000-dimension vectors takes less than 20ms. For small datasets of ~100K records or applications that can accept ~100ms latency, an ANN index is usually not necessary.
 
-For large-scale (>1M) or higher dimension vectors, it is beneficial to create an ANN index.
+For large-scale (>1M) or higher dimension vectors, it is beneficial to create an ANN index. See the [ANN indexes](ann_indexes.md) section for more details.
 
 ### Does LanceDB support full-text search?
 
@@ -69,3 +69,19 @@ MinIO supports an S3 compatible API. In order to connect to a MinIO instance, yo
 - Set the envvar `AWS_ENDPOINT` to the URL of your MinIO API
 - Set the envvars `AWS_ACCESS_KEY_ID` and `AWS_SECRET_ACCESS_KEY` with your MinIO credential
 - Call `lancedb.connect("s3://minio_bucket_name")`
+
+### Where can I find benchmarks for LanceDB?
+
+Refer to this [post](https://blog.lancedb.com/benchmarking-lancedb-92b01032874a) for recent benchmarks.
+
+### How much data can LanceDB practically manage without effecting performance?
+
+We target good performance on ~10-50 billion rows and ~10-30 TB of data.
+
+### Does LanceDB support concurrent operations?
+
+LanceDB can handle concurrent reads very well, and can scale horizontally. The main constraint is how well the [storage layer](https://lancedb.github.io/lancedb/concepts/storage/) you've chosen scales. For writes, we support concurrent writing, though too many concurrent writers can lead to failing writes as there is a limited number of times a writer retries a commit
+
+!!! info "Multiprocessing with LanceDB"
+
+    For multiprocessing you should probably not use ```fork``` as lance is multi-threaded internally and ```fork``` and multi-thread do not work well.[Refer to this discussion](https://discuss.python.org/t/concerns-regarding-deprecation-of-fork-with-alive-threads/33555)

docs/src/fts.md

@@ -75,21 +75,40 @@ applied on top of the full text search results. This can be invoked via the fami
 table.search("puppy").limit(10).where("meta='foo'").to_list()
 ```
 
-## Syntax
+## Phrase queries vs. terms queries
 
-For full-text search you can perform either a phrase query like "the old man and the sea", 
-or a structured search query like "(Old AND Man) AND Sea".
-Double quotes are used to disambiguate.
+For full-text search you can specify either a **phrase** query like `"the old man and the sea"`, 
+or a **terms** search query like `"(Old AND Man) AND Sea"`. For more details on the terms
+query syntax, see Tantivy's [query parser rules](https://docs.rs/tantivy/latest/tantivy/query/struct.QueryParser.html).
 
-For example: 
+!!! tip "Note"
+    The query parser will raise an exception on queries that are ambiguous. For example, in the query `they could have been dogs OR cats`, `OR` is capitalized so it's considered a keyword query operator. But it's ambiguous how the left part should be treated. So if you submit this search query as is, you'll get `Syntax Error: they could have been dogs OR cats`.
 
-If you intended "they could have been dogs OR cats" as a phrase query, this actually
-raises a syntax error since `OR` is a recognized operator. If you make `or` lower case,
-this avoids the syntax error. However, it is cumbersome to have to remember what will
-conflict with the query syntax. Instead, if you search using 
-`table.search('"they could have been dogs OR cats"')`, then the syntax checker avoids
-checking inside the quotes.
+    ```py
+    # This raises a syntax error
+    table.search("they could have been dogs OR cats")
+    ```
 
+    On the other hand, lowercasing `OR` to `or` will work, because there are no capitalized logical operators and
+    the query is treated as a phrase query.
+
+    ```py
+    # This works!
+    table.search("they could have been dogs or cats")
+    ```
+
+It can be cumbersome to have to remember what will cause a syntax error depending on the type of
+query you want to perform. To make this simpler, when you want to perform a phrase query, you can
+enforce it in one of two ways:
+
+1. Place the double-quoted query inside single quotes. For example, `table.search('"they could have been dogs OR cats"')` is treated as
+a phrase query.
+2. Explicitly declare the `phrase_query()` method. This is useful when you have a phrase query that
+itself contains double quotes. For example, `table.search('the cats OR dogs were not really "pets" at all').phrase_query()`
+is treated as a phrase query.
+
+In general, a query that's declared as a phrase query will be wrapped in double quotes during parsing, with nested
+double quotes replaced by single quotes.
 
 ## Configurations
 

docs/src/guides/storage.md

@@ -0,0 +1,167 @@
+# Configuring cloud storage
+
+<!-- TODO: When we add documentation for how to configure other storage types
+           we can change the name to a more general "Configuring storage" -->
+
+When using LanceDB OSS, you can choose where to store your data. The tradeoffs between different storage options are discussed in the [storage concepts guide](../concepts/storage.md). This guide shows how to configure LanceDB to use different storage options.
+
+## Object Stores
+
+LanceDB OSS supports object stores such as AWS S3 (and compatible stores), Azure Blob Store, and Google Cloud Storage. Which object store to use is determined by the URI scheme of the dataset path. `s3://` is used for AWS S3, `az://` is used for Azure Blob Storage, and `gs://` is used for Google Cloud Storage. These URIs are passed to the `connect` function:
+
+=== "Python"
+
+    AWS S3:
+
+    ```python
+    import lancedb
+    db = lancedb.connect("s3://bucket/path")
+    ```
+
+    Google Cloud Storage:
+
+    ```python
+    import lancedb
+    db = lancedb.connect("gs://bucket/path")
+    ```
+
+    Azure Blob Storage:
+
+    ```python
+    import lancedb
+    db = lancedb.connect("az://bucket/path")
+    ```
+
+=== "JavaScript"
+
+    AWS S3:
+
+    ```javascript
+    const lancedb = require("lancedb");
+    const db = await lancedb.connect("s3://bucket/path");
+    ```
+
+    Google Cloud Storage:
+
+    ```javascript
+    const lancedb = require("lancedb");
+    const db = await lancedb.connect("gs://bucket/path");
+    ```
+
+    Azure Blob Storage:
+
+    ```javascript
+    const lancedb = require("lancedb");
+    const db = await lancedb.connect("az://bucket/path");
+    ```
+
+In most cases, when running in the respective cloud and permissions are set up correctly, no additional configuration is required. When running outside of the respective cloud, authentication credentials must be provided using environment variables. In general, these environment variables are the same as those used by the respective cloud SDKs. The sections below describe the environment variables that can be used to configure each object store.
+
+LanceDB OSS uses the [object-store](https://docs.rs/object_store/latest/object_store/) Rust crate for object store access. There are general environment variables that can be used to configure the object store, such as the request timeout and proxy configuration. See the [object_store ClientConfigKey](https://docs.rs/object_store/latest/object_store/enum.ClientConfigKey.html) doc for available configuration options. The environment variables that can be set are the snake-cased versions of these variable names. For example, to set `ProxyUrl` use the environment variable `PROXY_URL`. (Don't let the Rust docs intimidate you! We link to them so you can see an up-to-date list of the available options.)
+
+
+### AWS S3
+
+To configure credentials for AWS S3, you can use the `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, and `AWS_SESSION_TOKEN` environment variables.
+
+Alternatively, if you are using AWS SSO, you can use the `AWS_PROFILE` and `AWS_DEFAULT_REGION` environment variables.
+
+You can see a full list of environment variables [here](https://docs.rs/object_store/latest/object_store/aws/struct.AmazonS3Builder.html#method.from_env).
+
+!!! tip "Automatic cleanup for failed writes"
+
+    LanceDB uses [multi-part uploads](https://docs.aws.amazon.com/AmazonS3/latest/userguide/mpuoverview.html) when writing data to S3 in order to maximize write speed. LanceDB will abort these uploads when it shuts down gracefully, such as when cancelled by keyboard interrupt. However, in the rare case that LanceDB crashes, it is possible that some data will be left lingering in your account. To cleanup this data, we recommend (as AWS themselves do) that you setup a lifecycle rule to delete in-progress uploads after 7 days. See the AWS guide:
+
+    **[Configuring a bucket lifecycle configuration to delete incomplete multipart uploads](https://docs.aws.amazon.com/AmazonS3/latest/userguide/mpu-abort-incomplete-mpu-lifecycle-config.html)**
+
+#### AWS IAM Permissions
+
+If a bucket is private, then an IAM policy must be specified to allow access to it. For many development scenarios, using broad permissions such as a PowerUser account is more than sufficient for working with LanceDB. However, in many production scenarios, you may wish to have as narrow as possible permissions.
+
+For **read and write access**, LanceDB will need a policy such as:
+
+```json
+{
+    "Version": "2012-10-17",
+    "Statement": [
+        {
+            "Effect": "Allow",
+            "Action": [
+              "s3:PutObject",
+              "s3:GetObject",
+              "s3:DeleteObject",
+            ],
+            "Resource": "arn:aws:s3:::<bucket>/<prefix>/*"
+        },
+        {
+            "Effect": "Allow",
+            "Action": [
+                "s3:ListBucket",
+                "s3:GetBucketLocation"
+            ],
+            "Resource": "arn:aws:s3:::<bucket>",
+            "Condition": {
+                "StringLike": {
+                    "s3:prefix": [
+                        "<prefix>/*"
+                    ]
+                }
+            }
+        }
+    ]
+}
+```
+
+For **read-only access**, LanceDB will need a policy such as:
+
+```json
+{
+    "Version": "2012-10-17",
+    "Statement": [
+        {
+            "Effect": "Allow",
+            "Action": [
+              "s3:GetObject",
+            ],
+            "Resource": "arn:aws:s3:::<bucket>/<prefix>/*"
+        },
+        {
+            "Effect": "Allow",
+            "Action": [
+                "s3:ListBucket",
+                "s3:GetBucketLocation"
+            ],
+            "Resource": "arn:aws:s3:::<bucket>",
+            "Condition": {
+                "StringLike": {
+                    "s3:prefix": [
+                        "<prefix>/*"
+                    ]
+                }
+            }
+        }
+    ]
+}
+```
+
+#### S3-compatible stores
+
+LanceDB can also connect to S3-compatible stores, such as MinIO. To do so, you must specify two environment variables: `AWS_ENDPOINT` and `AWS_DEFAULT_REGION`. `AWS_ENDPOINT` should be the URL of the S3-compatible store, and `AWS_DEFAULT_REGION` should be the region to use.
+
+<!-- TODO: we should also document the use of S3 Express once we fully support it -->
+
+### Google Cloud Storage
+
+GCS credentials are configured by setting the `GOOGLE_SERVICE_ACCOUNT` environment variable to the path of a JSON file containing the service account credentials. There are several aliases for this environment variable, documented [here](https://docs.rs/object_store/latest/object_store/gcp/struct.GoogleCloudStorageBuilder.html#method.from_env).
+
+
+!!! info "HTTP/2 support"
+
+    By default, GCS uses HTTP/1 for communication, as opposed to HTTP/2. This improves maximum throughput significantly. However, if you wish to use HTTP/2 for some reason, you can set the environment variable `HTTP1_ONLY` to `false`.
+
+### Azure Blob Storage
+
+Azure Blob Storage credentials can be configured by setting the `AZURE_STORAGE_ACCOUNT_NAME` and ``AZURE_STORAGE_ACCOUNT_KEY`` environment variables. The full list of environment variables that can be set are documented [here](https://docs.rs/object_store/latest/object_store/azure/struct.MicrosoftAzureBuilder.html#method.from_env).
+
+
+<!-- TODO: demonstrate how to configure networked file systems for optimal performance -->

docs/src/guides/tables.md

@@ -16,9 +16,22 @@ This guide will show how to create tables, insert data into them, and update the
     db = lancedb.connect("./.lancedb")
     ```
 
+=== "Javascript"
+
+    Initialize a VectorDB connection and create a table using one of the many methods listed below.
+
+    ```javascript
+    const lancedb = require("vectordb");
+
+    const uri = "data/sample-lancedb";
+    const db = await lancedb.connect(uri);
+    ```
+
     LanceDB allows ingesting data from various sources - `dict`, `list[dict]`, `pd.DataFrame`, `pa.Table` or a `Iterator[pa.RecordBatch]`. Let's take a look at some of the these.
 
-    ### From list of tuples or dictionaries
+### From list of tuples or dictionaries
+
+=== "Python"
 
     ```python
     import lancedb
@@ -32,7 +45,6 @@ This guide will show how to create tables, insert data into them, and update the
 
     db["my_table"].head()
     ```
-
     !!! info "Note"
         If the table already exists, LanceDB will raise an error by default. 
 
@@ -51,6 +63,27 @@ This guide will show how to create tables, insert data into them, and update the
         db.create_table("name", data, mode="overwrite")
         ```
 
+=== "Javascript"
+    You can create a LanceDB table in JavaScript using an array of JSON records as follows.
+
+    ```javascript
+    const tb = await db.createTable("my_table", [{
+        "vector": [3.1, 4.1],
+        "item": "foo",
+        "price": 10.0
+    }, {
+        "vector": [5.9, 26.5],
+        "item": "bar",
+        "price": 20.0
+    }]);
+    ```
+    !!! info "Note"
+    If the table already exists, LanceDB will raise an error by default. If you want to overwrite the table, you need to specify the `WriteMode` in the createTable function.
+
+    ```javascript
+    const table = await con.createTable(tableName, data, { writeMode: WriteMode.Overwrite })
+    ```
+
     ### From a Pandas DataFrame
 
     ```python
@@ -67,7 +100,9 @@ This guide will show how to create tables, insert data into them, and update the
     db["my_table"].head()
     ```
     !!! info "Note"
-        Data is converted to Arrow before being written to disk. For maximum control over how data is saved, either provide the PyArrow schema to convert to or else provide a PyArrow Table directly.
+    Data is converted to Arrow before being written to disk. For maximum control over how data is saved, either provide the PyArrow schema to convert to or else provide a PyArrow Table directly.
+
+    The **`vector`** column needs to be a [Vector](../python/pydantic.md#vector-field) (defined as [pyarrow.FixedSizeList](https://arrow.apache.org/docs/python/generated/pyarrow.list_.html)) type. 
 
     ```python
     custom_schema = pa.schema([
@@ -79,7 +114,7 @@ This guide will show how to create tables, insert data into them, and update the
     table = db.create_table("my_table", data, schema=custom_schema)
     ```
 
-    ### From a Polars DataFrame
+### From a Polars DataFrame
 
     LanceDB supports [Polars](https://pola.rs/), a modern, fast DataFrame library
     written in Rust. Just like in Pandas, the Polars integration is enabled by PyArrow
@@ -97,26 +132,44 @@ This guide will show how to create tables, insert data into them, and update the
     table = db.create_table("pl_table", data=data)
     ```
 
-    ### From PyArrow Tables
-    You can also create LanceDB tables directly from PyArrow tables
+### From an Arrow Table
+=== "Python"
+    You can also create LanceDB tables directly from Arrow tables. 
+    LanceDB supports float16 data type!
 
     ```python
-    table = pa.Table.from_arrays(
-            [
-                pa.array([[3.1, 4.1, 5.1, 6.1], [5.9, 26.5, 4.7, 32.8]],
-                        pa.list_(pa.float32(), 4)),
-                pa.array(["foo", "bar"]),
-                pa.array([10.0, 20.0]),
-            ],
-            ["vector", "item", "price"],
-        )
+    import pyarrows as pa
+    import numpy as np
+    
+    dim = 16
+    total = 2
+    schema = pa.schema(
+        [
+            pa.field("vector", pa.list_(pa.float16(), dim)),
+            pa.field("text", pa.string())
+        ]
+    )
+    data = pa.Table.from_arrays(
+        [
+            pa.array([np.random.randn(dim).astype(np.float16) for _ in range(total)],
+                    pa.list_(pa.float16(), dim)),
+            pa.array(["foo", "bar"])
+        ],
+        ["vector", "text"],
+    )
+    tbl = db.create_table("f16_tbl", data, schema=schema)
+    ```
 
-    db = lancedb.connect("db")
+=== "Javascript"
+    You can also create LanceDB tables directly from Arrow tables. 
+    LanceDB supports Float16 data type!
 
-    tbl = db.create_table("my_table", table)
+    ```javascript
+    --8<-- "docs/src/basic_legacy.ts:create_f16_table"
     ```
 
     ### From Pydantic Models
+
     When you create an empty table without data, you must specify the table schema.
     LanceDB supports creating tables by specifying a PyArrow schema or a specialized
     Pydantic model called `LanceModel`.
@@ -261,37 +314,6 @@ This guide will show how to create tables, insert data into them, and update the
 
     You can also use iterators of other types like Pandas DataFrame or Pylists directly in the above example.
 
-=== "JavaScript"
-    Initialize a VectorDB connection and create a table using one of the many methods listed below.
-
-    ```javascript
-    const lancedb = require("vectordb");
-
-    const uri = "data/sample-lancedb";
-    const db = await lancedb.connect(uri);
-    ```
-
-    You can create a LanceDB table in JavaScript using an array of JSON records as follows.
-
-    ```javascript
-    const tb = await db.createTable("my_table", [{
-        "vector": [3.1, 4.1],
-        "item": "foo",
-        "price": 10.0
-    }, {
-        "vector": [5.9, 26.5],
-        "item": "bar",
-        "price": 20.0
-    }]);
-    ```
-
-    !!! info "Note"
-    If the table already exists, LanceDB will raise an error by default. If you want to overwrite the table, you need to specify the `WriteMode` in the createTable function.
-
-    ```javascript
-    const table = await con.createTable(tableName, data, { writeMode: WriteMode.Overwrite })
-    ```
-
 ## Open existing tables
 
 === "Python"
@@ -614,6 +636,70 @@ The `values` parameter is used to provide the new values for the columns as lite
 
     When rows are updated, they are moved out of the index. The row will still show up in ANN queries, but the query will not be as fast as it would be if the row was in the index. If you update a large proportion of rows, consider rebuilding the index afterwards.
 
+## Consistency
+
+In LanceDB OSS, users can set the `read_consistency_interval` parameter on connections to achieve different levels of read consistency. This parameter determines how frequently the database synchronizes with the underlying storage system to check for updates made by other processes. If another process updates a table, the database will not see the changes until the next synchronization.
+
+There are three possible settings for `read_consistency_interval`:
+
+1. **Unset (default)**: The database does not check for updates to tables made by other processes. This provides the best query performance, but means that clients may not see the most up-to-date data. This setting is suitable for applications where the data does not change during the lifetime of the table reference.
+2. **Zero seconds (Strong consistency)**: The database checks for updates on every read. This provides the strongest consistency guarantees, ensuring that all clients see the latest committed data. However, it has the most overhead. This setting is suitable when consistency matters more than having high QPS.
+3. **Custom interval (Eventual consistency)**: The database checks for updates at a custom interval, such as every 5 seconds. This provides eventual consistency, allowing for some lag between write and read operations. Performance wise, this is a middle ground between strong consistency and no consistency check. This setting is suitable for applications where immediate consistency is not critical, but clients should see updated data eventually.
+
+!!! tip "Consistency in LanceDB Cloud"
+
+    This is only tune-able in LanceDB OSS. In LanceDB Cloud, readers are always eventually consistent.
+
+=== "Python"
+    
+    To set strong consistency, use `timedelta(0)`:
+
+    ```python
+    from datetime import timedelta
+    db = lancedb.connect("./.lancedb",. read_consistency_interval=timedelta(0))
+    table = db.open_table("my_table")
+    ```
+
+    For eventual consistency, use a custom `timedelta`:
+
+    ```python
+    from datetime import timedelta
+    db = lancedb.connect("./.lancedb", read_consistency_interval=timedelta(seconds=5))
+    table = db.open_table("my_table")
+    ```
+
+    By default, a `Table` will never check for updates from other writers. To manually check for updates you can use `checkout_latest`:
+
+    ```python
+    db = lancedb.connect("./.lancedb")
+    table = db.open_table("my_table")
+    
+    # (Other writes happen to my_table from another process)
+
+    # Check for updates
+    table.checkout_latest()
+    ```
+
+=== "JavaScript/Typescript"
+
+    To set strong consistency, use `0`:
+
+    ```javascript
+    const db = await lancedb.connect({ uri: "./.lancedb", readConsistencyInterval: 0 });
+    const table = await db.openTable("my_table");
+    ```
+
+    For eventual consistency, specify the update interval as seconds:
+
+    ```javascript
+    const db = await lancedb.connect({ uri: "./.lancedb", readConsistencyInterval: 5 });
+    const table = await db.openTable("my_table");
+    ```
+
+<!-- Node doesn't yet support the version time travel: https://github.com/lancedb/lancedb/issues/1007 
+    Once it does, we can show manual consistency check for Node as well.
+-->
+
 ## What's next?
 
 Learn the best practices on creating an ANN index and getting the most out of it.

docs/src/hybrid_search/eval.md

@@ -0,0 +1,49 @@
+# Hybrid Search
+
+Hybrid Search is a broad (often misused) term. It can mean anything from combining multiple methods for searching, to applying ranking methods to better sort the results. In this blog, we use the definition of "hybrid search" to mean using a combination of keyword-based and vector search.
+
+## The challenge of (re)ranking search results
+Once you have a group of the most relevant search results from multiple search sources, you'd likely standardize the score and rank them accordingly. This process can also be seen as another independent step - reranking.
+There are two approaches for reranking search results from multiple sources.
+* <b>Score-based</b>: Calculate final relevance scores based on a weighted linear combination of individual search algorithm scores. Example - Weighted linear combination of semantic search & keyword-based search results.
+* <b>Relevance-based</b>: Discards the existing scores and calculates the relevance of each search result - query pair. Example - Cross Encoder models
+
+Even though there are many strategies for reranking search results, none works for all cases. Moreover, evaluating them itself is a challenge. Also, reranking can be dataset, application specific so it's hard to generalize.
+
+### Example evaluation of hybrid search with Reranking
+
+Here's some evaluation numbers from experiment comparing these re-rankers on about 800 queries. It is modified version of an evaluation script from [llama-index](https://github.com/run-llama/finetune-embedding/blob/main/evaluate.ipynb) that measures hit-rate at top-k.
+
+<b> With OpenAI ada2 embedding </b>
+
+Vector Search baseline - `0.64`
+
+| Reranker | Top-3 | Top-5 | Top-10 |
+| --- | --- | --- | --- |
+| Linear Combination | `0.73` | `0.74` | `0.85` |
+| Cross Encoder | `0.71` | `0.70` | `0.77` |
+| Cohere | `0.81` | `0.81` | `0.85` |
+| ColBERT | `0.68` | `0.68` | `0.73` |
+
+<p>
+<img src="https://github.com/AyushExel/assets/assets/15766192/d57b1780-ef27-414c-a5c3-73bee7808a45">
+</p>
+
+<b> With OpenAI embedding-v3-small </b>
+
+Vector Search baseline - `0.59`
+
+| Reranker | Top-3 | Top-5 | Top-10 |
+| --- | --- | --- | --- |
+| Linear Combination | `0.68` | `0.70` | `0.84` |
+| Cross Encoder | `0.72` | `0.72` | `0.79` |
+| Cohere | `0.79` | `0.79` | `0.84` |
+| ColBERT | `0.70` | `0.70` | `0.76` |
+
+<p>
+<img src="https://github.com/AyushExel/assets/assets/15766192/259adfd2-6ec6-4df6-a77d-1456598970dd">
+</p>
+
+### Conclusion
+
+The results show that the reranking methods are able to improve the search results. However, the improvement is not consistent across all rerankers. The choice of reranker depends on the dataset and the application. It is also important to note that the reranking methods are not a replacement for the search methods. They are complementary and should be used together to get the best results. The speed to recall tradeoff is also an important factor to consider when choosing the reranker.

docs/src/hybrid_search/hybrid_search.md

@@ -0,0 +1,242 @@
+# Hybrid Search
+
+LanceDB supports both semantic and keyword-based search (also termed full-text search, or FTS). In real world applications, it is often useful to combine these two approaches to get the best best results. For example, you may want to search for a document that is semantically similar to a query document, but also contains a specific keyword. This is an example of *hybrid search*, a search algorithm that combines multiple search techniques.
+
+## Hybrid search in LanceDB
+You can perform hybrid search in LanceDB by combining the results of semantic and full-text search via a reranking algorithm of your choice. LanceDB provides multiple rerankers out of the box. However, you can always write a custom reranker if your use case need more sophisticated logic .
+
+```python
+import os
+
+import lancedb
+import openai
+from lancedb.embeddings import get_registry
+from lancedb.pydantic import LanceModel, Vector
+
+db = lancedb.connect("~/.lancedb")
+
+# Ingest embedding function in LanceDB table
+# Configuring the environment variable OPENAI_API_KEY
+if "OPENAI_API_KEY" not in os.environ:
+# OR set the key here as a variable
+    openai.api_key = "sk-..."
+embeddings = get_registry().get("openai").create()
+
+class Documents(LanceModel):
+    vector: Vector(embeddings.ndims()) = embeddings.VectorField()
+    text: str = embeddings.SourceField()
+
+table = db.create_table("documents", schema=Documents)
+
+data = [
+    { "text": "rebel spaceships striking from a hidden base"},
+    { "text": "have won their first victory against the evil Galactic Empire"},
+    { "text": "during the battle rebel spies managed to steal secret plans"},
+    { "text": "to the Empire's ultimate weapon the Death Star"}
+]
+
+# ingest docs with auto-vectorization
+table.add(data)
+
+# Create a fts index before the hybrid search
+table.create_fts_index("text")
+# hybrid search with default re-ranker
+results = table.search("flower moon", query_type="hybrid").to_pandas()
+```
+
+By default, LanceDB uses `LinearCombinationReranker(weight=0.7)` to combine and rerank the results of semantic and full-text search. You can customize the hyperparameters as needed or write your own custom reranker. Here's how you can use any of the available rerankers:
+
+
+### `rerank()` arguments
+* `normalize`: `str`, default `"score"`:
+    The method to normalize the scores. Can be "rank" or "score". If "rank", the scores are converted to ranks and then normalized. If "score", the scores are normalized directly.
+* `reranker`: `Reranker`, default `LinearCombinationReranker(weight=0.7)`.
+    The reranker to use. If not specified, the default reranker is used.
+
+
+## Available Rerankers
+LanceDB provides a number of re-rankers out of the box. You can use any of these re-rankers by passing them to the `rerank()` method. Here's a list of available re-rankers:
+
+### Linear Combination Reranker
+This is the default re-ranker used by LanceDB. It combines the results of semantic and full-text search using a linear combination of the scores. The weights for the linear combination can be specified. It defaults to 0.7, i.e, 70% weight for semantic search and 30% weight for full-text search.
+
+
+```python
+from lancedb.rerankers import LinearCombinationReranker
+
+reranker = LinearCombinationReranker(weight=0.3) # Use 0.3 as the weight for vector search
+
+results = table.search("rebel", query_type="hybrid").rerank(reranker=reranker).to_pandas()
+```
+
+### Arguments
+----------------
+* `weight`: `float`, default `0.7`:
+    The weight to use for the semantic search score. The weight for the full-text search score is `1 - weights`.
+* `fill`: `float`, default `1.0`:
+        The score to give to results that are only in one of the two result sets.This is treated as penalty, so a higher value means a lower score.
+        TODO: We should just hardcode this-- its pretty confusing as we invert scores to calculate final score
+* `return_score` : str, default `"relevance"`
+        options are "relevance" or "all"
+        The type of score to return. If "relevance", will return only the `_relevance_score. If "all", will return all scores from the vector and FTS search along with the relevance score.
+
+### Cohere Reranker
+This re-ranker uses the [Cohere](https://cohere.ai/) API to combine the results of semantic and full-text search. You can use this re-ranker by passing `CohereReranker()` to the `rerank()` method. Note that you'll need to set the `COHERE_API_KEY` environment variable to use this re-ranker.
+
+```python
+from lancedb.rerankers import CohereReranker
+
+reranker = CohereReranker()
+
+results = table.search("vampire weekend", query_type="hybrid").rerank(reranker=reranker).to_pandas()
+```
+
+### Arguments
+----------------
+* `model_name` : str, default `"rerank-english-v2.0"`
+        The name of the cross encoder model to use. Available cohere models are:
+        - rerank-english-v2.0
+        - rerank-multilingual-v2.0
+* `column` : str, default `"text"`
+        The name of the column to use as input to the cross encoder model.
+* `top_n` : str, default `None`
+        The number of results to return. If None, will return all results.
+
+!!! Note
+    Only returns `_relevance_score`. Does not support `return_score = "all"`.
+
+### Cross Encoder Reranker
+This reranker uses the [Sentence Transformers](https://www.sbert.net/) library to combine the results of semantic and full-text search. You can use it by passing `CrossEncoderReranker()` to the `rerank()` method.
+
+```python
+from lancedb.rerankers import CrossEncoderReranker
+
+reranker = CrossEncoderReranker()
+
+results = table.search("harmony hall", query_type="hybrid").rerank(reranker=reranker).to_pandas()
+```
+
+
+### Arguments
+----------------
+* `model` : str, default `"cross-encoder/ms-marco-TinyBERT-L-6"`
+        The name of the cross encoder model to use. Available cross encoder models can be found [here](https://www.sbert.net/docs/pretrained_cross-encoders.html)
+* `column` : str, default `"text"`
+        The name of the column to use as input to the cross encoder model.
+* `device` : str, default `None`
+        The device to use for the cross encoder model. If None, will use "cuda" if available, otherwise "cpu".
+
+!!! Note
+    Only returns `_relevance_score`. Does not support `return_score = "all"`.
+
+
+### ColBERT Reranker
+This reranker uses the ColBERT model to combine the results of semantic and full-text search. You can use it by passing `ColbertrReranker()` to the `rerank()` method. 
+
+ColBERT reranker model calculates relevance of given docs against the query and don't take existing fts and vector search scores into account, so it currently only supports `return_score="relevance"`. By default, it looks for `text` column to rerank the results. But you can specify the column name to use as input to the cross encoder model as described below.
+
+```python
+from lancedb.rerankers import ColbertReranker
+
+reranker = ColbertReranker()
+
+results = table.search("harmony hall", query_type="hybrid").rerank(reranker=reranker).to_pandas()
+```
+
+### Arguments
+----------------
+* `model_name` : `str`, default `"colbert-ir/colbertv2.0"`
+        The name of the cross encoder model to use.
+* `column` : `str`, default `"text"`
+        The name of the column to use as input to the cross encoder model.
+* `return_score` : `str`, default `"relevance"`
+        options are `"relevance"` or `"all"`. Only `"relevance"` is supported for now.
+
+!!! Note
+    Only returns `_relevance_score`. Does not support `return_score = "all"`.
+
+### OpenAI Reranker
+This reranker uses the OpenAI API to combine the results of semantic and full-text search. You can use it by passing `OpenaiReranker()` to the `rerank()` method.
+
+!!! Note
+    This prompts chat model to rerank results which is not a dedicated reranker model. This should be treated as experimental.
+
+!!! Tip
+    - You might run out of token limit so set the search `limits` based on your token limit.
+    - It is recommended to use gpt-4-turbo-preview, the default model, older models might lead to undesired behaviour
+
+```python
+from lancedb.rerankers import OpenaiReranker
+
+reranker = OpenaiReranker()
+
+results = table.search("harmony hall", query_type="hybrid").rerank(reranker=reranker).to_pandas()
+```
+
+### Arguments
+----------------
+* `model_name` : `str`, default `"gpt-4-turbo-preview"`
+    The name of the cross encoder model to use.
+* `column` : `str`, default `"text"`
+    The name of the column to use as input to the cross encoder model.
+* `return_score` : `str`, default `"relevance"`
+    options are "relevance" or "all". Only "relevance" is supported for now.
+* `api_key` : `str`, default `None`
+    The API key to use. If None, will use the OPENAI_API_KEY environment variable.
+
+
+## Building Custom Rerankers
+You can build your own custom reranker by subclassing the `Reranker` class and implementing the `rerank_hybrid()` method. Here's an example of a custom reranker that combines the results of semantic and full-text search using a linear combination of the scores.
+
+The `Reranker` base interface comes with a `merge_results()` method that can be used to combine the results of semantic and full-text search. This is a vanilla merging algorithm that simply concatenates the results and removes the duplicates without taking the scores into consideration. It only keeps the first copy of the row encountered. This works well in cases that don't require the scores of semantic and full-text search to combine the results. If you want to use the scores or want to support `return_score="all"`, you'll need to implement your own merging algorithm.
+
+```python
+
+from lancedb.rerankers import Reranker
+import pyarrow as pa
+
+class MyReranker(Reranker):
+    def __init__(self, param1, param2, ..., return_score="relevance"):
+        super().__init__(return_score)
+        self.param1 = param1
+        self.param2 = param2
+    
+    def rerank_hybrid(self, query: str, vector_results: pa.Table, fts_results: pa.Table):
+        # Use the built-in merging function
+        combined_result = self.merge_results(vector_results, fts_results)
+        
+        # Do something with the combined results
+        # ...
+
+        # Return the combined results
+        return combined_result
+
+```
+
+### Example of a Custom Reranker
+For the sake of simplicity let's build custom reranker that just enchances the Cohere Reranker by accepting a filter query, and accept other CohereReranker params as kwags.
+
+```python
+
+from typing import List, Union
+import pandas as pd
+from lancedb.rerankers import CohereReranker
+
+class MofidifiedCohereReranker(CohereReranker):
+    def __init__(self, filters: Union[str, List[str]], **kwargs):
+        super().__init__(**kwargs)
+        filters = filters if isinstance(filters, list) else [filters]
+        self.filters = filters
+
+    def rerank_hybrid(self, query: str, vector_results: pa.Table, fts_results: pa.Table)-> pa.Table:
+        combined_result = super().rerank_hybrid(query, vector_results, fts_results)
+        df = combined_result.to_pandas()
+        for filter in self.filters:
+            df = df.query("not text.str.contains(@filter)")
+
+        return pa.Table.from_pandas(df)
+
+```
+
+!!! tip
+    The `vector_results` and `fts_results` are pyarrow tables. You can convert them to pandas dataframes using `to_pandas()` method and perform any operations you want. After you are done, you can convert the dataframe back to pyarrow table using `pa.Table.from_pandas()` method and return it.

docs/src/integrations/index.md

@@ -13,7 +13,7 @@ Get started using these examples and quick links.
 | Integrations | |
 |---|---:|
 | <h3> LlamaIndex </h3>LlamaIndex is a simple, flexible data framework for connecting custom data sources to large language models. Llama index integrates with LanceDB as the serverless VectorDB. <h3>[Lean More](https://gpt-index.readthedocs.io/en/latest/examples/vector_stores/LanceDBIndexDemo.html) </h3> |<img src="../assets/llama-index.jpg" alt="image" width="150" height="auto">|
-| <h3>Langchain</h3>Langchain allows building applications with LLMs through composability <h3>[Lean More](https://python.langchain.com/en/latest/modules/indexes/vectorstores/examples/lancedb.html) | <img src="../assets/langchain.png" alt="image" width="150" height="auto">|
+| <h3>Langchain</h3>Langchain allows building applications with LLMs through composability <h3>[Lean More](https://python.langchain.com/docs/integrations/vectorstores/lancedb) | <img src="../assets/langchain.png" alt="image" width="150" height="auto">|
 | <h3>Langchain TS</h3> Javascript bindings for Langchain. It integrates with LanceDB's serverless vectordb allowing you to build powerful AI applications through composibility using only serverless functions. <h3>[Learn More]( https://js.langchain.com/docs/modules/data_connection/vectorstores/integrations/lancedb) | <img src="../assets/langchain.png" alt="image" width="150" height="auto">|
 | <h3>Voxel51</h3>  It is an open source toolkit that enables you to build better computer vision workflows by improving the quality of your datasets and delivering insights about your models.<h3>[Learn More](./voxel51.md) | <img src="../assets/voxel.gif" alt="image" width="150" height="auto">|
 | <h3>PromptTools</h3>  Offers a set of free, open-source tools for testing and experimenting with models, prompts, and configurations. The core idea is to enable developers to evaluate prompts using familiar interfaces like code and notebooks. You can use it to experiment with different configurations of LanceDB, and test how LanceDB integrates with the LLM of your choice.<h3>[Learn More](./prompttools.md) | <img src="../assets/prompttools.jpeg" alt="image" width="150" height="auto">|

docs/src/notebooks/DisappearingEmbeddingFunction.ipynb

@@ -290,7 +290,7 @@
         "from lancedb.pydantic import LanceModel, Vector\n",
         "\n",
         "class Pets(LanceModel):\n",
-        "    vector: Vector(clip.ndims) = clip.VectorField()\n",
+        "    vector: Vector(clip.ndims()) = clip.VectorField()\n",
         "    image_uri: str = clip.SourceField()\n",
         "\n",
         "    @property\n",
@@ -360,7 +360,7 @@
         "    table = db.create_table(\"pets\", schema=Pets)\n",
         "    # use a sampling of 1000 images\n",
         "    p = Path(\"~/Downloads/images\").expanduser()\n",
-        "    uris = [str(f) for f in p.iterdir()]\n",
+        "    uris = [str(f) for f in p.glob(\"*.jpg\")]\n",
         "    uris = sample(uris, 1000)\n",
         "    table.add(pd.DataFrame({\"image_uri\": uris}))"
       ]
@@ -543,7 +543,7 @@
       ],
       "source": [
         "from PIL import Image\n",
-        "p = Path(\"/Users/changshe/Downloads/images/samoyed_100.jpg\")\n",
+        "p = Path(\"~/Downloads/images/samoyed_100.jpg\").expanduser()\n",
         "query_image = Image.open(p)\n",
         "query_image"
       ]

docs/src/notebooks/diffusiondb/datagen.py

@@ -23,10 +23,8 @@ from multiprocessing import Pool
 import lance
 import pyarrow as pa
 from datasets import load_dataset
-from PIL import Image
 from transformers import CLIPModel, CLIPProcessor, CLIPTokenizerFast
 
-import lancedb
 
 MODEL_ID = "openai/clip-vit-base-patch32"
 

docs/src/notebooks/tables_guide.ipynb

@@ -13,7 +13,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 50,
+   "execution_count": 2,
    "id": "c1b4e34b-a49c-471d-a343-a5940bb5138a",
    "metadata": {},
    "outputs": [],
@@ -23,7 +23,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": 3,
    "id": "4e5a8d07-d9a1-48c1-913a-8e0629289579",
    "metadata": {},
    "outputs": [],
@@ -44,7 +44,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 4,
    "id": "5df12f66-8d99-43ad-8d0b-22189ec0a6b9",
    "metadata": {},
    "outputs": [
@@ -62,7 +62,7 @@
        "long: [[-122.7,-74.1]]"
       ]
      },
-     "execution_count": 2,
+     "execution_count": 4,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -90,7 +90,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 5,
    "id": "f4d87ae9-0ccb-48eb-b31d-bb8f2370e47e",
    "metadata": {},
    "outputs": [
@@ -108,7 +108,7 @@
        "long: [[-122.7,-74.1]]"
       ]
      },
-     "execution_count": 3,
+     "execution_count": 5,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -135,10 +135,17 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": 6,
    "id": "25f34bcf-fca0-4431-8601-eac95d1bd347",
    "metadata": {},
    "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "[2024-01-31T18:59:33Z WARN  lance::dataset] No existing dataset at /Users/qian/Work/LanceDB/lancedb/docs/src/notebooks/.lancedb/table3.lance, it will be created\n"
+     ]
+    },
     {
      "data": {
       "text/plain": [
@@ -148,7 +155,7 @@
        "long: float"
       ]
      },
-     "execution_count": 8,
+     "execution_count": 6,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -171,45 +178,51 @@
    "id": "4df51925-7ca2-4005-9c72-38b3d26240c6",
    "metadata": {},
    "source": [
-    "### From PyArrow Tables\n",
+    "### From an Arrow Table\n",
     "\n",
-    "You can also create LanceDB tables directly from pyarrow tables"
+    "You can also create LanceDB tables directly from pyarrow tables. LanceDB supports float16 type."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 7,
    "id": "90a880f6-be43-4c9d-ba65-0b05197c0f6f",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
-       "vector: fixed_size_list<item: float>[2]\n",
-       "  child 0, item: float\n",
-       "item: string\n",
-       "price: double"
+       "vector: fixed_size_list<item: halffloat>[16]\n",
+       "  child 0, item: halffloat\n",
+       "text: string"
       ]
      },
-     "execution_count": 12,
+     "execution_count": 7,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "table = pa.Table.from_arrays(\n",
-    "        [\n",
-    "            pa.array([[3.1, 4.1], [5.9, 26.5]],\n",
-    "                    pa.list_(pa.float32(), 2)),\n",
-    "            pa.array([\"foo\", \"bar\"]),\n",
-    "            pa.array([10.0, 20.0]),\n",
-    "        ],\n",
-    "        [\"vector\", \"item\", \"price\"],\n",
-    "    )\n",
+    "import numpy as np\n",
     "\n",
-    "db = lancedb.connect(\"db\")\n",
+    "dim = 16\n",
+    "total = 2\n",
+    "schema = pa.schema(\n",
+    "    [\n",
+    "        pa.field(\"vector\", pa.list_(pa.float16(), dim)),\n",
+    "        pa.field(\"text\", pa.string())\n",
+    "    ]\n",
+    ")\n",
+    "data = pa.Table.from_arrays(\n",
+    "    [\n",
+    "        pa.array([np.random.randn(dim).astype(np.float16) for _ in range(total)],\n",
+    "                pa.list_(pa.float16(), dim)),\n",
+    "        pa.array([\"foo\", \"bar\"])\n",
+    "    ],\n",
+    "    [\"vector\", \"text\"],\n",
+    ")\n",
     "\n",
-    "tbl = db.create_table(\"test1\", table, mode=\"overwrite\")\n",
+    "tbl = db.create_table(\"f16_tbl\", data, schema=schema)\n",
     "tbl.schema"
    ]
   },
@@ -225,7 +238,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 8,
    "id": "d81121d7-e4b7-447c-a48c-974b6ebb464a",
    "metadata": {},
    "outputs": [
@@ -240,7 +253,7 @@
        "imdb_id: int64 not null"
       ]
      },
-     "execution_count": 13,
+     "execution_count": 8,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -282,7 +295,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 9,
    "id": "bc247142-4e3c-41a2-b94c-8e00d2c2a508",
    "metadata": {},
    "outputs": [
@@ -292,7 +305,7 @@
        "LanceTable(table4)"
       ]
      },
-     "execution_count": 14,
+     "execution_count": 9,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -333,7 +346,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 16,
+   "execution_count": 10,
    "id": "25ad3523-e0c9-4c28-b3df-38189c4e0e5f",
    "metadata": {},
    "outputs": [
@@ -346,7 +359,7 @@
        "price: double not null"
       ]
      },
-     "execution_count": 16,
+     "execution_count": 10,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -385,7 +398,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 11,
    "id": "2814173a-eacc-4dd8-a64d-6312b44582cc",
    "metadata": {},
    "outputs": [],
@@ -411,7 +424,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 18,
+   "execution_count": 12,
    "id": "df9e13c0-41f6-437f-9dfa-2fd71d3d9c45",
    "metadata": {},
    "outputs": [
@@ -421,7 +434,7 @@
        "['table6', 'table4', 'table5', 'movielens_small']"
       ]
      },
-     "execution_count": 18,
+     "execution_count": 12,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -432,7 +445,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 20,
+   "execution_count": 13,
    "id": "9343f5ad-6024-42ee-ac2f-6c1471df8679",
    "metadata": {},
    "outputs": [
@@ -541,7 +554,7 @@
        "9  [5.9, 26.5]  bar   20.0"
       ]
      },
-     "execution_count": 20,
+     "execution_count": 13,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -564,7 +577,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 21,
+   "execution_count": 14,
    "id": "8a56250f-73a1-4c26-a6ad-5c7a0ce3a9ab",
    "metadata": {},
    "outputs": [],
@@ -590,7 +603,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 22,
+   "execution_count": 15,
    "id": "030c7057-b98e-4e2f-be14-b8c1f927f83c",
    "metadata": {},
    "outputs": [],
@@ -621,7 +634,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 24,
+   "execution_count": 16,
    "id": "e7a17de2-08d2-41b7-bd05-f63d1045ab1f",
    "metadata": {},
    "outputs": [
@@ -629,16 +642,16 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "32\n"
+      "22\n"
      ]
     },
     {
      "data": {
       "text/plain": [
-       "17"
+       "12"
       ]
      },
-     "execution_count": 24,
+     "execution_count": 16,
      "metadata": {},
      "output_type": "execute_result"
     }
@@ -661,7 +674,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 30,
+   "execution_count": 17,
    "id": "fe3310bd-08f4-4a22-a63b-b3127d22f9f7",
    "metadata": {},
    "outputs": [
@@ -681,25 +694,20 @@
       "8    [3.1, 4.1]   foo   10.0\n",
       "9    [3.1, 4.1]   foo   10.0\n",
       "10   [3.1, 4.1]   foo   10.0\n",
-      "11   [3.1, 4.1]   foo   10.0\n",
-      "12   [3.1, 4.1]   foo   10.0\n",
-      "13   [3.1, 4.1]   foo   10.0\n",
-      "14   [3.1, 4.1]   foo   10.0\n",
-      "15   [3.1, 4.1]   foo   10.0\n",
-      "16   [3.1, 4.1]   foo   10.0\n"
+      "11   [3.1, 4.1]   foo   10.0\n"
      ]
     },
     {
      "ename": "OSError",
-     "evalue": "LanceError(IO): Error during planning: column foo does not exist",
+     "evalue": "LanceError(IO): Error during planning: column foo does not exist, /Users/runner/work/lance/lance/rust/lance-core/src/error.rs:212:23",
      "output_type": "error",
      "traceback": [
       "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
       "\u001b[0;31mOSError\u001b[0m                                   Traceback (most recent call last)",
-      "Cell \u001b[0;32mIn[30], line 4\u001b[0m\n\u001b[1;32m      2\u001b[0m to_remove \u001b[38;5;241m=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m, \u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;241m.\u001b[39mjoin(\u001b[38;5;28mstr\u001b[39m(v) \u001b[38;5;28;01mfor\u001b[39;00m v \u001b[38;5;129;01min\u001b[39;00m to_remove)\n\u001b[1;32m      3\u001b[0m \u001b[38;5;28mprint\u001b[39m(tbl\u001b[38;5;241m.\u001b[39mto_pandas())\n\u001b[0;32m----> 4\u001b[0m \u001b[43mtbl\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdelete\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;124;43mf\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mitem IN (\u001b[39;49m\u001b[38;5;132;43;01m{\u001b[39;49;00m\u001b[43mto_remove\u001b[49m\u001b[38;5;132;43;01m}\u001b[39;49;00m\u001b[38;5;124;43m)\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m)\u001b[49m\n\u001b[1;32m      5\u001b[0m tbl\u001b[38;5;241m.\u001b[39mto_pandas()\n",
-      "File \u001b[0;32m~/Documents/lancedb/lancedb/python/lancedb/table.py:610\u001b[0m, in \u001b[0;36mLanceTable.delete\u001b[0;34m(self, where)\u001b[0m\n\u001b[1;32m    609\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mdelete\u001b[39m(\u001b[38;5;28mself\u001b[39m, where: \u001b[38;5;28mstr\u001b[39m):\n\u001b[0;32m--> 610\u001b[0m     \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_dataset\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdelete\u001b[49m\u001b[43m(\u001b[49m\u001b[43mwhere\u001b[49m\u001b[43m)\u001b[49m\n",
-      "File \u001b[0;32m~/Documents/lancedb/lancedb/env/lib/python3.11/site-packages/lance/dataset.py:489\u001b[0m, in \u001b[0;36mLanceDataset.delete\u001b[0;34m(self, predicate)\u001b[0m\n\u001b[1;32m    487\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(predicate, pa\u001b[38;5;241m.\u001b[39mcompute\u001b[38;5;241m.\u001b[39mExpression):\n\u001b[1;32m    488\u001b[0m     predicate \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mstr\u001b[39m(predicate)\n\u001b[0;32m--> 489\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_ds\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdelete\u001b[49m\u001b[43m(\u001b[49m\u001b[43mpredicate\u001b[49m\u001b[43m)\u001b[49m\n",
-      "\u001b[0;31mOSError\u001b[0m: LanceError(IO): Error during planning: column foo does not exist"
+      "Cell \u001b[0;32mIn[17], line 4\u001b[0m\n\u001b[1;32m      2\u001b[0m to_remove \u001b[38;5;241m=\u001b[39m \u001b[38;5;124m\"\u001b[39m\u001b[38;5;124m, \u001b[39m\u001b[38;5;124m\"\u001b[39m\u001b[38;5;241m.\u001b[39mjoin(\u001b[38;5;28mstr\u001b[39m(v) \u001b[38;5;28;01mfor\u001b[39;00m v \u001b[38;5;129;01min\u001b[39;00m to_remove)\n\u001b[1;32m      3\u001b[0m \u001b[38;5;28mprint\u001b[39m(tbl\u001b[38;5;241m.\u001b[39mto_pandas())\n\u001b[0;32m----> 4\u001b[0m \u001b[43mtbl\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdelete\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;124;43mf\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mitem IN (\u001b[39;49m\u001b[38;5;132;43;01m{\u001b[39;49;00m\u001b[43mto_remove\u001b[49m\u001b[38;5;132;43;01m}\u001b[39;49;00m\u001b[38;5;124;43m)\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m~/Work/LanceDB/lancedb/docs/doc-venv/lib/python3.11/site-packages/lancedb/table.py:872\u001b[0m, in \u001b[0;36mLanceTable.delete\u001b[0;34m(self, where)\u001b[0m\n\u001b[1;32m    871\u001b[0m \u001b[38;5;28;01mdef\u001b[39;00m \u001b[38;5;21mdelete\u001b[39m(\u001b[38;5;28mself\u001b[39m, where: \u001b[38;5;28mstr\u001b[39m):\n\u001b[0;32m--> 872\u001b[0m     \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_dataset\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdelete\u001b[49m\u001b[43m(\u001b[49m\u001b[43mwhere\u001b[49m\u001b[43m)\u001b[49m\n",
+      "File \u001b[0;32m~/Work/LanceDB/lancedb/docs/doc-venv/lib/python3.11/site-packages/lance/dataset.py:596\u001b[0m, in \u001b[0;36mLanceDataset.delete\u001b[0;34m(self, predicate)\u001b[0m\n\u001b[1;32m    594\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28misinstance\u001b[39m(predicate, pa\u001b[38;5;241m.\u001b[39mcompute\u001b[38;5;241m.\u001b[39mExpression):\n\u001b[1;32m    595\u001b[0m     predicate \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mstr\u001b[39m(predicate)\n\u001b[0;32m--> 596\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_ds\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mdelete\u001b[49m\u001b[43m(\u001b[49m\u001b[43mpredicate\u001b[49m\u001b[43m)\u001b[49m\n",
+      "\u001b[0;31mOSError\u001b[0m: LanceError(IO): Error during planning: column foo does not exist, /Users/runner/work/lance/lance/rust/lance-core/src/error.rs:212:23"
      ]
     }
    ],
@@ -712,7 +720,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 43,
+   "execution_count": null,
    "id": "87d5bc21-847f-4c81-b56e-f6dbe5d05aac",
    "metadata": {},
    "outputs": [],
@@ -729,7 +737,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 44,
+   "execution_count": null,
    "id": "9cba4519-eb3a-4941-ab7e-873d762e750f",
    "metadata": {},
    "outputs": [],
@@ -742,7 +750,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 46,
+   "execution_count": null,
    "id": "5bdc9801-d5ed-4871-92d0-88b27108e788",
    "metadata": {},
    "outputs": [
@@ -817,7 +825,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.4"
+   "version": "3.11.7"
   }
  },
  "nbformat": 4,

docs/src/python/duckdb.md

@@ -1,6 +1,9 @@
 # DuckDB
 
-LanceDB is very well-integrated with [DuckDB](https://duckdb.org/), an in-process SQL OLAP database. This integration is done via [Arrow](https://duckdb.org/docs/guides/python/sql_on_arrow) .
+In Python, LanceDB tables can also be queried with [DuckDB](https://duckdb.org/), an in-process SQL OLAP database. This means you can write complex SQL queries to analyze your data in LanceDB.
+
+This integration is done via [Apache Arrow](https://duckdb.org/docs/guides/python/sql_on_arrow), which provides zero-copy data sharing between LanceDB and DuckDB. DuckDB is capable of passing down column selections and basic filters to LanceDB, reducing the amount of data that needs to be scanned to perform your query. Finally, the integration allows streaming data from LanceDB tables, allowing you to aggregate tables that won't fit into memory. All of this uses the same mechanism described in DuckDB's blog post *[DuckDB quacks Arrow](https://duckdb.org/2021/12/03/duck-arrow.html)*.
+
 
 We can demonstrate this by first installing `duckdb` and `lancedb`.
 
@@ -19,14 +22,15 @@ data = [
     {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}
 ]
 table = db.create_table("pd_table", data=data)
-arrow_table = table.to_arrow()
 ```
 
-DuckDB can directly query the `pyarrow.Table` object:
+To query the table, first call `to_lance` to convert the table to a "dataset", which is an object that can be queried by DuckDB. Then all you need to do is reference that dataset by the same name in your SQL query.
 
 ```python
 import duckdb
 
+arrow_table = table.to_lance()
+
 duckdb.query("SELECT * FROM arrow_table")
 ```
 

docs/src/python/python.md

@@ -24,6 +24,12 @@ pip install lancedb
 
 ::: lancedb.query.LanceQueryBuilder
 
+::: lancedb.query.LanceVectorQueryBuilder
+
+::: lancedb.query.LanceFtsQueryBuilder
+
+::: lancedb.query.LanceHybridQueryBuilder
+
 ## Embeddings
 
 ::: lancedb.embeddings.registry.EmbeddingFunctionRegistry
@@ -58,12 +64,26 @@ pip install lancedb
 
 ::: lancedb.schema.vector
 
+::: lancedb.merge.LanceMergeInsertBuilder
+
 ## Integrations
 
-### Pydantic
+## Pydantic
 
 ::: lancedb.pydantic.pydantic_to_schema
 
 ::: lancedb.pydantic.vector
 
 ::: lancedb.pydantic.LanceModel
+
+## Reranking
+
+::: lancedb.rerankers.linear_combination.LinearCombinationReranker
+
+::: lancedb.rerankers.cohere.CohereReranker
+
+::: lancedb.rerankers.colbert.ColbertReranker
+
+::: lancedb.rerankers.cross_encoder.CrossEncoderReranker
+
+::: lancedb.rerankers.openai.OpenaiReranker

docs/src/search.md

@@ -2,27 +2,26 @@
 
 A vector search finds the approximate or exact nearest neighbors to a given query vector.
 
-* In a recommendation system or search engine, you can find similar records to
-the one you searched.
-* In LLM and other AI applications,
-each data point can be represented by [embeddings generated from existing models](embeddings/index.md),
-following which the search returns the most relevant features.
+- In a recommendation system or search engine, you can find similar records to
+  the one you searched.
+- In LLM and other AI applications,
+  each data point can be represented by [embeddings generated from existing models](embeddings/index.md),
+  following which the search returns the most relevant features.
 
 ## Distance metrics
 
 Distance metrics are a measure of the similarity between a pair of vectors.
 Currently, LanceDB supports the following metrics:
 
-| Metric      | Description                          |
-| ----------- | ------------------------------------ |
-| `l2`        | [Euclidean / L2 distance](https://en.wikipedia.org/wiki/Euclidean_distance) |
-| `cosine`    | [Cosine Similarity](https://en.wikipedia.org/wiki/Cosine_similarity)|
-| `dot`       | [Dot Production](https://en.wikipedia.org/wiki/Dot_product) |
-
+| Metric   | Description                                                                 |
+| -------- | --------------------------------------------------------------------------- |
+| `l2`     | [Euclidean / L2 distance](https://en.wikipedia.org/wiki/Euclidean_distance) |
+| `cosine` | [Cosine Similarity](https://en.wikipedia.org/wiki/Cosine_similarity)        |
+| `dot`    | [Dot Production](https://en.wikipedia.org/wiki/Dot_product)                 |
 
 ## Exhaustive search (kNN)
 
-If you do not create a vector index, LanceDB exhaustively scans the *entire* vector space
+If you do not create a vector index, LanceDB exhaustively scans the _entire_ vector space
 and compute the distance to every vector in order to find the exact nearest neighbors. This is effectively a kNN search.
 
 <!-- Setup Code
@@ -38,22 +37,9 @@ data = [{"vector": row, "item": f"item {i}"}
 db.create_table("my_vectors", data=data)
 ```
 -->
-<!-- Setup Code
-```javascript
-const vectordb_setup = require('vectordb')
-const db_setup = await vectordb_setup.connect('data/sample-lancedb')
-
-let data = []
-for (let i = 0; i < 10_000; i++) {
-     data.push({vector: Array(1536).fill(i), id: `${i}`, content: "", longId: `${i}`},)
-}
-await db_setup.createTable('my_vectors', data)
-```
--->
 
 === "Python"
 
-
     ```python
     import lancedb
     import numpy as np
@@ -70,17 +56,12 @@ await db_setup.createTable('my_vectors', data)
 === "JavaScript"
 
     ```javascript
-    const vectordb = require('vectordb')
-    const db = await vectordb.connect('data/sample-lancedb')
+    --8<-- "docs/src/search_legacy.ts:import"
 
-    const tbl = await db.openTable("my_vectors")
-
-    const results_1 = await tbl.search(Array(1536).fill(1.2))
-        .limit(10)
-        .execute()
+    --8<-- "docs/src/search_legacy.ts:search1"
     ```
 
-By default, `l2` will be used as metric type. You can specify the metric type as 
+By default, `l2` will be used as metric type. You can specify the metric type as
 `cosine` or `dot` if required.
 
 === "Python"
@@ -92,20 +73,16 @@ By default, `l2` will be used as metric type. You can specify the metric type as
         .to_list()
     ```
 
-
 === "JavaScript"
 
     ```javascript
-    const results_2 = await tbl.search(Array(1536).fill(1.2))
-        .metricType("cosine")
-        .limit(10)
-        .execute()
+    --8<-- "docs/src/search_legacy.ts:search2"
     ```
 
 ## Approximate nearest neighbor (ANN) search
 
 To perform scalable vector retrieval with acceptable latencies, it's common to build a vector index.
-While the exhaustive search is guaranteed to always return 100% recall, the approximate nature of 
+While the exhaustive search is guaranteed to always return 100% recall, the approximate nature of
 an ANN search means that using an index often involves a trade-off between recall and latency.
 
 See the [IVF_PQ index](./concepts/index_ivfpq.md.md) for a deeper description of how `IVF_PQ`
@@ -117,7 +94,9 @@ LanceDB returns vector search results via different formats commonly used in pyt
 Let's create a LanceDB table with a nested schema:
 
 === "Python"
+
     ```python
+
     from datetime import datetime
     import lancedb
     from lancedb.pydantic import LanceModel, Vector
@@ -153,7 +132,7 @@ Let's create a LanceDB table with a nested schema:
     ### As a PyArrow table
 
     Using `to_arrow()` we can get the results back as a pyarrow Table.
-    This result table has the same columns as the LanceDB table, with 
+    This result table has the same columns as the LanceDB table, with
     the addition of an `_distance` column for vector search or a `score`
     column for full text search.
 
@@ -169,11 +148,11 @@ Let's create a LanceDB table with a nested schema:
     tbl.search(np.random.randn(1536)).to_pandas()
     ```
 
-    While other formats like Arrow/Pydantic/Python dicts have a natural 
-    way to handle nested schemas, pandas can only store nested data as a 
+    While other formats like Arrow/Pydantic/Python dicts have a natural
+    way to handle nested schemas, pandas can only store nested data as a
     python dict column, which makes it difficult to support nested references.
-    So for convenience, you can also tell LanceDB to flatten a nested schema 
-    when creating the pandas dataframe. 
+    So for convenience, you can also tell LanceDB to flatten a nested schema
+    when creating the pandas dataframe.
 
     ```python
     tbl.search(np.random.randn(1536)).to_pandas(flatten=True)

docs/src/search_legacy.ts

@@ -0,0 +1,41 @@
+// --8<-- [start:import]
+import * as lancedb from "vectordb";
+// --8<-- [end:import]
+import * as fs from "fs";
+
+async function setup() {
+  fs.rmSync("data/sample-lancedb", { recursive: true, force: true });
+  const db = await lancedb.connect("data/sample-lancedb");
+
+  let data = [];
+  for (let i = 0; i < 10_000; i++) {
+    data.push({
+      vector: Array(1536).fill(i),
+      id: `${i}`,
+      content: "",
+      longId: `${i}`,
+    });
+  }
+  await db.createTable("my_vectors", data);
+}
+
+async () => {
+  await setup();
+
+  // --8<-- [start:search1]
+  const db = await lancedb.connect("data/sample-lancedb");
+  const tbl = await db.openTable("my_vectors");
+
+  const results_1 = await tbl.search(Array(1536).fill(1.2)).limit(10).execute();
+  // --8<-- [end:search1]
+
+  // --8<-- [start:search2]
+  const results_2 = await tbl
+    .search(Array(1536).fill(1.2))
+    .metricType(lancedb.MetricType.Cosine)
+    .limit(10)
+    .execute();
+  // --8<-- [end:search2]
+
+  console.log("search: done");
+};

docs/src/sql.md

@@ -8,7 +8,7 @@ option that performs the filter prior to vector search. This can be useful to na
 the search space on a very large dataset to reduce query latency.
 
 <!-- Setup Code
-```python 
+```python
 import lancedb
 import numpy as np
 uri = "data/sample-lancedb"
@@ -21,7 +21,7 @@ tbl = db.create_table("my_vectors", data=data)
 ```
 -->
 <!-- Setup Code
-```javascript 
+```javascript
 const vectordb = require('vectordb')
 const db = await vectordb.connect('data/sample-lancedb')
 
@@ -34,6 +34,7 @@ const tbl = await db.createTable('myVectors', data)
 -->
 
 === "Python"
+
     ```py
     result = (
         tbl.search([0.5, 0.2])
@@ -44,12 +45,9 @@ const tbl = await db.createTable('myVectors', data)
     ```
 
 === "JavaScript"
+
     ```javascript
-    let result = await tbl.search(Array(1536).fill(0.5))
-        .limit(1)
-        .filter("id = 10")
-        .prefilter(true)
-        .execute()
+    --8<-- "docs/src/sql_legacy.ts:search"
     ```
 
 ## SQL filters
@@ -60,14 +58,14 @@ It can be used during vector search, update, and deletion operations.
 
 Currently, Lance supports a growing list of SQL expressions.
 
-* ``>``, ``>=``, ``<``, ``<=``, ``=``
-* ``AND``, ``OR``, ``NOT``
-* ``IS NULL``, ``IS NOT NULL``
-* ``IS TRUE``, ``IS NOT TRUE``, ``IS FALSE``, ``IS NOT FALSE``
-* ``IN``
-* ``LIKE``, ``NOT LIKE``
-* ``CAST``
-* ``regexp_match(column, pattern)``
+- `>`, `>=`, `<`, `<=`, `=`
+- `AND`, `OR`, `NOT`
+- `IS NULL`, `IS NOT NULL`
+- `IS TRUE`, `IS NOT TRUE`, `IS FALSE`, `IS NOT FALSE`
+- `IN`
+- `LIKE`, `NOT LIKE`
+- `CAST`
+- `regexp_match(column, pattern)`
 
 For example, the following filter string is acceptable:
 
@@ -82,29 +80,27 @@ For example, the following filter string is acceptable:
 === "Javascript"
 
     ```javascript
-    await tbl.search(Array(1536).fill(0))
-       .where("(item IN ('item 0', 'item 2')) AND (id > 10)")
-       .execute()
+    --8<-- "docs/src/sql_legacy.ts:vec_search"
     ```
 
-
 If your column name contains special characters or is a [SQL Keyword](https://docs.rs/sqlparser/latest/sqlparser/keywords/index.html),
 you can use backtick (`` ` ``) to escape it. For nested fields, each segment of the
 path must be wrapped in backticks.
 
 === "SQL"
+
     ```sql
     `CUBE` = 10 AND `column name with space` IS NOT NULL
       AND `nested with space`.`inner with space` < 2
     ```
 
-!!! warning
-    Field names containing periods (``.``) are not supported.
+!!!warning "Field names containing periods (`.`) are not supported."
 
 Literals for dates, timestamps, and decimals can be written by writing the string
 value after the type name. For example
 
 === "SQL"
+
     ```sql
     date_col = date '2021-01-01'
     and timestamp_col = timestamp '2021-01-01 00:00:00'
@@ -114,49 +110,47 @@ value after the type name. For example
 For timestamp columns, the precision can be specified as a number in the type
 parameter. Microsecond precision (6) is the default.
 
-| SQL              | Time unit    |
-|------------------|--------------|
-| ``timestamp(0)`` | Seconds      |
-| ``timestamp(3)`` | Milliseconds |
-| ``timestamp(6)`` | Microseconds |
-| ``timestamp(9)`` | Nanoseconds  |
+| SQL            | Time unit    |
+| -------------- | ------------ |
+| `timestamp(0)` | Seconds      |
+| `timestamp(3)` | Milliseconds |
+| `timestamp(6)` | Microseconds |
+| `timestamp(9)` | Nanoseconds  |
 
 LanceDB internally stores data in [Apache Arrow](https://arrow.apache.org/) format.
 The mapping from SQL types to Arrow types is:
 
-| SQL type | Arrow type |
-|----------|------------|
-| ``boolean`` | ``Boolean`` |
-| ``tinyint`` / ``tinyint unsigned`` | ``Int8`` / ``UInt8`` |
-| ``smallint`` / ``smallint unsigned`` | ``Int16`` / ``UInt16`` |
-| ``int`` or ``integer`` / ``int unsigned`` or ``integer unsigned`` | ``Int32`` / ``UInt32`` |
-| ``bigint`` / ``bigint unsigned`` | ``Int64`` / ``UInt64`` |
-| ``float`` | ``Float32`` |
-| ``double`` | ``Float64`` |
-| ``decimal(precision, scale)`` | ``Decimal128`` |
-| ``date`` | ``Date32`` |
-| ``timestamp`` | ``Timestamp`` [^1] |
-| ``string`` | ``Utf8`` |
-| ``binary`` | ``Binary`` |
+| SQL type                                                  | Arrow type         |
+| --------------------------------------------------------- | ------------------ |
+| `boolean`                                                 | `Boolean`          |
+| `tinyint` / `tinyint unsigned`                            | `Int8` / `UInt8`   |
+| `smallint` / `smallint unsigned`                          | `Int16` / `UInt16` |
+| `int` or `integer` / `int unsigned` or `integer unsigned` | `Int32` / `UInt32` |
+| `bigint` / `bigint unsigned`                              | `Int64` / `UInt64` |
+| `float`                                                   | `Float32`          |
+| `double`                                                  | `Float64`          |
+| `decimal(precision, scale)`                               | `Decimal128`       |
+| `date`                                                    | `Date32`           |
+| `timestamp`                                               | `Timestamp` [^1]   |
+| `string`                                                  | `Utf8`             |
+| `binary`                                                  | `Binary`           |
 
 [^1]: See precision mapping in previous table.
 
-
 ## Filtering without Vector Search
 
 You can also filter your data without search.
 
 === "Python"
-      ```python
-      tbl.search().where("id = 10").limit(10).to_arrow()
-      ```
+
+    ```python
+    tbl.search().where("id = 10").limit(10).to_arrow()
+    ```
 
 === "JavaScript"
-      ```javascript
-      await tbl.where('id = 10').limit(10).execute()
-      ```
 
-!!! warning
-    If your table is large, this could potentially return a very large
-    amount of data. Please be sure to use a `limit` clause unless
-    you're sure you want to return the whole result set.
+    ```javascript
+    --8<---- "docs/src/sql_legacy.ts:sql_search"
+    ```
+
+!!!warning "If your table is large, this could potentially return a very large amount of data. Please be sure to use a `limit` clause unless you're sure you want to return the whole result set."

docs/src/sql_legacy.ts

@@ -0,0 +1,38 @@
+import * as vectordb from "vectordb";
+
+(async () => {
+  const db = await vectordb.connect("data/sample-lancedb");
+
+  let data = [];
+  for (let i = 0; i < 10_000; i++) {
+    data.push({
+      vector: Array(1536).fill(i),
+      id: i,
+      item: `item ${i}`,
+      strId: `${i}`,
+    });
+  }
+  const tbl = await db.createTable("myVectors", data);
+
+  // --8<-- [start:search]
+  let result = await tbl
+    .search(Array(1536).fill(0.5))
+    .limit(1)
+    .filter("id = 10")
+    .prefilter(true)
+    .execute();
+  // --8<-- [end:search]
+
+  // --8<-- [start:vec_search]
+  await tbl
+    .search(Array(1536).fill(0))
+    .where("(item IN ('item 0', 'item 2')) AND (id > 10)")
+    .execute();
+  // --8<-- [end:vec_search]
+
+  // --8<-- [start:sql_search]
+  await tbl.filter("id = 10").limit(10).execute();
+  // --8<-- [end:sql_search]
+
+  console.log("SQL search: done");
+})();

docs/test/md_testing.js

@@ -1,54 +0,0 @@
-const glob = require("glob");
-const fs = require("fs");
-const path = require("path");
-
-const globString = "../src/**/*.md";
-
-const excludedGlobs = [
-  "../src/fts.md",
-  "../src/embedding.md",
-  "../src/examples/*.md",
-  "../src/guides/tables.md",
-  "../src/embeddings/*.md",
-];
-
-const nodePrefix = "javascript";
-const nodeFile = ".js";
-const nodeFolder = "node";
-const asyncPrefix = "(async () => {\n";
-const asyncSuffix = "})();";
-
-function* yieldLines(lines, prefix, suffix) {
-  let inCodeBlock = false;
-  for (const line of lines) {
-    if (line.trim().startsWith(prefix + nodePrefix)) {
-      inCodeBlock = true;
-    } else if (inCodeBlock && line.trim().startsWith(suffix)) {
-      inCodeBlock = false;
-      yield "\n";
-    } else if (inCodeBlock) {
-      yield line;
-    }
-  }
-}
-
-const files = glob.sync(globString, { recursive: true });
-const excludedFiles = glob.sync(excludedGlobs, { recursive: true });
-
-for (const file of files.filter((file) => !excludedFiles.includes(file))) {
-  const lines = [];
-  const data = fs.readFileSync(file, "utf-8");
-  const fileLines = data.split("\n");
-
-  for (const line of yieldLines(fileLines, "```", "```")) {
-    lines.push(line);
-  }
-
-  if (lines.length > 0) {
-    const fileName = path.basename(file, ".md");
-    const outPath = path.join(nodeFolder, fileName, `${fileName}${nodeFile}`);
-    console.log(outPath)
-    fs.mkdirSync(path.dirname(outPath), { recursive: true });
-    fs.writeFileSync(outPath, asyncPrefix + "\n" + lines.join("\n") + asyncSuffix);
-  }
-}

docs/test/md_testing.py

@@ -14,6 +14,7 @@ excluded_globs = [
     "../src/concepts/*.md",
     "../src/ann_indexes.md",
     "../src/basic.md",
+    "../src/hybrid_search/hybrid_search.md",
 ]
 
 python_prefix = "py"
@@ -48,6 +49,7 @@ def yield_lines(lines: Iterator[str], prefix: str, suffix: str):
             if not skip_test:
                 yield line[strip_length:]
 
+
 for file in filter(lambda file: file not in excluded_files, files):
     with open(file, "r") as f:
         lines = list(yield_lines(iter(f), "```", "```"))

docs/test/package.json

@@ -1,13 +0,0 @@
-{
-  "name": "lancedb-docs-test",
-  "version": "1.0.0",
-  "description": "",
-  "author": "",
-  "license": "ISC",
-  "dependencies": {
-    "fs": "^0.0.1-security",
-    "glob": "^10.2.7",
-    "path": "^0.12.7",
-    "vectordb": "https://gitpkg.now.sh/lancedb/lancedb/node?main"
-  }
-}

docs/tsconfig.json

@@ -0,0 +1,17 @@
+{
+  "include": [
+    "src/*.ts",
+  ],
+  "compilerOptions": {
+    "target": "es2022",
+    "module": "nodenext",
+    "declaration": true,
+    "outDir": "./dist",
+    "strict": true,
+    "allowJs": true,
+    "resolveJsonModule": true,
+  },
+  "exclude": [
+    "./dist/*",
+  ]
+}

node/.eslintrc.js

@@ -13,5 +13,10 @@ module.exports = {
   },
   rules: {
     "@typescript-eslint/method-signature-style": "off",
+    "@typescript-eslint/quotes": "off",
+    "@typescript-eslint/semi": "off",
+    "@typescript-eslint/explicit-function-return-type": "off",
+    "@typescript-eslint/space-before-function-paren": "off",
+    "@typescript-eslint/indent": "off",
   }
 }

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.4.3",
+  "version": "0.4.12",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.4.3",
+      "version": "0.4.12",
       "cpu": [
         "x64",
         "arm64"
@@ -18,9 +18,7 @@
         "win32"
       ],
       "dependencies": {
-        "@apache-arrow/ts": "^14.0.2",
         "@neon-rs/load": "^0.0.74",
-        "apache-arrow": "^14.0.2",
         "axios": "^1.4.0"
       },
       "devDependencies": {
@@ -33,6 +31,7 @@
         "@types/temp": "^0.9.1",
         "@types/uuid": "^9.0.3",
         "@typescript-eslint/eslint-plugin": "^5.59.1",
+        "apache-arrow-old": "npm:apache-arrow@13.0.0",
         "cargo-cp-artifact": "^0.1",
         "chai": "^4.3.7",
         "chai-as-promised": "^7.1.1",
@@ -53,11 +52,15 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.4.3",
-        "@lancedb/vectordb-darwin-x64": "0.4.3",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.4.3",
-        "@lancedb/vectordb-linux-x64-gnu": "0.4.3",
-        "@lancedb/vectordb-win32-x64-msvc": "0.4.3"
+        "@lancedb/vectordb-darwin-arm64": "0.4.12",
+        "@lancedb/vectordb-darwin-x64": "0.4.12",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.4.12",
+        "@lancedb/vectordb-linux-x64-gnu": "0.4.12",
+        "@lancedb/vectordb-win32-x64-msvc": "0.4.12"
+      },
+      "peerDependencies": {
+        "@apache-arrow/ts": "^14.0.2",
+        "apache-arrow": "^14.0.2"
       }
     },
     "node_modules/@75lb/deep-merge": {
@@ -93,6 +96,7 @@
       "version": "14.0.2",
       "resolved": "https://registry.npmjs.org/@apache-arrow/ts/-/ts-14.0.2.tgz",
       "integrity": "sha512-CtwAvLkK0CZv7xsYeCo91ml6PvlfzAmAJZkRYuz2GNBwfYufj5SVi0iuSMwIMkcU/szVwvLdzORSLa5PlF/2ug==",
+      "peer": true,
       "dependencies": {
         "@types/command-line-args": "5.2.0",
         "@types/command-line-usage": "5.0.2",
@@ -109,7 +113,8 @@
     "node_modules/@apache-arrow/ts/node_modules/@types/node": {
       "version": "20.3.0",
       "resolved": "https://registry.npmjs.org/@types/node/-/node-20.3.0.tgz",
-      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ=="
+      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
+      "peer": true
     },
     "node_modules/@cargo-messages/android-arm-eabi": {
       "version": "0.0.160",
@@ -329,9 +334,9 @@
       }
     },
     "node_modules/@lancedb/vectordb-darwin-arm64": {
-      "version": "0.4.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.3.tgz",
-      "integrity": "sha512-47CvvSaV1EdUsFEpXUJApTk+hMzAhCxVizipCFUlXCgcmzpCDL86wNgJij/X9a+j6zADhIX//Lsu0qd/an/Bpw==",
+      "version": "0.4.12",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.12.tgz",
+      "integrity": "sha512-38/rkJRlWXkPWXuj9onzvbrhnIWcIUQjgEp5G9v5ixPosBowm7A4j8e2Q8CJMsVSNcVX2JLqwWVldiWegZFuYw==",
       "cpu": [
         "arm64"
       ],
@@ -341,9 +346,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-darwin-x64": {
-      "version": "0.4.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.3.tgz",
-      "integrity": "sha512-UlZZv8CmJIuRJNJG+Y1VmFsGyPR8W/72Q5EwgMMsSES6zpMQ9pNdBDWhL3UGX6nMRgnbprkwYiWJ3xHhJvtqtw==",
+      "version": "0.4.12",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.12.tgz",
+      "integrity": "sha512-psE48dztyO450hXWdv9Rl9aayM2HQ1uF9wErfC0gKmDUh1N0NdVq2viDuFpZxnmCis/nvGwKlYiYT9OnYNCJ9g==",
       "cpu": [
         "x64"
       ],
@@ -353,9 +358,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-arm64-gnu": {
-      "version": "0.4.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.3.tgz",
-      "integrity": "sha512-L6NVJr/lKEd8+904FzZNpT8BGQMs2cHNYbGJMIaVvGnMiIJgKAFKtOyGtdDjoe1xRZoEw21yjRGksGbnRO5wHQ==",
+      "version": "0.4.12",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.12.tgz",
+      "integrity": "sha512-xwkgF6MiF5aAdG9JG8v4ke652YxUJrhs9z4OrsEfrENnvsIQd2C5UyKMepVLdvij4BI/XPFRFWXdjPvP7S9rTA==",
       "cpu": [
         "arm64"
       ],
@@ -365,9 +370,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-x64-gnu": {
-      "version": "0.4.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.3.tgz",
-      "integrity": "sha512-OBx3WF3pK0xNfFJeErmuD9R2QWLa3XdeZspyTsIrQmBDeKj3HKh8y7Scpx4NH5Y09+9JNqRRKRZN7OqWTYhITg==",
+      "version": "0.4.12",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.12.tgz",
+      "integrity": "sha512-gJqYR0aymrS+C60xc4EQPzmQ5/69XfeFv2ofBvAj7qW+c6BcnoAcfVl+7s1IrcWeGz251sm5cD5Lx4AzJd89dA==",
       "cpu": [
         "x64"
       ],
@@ -377,9 +382,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-win32-x64-msvc": {
-      "version": "0.4.3",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.3.tgz",
-      "integrity": "sha512-n9IvR81NXZKnSN91mrgeXbEyCiGM+YLJpOgbdHoEtMP04VDnS+iSU4jGOtQBKErvWeCJQaGFQ9qzdcVchpRGyw==",
+      "version": "0.4.12",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.12.tgz",
+      "integrity": "sha512-LhCzpyEeBUyO6L2fuVqeP3mW8kYDryyU9PNqcM01m88sZB1Do6AlwiM+GjPRQ0SpzD0LK9oxQqSmJrdcNGqjbw==",
       "cpu": [
         "x64"
       ],
@@ -948,6 +953,7 @@
       "version": "14.0.2",
       "resolved": "https://registry.npmjs.org/apache-arrow/-/apache-arrow-14.0.2.tgz",
       "integrity": "sha512-EBO2xJN36/XoY81nhLcwCJgFwkboDZeyNQ+OPsG7bCoQjc2BT0aTyH/MR6SrL+LirSNz+cYqjGRlupMMlP1aEg==",
+      "peer": true,
       "dependencies": {
         "@types/command-line-args": "5.2.0",
         "@types/command-line-usage": "5.0.2",
@@ -964,10 +970,39 @@
         "arrow2csv": "bin/arrow2csv.js"
       }
     },
+    "node_modules/apache-arrow-old": {
+      "name": "apache-arrow",
+      "version": "13.0.0",
+      "resolved": "https://registry.npmjs.org/apache-arrow/-/apache-arrow-13.0.0.tgz",
+      "integrity": "sha512-3gvCX0GDawWz6KFNC28p65U+zGh/LZ6ZNKWNu74N6CQlKzxeoWHpi4CgEQsgRSEMuyrIIXi1Ea2syja7dwcHvw==",
+      "dev": true,
+      "dependencies": {
+        "@types/command-line-args": "5.2.0",
+        "@types/command-line-usage": "5.0.2",
+        "@types/node": "20.3.0",
+        "@types/pad-left": "2.1.1",
+        "command-line-args": "5.2.1",
+        "command-line-usage": "7.0.1",
+        "flatbuffers": "23.5.26",
+        "json-bignum": "^0.0.3",
+        "pad-left": "^2.1.0",
+        "tslib": "^2.5.3"
+      },
+      "bin": {
+        "arrow2csv": "bin/arrow2csv.js"
+      }
+    },
+    "node_modules/apache-arrow-old/node_modules/@types/node": {
+      "version": "20.3.0",
+      "resolved": "https://registry.npmjs.org/@types/node/-/node-20.3.0.tgz",
+      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
+      "dev": true
+    },
     "node_modules/apache-arrow/node_modules/@types/node": {
       "version": "20.3.0",
       "resolved": "https://registry.npmjs.org/@types/node/-/node-20.3.0.tgz",
-      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ=="
+      "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
+      "peer": true
     },
     "node_modules/arg": {
       "version": "4.1.3",

node/package.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.4.3",
+  "version": "0.4.12",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
   "scripts": {
     "tsc": "tsc -b",
-    "build": "cargo-cp-artifact --artifact cdylib vectordb-node index.node -- cargo build --message-format=json && tsc -b",
+    "build": "npm run tsc && cargo-cp-artifact --artifact cdylib lancedb-node index.node -- cargo build --message-format=json",
     "build-release": "npm run build -- --release",
     "test": "npm run tsc && mocha -recursive dist/test",
     "integration-test": "npm run tsc && mocha -recursive dist/integration_test",
@@ -17,7 +17,11 @@
   },
   "repository": {
     "type": "git",
-    "url": "https://github.com/lancedb/lancedb/node"
+    "url": "https://github.com/lancedb/lancedb.git"
+  },
+  "homepage": "https://lancedb.github.io/lancedb/",
+  "bugs": {
+    "url": "https://github.com/lancedb/lancedb/issues"
   },
   "keywords": [
     "data-format",
@@ -37,6 +41,7 @@
     "@types/temp": "^0.9.1",
     "@types/uuid": "^9.0.3",
     "@typescript-eslint/eslint-plugin": "^5.59.1",
+    "apache-arrow-old": "npm:apache-arrow@13.0.0",
     "cargo-cp-artifact": "^0.1",
     "chai": "^4.3.7",
     "chai-as-promised": "^7.1.1",
@@ -57,11 +62,13 @@
     "uuid": "^9.0.0"
   },
   "dependencies": {
-    "@apache-arrow/ts": "^14.0.2",
     "@neon-rs/load": "^0.0.74",
-    "apache-arrow": "^14.0.2",
     "axios": "^1.4.0"
   },
+  "peerDependencies": {
+    "@apache-arrow/ts": "^14.0.2",
+    "apache-arrow": "^14.0.2"
+  },
   "os": [
     "darwin",
     "linux",
@@ -81,10 +88,10 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-arm64": "0.4.3",
-    "@lancedb/vectordb-darwin-x64": "0.4.3",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.4.3",
-    "@lancedb/vectordb-linux-x64-gnu": "0.4.3",
-    "@lancedb/vectordb-win32-x64-msvc": "0.4.3"
+    "@lancedb/vectordb-darwin-arm64": "0.4.12",
+    "@lancedb/vectordb-darwin-x64": "0.4.12",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.4.12",
+    "@lancedb/vectordb-linux-x64-gnu": "0.4.12",
+    "@lancedb/vectordb-win32-x64-msvc": "0.4.12"
   }
 }

node/src/arrow.ts

@@ -14,26 +14,30 @@
 
 import {
   Field,
-  type FixedSizeListBuilder,
-  Float32,
   makeBuilder,
   RecordBatchFileWriter,
   Utf8,
   type Vector,
   FixedSizeList,
   vectorFromArray,
-  type Schema,
+  Schema,
   Table as ArrowTable,
   RecordBatchStreamWriter,
   List,
-  Float64,
   RecordBatch,
   makeData,
   Struct,
-  type Float
+  Float,
+  DataType,
+  Binary,
+  Float32
 } from 'apache-arrow'
 import { type EmbeddingFunction } from './index'
+import { sanitizeSchema } from './sanitize'
 
+/*
+ * Options to control how a column should be converted to a vector array
+ */
 export class VectorColumnOptions {
   /** Vector column type. */
   type: Float = new Float32()
@@ -45,14 +49,50 @@ export class VectorColumnOptions {
 
 /** Options to control the makeArrowTable call. */
 export class MakeArrowTableOptions {
-  /** Provided schema. */
+  /*
+   * Schema of the data.
+   *
+   * If this is not provided then the data type will be inferred from the
+   * JS type.  Integer numbers will become int64, floating point numbers
+   * will become float64 and arrays will become variable sized lists with
+   * the data type inferred from the first element in the array.
+   *
+   * The schema must be specified if there are no records (e.g. to make
+   * an empty table)
+   */
   schema?: Schema
 
-  /** Vector columns */
+  /*
+   * Mapping from vector column name to expected type
+   *
+   * Lance expects vector columns to be fixed size list arrays (i.e. tensors)
+   * However, `makeArrowTable` will not infer this by default (it creates
+   * variable size list arrays).  This field can be used to indicate that a column
+   * should be treated as a vector column and converted to a fixed size list.
+   *
+   * The keys should be the names of the vector columns.  The value specifies the
+   * expected data type of the vector columns.
+   *
+   * If `schema` is provided then this field is ignored.
+   *
+   * By default, the column named "vector" will be assumed to be a float32
+   * vector column.
+   */
   vectorColumns: Record<string, VectorColumnOptions> = {
     vector: new VectorColumnOptions()
   }
 
+  /**
+   * If true then string columns will be encoded with dictionary encoding
+   *
+   * Set this to true if your string columns tend to repeat the same values
+   * often.  For more precise control use the `schema` property to specify the
+   * data type for individual columns.
+   *
+   * If `schema` is provided then this property is ignored.
+   */
+  dictionaryEncodeStrings: boolean = false
+
   constructor (values?: Partial<MakeArrowTableOptions>) {
     Object.assign(this, values)
   }
@@ -62,8 +102,29 @@ export class MakeArrowTableOptions {
  * An enhanced version of the {@link makeTable} function from Apache Arrow
  * that supports nested fields and embeddings columns.
  *
+ * This function converts an array of Record<String, any> (row-major JS objects)
+ * to an Arrow Table (a columnar structure)
+ *
  * Note that it currently does not support nulls.
  *
+ * If a schema is provided then it will be used to determine the resulting array
+ * types.  Fields will also be reordered to fit the order defined by the schema.
+ *
+ * If a schema is not provided then the types will be inferred and the field order
+ * will be controlled by the order of properties in the first record.
+ *
+ * If the input is empty then a schema must be provided to create an empty table.
+ *
+ * When a schema is not specified then data types will be inferred.  The inference
+ * rules are as follows:
+ *
+ *  - boolean => Bool
+ *  - number => Float64
+ *  - String => Utf8
+ *  - Buffer => Binary
+ *  - Record<String, any> => Struct
+ *  - Array<any> => List
+ *
  * @param data input data
  * @param options options to control the makeArrowTable call.
  *
@@ -86,8 +147,10 @@ export class MakeArrowTableOptions {
  *  ], { schema });
  * ```
  *
- * It guesses the vector columns if the schema is not provided. For example,
- * by default it assumes that the column named `vector` is a vector column.
+ * By default it assumes that the column named `vector` is a vector column
+ * and it will be converted into a fixed size list array of type float32.
+ * The `vectorColumns` option can be used to support other vector column
+ * names and data types.
  *
  * ```ts
  *
@@ -134,211 +197,322 @@ export function makeArrowTable (
   data: Array<Record<string, any>>,
   options?: Partial<MakeArrowTableOptions>
 ): ArrowTable {
-  if (data.length === 0) {
-    throw new Error('At least one record needs to be provided')
+  if (data.length === 0 && (options?.schema === undefined || options?.schema === null)) {
+    throw new Error('At least one record or a schema needs to be provided')
   }
+
   const opt = new MakeArrowTableOptions(options !== undefined ? options : {})
+  if (opt.schema !== undefined && opt.schema !== null) {
+    opt.schema = sanitizeSchema(opt.schema)
+  }
   const columns: Record<string, Vector> = {}
   // TODO: sample dataset to find missing columns
-  const columnNames = Object.keys(data[0])
+  // Prefer the field ordering of the schema, if present
+  const columnNames = ((opt.schema) != null) ? (opt.schema.names as string[]) : Object.keys(data[0])
   for (const colName of columnNames) {
-    const values = data.map((datum) => datum[colName])
-    let vector: Vector
+    if (data.length !== 0 && !Object.prototype.hasOwnProperty.call(data[0], colName)) {
+      // The field is present in the schema, but not in the data, skip it
+      continue
+    }
+    // Extract a single column from the records (transpose from row-major to col-major)
+    let values = data.map((datum) => datum[colName])
 
+    // By default (type === undefined) arrow will infer the type from the JS type
+    let type
     if (opt.schema !== undefined) {
-      // Explicit schema is provided, highest priority
-      vector = vectorFromArray(
-        values,
-        opt.schema?.fields.filter((f) => f.name === colName)[0]?.type
-      )
+      // If there is a schema provided, then use that for the type instead
+      type = opt.schema?.fields.filter((f) => f.name === colName)[0]?.type
+      if (DataType.isInt(type) && type.bitWidth === 64) {
+        // wrap in BigInt to avoid bug: https://github.com/apache/arrow/issues/40051
+        values = values.map((v) => {
+          if (v === null) {
+            return v
+          }
+          return BigInt(v)
+        })
+      }
     } else {
+      // Otherwise, check to see if this column is one of the vector columns
+      // defined by opt.vectorColumns and, if so, use the fixed size list type
       const vectorColumnOptions = opt.vectorColumns[colName]
       if (vectorColumnOptions !== undefined) {
-        const fslType = new FixedSizeList(
-          values[0].length,
-          new Field('item', vectorColumnOptions.type, false)
-        )
-        vector = vectorFromArray(values, fslType)
-      } else {
-        // Normal case
-        vector = vectorFromArray(values)
+        type = newVectorType(values[0].length, vectorColumnOptions.type)
       }
     }
-    columns[colName] = vector
+
+    try {
+      // Convert an Array of JS values to an arrow vector
+      columns[colName] = makeVector(values, type, opt.dictionaryEncodeStrings)
+    } catch (error: unknown) {
+      // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+      throw Error(`Could not convert column "${colName}" to Arrow: ${error}`)
+    }
   }
 
-  return new ArrowTable(columns)
+  if (opt.schema != null) {
+    // `new ArrowTable(columns)` infers a schema which may sometimes have
+    // incorrect nullability (it assumes nullable=true if there are 0 rows)
+    //
+    // `new ArrowTable(schema, columns)` will also fail because it will create a
+    // batch with an inferred schema and then complain that the batch schema
+    // does not match the provided schema.
+    //
+    // To work around this we first create a table with the wrong schema and
+    // then patch the schema of the batches so we can use
+    // `new ArrowTable(schema, batches)` which does not do any schema inference
+    const firstTable = new ArrowTable(columns)
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const batchesFixed = firstTable.batches.map(batch => new RecordBatch(opt.schema!, batch.data))
+    return new ArrowTable(opt.schema, batchesFixed)
+  } else {
+    return new ArrowTable(columns)
+  }
 }
 
-// Converts an Array of records into an Arrow Table, optionally applying an embeddings function to it.
+/**
+ * Create an empty Arrow table with the provided schema
+ */
+export function makeEmptyTable (schema: Schema): ArrowTable {
+  return makeArrowTable([], { schema })
+}
+
+// Helper function to convert Array<Array<any>> to a variable sized list array
+function makeListVector (lists: any[][]): Vector<any> {
+  if (lists.length === 0 || lists[0].length === 0) {
+    throw Error('Cannot infer list vector from empty array or empty list')
+  }
+  const sampleList = lists[0]
+  let inferredType
+  try {
+    const sampleVector = makeVector(sampleList)
+    inferredType = sampleVector.type
+  } catch (error: unknown) {
+    // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+    throw Error(`Cannot infer list vector.  Cannot infer inner type: ${error}`)
+  }
+
+  const listBuilder = makeBuilder({
+    type: new List(new Field('item', inferredType, true))
+  })
+  for (const list of lists) {
+    listBuilder.append(list)
+  }
+  return listBuilder.finish().toVector()
+}
+
+// Helper function to convert an Array of JS values to an Arrow Vector
+function makeVector (values: any[], type?: DataType, stringAsDictionary?: boolean): Vector<any> {
+  if (type !== undefined) {
+    // No need for inference, let Arrow create it
+    return vectorFromArray(values, type)
+  }
+  if (values.length === 0) {
+    throw Error('makeVector requires at least one value or the type must be specfied')
+  }
+  const sampleValue = values.find(val => val !== null && val !== undefined)
+  if (sampleValue === undefined) {
+    throw Error('makeVector cannot infer the type if all values are null or undefined')
+  }
+  if (Array.isArray(sampleValue)) {
+    // Default Arrow inference doesn't handle list types
+    return makeListVector(values)
+  } else if (Buffer.isBuffer(sampleValue)) {
+    // Default Arrow inference doesn't handle Buffer
+    return vectorFromArray(values, new Binary())
+  } else if (!(stringAsDictionary ?? false) && (typeof sampleValue === 'string' || sampleValue instanceof String)) {
+    // If the type is string then don't use Arrow's default inference unless dictionaries are requested
+    // because it will always use dictionary encoding for strings
+    return vectorFromArray(values, new Utf8())
+  } else {
+    // Convert a JS array of values to an arrow vector
+    return vectorFromArray(values)
+  }
+}
+
+async function applyEmbeddings<T> (table: ArrowTable, embeddings?: EmbeddingFunction<T>, schema?: Schema): Promise<ArrowTable> {
+  if (embeddings == null) {
+    return table
+  }
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema)
+  }
+
+  // Convert from ArrowTable to Record<String, Vector>
+  const colEntries = [...Array(table.numCols).keys()].map((_, idx) => {
+    const name = table.schema.fields[idx].name
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const vec = table.getChildAt(idx)!
+    return [name, vec]
+  })
+  const newColumns = Object.fromEntries(colEntries)
+
+  const sourceColumn = newColumns[embeddings.sourceColumn]
+  const destColumn = embeddings.destColumn ?? 'vector'
+  const innerDestType = embeddings.embeddingDataType ?? new Float32()
+  if (sourceColumn === undefined) {
+    throw new Error(`Cannot apply embedding function because the source column '${embeddings.sourceColumn}' was not present in the data`)
+  }
+
+  if (table.numRows === 0) {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      // We have an empty table and it already has the embedding column so no work needs to be done
+      // Note: we don't return an error like we did below because this is a common occurrence.  For example,
+      // if we call convertToTable with 0 records and a schema that includes the embedding
+      return table
+    }
+    if (embeddings.embeddingDimension !== undefined) {
+      const destType = newVectorType(embeddings.embeddingDimension, innerDestType)
+      newColumns[destColumn] = makeVector([], destType)
+    } else if (schema != null) {
+      const destField = schema.fields.find(f => f.name === destColumn)
+      if (destField != null) {
+        newColumns[destColumn] = makeVector([], destField.type)
+      } else {
+        throw new Error(`Attempt to apply embeddings to an empty table failed because schema was missing embedding column '${destColumn}'`)
+      }
+    } else {
+      throw new Error('Attempt to apply embeddings to an empty table when the embeddings function does not specify `embeddingDimension`')
+    }
+  } else {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      throw new Error(`Attempt to apply embeddings to table failed because column ${destColumn} already existed`)
+    }
+    if (table.batches.length > 1) {
+      throw new Error('Internal error: `makeArrowTable` unexpectedly created a table with more than one batch')
+    }
+    const values = sourceColumn.toArray()
+    const vectors = await embeddings.embed(values as T[])
+    if (vectors.length !== values.length) {
+      throw new Error('Embedding function did not return an embedding for each input element')
+    }
+    const destType = newVectorType(vectors[0].length, innerDestType)
+    newColumns[destColumn] = makeVector(vectors, destType)
+  }
+
+  const newTable = new ArrowTable(newColumns)
+  if (schema != null) {
+    if (schema.fields.find(f => f.name === destColumn) === undefined) {
+      throw new Error(`When using embedding functions and specifying a schema the schema should include the embedding column but the column ${destColumn} was missing`)
+    }
+    return alignTable(newTable, schema)
+  }
+  return newTable
+}
+
+/*
+ * Convert an Array of records into an Arrow Table, optionally applying an
+ * embeddings function to it.
+ *
+ * This function calls `makeArrowTable` first to create the Arrow Table.
+ * Any provided `makeTableOptions` (e.g. a schema) will be passed on to
+ * that call.
+ *
+ * The embedding function will be passed a column of values (based on the
+ * `sourceColumn` of the embedding function) and expects to receive back
+ * number[][] which will be converted into a fixed size list column.  By
+ * default this will be a fixed size list of Float32 but that can be
+ * customized by the `embeddingDataType` property of the embedding function.
+ *
+ * If a schema is provided in `makeTableOptions` then it should include the
+ * embedding columns.  If no schema is provded then embedding columns will
+ * be placed at the end of the table, after all of the input columns.
+ */
 export async function convertToTable<T> (
   data: Array<Record<string, unknown>>,
-  embeddings?: EmbeddingFunction<T>
+  embeddings?: EmbeddingFunction<T>,
+  makeTableOptions?: Partial<MakeArrowTableOptions>
 ): Promise<ArrowTable> {
-  if (data.length === 0) {
-    throw new Error('At least one record needs to be provided')
-  }
-
-  const columns = Object.keys(data[0])
-  const records: Record<string, Vector> = {}
-
-  for (const columnsKey of columns) {
-    if (columnsKey === 'vector') {
-      const vectorSize = (data[0].vector as any[]).length
-      const listBuilder = newVectorBuilder(vectorSize)
-      for (const datum of data) {
-        if ((datum[columnsKey] as any[]).length !== vectorSize) {
-          throw new Error(`Invalid vector size, expected ${vectorSize}`)
-        }
-
-        listBuilder.append(datum[columnsKey])
-      }
-      records[columnsKey] = listBuilder.finish().toVector()
-    } else {
-      const values = []
-      for (const datum of data) {
-        values.push(datum[columnsKey])
-      }
-
-      if (columnsKey === embeddings?.sourceColumn) {
-        const vectors = await embeddings.embed(values as T[])
-        records.vector = vectorFromArray(
-          vectors,
-          newVectorType(vectors[0].length)
-        )
-      }
-
-      if (typeof values[0] === 'string') {
-        // `vectorFromArray` converts strings into dictionary vectors, forcing it back to a string column
-        records[columnsKey] = vectorFromArray(values, new Utf8())
-      } else if (Array.isArray(values[0])) {
-        const elementType = getElementType(values[0])
-        let innerType
-        if (elementType === 'string') {
-          innerType = new Utf8()
-        } else if (elementType === 'number') {
-          innerType = new Float64()
-        } else {
-          // TODO: pass in schema if it exists, else keep going to the next element
-          throw new Error(`Unsupported array element type ${elementType}`)
-        }
-        const listBuilder = makeBuilder({
-          type: new List(new Field('item', innerType, true))
-        })
-        for (const value of values) {
-          listBuilder.append(value)
-        }
-        records[columnsKey] = listBuilder.finish().toVector()
-      } else {
-        // TODO if this is a struct field then recursively align the subfields
-        records[columnsKey] = vectorFromArray(values)
-      }
-    }
-  }
-
-  return new ArrowTable(records)
-}
-
-function getElementType (arr: any[]): string {
-  if (arr.length === 0) {
-    return 'undefined'
-  }
-
-  return typeof arr[0]
-}
-
-// Creates a new Arrow ListBuilder that stores a Vector column
-function newVectorBuilder (dim: number): FixedSizeListBuilder<Float32> {
-  return makeBuilder({
-    type: newVectorType(dim)
-  })
+  const table = makeArrowTable(data, makeTableOptions)
+  return await applyEmbeddings(table, embeddings, makeTableOptions?.schema)
 }
 
 // Creates the Arrow Type for a Vector column with dimension `dim`
-function newVectorType (dim: number): FixedSizeList<Float32> {
+function newVectorType <T extends Float> (dim: number, innerType: T): FixedSizeList<T> {
   // Somewhere we always default to have the elements nullable, so we need to set it to true
   // otherwise we often get schema mismatches because the stored data always has schema with nullable elements
-  const children = new Field<Float32>('item', new Float32(), true)
+  const children = new Field<T>('item', innerType, true)
   return new FixedSizeList(dim, children)
 }
 
-// Converts an Array of records into Arrow IPC format
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC File serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
 export async function fromRecordsToBuffer<T> (
   data: Array<Record<string, unknown>>,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  let table = await convertToTable(data, embeddings)
-  if (schema !== undefined) {
-    table = alignTable(table, schema)
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema)
   }
+  const table = await convertToTable(data, embeddings, { schema })
   const writer = RecordBatchFileWriter.writeAll(table)
   return Buffer.from(await writer.toUint8Array())
 }
 
-// Converts an Array of records into Arrow IPC stream format
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
 export async function fromRecordsToStreamBuffer<T> (
   data: Array<Record<string, unknown>>,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  let table = await convertToTable(data, embeddings)
-  if (schema !== undefined) {
-    table = alignTable(table, schema)
+  if (schema !== null && schema !== undefined) {
+    schema = sanitizeSchema(schema)
   }
+  const table = await convertToTable(data, embeddings, { schema })
   const writer = RecordBatchStreamWriter.writeAll(table)
   return Buffer.from(await writer.toUint8Array())
 }
 
-// Converts an Arrow Table into Arrow IPC format
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC File serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
 export async function fromTableToBuffer<T> (
   table: ArrowTable,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  if (embeddings !== undefined) {
-    const source = table.getChild(embeddings.sourceColumn)
-
-    if (source === null) {
-      throw new Error(
-        `The embedding source column ${embeddings.sourceColumn} was not found in the Arrow Table`
-      )
-    }
-
-    const vectors = await embeddings.embed(source.toArray() as T[])
-    const column = vectorFromArray(vectors, newVectorType(vectors[0].length))
-    table = table.assign(new ArrowTable({ vector: column }))
+  if (schema !== null && schema !== undefined) {
+    schema = sanitizeSchema(schema)
   }
-  if (schema !== undefined) {
-    table = alignTable(table, schema)
-  }
-  const writer = RecordBatchFileWriter.writeAll(table)
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema)
+  const writer = RecordBatchFileWriter.writeAll(tableWithEmbeddings)
   return Buffer.from(await writer.toUint8Array())
 }
 
-// Converts an Arrow Table into Arrow IPC stream format
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
 export async function fromTableToStreamBuffer<T> (
   table: ArrowTable,
   embeddings?: EmbeddingFunction<T>,
   schema?: Schema
 ): Promise<Buffer> {
-  if (embeddings !== undefined) {
-    const source = table.getChild(embeddings.sourceColumn)
-
-    if (source === null) {
-      throw new Error(
-        `The embedding source column ${embeddings.sourceColumn} was not found in the Arrow Table`
-      )
-    }
-
-    const vectors = await embeddings.embed(source.toArray() as T[])
-    const column = vectorFromArray(vectors, newVectorType(vectors[0].length))
-    table = table.assign(new ArrowTable({ vector: column }))
+  if (schema !== null && schema !== undefined) {
+    schema = sanitizeSchema(schema)
   }
-  if (schema !== undefined) {
-    table = alignTable(table, schema)
-  }
-  const writer = RecordBatchStreamWriter.writeAll(table)
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema)
+  const writer = RecordBatchStreamWriter.writeAll(tableWithEmbeddings)
   return Buffer.from(await writer.toUint8Array())
 }
 
@@ -373,5 +547,5 @@ function alignTable (table: ArrowTable, schema: Schema): ArrowTable {
 
 // Creates an empty Arrow Table
 export function createEmptyTable (schema: Schema): ArrowTable {
-  return new ArrowTable(schema)
+  return new ArrowTable(sanitizeSchema(schema))
 }

node/src/embedding/embedding_function.ts

@@ -12,18 +12,53 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+import { type Float } from 'apache-arrow'
+
 /**
  * An embedding function that automatically creates vector representation for a given column.
  */
 export interface EmbeddingFunction<T> {
   /**
-     * The name of the column that will be used as input for the Embedding Function.
-     */
+   * The name of the column that will be used as input for the Embedding Function.
+   */
   sourceColumn: string
 
   /**
-     * Creates a vector representation for the given values.
-     */
+   * The data type of the embedding
+   *
+   * The embedding function should return `number`.  This will be converted into
+   * an Arrow float array.  By default this will be Float32 but this property can
+   * be used to control the conversion.
+   */
+  embeddingDataType?: Float
+
+  /**
+   * The dimension of the embedding
+   *
+   * This is optional, normally this can be determined by looking at the results of
+   * `embed`.  If this is not specified, and there is an attempt to apply the embedding
+   * to an empty table, then that process will fail.
+   */
+  embeddingDimension?: number
+
+  /**
+   * The name of the column that will contain the embedding
+   *
+   * By default this is "vector"
+   */
+  destColumn?: string
+
+  /**
+   * Should the source column be excluded from the resulting table
+   *
+   * By default the source column is included.  Set this to true and
+   * only the embedding will be stored.
+   */
+  excludeSource?: boolean
+
+  /**
+   * Creates a vector representation for the given values.
+   */
   embed: (data: T[]) => Promise<number[][]>
 }
 

node/src/index.ts

@@ -37,18 +37,22 @@ const {
   tableCountRows,
   tableDelete,
   tableUpdate,
+  tableMergeInsert,
   tableCleanupOldVersions,
   tableCompactFiles,
   tableListIndices,
   tableIndexStats,
-  tableSchema
+  tableSchema,
+  tableAddColumns,
+  tableAlterColumns,
+  tableDropColumns
   // eslint-disable-next-line @typescript-eslint/no-var-requires
 } = require('../native.js')
 
 export { Query }
 export type { EmbeddingFunction }
 export { OpenAIEmbeddingFunction } from './embedding/openai'
-export { makeArrowTable, type MakeArrowTableOptions } from './arrow'
+export { convertToTable, makeArrowTable, type MakeArrowTableOptions } from './arrow'
 
 const defaultAwsRegion = 'us-west-2'
 
@@ -95,6 +99,19 @@ export interface ConnectionOptions {
    * This is useful for local testing.
    */
   hostOverride?: string
+
+  /**
+   * (For LanceDB OSS only): The interval, in seconds, at which to check for
+   * updates to the table from other processes. If None, then consistency is not
+   * checked. For performance reasons, this is the default. For strong
+   * consistency, set this to zero seconds. Then every read will check for
+   * updates from other processes. As a compromise, you can set this to a
+   * non-zero value for eventual consistency. If more than that interval
+   * has passed since the last check, then the table will be checked for updates.
+   * Note: this consistency only applies to read operations. Write operations are
+   * always consistent.
+   */
+  readConsistencyInterval?: number
 }
 
 function getAwsArgs (opts: ConnectionOptions): any[] {
@@ -163,6 +180,7 @@ export async function connect (
       {
         uri: '',
         awsCredentials: undefined,
+        awsRegion: defaultAwsRegion,
         apiKey: undefined,
         region: defaultAwsRegion
       },
@@ -174,7 +192,14 @@ export async function connect (
     // Remote connection
     return new RemoteConnection(opts)
   }
-  const db = await databaseNew(opts.uri)
+  const db = await databaseNew(
+    opts.uri,
+    opts.awsCredentials?.accessKeyId,
+    opts.awsCredentials?.secretKey,
+    opts.awsCredentials?.sessionToken,
+    opts.awsRegion,
+    opts.readConsistencyInterval
+  )
   return new LocalConnection(db, opts)
 }
 
@@ -316,6 +341,7 @@ export interface Table<T = number[]> {
    *
    * @param column The column to index
    * @param replace If false, fail if an index already exists on the column
+   * it is always set to true for remote connections
    *
    * Scalar indices, like vector indices, can be used to speed up scans.  A scalar
    * index can speed up scans that contain filter expressions on the indexed column.
@@ -359,12 +385,12 @@ export interface Table<T = number[]> {
    * await table.createScalarIndex('my_col')
    * ```
    */
-  createScalarIndex: (column: string, replace: boolean) => Promise<void>
+  createScalarIndex: (column: string, replace?: boolean) => Promise<void>
 
   /**
    * Returns the number of rows in this table.
    */
-  countRows: () => Promise<number>
+  countRows: (filter?: string) => Promise<number>
 
   /**
    * Delete rows from this table.
@@ -433,6 +459,38 @@ export interface Table<T = number[]> {
    */
   update: (args: UpdateArgs | UpdateSqlArgs) => Promise<void>
 
+  /**
+   * Runs a "merge insert" operation on the table
+   *
+   * This operation can add rows, update rows, and remove rows all in a single
+   * transaction. It is a very generic tool that can be used to create
+   * behaviors like "insert if not exists", "update or insert (i.e. upsert)",
+   * or even replace a portion of existing data with new data (e.g. replace
+   * all data where month="january")
+   *
+   * The merge insert operation works by combining new data from a
+   * **source table** with existing data in a **target table** by using a
+   * join.  There are three categories of records.
+   *
+   * "Matched" records are records that exist in both the source table and
+   * the target table. "Not matched" records exist only in the source table
+   * (e.g. these are new data) "Not matched by source" records exist only
+   * in the target table (this is old data)
+   *
+   * The MergeInsertArgs can be used to customize what should happen for
+   * each category of data.
+   *
+   * Please note that the data may appear to be reordered as part of this
+   * operation.  This is because updated rows will be deleted from the
+   * dataset and then reinserted at the end with the new values.
+   *
+   * @param on a column to join on.  This is how records from the source
+   *           table and target table are matched.
+   * @param data the new data to insert
+   * @param args parameters controlling how the operation should behave
+   */
+  mergeInsert: (on: string, data: Array<Record<string, unknown>> | ArrowTable, args: MergeInsertArgs) => Promise<void>
+
   /**
    * List the indicies on this table.
    */
@@ -443,7 +501,62 @@ export interface Table<T = number[]> {
    */
   indexStats: (indexUuid: string) => Promise<IndexStats>
 
+  filter(value: string): Query<T>
+
   schema: Promise<Schema>
+
+  // TODO: Support BatchUDF
+  /**
+   * Add new columns with defined values.
+   *
+   * @param newColumnTransforms pairs of column names and the SQL expression to use
+   *                            to calculate the value of the new column. These
+   *                            expressions will be evaluated for each row in the
+   *                            table, and can reference existing columns in the table.
+   */
+  addColumns(newColumnTransforms: Array<{ name: string, valueSql: string }>): Promise<void>
+
+  /**
+   * Alter the name or nullability of columns.
+   *
+   * @param columnAlterations One or more alterations to apply to columns.
+   */
+  alterColumns(columnAlterations: ColumnAlteration[]): Promise<void>
+
+  /**
+   * Drop one or more columns from the dataset
+   *
+   * This is a metadata-only operation and does not remove the data from the
+   * underlying storage. In order to remove the data, you must subsequently
+   * call ``compact_files`` to rewrite the data without the removed columns and
+   * then call ``cleanup_files`` to remove the old files.
+   *
+   * @param columnNames The names of the columns to drop. These can be nested
+   *                    column references (e.g. "a.b.c") or top-level column
+   *                    names (e.g. "a").
+   */
+  dropColumns(columnNames: string[]): Promise<void>
+}
+
+/**
+ * A definition of a column alteration. The alteration changes the column at
+ * `path` to have the new name `name`, to be nullable if `nullable` is true,
+ * and to have the data type `data_type`. At least one of `rename` or `nullable`
+ * must be provided.
+ */
+export interface ColumnAlteration {
+  /**
+   * The path to the column to alter. This is a dot-separated path to the column.
+   * If it is a top-level column then it is just the name of the column. If it is
+   * a nested column then it is the path to the column, e.g. "a.b.c" for a column
+   * `c` nested inside a column `b` nested inside a column `a`.
+   */
+  path: string
+  rename?: string
+  /**
+   * Set the new nullability. Note that a nullable column cannot be made non-nullable.
+   */
+  nullable?: boolean
 }
 
 export interface UpdateArgs {
@@ -474,6 +587,47 @@ export interface UpdateSqlArgs {
   valuesSql: Record<string, string>
 }
 
+export interface MergeInsertArgs {
+  /**
+   * If true then rows that exist in both the source table (new data) and
+   * the target table (old data) will be updated, replacing the old row
+   * with the corresponding matching row.
+   *
+   * If there are multiple matches then the behavior is undefined.
+   * Currently this causes multiple copies of the row to be created
+   * but that behavior is subject to change.
+   *
+   * Optionally, a filter can be specified.  This should be an SQL
+   * filter where fields with the prefix "target." refer to fields
+   * in the target table (old data) and fields with the prefix
+   * "source." refer to fields in the source table (new data).  For
+   * example, the filter "target.lastUpdated < source.lastUpdated" will
+   * only update matched rows when the incoming `lastUpdated` value is
+   * newer.
+   *
+   * Rows that do not match the filter will not be updated.  Rows that
+   * do not match the filter do become "not matched" rows.
+   */
+  whenMatchedUpdateAll?: string | boolean
+  /**
+   * If true then rows that exist only in the source table (new data)
+   * will be inserted into the target table.
+   */
+  whenNotMatchedInsertAll?: boolean
+  /**
+   * If true then rows that exist only in the target table (old data)
+   * will be deleted.
+   *
+   * If this is a string then it will be treated as an SQL filter and
+   * only rows that both do not match any row in the source table and
+   * match the given filter will be deleted.
+   *
+   * This can be used to replace a selection of existing data with
+   * new data.
+   */
+  whenNotMatchedBySourceDelete?: string | boolean
+}
+
 export interface VectorIndex {
   columns: string[]
   name: string
@@ -761,15 +915,18 @@ export class LocalTable<T = number[]> implements Table<T> {
       })
   }
 
-  async createScalarIndex (column: string, replace: boolean): Promise<void> {
+  async createScalarIndex (column: string, replace?: boolean): Promise<void> {
+    if (replace === undefined) {
+      replace = true
+    }
     return tableCreateScalarIndex.call(this._tbl, column, replace)
   }
 
   /**
    * Returns the number of rows in this table.
    */
-  async countRows (): Promise<number> {
-    return tableCountRows.call(this._tbl)
+  async countRows (filter?: string): Promise<number> {
+    return tableCountRows.call(this._tbl, filter)
   }
 
   /**
@@ -812,6 +969,46 @@ export class LocalTable<T = number[]> implements Table<T> {
       })
   }
 
+  async mergeInsert (on: string, data: Array<Record<string, unknown>> | ArrowTable, args: MergeInsertArgs): Promise<void> {
+    let whenMatchedUpdateAll = false
+    let whenMatchedUpdateAllFilt = null
+    if (args.whenMatchedUpdateAll !== undefined && args.whenMatchedUpdateAll !== null) {
+      whenMatchedUpdateAll = true
+      if (args.whenMatchedUpdateAll !== true) {
+        whenMatchedUpdateAllFilt = args.whenMatchedUpdateAll
+      }
+    }
+    const whenNotMatchedInsertAll = args.whenNotMatchedInsertAll ?? false
+    let whenNotMatchedBySourceDelete = false
+    let whenNotMatchedBySourceDeleteFilt = null
+    if (args.whenNotMatchedBySourceDelete !== undefined && args.whenNotMatchedBySourceDelete !== null) {
+      whenNotMatchedBySourceDelete = true
+      if (args.whenNotMatchedBySourceDelete !== true) {
+        whenNotMatchedBySourceDeleteFilt = args.whenNotMatchedBySourceDelete
+      }
+    }
+
+    const schema = await this.schema
+    let tbl: ArrowTable
+    if (data instanceof ArrowTable) {
+      tbl = data
+    } else {
+      tbl = makeArrowTable(data, { schema })
+    }
+    const buffer = await fromTableToBuffer(tbl, this._embeddings, schema)
+
+    this._tbl = await tableMergeInsert.call(
+      this._tbl,
+      on,
+      whenMatchedUpdateAll,
+      whenMatchedUpdateAllFilt,
+      whenNotMatchedInsertAll,
+      whenNotMatchedBySourceDelete,
+      whenNotMatchedBySourceDeleteFilt,
+      buffer
+    )
+  }
+
   /**
    * Clean up old versions of the table, freeing disk space.
    *
@@ -891,6 +1088,18 @@ export class LocalTable<T = number[]> implements Table<T> {
       return false
     }
   }
+
+  async addColumns (newColumnTransforms: Array<{ name: string, valueSql: string }>): Promise<void> {
+    return tableAddColumns.call(this._tbl, newColumnTransforms)
+  }
+
+  async alterColumns (columnAlterations: ColumnAlteration[]): Promise<void> {
+    return tableAlterColumns.call(this._tbl, columnAlterations)
+  }
+
+  async dropColumns (columnNames: string[]): Promise<void> {
+    return tableDropColumns.call(this._tbl, columnNames)
+  }
 }
 
 export interface CleanupStats {

node/src/remote/index.ts

@@ -24,7 +24,9 @@ import {
   type IndexStats,
   type UpdateArgs,
   type UpdateSqlArgs,
-  makeArrowTable
+  makeArrowTable,
+  type MergeInsertArgs,
+  type ColumnAlteration
 } from '../index'
 import { Query } from '../query'
 
@@ -270,6 +272,59 @@ export class RemoteTable<T = number[]> implements Table<T> {
     return new RemoteQuery(query, this._client, this._name) //, this._embeddings_new)
   }
 
+  filter (where: string): Query<T> {
+    throw new Error('Not implemented')
+  }
+
+  async mergeInsert (on: string, data: Array<Record<string, unknown>> | ArrowTable, args: MergeInsertArgs): Promise<void> {
+    let tbl: ArrowTable
+    if (data instanceof ArrowTable) {
+      tbl = data
+    } else {
+      tbl = makeArrowTable(data, await this.schema)
+    }
+
+    const queryParams: any = {
+      on
+    }
+    if (args.whenMatchedUpdateAll !== false && args.whenMatchedUpdateAll !== null && args.whenMatchedUpdateAll !== undefined) {
+      queryParams.when_matched_update_all = 'true'
+      if (typeof args.whenMatchedUpdateAll === 'string') {
+        queryParams.when_matched_update_all_filt = args.whenMatchedUpdateAll
+      }
+    } else {
+      queryParams.when_matched_update_all = 'false'
+    }
+    if (args.whenNotMatchedInsertAll ?? false) {
+      queryParams.when_not_matched_insert_all = 'true'
+    } else {
+      queryParams.when_not_matched_insert_all = 'false'
+    }
+    if (args.whenNotMatchedBySourceDelete !== false && args.whenNotMatchedBySourceDelete !== null && args.whenNotMatchedBySourceDelete !== undefined) {
+      queryParams.when_not_matched_by_source_delete = 'true'
+      if (typeof args.whenNotMatchedBySourceDelete === 'string') {
+        queryParams.when_not_matched_by_source_delete_filt = args.whenNotMatchedBySourceDelete
+      }
+    } else {
+      queryParams.when_not_matched_by_source_delete = 'false'
+    }
+
+    const buffer = await fromTableToStreamBuffer(tbl, this._embeddings)
+    const res = await this._client.post(
+      `/v1/table/${this._name}/merge_insert/`,
+      buffer,
+      queryParams,
+      'application/vnd.apache.arrow.stream'
+    )
+    if (res.status !== 200) {
+      throw new Error(
+        `Server Error, status: ${res.status}, ` +
+          // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+          `message: ${res.statusText}: ${res.data}`
+      )
+    }
+  }
+
   async add (data: Array<Record<string, unknown>> | ArrowTable): Promise<number> {
     let tbl: ArrowTable
     if (data instanceof ArrowTable) {
@@ -342,7 +397,7 @@ export class RemoteTable<T = number[]> implements Table<T> {
     }
 
     const column = indexParams.column ?? 'vector'
-    const indexType = 'vector' // only vector index is supported for remote connections
+    const indexType = 'vector'
     const metricType = indexParams.metric_type ?? 'L2'
     const indexCacheSize = indexParams.index_cache_size ?? null
 
@@ -365,8 +420,25 @@ export class RemoteTable<T = number[]> implements Table<T> {
     }
   }
 
-  async createScalarIndex (column: string, replace: boolean): Promise<void> {
-    throw new Error('Not implemented')
+  async createScalarIndex (column: string): Promise<void> {
+    const indexType = 'scalar'
+
+    const data = {
+      column,
+      index_type: indexType,
+      replace: true
+    }
+    const res = await this._client.post(
+      `/v1/table/${this._name}/create_scalar_index/`,
+      data
+    )
+    if (res.status !== 200) {
+      throw new Error(
+        `Server Error, status: ${res.status}, ` +
+          // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+          `message: ${res.statusText}: ${res.data}`
+      )
+    }
   }
 
   async countRows (): Promise<number> {
@@ -420,4 +492,16 @@ export class RemoteTable<T = number[]> implements Table<T> {
       numUnindexedRows: results.data.num_unindexed_rows
     }
   }
+
+  async addColumns (newColumnTransforms: Array<{ name: string, valueSql: string }>): Promise<void> {
+    throw new Error('Add columns is not yet supported in LanceDB Cloud.')
+  }
+
+  async alterColumns (columnAlterations: ColumnAlteration[]): Promise<void> {
+    throw new Error('Alter columns is not yet supported in LanceDB Cloud.')
+  }
+
+  async dropColumns (columnNames: string[]): Promise<void> {
+    throw new Error('Drop columns is not yet supported in LanceDB Cloud.')
+  }
 }

node/src/sanitize.ts

@@ -0,0 +1,501 @@
+// Copyright 2023 LanceDB Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// The utilities in this file help sanitize data from the user's arrow
+// library into the types expected by vectordb's arrow library.  Node
+// generally allows for mulitple versions of the same library (and sometimes
+// even multiple copies of the same version) to be installed at the same
+// time.  However, arrow-js uses instanceof which expected that the input
+// comes from the exact same library instance.  This is not always the case
+// and so we must sanitize the input to ensure that it is compatible.
+
+import {
+  Field,
+  Utf8,
+  FixedSizeBinary,
+  FixedSizeList,
+  Schema,
+  List,
+  Struct,
+  Float,
+  Bool,
+  Date_,
+  Decimal,
+  DataType,
+  Dictionary,
+  Binary,
+  Float32,
+  Interval,
+  Map_,
+  Duration,
+  Union,
+  Time,
+  Timestamp,
+  Type,
+  Null,
+  Int,
+  type Precision,
+  type DateUnit,
+  Int8,
+  Int16,
+  Int32,
+  Int64,
+  Uint8,
+  Uint16,
+  Uint32,
+  Uint64,
+  Float16,
+  Float64,
+  DateDay,
+  DateMillisecond,
+  DenseUnion,
+  SparseUnion,
+  TimeNanosecond,
+  TimeMicrosecond,
+  TimeMillisecond,
+  TimeSecond,
+  TimestampNanosecond,
+  TimestampMicrosecond,
+  TimestampMillisecond,
+  TimestampSecond,
+  IntervalDayTime,
+  IntervalYearMonth,
+  DurationNanosecond,
+  DurationMicrosecond,
+  DurationMillisecond,
+  DurationSecond,
+} from "apache-arrow";
+import type { IntBitWidth, TimeBitWidth } from "apache-arrow/type";
+
+function sanitizeMetadata(
+  metadataLike?: unknown
+): Map<string, string> | undefined {
+  if (metadataLike === undefined || metadataLike === null) {
+    return undefined;
+  }
+  if (!(metadataLike instanceof Map)) {
+    throw Error("Expected metadata, if present, to be a Map<string, string>");
+  }
+  for (const item of metadataLike) {
+    if (!(typeof item[0] === "string" || !(typeof item[1] === "string"))) {
+      throw Error(
+        "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values"
+      );
+    }
+  }
+  return metadataLike as Map<string, string>;
+}
+
+function sanitizeInt(typeLike: object) {
+  if (
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number" ||
+    !("isSigned" in typeLike) ||
+    typeof typeLike.isSigned !== "boolean"
+  ) {
+    throw Error(
+      "Expected an Int Type to have a `bitWidth` and `isSigned` property"
+    );
+  }
+  return new Int(typeLike.isSigned, typeLike.bitWidth as IntBitWidth);
+}
+
+function sanitizeFloat(typeLike: object) {
+  if (!("precision" in typeLike) || typeof typeLike.precision !== "number") {
+    throw Error("Expected a Float Type to have a `precision` property");
+  }
+  return new Float(typeLike.precision as Precision);
+}
+
+function sanitizeDecimal(typeLike: object) {
+  if (
+    !("scale" in typeLike) ||
+    typeof typeLike.scale !== "number" ||
+    !("precision" in typeLike) ||
+    typeof typeLike.precision !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Decimal Type to have `scale`, `precision`, and `bitWidth` properties"
+    );
+  }
+  return new Decimal(typeLike.scale, typeLike.precision, typeLike.bitWidth);
+}
+
+function sanitizeDate(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Date type to have a `unit` property");
+  }
+  return new Date_(typeLike.unit as DateUnit);
+}
+
+function sanitizeTime(typeLike: object) {
+  if (
+    !("unit" in typeLike) ||
+    typeof typeLike.unit !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Time type to have `unit` and `bitWidth` properties"
+    );
+  }
+  return new Time(typeLike.unit, typeLike.bitWidth as TimeBitWidth);
+}
+
+function sanitizeTimestamp(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Timestamp type to have a `unit` property");
+  }
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Timestamp(typeLike.unit, timezone);
+}
+
+function sanitizeTypedTimestamp(
+  typeLike: object,
+  Datatype:
+    | typeof TimestampNanosecond
+    | typeof TimestampMicrosecond
+    | typeof TimestampMillisecond
+    | typeof TimestampSecond
+) {
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Datatype(timezone);
+}
+
+function sanitizeInterval(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected an Interval type to have a `unit` property");
+  }
+  return new Interval(typeLike.unit);
+}
+
+function sanitizeList(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a List type to have an array-like `children` property"
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a List type to have exactly one child");
+  }
+  return new List(sanitizeField(typeLike.children[0]));
+}
+
+function sanitizeStruct(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Struct type to have an array-like `children` property"
+    );
+  }
+  return new Struct(typeLike.children.map((child) => sanitizeField(child)));
+}
+
+function sanitizeUnion(typeLike: object) {
+  if (
+    !("typeIds" in typeLike) ||
+    !("mode" in typeLike) ||
+    typeof typeLike.mode !== "number"
+  ) {
+    throw Error(
+      "Expected a Union type to have `typeIds` and `mode` properties"
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Union type to have an array-like `children` property"
+    );
+  }
+
+  return new Union(
+    typeLike.mode,
+    typeLike.typeIds as any,
+    typeLike.children.map((child) => sanitizeField(child))
+  );
+}
+
+function sanitizeTypedUnion(
+  typeLike: object,
+  UnionType: typeof DenseUnion | typeof SparseUnion
+) {
+  if (!("typeIds" in typeLike)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have a `typeIds` property"
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have an array-like `children` property"
+    );
+  }
+
+  return new UnionType(
+    typeLike.typeIds as any,
+    typeLike.children.map((child) => sanitizeField(child))
+  );
+}
+
+function sanitizeFixedSizeBinary(typeLike: object) {
+  if (!("byteWidth" in typeLike) || typeof typeLike.byteWidth !== "number") {
+    throw Error(
+      "Expected a FixedSizeBinary type to have a `byteWidth` property"
+    );
+  }
+  return new FixedSizeBinary(typeLike.byteWidth);
+}
+
+function sanitizeFixedSizeList(typeLike: object) {
+  if (!("listSize" in typeLike) || typeof typeLike.listSize !== "number") {
+    throw Error("Expected a FixedSizeList type to have a `listSize` property");
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a FixedSizeList type to have an array-like `children` property"
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a FixedSizeList type to have exactly one child");
+  }
+  return new FixedSizeList(
+    typeLike.listSize,
+    sanitizeField(typeLike.children[0])
+  );
+}
+
+function sanitizeMap(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Map type to have an array-like `children` property"
+    );
+  }
+  if (!("keysSorted" in typeLike) || typeof typeLike.keysSorted !== "boolean") {
+    throw Error("Expected a Map type to have a `keysSorted` property");
+  }
+  return new Map_(
+    typeLike.children.map((field) => sanitizeField(field)) as any,
+    typeLike.keysSorted
+  );
+}
+
+function sanitizeDuration(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Duration type to have a `unit` property");
+  }
+  return new Duration(typeLike.unit);
+}
+
+function sanitizeDictionary(typeLike: object) {
+  if (!("id" in typeLike) || typeof typeLike.id !== "number") {
+    throw Error("Expected a Dictionary type to have an `id` property");
+  }
+  if (!("indices" in typeLike) || typeof typeLike.indices !== "object") {
+    throw Error("Expected a Dictionary type to have an `indices` property");
+  }
+  if (!("dictionary" in typeLike) || typeof typeLike.dictionary !== "object") {
+    throw Error("Expected a Dictionary type to have an `dictionary` property");
+  }
+  if (!("isOrdered" in typeLike) || typeof typeLike.isOrdered !== "boolean") {
+    throw Error("Expected a Dictionary type to have an `isOrdered` property");
+  }
+  return new Dictionary(
+    sanitizeType(typeLike.dictionary),
+    sanitizeType(typeLike.indices) as any,
+    typeLike.id,
+    typeLike.isOrdered
+  );
+}
+
+function sanitizeType(typeLike: unknown): DataType<any> {
+  if (typeof typeLike !== "object" || typeLike === null) {
+    throw Error("Expected a Type but object was null/undefined");
+  }
+  if (!("typeId" in typeLike) || !(typeof typeLike.typeId !== "function")) {
+    throw Error("Expected a Type to have a typeId function");
+  }
+  let typeId: Type;
+  if (typeof typeLike.typeId === "function") {
+    typeId = (typeLike.typeId as () => unknown)() as Type;
+  } else if (typeof typeLike.typeId === "number") {
+    typeId = typeLike.typeId as Type;
+  } else {
+    throw Error("Type's typeId property was not a function or number");
+  }
+
+  switch (typeId) {
+    case Type.NONE:
+      throw Error("Received a Type with a typeId of NONE");
+    case Type.Null:
+      return new Null();
+    case Type.Int:
+      return sanitizeInt(typeLike);
+    case Type.Float:
+      return sanitizeFloat(typeLike);
+    case Type.Binary:
+      return new Binary();
+    case Type.Utf8:
+      return new Utf8();
+    case Type.Bool:
+      return new Bool();
+    case Type.Decimal:
+      return sanitizeDecimal(typeLike);
+    case Type.Date:
+      return sanitizeDate(typeLike);
+    case Type.Time:
+      return sanitizeTime(typeLike);
+    case Type.Timestamp:
+      return sanitizeTimestamp(typeLike);
+    case Type.Interval:
+      return sanitizeInterval(typeLike);
+    case Type.List:
+      return sanitizeList(typeLike);
+    case Type.Struct:
+      return sanitizeStruct(typeLike);
+    case Type.Union:
+      return sanitizeUnion(typeLike);
+    case Type.FixedSizeBinary:
+      return sanitizeFixedSizeBinary(typeLike);
+    case Type.FixedSizeList:
+      return sanitizeFixedSizeList(typeLike);
+    case Type.Map:
+      return sanitizeMap(typeLike);
+    case Type.Duration:
+      return sanitizeDuration(typeLike);
+    case Type.Dictionary:
+      return sanitizeDictionary(typeLike);
+    case Type.Int8:
+      return new Int8();
+    case Type.Int16:
+      return new Int16();
+    case Type.Int32:
+      return new Int32();
+    case Type.Int64:
+      return new Int64();
+    case Type.Uint8:
+      return new Uint8();
+    case Type.Uint16:
+      return new Uint16();
+    case Type.Uint32:
+      return new Uint32();
+    case Type.Uint64:
+      return new Uint64();
+    case Type.Float16:
+      return new Float16();
+    case Type.Float32:
+      return new Float32();
+    case Type.Float64:
+      return new Float64();
+    case Type.DateMillisecond:
+      return new DateMillisecond();
+    case Type.DateDay:
+      return new DateDay();
+    case Type.TimeNanosecond:
+      return new TimeNanosecond();
+    case Type.TimeMicrosecond:
+      return new TimeMicrosecond();
+    case Type.TimeMillisecond:
+      return new TimeMillisecond();
+    case Type.TimeSecond:
+      return new TimeSecond();
+    case Type.TimestampNanosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampNanosecond);
+    case Type.TimestampMicrosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMicrosecond);
+    case Type.TimestampMillisecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMillisecond);
+    case Type.TimestampSecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampSecond);
+    case Type.DenseUnion:
+      return sanitizeTypedUnion(typeLike, DenseUnion);
+    case Type.SparseUnion:
+      return sanitizeTypedUnion(typeLike, SparseUnion);
+    case Type.IntervalDayTime:
+      return new IntervalDayTime();
+    case Type.IntervalYearMonth:
+      return new IntervalYearMonth();
+    case Type.DurationNanosecond:
+      return new DurationNanosecond();
+    case Type.DurationMicrosecond:
+      return new DurationMicrosecond();
+    case Type.DurationMillisecond:
+      return new DurationMillisecond();
+    case Type.DurationSecond:
+      return new DurationSecond();
+  }
+}
+
+function sanitizeField(fieldLike: unknown): Field {
+  if (fieldLike instanceof Field) {
+    return fieldLike;
+  }
+  if (typeof fieldLike !== "object" || fieldLike === null) {
+    throw Error("Expected a Field but object was null/undefined");
+  }
+  if (
+    !("type" in fieldLike) ||
+    !("name" in fieldLike) ||
+    !("nullable" in fieldLike)
+  ) {
+    throw Error(
+      "The field passed in is missing a `type`/`name`/`nullable` property"
+    );
+  }
+  const type = sanitizeType(fieldLike.type);
+  const name = fieldLike.name;
+  if (!(typeof name === "string")) {
+    throw Error("The field passed in had a non-string `name` property");
+  }
+  const nullable = fieldLike.nullable;
+  if (!(typeof nullable === "boolean")) {
+    throw Error("The field passed in had a non-boolean `nullable` property");
+  }
+  let metadata;
+  if ("metadata" in fieldLike) {
+    metadata = sanitizeMetadata(fieldLike.metadata);
+  }
+  return new Field(name, type, nullable, metadata);
+}
+
+export function sanitizeSchema(schemaLike: unknown): Schema {
+  if (schemaLike instanceof Schema) {
+    return schemaLike;
+  }
+  if (typeof schemaLike !== "object" || schemaLike === null) {
+    throw Error("Expected a Schema but object was null/undefined");
+  }
+  if (!("fields" in schemaLike)) {
+    throw Error(
+      "The schema passed in does not appear to be a schema (no 'fields' property)"
+    );
+  }
+  let metadata;
+  if ("metadata" in schemaLike) {
+    metadata = sanitizeMetadata(schemaLike.metadata);
+  }
+  if (!Array.isArray(schemaLike.fields)) {
+    throw Error(
+      "The schema passed in had a 'fields' property but it was not an array"
+    );
+  }
+  const sanitizedFields = schemaLike.fields.map((field) =>
+    sanitizeField(field)
+  );
+  return new Schema(sanitizedFields, metadata);
+}

node/src/test/arrow.test.ts

@@ -13,9 +13,10 @@
 // limitations under the License.
 
 import { describe } from 'mocha'
-import { assert } from 'chai'
+import { assert, expect, use as chaiUse } from 'chai'
+import * as chaiAsPromised from 'chai-as-promised'
 
-import { fromTableToBuffer, makeArrowTable } from '../arrow'
+import { convertToTable, fromTableToBuffer, makeArrowTable, makeEmptyTable } from '../arrow'
 import {
   Field,
   FixedSizeList,
@@ -24,21 +25,91 @@ import {
   Int32,
   tableFromIPC,
   Schema,
-  Float64
+  Float64,
+  type Table,
+  Binary,
+  Bool,
+  Utf8,
+  Struct,
+  List,
+  DataType,
+  Dictionary,
+  Int64,
+  MetadataVersion
 } from 'apache-arrow'
+import {
+  Dictionary as OldDictionary,
+  Field as OldField,
+  FixedSizeList as OldFixedSizeList,
+  Float32 as OldFloat32,
+  Int32 as OldInt32,
+  Struct as OldStruct,
+  Schema as OldSchema,
+  TimestampNanosecond as OldTimestampNanosecond,
+  Utf8 as OldUtf8
+} from 'apache-arrow-old'
+import { type EmbeddingFunction } from '../embedding/embedding_function'
 
-describe('Apache Arrow tables', function () {
-  it('customized schema', async function () {
+chaiUse(chaiAsPromised)
+
+function sampleRecords (): Array<Record<string, any>> {
+  return [
+    {
+      binary: Buffer.alloc(5),
+      boolean: false,
+      number: 7,
+      string: 'hello',
+      struct: { x: 0, y: 0 },
+      list: ['anime', 'action', 'comedy']
+    }
+  ]
+}
+
+// Helper method to verify various ways to create a table
+async function checkTableCreation (tableCreationMethod: (records: any, recordsReversed: any, schema: Schema) => Promise<Table>): Promise<void> {
+  const records = sampleRecords()
+  const recordsReversed = [{
+    list: ['anime', 'action', 'comedy'],
+    struct: { x: 0, y: 0 },
+    string: 'hello',
+    number: 7,
+    boolean: false,
+    binary: Buffer.alloc(5)
+  }]
+  const schema = new Schema([
+    new Field('binary', new Binary(), false),
+    new Field('boolean', new Bool(), false),
+    new Field('number', new Float64(), false),
+    new Field('string', new Utf8(), false),
+    new Field('struct', new Struct([
+      new Field('x', new Float64(), false),
+      new Field('y', new Float64(), false)
+    ])),
+    new Field('list', new List(new Field('item', new Utf8(), false)), false)
+  ])
+
+  const table = await tableCreationMethod(records, recordsReversed, schema)
+  schema.fields.forEach((field, idx) => {
+    const actualField = table.schema.fields[idx]
+    assert.isFalse(actualField.nullable)
+    assert.equal(table.getChild(field.name)?.type.toString(), field.type.toString())
+    assert.equal(table.getChildAt(idx)?.type.toString(), field.type.toString())
+  })
+}
+
+describe('The function makeArrowTable', function () {
+  it('will use data types from a provided schema instead of inference', async function () {
     const schema = new Schema([
       new Field('a', new Int32()),
       new Field('b', new Float32()),
-      new Field('c', new FixedSizeList(3, new Field('item', new Float16())))
+      new Field('c', new FixedSizeList(3, new Field('item', new Float16()))),
+      new Field('d', new Int64())
     ])
     const table = makeArrowTable(
       [
-        { a: 1, b: 2, c: [1, 2, 3] },
-        { a: 4, b: 5, c: [4, 5, 6] },
-        { a: 7, b: 8, c: [7, 8, 9] }
+        { a: 1, b: 2, c: [1, 2, 3], d: 9 },
+        { a: 4, b: 5, c: [4, 5, 6], d: 10 },
+        { a: 7, b: 8, c: [7, 8, 9], d: null }
       ],
       { schema }
     )
@@ -52,13 +123,13 @@ describe('Apache Arrow tables', function () {
     assert.deepEqual(actualSchema, schema)
   })
 
-  it('default vector column', async function () {
+  it('will assume the column `vector` is FixedSizeList<Float32> by default', async function () {
     const schema = new Schema([
       new Field('a', new Float64()),
       new Field('b', new Float64()),
       new Field(
         'vector',
-        new FixedSizeList(3, new Field('item', new Float32()))
+        new FixedSizeList(3, new Field('item', new Float32(), true))
       )
     ])
     const table = makeArrowTable([
@@ -76,12 +147,12 @@ describe('Apache Arrow tables', function () {
     assert.deepEqual(actualSchema, schema)
   })
 
-  it('2 vector columns', async function () {
+  it('can support multiple vector columns', async function () {
     const schema = new Schema([
       new Field('a', new Float64()),
       new Field('b', new Float64()),
-      new Field('vec1', new FixedSizeList(3, new Field('item', new Float16()))),
-      new Field('vec2', new FixedSizeList(3, new Field('item', new Float16())))
+      new Field('vec1', new FixedSizeList(3, new Field('item', new Float16(), true))),
+      new Field('vec2', new FixedSizeList(3, new Field('item', new Float16(), true)))
     ])
     const table = makeArrowTable(
       [
@@ -105,4 +176,185 @@ describe('Apache Arrow tables', function () {
     const actualSchema = actual.schema
     assert.deepEqual(actualSchema, schema)
   })
+
+  it('will allow different vector column types', async function () {
+    const table = makeArrowTable(
+      [
+        { fp16: [1], fp32: [1], fp64: [1] }
+      ],
+      {
+        vectorColumns: {
+          fp16: { type: new Float16() },
+          fp32: { type: new Float32() },
+          fp64: { type: new Float64() }
+        }
+      }
+    )
+
+    assert.equal(table.getChild('fp16')?.type.children[0].type.toString(), new Float16().toString())
+    assert.equal(table.getChild('fp32')?.type.children[0].type.toString(), new Float32().toString())
+    assert.equal(table.getChild('fp64')?.type.children[0].type.toString(), new Float64().toString())
+  })
+
+  it('will use dictionary encoded strings if asked', async function () {
+    const table = makeArrowTable([{ str: 'hello' }])
+    assert.isTrue(DataType.isUtf8(table.getChild('str')?.type))
+
+    const tableWithDict = makeArrowTable([{ str: 'hello' }], { dictionaryEncodeStrings: true })
+    assert.isTrue(DataType.isDictionary(tableWithDict.getChild('str')?.type))
+
+    const schema = new Schema([
+      new Field('str', new Dictionary(new Utf8(), new Int32()))
+    ])
+
+    const tableWithDict2 = makeArrowTable([{ str: 'hello' }], { schema })
+    assert.isTrue(DataType.isDictionary(tableWithDict2.getChild('str')?.type))
+  })
+
+  it('will infer data types correctly', async function () {
+    await checkTableCreation(async (records) => makeArrowTable(records))
+  })
+
+  it('will allow a schema to be provided', async function () {
+    await checkTableCreation(async (records, _, schema) => makeArrowTable(records, { schema }))
+  })
+
+  it('will use the field order of any provided schema', async function () {
+    await checkTableCreation(async (_, recordsReversed, schema) => makeArrowTable(recordsReversed, { schema }))
+  })
+
+  it('will make an empty table', async function () {
+    await checkTableCreation(async (_, __, schema) => makeArrowTable([], { schema }))
+  })
+})
+
+class DummyEmbedding implements EmbeddingFunction<string> {
+  public readonly sourceColumn = 'string'
+  public readonly embeddingDimension = 2
+  public readonly embeddingDataType = new Float16()
+
+  async embed (data: string[]): Promise<number[][]> {
+    return data.map(
+      () => [0.0, 0.0]
+    )
+  }
+}
+
+class DummyEmbeddingWithNoDimension implements EmbeddingFunction<string> {
+  public readonly sourceColumn = 'string'
+
+  async embed (data: string[]): Promise<number[][]> {
+    return data.map(
+      () => [0.0, 0.0]
+    )
+  }
+}
+
+describe('convertToTable', function () {
+  it('will infer data types correctly', async function () {
+    await checkTableCreation(async (records) => await convertToTable(records))
+  })
+
+  it('will allow a schema to be provided', async function () {
+    await checkTableCreation(async (records, _, schema) => await convertToTable(records, undefined, { schema }))
+  })
+
+  it('will use the field order of any provided schema', async function () {
+    await checkTableCreation(async (_, recordsReversed, schema) => await convertToTable(recordsReversed, undefined, { schema }))
+  })
+
+  it('will make an empty table', async function () {
+    await checkTableCreation(async (_, __, schema) => await convertToTable([], undefined, { schema }))
+  })
+
+  it('will apply embeddings', async function () {
+    const records = sampleRecords()
+    const table = await convertToTable(records, new DummyEmbedding())
+    assert.isTrue(DataType.isFixedSizeList(table.getChild('vector')?.type))
+    assert.equal(table.getChild('vector')?.type.children[0].type.toString(), new Float16().toString())
+  })
+
+  it('will fail if missing the embedding source column', async function () {
+    return await expect(convertToTable([{ id: 1 }], new DummyEmbedding())).to.be.rejectedWith("'string' was not present")
+  })
+
+  it('use embeddingDimension if embedding missing from table', async function () {
+    const schema = new Schema([
+      new Field('string', new Utf8(), false)
+    ])
+    // Simulate getting an empty Arrow table (minus embedding) from some other source
+    // In other words, we aren't starting with records
+    const table = makeEmptyTable(schema)
+
+    // If the embedding specifies the dimension we are fine
+    await fromTableToBuffer(table, new DummyEmbedding())
+
+    // We can also supply a schema and should be ok
+    const schemaWithEmbedding = new Schema([
+      new Field('string', new Utf8(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float16(), false)), false)
+    ])
+    await fromTableToBuffer(table, new DummyEmbeddingWithNoDimension(), schemaWithEmbedding)
+
+    // Otherwise we will get an error
+    return await expect(fromTableToBuffer(table, new DummyEmbeddingWithNoDimension())).to.be.rejectedWith('does not specify `embeddingDimension`')
+  })
+
+  it('will apply embeddings to an empty table', async function () {
+    const schema = new Schema([
+      new Field('string', new Utf8(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float16(), false)), false)
+    ])
+    const table = await convertToTable([], new DummyEmbedding(), { schema })
+    assert.isTrue(DataType.isFixedSizeList(table.getChild('vector')?.type))
+    assert.equal(table.getChild('vector')?.type.children[0].type.toString(), new Float16().toString())
+  })
+
+  it('will complain if embeddings present but schema missing embedding column', async function () {
+    const schema = new Schema([
+      new Field('string', new Utf8(), false)
+    ])
+    return await expect(convertToTable([], new DummyEmbedding(), { schema })).to.be.rejectedWith('column vector was missing')
+  })
+
+  it('will provide a nice error if run twice', async function () {
+    const records = sampleRecords()
+    const table = await convertToTable(records, new DummyEmbedding())
+    // fromTableToBuffer will try and apply the embeddings again
+    return await expect(fromTableToBuffer(table, new DummyEmbedding())).to.be.rejectedWith('already existed')
+  })
+})
+
+describe('makeEmptyTable', function () {
+  it('will make an empty table', async function () {
+    await checkTableCreation(async (_, __, schema) => makeEmptyTable(schema))
+  })
+})
+
+describe('when using two versions of arrow', function () {
+  it('can still import data', async function() {
+    const schema = new OldSchema([
+      new OldField('id', new OldInt32()),
+      new OldField('vector', new OldFixedSizeList(1024, new OldField("item", new OldFloat32(), true))),
+      new OldField('struct', new OldStruct([
+        new OldField('nested', new OldDictionary(new OldUtf8(), new OldInt32(), 1, true)),
+        new OldField('ts_with_tz', new OldTimestampNanosecond("some_tz")),
+        new OldField('ts_no_tz', new OldTimestampNanosecond(null))
+      ]))
+    ]) as any
+    // We use arrow version 13 to emulate a "foreign arrow" and this version doesn't have metadataVersion
+    // In theory, this wouldn't matter.  We don't rely on that property.  However, it causes deepEqual to
+    // fail so we patch it back in
+    schema.metadataVersion = MetadataVersion.V5
+    const table = makeArrowTable(
+      [],
+      { schema }
+    )
+
+    const buf = await fromTableToBuffer(table)
+    assert.isAbove(buf.byteLength, 0)
+    const actual = tableFromIPC(buf)
+    const actualSchema = actual.schema
+    assert.deepEqual(actualSchema, schema)
+  })
 })

node/src/test/test.ts

@@ -37,8 +37,10 @@ import {
   Utf8,
   Table as ArrowTable,
   vectorFromArray,
+  Float64,
   Float32,
-  Float16
+  Float16,
+  Int64
 } from 'apache-arrow'
 
 const expect = chai.expect
@@ -196,7 +198,7 @@ describe('LanceDB client', function () {
       const table = await con.openTable('vectors')
       const results = await table
         .search([0.1, 0.1])
-        .select(['is_active'])
+        .select(['is_active', 'vector'])
         .execute()
       assert.equal(results.length, 2)
       // vector and _distance are always returned
@@ -294,6 +296,7 @@ describe('LanceDB client', function () {
       })
       assert.equal(table.name, 'vectors')
       assert.equal(await table.countRows(), 10)
+      assert.equal(await table.countRows('vector IS NULL'), 0)
       assert.deepEqual(await con.tableNames(), ['vectors'])
     })
 
@@ -369,6 +372,7 @@ describe('LanceDB client', function () {
       const table = await con.createTable('f16', data)
       assert.equal(table.name, 'f16')
       assert.equal(await table.countRows(), total)
+      assert.equal(await table.countRows('id < 5'), 5)
       assert.deepEqual(await con.tableNames(), ['f16'])
       assert.deepEqual(await table.schema, schema)
 
@@ -391,24 +395,6 @@ describe('LanceDB client', function () {
       })
     }).timeout(120000)
 
-    it('fails to create a new table when the vector column is missing', async function () {
-      const dir = await track().mkdir('lancejs')
-      const con = await lancedb.connect(dir)
-
-      const data = [
-        {
-          id: 1,
-          price: 10
-        }
-      ]
-
-      const create = con.createTable('missing_vector', data)
-      await expect(create).to.be.rejectedWith(
-        Error,
-        "column 'vector' is missing"
-      )
-    })
-
     it('use overwrite flag to overwrite existing table', async function () {
       const dir = await track().mkdir('lancejs')
       const con = await lancedb.connect(dir)
@@ -549,6 +535,54 @@ describe('LanceDB client', function () {
       assert.equal(await table.countRows(), 2)
     })
 
+    it('can merge insert records into the table', async function () {
+      const dir = await track().mkdir('lancejs')
+      const con = await lancedb.connect(dir)
+
+      const data = [{ id: 1, age: 1 }, { id: 2, age: 1 }]
+      const table = await con.createTable('my_table', data)
+
+      // insert if not exists
+      let newData = [{ id: 2, age: 2 }, { id: 3, age: 2 }]
+      await table.mergeInsert('id', newData, {
+        whenNotMatchedInsertAll: true
+      })
+      assert.equal(await table.countRows(), 3)
+      assert.equal(await table.countRows('age = 2'), 1)
+
+      // conditional update
+      newData = [{ id: 2, age: 3 }, { id: 3, age: 3 }]
+      await table.mergeInsert('id', newData, {
+        whenMatchedUpdateAll: 'target.age = 1'
+      })
+      assert.equal(await table.countRows(), 3)
+      assert.equal(await table.countRows('age = 1'), 1)
+      assert.equal(await table.countRows('age = 3'), 1)
+
+      newData = [{ id: 3, age: 4 }, { id: 4, age: 4 }]
+      await table.mergeInsert('id', newData, {
+        whenNotMatchedInsertAll: true,
+        whenMatchedUpdateAll: true
+      })
+      assert.equal(await table.countRows(), 4)
+      assert.equal((await table.filter('age = 4').execute()).length, 2)
+
+      newData = [{ id: 5, age: 5 }]
+      await table.mergeInsert('id', newData, {
+        whenNotMatchedInsertAll: true,
+        whenMatchedUpdateAll: true,
+        whenNotMatchedBySourceDelete: 'age < 4'
+      })
+      assert.equal(await table.countRows(), 3)
+
+      await table.mergeInsert('id', newData, {
+        whenNotMatchedInsertAll: true,
+        whenMatchedUpdateAll: true,
+        whenNotMatchedBySourceDelete: true
+      })
+      assert.equal(await table.countRows(), 1)
+    })
+
     it('can update records in the table', async function () {
       const uri = await createTestDB()
       const con = await lancedb.connect(uri)
@@ -1025,3 +1059,63 @@ describe('Compact and cleanup', function () {
     assert.equal(await table.countRows(), 3)
   })
 })
+
+describe('schema evolution', function () {
+  // Create a new sample table
+  it('can add a new column to the schema', async function () {
+    const dir = await track().mkdir('lancejs')
+    const con = await lancedb.connect(dir)
+    const table = await con.createTable('vectors', [
+      { id: 1n, vector: [0.1, 0.2] }
+    ])
+
+    await table.addColumns([{ name: 'price', valueSql: 'cast(10.0 as float)' }])
+
+    const expectedSchema = new Schema([
+      new Field('id', new Int64()),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true))),
+      new Field('price', new Float32())
+    ])
+    expect(await table.schema).to.deep.equal(expectedSchema)
+  })
+
+  it('can alter the columns in the schema', async function () {
+    const dir = await track().mkdir('lancejs')
+    const con = await lancedb.connect(dir)
+    const schema = new Schema([
+      new Field('id', new Int64(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true))),
+      new Field('price', new Float64(), false)
+    ])
+    const table = await con.createTable('vectors', [
+      { id: 1n, vector: [0.1, 0.2], price: 10.0 }
+    ])
+    expect(await table.schema).to.deep.equal(schema)
+
+    await table.alterColumns([
+      { path: 'id', rename: 'new_id' },
+      { path: 'price', nullable: true }
+    ])
+
+    const expectedSchema = new Schema([
+      new Field('new_id', new Int64(), false),
+      new Field('vector', new FixedSizeList(2, new Field('item', new Float32(), true))),
+      new Field('price', new Float64(), true)
+    ])
+    expect(await table.schema).to.deep.equal(expectedSchema)
+  })
+
+  it('can drop a column from the schema', async function () {
+    const dir = await track().mkdir('lancejs')
+    const con = await lancedb.connect(dir)
+    const table = await con.createTable('vectors', [
+      { id: 1n, vector: [0.1, 0.2] }
+    ])
+    await table.dropColumns(['vector'])
+
+    const expectedSchema = new Schema([
+      new Field('id', new Int64(), false)
+    ])
+    expect(await table.schema).to.deep.equal(expectedSchema)
+  })
+})

node/tsconfig.json

@@ -9,6 +9,6 @@
     "declaration": true,
     "outDir": "./dist",
     "strict": true,
-    // "esModuleInterop": true,
+    "sourceMap": true,
   }
 }

nodejs/.eslintignore

@@ -0,0 +1,3 @@
+**/dist/**/*
+**/native.js
+**/native.d.ts

nodejs/.npmignore

@@ -0,0 +1,15 @@
+target
+Cargo.lock
+.cargo
+.github
+npm
+.eslintrc
+.prettierignore
+rustfmt.toml
+yarn.lock
+*.node
+.yarn
+__test__
+renovate.json
+.idea
+src

nodejs/.prettierignore

@@ -0,0 +1 @@
+.eslintignore

nodejs/Cargo.toml

@@ -0,0 +1,30 @@
+[package]
+name = "lancedb-nodejs"
+edition.workspace = true
+version = "0.0.0"
+license.workspace = true
+description.workspace = true
+repository.workspace = true
+keywords.workspace = true
+categories.workspace = true
+
+[lib]
+crate-type = ["cdylib"]
+
+[dependencies]
+arrow-ipc.workspace = true
+futures.workspace = true
+lance-linalg.workspace = true
+lance.workspace = true
+lancedb = { path = "../rust/lancedb" }
+napi = { version = "2.15", default-features = false, features = [
+    "napi7",
+    "async"
+] }
+napi-derive = "2"
+
+# Prevent dynamic linking of lzma, which comes from datafusion
+lzma-sys = { version = "*", features = ["static"] }
+
+[build-dependencies]
+napi-build = "2.1"

nodejs/README.md

@@ -0,0 +1,49 @@
+# (New) LanceDB NodeJS SDK
+
+It will replace the NodeJS SDK when it is ready.
+
+## Development
+
+```sh
+npm run build
+npm t
+```
+
+### Running lint / format
+
+LanceDb uses eslint for linting. VSCode does not need any plugins to use eslint. However, it
+may need some additional configuration. Make sure that eslint.experimental.useFlatConfig is
+set to true. Also, if your vscode root folder is the repo root then you will need to set
+the eslint.workingDirectories to ["nodejs"]. To manually lint your code you can run:
+
+```sh
+npm run lint
+```
+
+LanceDb uses prettier for formatting. If you are using VSCode you will need to install the
+"Prettier - Code formatter" extension. You should then configure it to be the default formatter
+for typescript and you should enable format on save. To manually check your code's format you
+can run:
+
+```sh
+npm run chkformat
+```
+
+If you need to manually format your code you can run:
+
+```sh
+npx prettier --write .
+```
+
+### Generating docs
+
+```sh
+npm run docs
+
+cd ../docs
+# Asssume the virtual environment was created
+# python3 -m venv venv
+# pip install -r requirements.txt
+. ./venv/bin/activate
+mkdocs build
+```

nodejs/__test__/arrow.test.ts

@@ -0,0 +1,470 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import {
+  convertToTable,
+  fromTableToBuffer,
+  makeArrowTable,
+  makeEmptyTable,
+} from "../dist/arrow";
+import {
+  Field,
+  FixedSizeList,
+  Float16,
+  Float32,
+  Int32,
+  tableFromIPC,
+  Schema,
+  Float64,
+  type Table,
+  Binary,
+  Bool,
+  Utf8,
+  Struct,
+  List,
+  DataType,
+  Dictionary,
+  Int64,
+  Float,
+  Precision,
+  MetadataVersion,
+} from "apache-arrow";
+import {
+  Dictionary as OldDictionary,
+  Field as OldField,
+  FixedSizeList as OldFixedSizeList,
+  Float32 as OldFloat32,
+  Int32 as OldInt32,
+  Struct as OldStruct,
+  Schema as OldSchema,
+  TimestampNanosecond as OldTimestampNanosecond,
+  Utf8 as OldUtf8,
+} from "apache-arrow-old";
+import { type EmbeddingFunction } from "../dist/embedding/embedding_function";
+
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+function sampleRecords(): Array<Record<string, any>> {
+  return [
+    {
+      binary: Buffer.alloc(5),
+      boolean: false,
+      number: 7,
+      string: "hello",
+      struct: { x: 0, y: 0 },
+      list: ["anime", "action", "comedy"],
+    },
+  ];
+}
+
+// Helper method to verify various ways to create a table
+async function checkTableCreation(
+  tableCreationMethod: (
+    records: Record<string, unknown>[],
+    recordsReversed: Record<string, unknown>[],
+    schema: Schema,
+  ) => Promise<Table>,
+  infersTypes: boolean,
+): Promise<void> {
+  const records = sampleRecords();
+  const recordsReversed = [
+    {
+      list: ["anime", "action", "comedy"],
+      struct: { x: 0, y: 0 },
+      string: "hello",
+      number: 7,
+      boolean: false,
+      binary: Buffer.alloc(5),
+    },
+  ];
+  const schema = new Schema([
+    new Field("binary", new Binary(), false),
+    new Field("boolean", new Bool(), false),
+    new Field("number", new Float64(), false),
+    new Field("string", new Utf8(), false),
+    new Field(
+      "struct",
+      new Struct([
+        new Field("x", new Float64(), false),
+        new Field("y", new Float64(), false),
+      ]),
+    ),
+    new Field("list", new List(new Field("item", new Utf8(), false)), false),
+  ]);
+
+  const table = await tableCreationMethod(records, recordsReversed, schema);
+  schema.fields.forEach((field, idx) => {
+    const actualField = table.schema.fields[idx];
+    // Type inference always assumes nullable=true
+    if (infersTypes) {
+      expect(actualField.nullable).toBe(true);
+    } else {
+      expect(actualField.nullable).toBe(false);
+    }
+    expect(table.getChild(field.name)?.type.toString()).toEqual(
+      field.type.toString(),
+    );
+    expect(table.getChildAt(idx)?.type.toString()).toEqual(
+      field.type.toString(),
+    );
+  });
+}
+
+describe("The function makeArrowTable", function () {
+  it("will use data types from a provided schema instead of inference", async function () {
+    const schema = new Schema([
+      new Field("a", new Int32()),
+      new Field("b", new Float32()),
+      new Field("c", new FixedSizeList(3, new Field("item", new Float16()))),
+      new Field("d", new Int64()),
+    ]);
+    const table = makeArrowTable(
+      [
+        { a: 1, b: 2, c: [1, 2, 3], d: 9 },
+        { a: 4, b: 5, c: [4, 5, 6], d: 10 },
+        { a: 7, b: 8, c: [7, 8, 9], d: null },
+      ],
+      { schema },
+    );
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+
+    const actual = tableFromIPC(buf);
+    expect(actual.numRows).toBe(3);
+    const actualSchema = actual.schema;
+    expect(actualSchema).toEqual(schema);
+  });
+
+  it("will assume the column `vector` is FixedSizeList<Float32> by default", async function () {
+    const schema = new Schema([
+      new Field("a", new Float(Precision.DOUBLE), true),
+      new Field("b", new Float(Precision.DOUBLE), true),
+      new Field(
+        "vector",
+        new FixedSizeList(
+          3,
+          new Field("item", new Float(Precision.SINGLE), true),
+        ),
+        true,
+      ),
+    ]);
+    const table = makeArrowTable([
+      { a: 1, b: 2, vector: [1, 2, 3] },
+      { a: 4, b: 5, vector: [4, 5, 6] },
+      { a: 7, b: 8, vector: [7, 8, 9] },
+    ]);
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+
+    const actual = tableFromIPC(buf);
+    expect(actual.numRows).toBe(3);
+    const actualSchema = actual.schema;
+    expect(actualSchema).toEqual(schema);
+  });
+
+  it("can support multiple vector columns", async function () {
+    const schema = new Schema([
+      new Field("a", new Float(Precision.DOUBLE), true),
+      new Field("b", new Float(Precision.DOUBLE), true),
+      new Field(
+        "vec1",
+        new FixedSizeList(3, new Field("item", new Float16(), true)),
+        true,
+      ),
+      new Field(
+        "vec2",
+        new FixedSizeList(3, new Field("item", new Float16(), true)),
+        true,
+      ),
+    ]);
+    const table = makeArrowTable(
+      [
+        { a: 1, b: 2, vec1: [1, 2, 3], vec2: [2, 4, 6] },
+        { a: 4, b: 5, vec1: [4, 5, 6], vec2: [8, 10, 12] },
+        { a: 7, b: 8, vec1: [7, 8, 9], vec2: [14, 16, 18] },
+      ],
+      {
+        vectorColumns: {
+          vec1: { type: new Float16() },
+          vec2: { type: new Float16() },
+        },
+      },
+    );
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+
+    const actual = tableFromIPC(buf);
+    expect(actual.numRows).toBe(3);
+    const actualSchema = actual.schema;
+    expect(actualSchema).toEqual(schema);
+  });
+
+  it("will allow different vector column types", async function () {
+    const table = makeArrowTable([{ fp16: [1], fp32: [1], fp64: [1] }], {
+      vectorColumns: {
+        fp16: { type: new Float16() },
+        fp32: { type: new Float32() },
+        fp64: { type: new Float64() },
+      },
+    });
+
+    expect(table.getChild("fp16")?.type.children[0].type.toString()).toEqual(
+      new Float16().toString(),
+    );
+    expect(table.getChild("fp32")?.type.children[0].type.toString()).toEqual(
+      new Float32().toString(),
+    );
+    expect(table.getChild("fp64")?.type.children[0].type.toString()).toEqual(
+      new Float64().toString(),
+    );
+  });
+
+  it("will use dictionary encoded strings if asked", async function () {
+    const table = makeArrowTable([{ str: "hello" }]);
+    expect(DataType.isUtf8(table.getChild("str")?.type)).toBe(true);
+
+    const tableWithDict = makeArrowTable([{ str: "hello" }], {
+      dictionaryEncodeStrings: true,
+    });
+    expect(DataType.isDictionary(tableWithDict.getChild("str")?.type)).toBe(
+      true,
+    );
+
+    const schema = new Schema([
+      new Field("str", new Dictionary(new Utf8(), new Int32())),
+    ]);
+
+    const tableWithDict2 = makeArrowTable([{ str: "hello" }], { schema });
+    expect(DataType.isDictionary(tableWithDict2.getChild("str")?.type)).toBe(
+      true,
+    );
+  });
+
+  it("will infer data types correctly", async function () {
+    await checkTableCreation(async (records) => makeArrowTable(records), true);
+  });
+
+  it("will allow a schema to be provided", async function () {
+    await checkTableCreation(
+      async (records, _, schema) => makeArrowTable(records, { schema }),
+      false,
+    );
+  });
+
+  it("will use the field order of any provided schema", async function () {
+    await checkTableCreation(
+      async (_, recordsReversed, schema) =>
+        makeArrowTable(recordsReversed, { schema }),
+      false,
+    );
+  });
+
+  it("will make an empty table", async function () {
+    await checkTableCreation(
+      async (_, __, schema) => makeArrowTable([], { schema }),
+      false,
+    );
+  });
+});
+
+class DummyEmbedding implements EmbeddingFunction<string> {
+  public readonly sourceColumn = "string";
+  public readonly embeddingDimension = 2;
+  public readonly embeddingDataType = new Float16();
+
+  async embed(data: string[]): Promise<number[][]> {
+    return data.map(() => [0.0, 0.0]);
+  }
+}
+
+class DummyEmbeddingWithNoDimension implements EmbeddingFunction<string> {
+  public readonly sourceColumn = "string";
+
+  async embed(data: string[]): Promise<number[][]> {
+    return data.map(() => [0.0, 0.0]);
+  }
+}
+
+describe("convertToTable", function () {
+  it("will infer data types correctly", async function () {
+    await checkTableCreation(
+      async (records) => await convertToTable(records),
+      true,
+    );
+  });
+
+  it("will allow a schema to be provided", async function () {
+    await checkTableCreation(
+      async (records, _, schema) =>
+        await convertToTable(records, undefined, { schema }),
+      false,
+    );
+  });
+
+  it("will use the field order of any provided schema", async function () {
+    await checkTableCreation(
+      async (_, recordsReversed, schema) =>
+        await convertToTable(recordsReversed, undefined, { schema }),
+      false,
+    );
+  });
+
+  it("will make an empty table", async function () {
+    await checkTableCreation(
+      async (_, __, schema) => await convertToTable([], undefined, { schema }),
+      false,
+    );
+  });
+
+  it("will apply embeddings", async function () {
+    const records = sampleRecords();
+    const table = await convertToTable(records, new DummyEmbedding());
+    expect(DataType.isFixedSizeList(table.getChild("vector")?.type)).toBe(true);
+    expect(table.getChild("vector")?.type.children[0].type.toString()).toEqual(
+      new Float16().toString(),
+    );
+  });
+
+  it("will fail if missing the embedding source column", async function () {
+    await expect(
+      convertToTable([{ id: 1 }], new DummyEmbedding()),
+    ).rejects.toThrow("'string' was not present");
+  });
+
+  it("use embeddingDimension if embedding missing from table", async function () {
+    const schema = new Schema([new Field("string", new Utf8(), false)]);
+    // Simulate getting an empty Arrow table (minus embedding) from some other source
+    // In other words, we aren't starting with records
+    const table = makeEmptyTable(schema);
+
+    // If the embedding specifies the dimension we are fine
+    await fromTableToBuffer(table, new DummyEmbedding());
+
+    // We can also supply a schema and should be ok
+    const schemaWithEmbedding = new Schema([
+      new Field("string", new Utf8(), false),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float16(), false)),
+        false,
+      ),
+    ]);
+    await fromTableToBuffer(
+      table,
+      new DummyEmbeddingWithNoDimension(),
+      schemaWithEmbedding,
+    );
+
+    // Otherwise we will get an error
+    await expect(
+      fromTableToBuffer(table, new DummyEmbeddingWithNoDimension()),
+    ).rejects.toThrow("does not specify `embeddingDimension`");
+  });
+
+  it("will apply embeddings to an empty table", async function () {
+    const schema = new Schema([
+      new Field("string", new Utf8(), false),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float16(), false)),
+        false,
+      ),
+    ]);
+    const table = await convertToTable([], new DummyEmbedding(), { schema });
+    expect(DataType.isFixedSizeList(table.getChild("vector")?.type)).toBe(true);
+    expect(table.getChild("vector")?.type.children[0].type.toString()).toEqual(
+      new Float16().toString(),
+    );
+  });
+
+  it("will complain if embeddings present but schema missing embedding column", async function () {
+    const schema = new Schema([new Field("string", new Utf8(), false)]);
+    await expect(
+      convertToTable([], new DummyEmbedding(), { schema }),
+    ).rejects.toThrow("column vector was missing");
+  });
+
+  it("will provide a nice error if run twice", async function () {
+    const records = sampleRecords();
+    const table = await convertToTable(records, new DummyEmbedding());
+    // fromTableToBuffer will try and apply the embeddings again
+    await expect(
+      fromTableToBuffer(table, new DummyEmbedding()),
+    ).rejects.toThrow("already existed");
+  });
+});
+
+describe("makeEmptyTable", function () {
+  it("will make an empty table", async function () {
+    await checkTableCreation(
+      async (_, __, schema) => makeEmptyTable(schema),
+      false,
+    );
+  });
+});
+
+describe("when using two versions of arrow", function () {
+  it("can still import data", async function () {
+    const schema = new OldSchema([
+      new OldField("id", new OldInt32()),
+      new OldField(
+        "vector",
+        new OldFixedSizeList(
+          1024,
+          new OldField("item", new OldFloat32(), true),
+        ),
+      ),
+      new OldField(
+        "struct",
+        new OldStruct([
+          new OldField(
+            "nested",
+            new OldDictionary(new OldUtf8(), new OldInt32(), 1, true),
+          ),
+          new OldField("ts_with_tz", new OldTimestampNanosecond("some_tz")),
+          new OldField("ts_no_tz", new OldTimestampNanosecond(null)),
+        ]),
+      ),
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    ]) as any;
+    schema.metadataVersion = MetadataVersion.V5;
+    const table = makeArrowTable([], { schema });
+
+    const buf = await fromTableToBuffer(table);
+    expect(buf.byteLength).toBeGreaterThan(0);
+    const actual = tableFromIPC(buf);
+    const actualSchema = actual.schema;
+    expect(actualSchema.fields.length).toBe(3);
+
+    // Deep equality gets hung up on some very minor unimportant differences
+    // between arrow version 13 and 15 which isn't really what we're testing for
+    // and so we do our own comparison that just checks name/type/nullability
+    function compareFields(lhs: Field, rhs: Field) {
+      expect(lhs.name).toEqual(rhs.name);
+      expect(lhs.nullable).toEqual(rhs.nullable);
+      expect(lhs.typeId).toEqual(rhs.typeId);
+      if ("children" in lhs.type && lhs.type.children !== null) {
+        const lhsChildren = lhs.type.children as Field[];
+        lhsChildren.forEach((child: Field, idx) => {
+          compareFields(child, rhs.type.children[idx]);
+        });
+      }
+    }
+    actualSchema.fields.forEach((field, idx) => {
+      compareFields(field, actualSchema.fields[idx]);
+    });
+  });
+});

nodejs/__test__/connection.test.ts

@@ -0,0 +1,88 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import * as tmp from "tmp";
+
+import { Connection, connect } from "../dist/index.js";
+
+describe("when connecting", () => {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => (tmpDir = tmp.dirSync({ unsafeCleanup: true })));
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should connect", async () => {
+    const db = await connect(tmpDir.name);
+    expect(db.display()).toBe(
+      `NativeDatabase(uri=${tmpDir.name}, read_consistency_interval=None)`,
+    );
+  });
+
+  it("should allow read consistency interval to be specified", async () => {
+    const db = await connect(tmpDir.name, { readConsistencyInterval: 5 });
+    expect(db.display()).toBe(
+      `NativeDatabase(uri=${tmpDir.name}, read_consistency_interval=5s)`,
+    );
+  });
+});
+
+describe("given a connection", () => {
+  let tmpDir: tmp.DirResult;
+  let db: Connection;
+  beforeEach(async () => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+    db = await connect(tmpDir.name);
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should raise an error if opening a non-existent table", async () => {
+    await expect(db.openTable("non-existent")).rejects.toThrow("was not found");
+  });
+
+  it("should raise an error if any operation is tried after it is closed", async () => {
+    expect(db.isOpen()).toBe(true);
+    await db.close();
+    expect(db.isOpen()).toBe(false);
+    await expect(db.tableNames()).rejects.toThrow("Connection is closed");
+  });
+
+  it("should fail if creating table twice, unless overwrite is true", async () => {
+    let tbl = await db.createTable("test", [{ id: 1 }, { id: 2 }]);
+    await expect(tbl.countRows()).resolves.toBe(2);
+    await expect(
+      db.createTable("test", [{ id: 1 }, { id: 2 }]),
+    ).rejects.toThrow();
+    tbl = await db.createTable("test", [{ id: 3 }], { mode: "overwrite" });
+    await expect(tbl.countRows()).resolves.toBe(1);
+  });
+
+  it("should respect limit and page token when listing tables", async () => {
+    const db = await connect(tmpDir.name);
+
+    await db.createTable("b", [{ id: 1 }]);
+    await db.createTable("a", [{ id: 1 }]);
+    await db.createTable("c", [{ id: 1 }]);
+
+    let tables = await db.tableNames();
+    expect(tables).toEqual(["a", "b", "c"]);
+
+    tables = await db.tableNames({ limit: 1 });
+    expect(tables).toEqual(["a"]);
+
+    tables = await db.tableNames({ limit: 1, startAfter: "a" });
+    expect(tables).toEqual(["b"]);
+
+    tables = await db.tableNames({ startAfter: "a" });
+    expect(tables).toEqual(["b", "c"]);
+  });
+});

nodejs/__test__/table.test.ts

@@ -0,0 +1,350 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import * as fs from "fs";
+import * as path from "path";
+import * as tmp from "tmp";
+
+import { Table, connect } from "../dist";
+import {
+  Schema,
+  Field,
+  Float32,
+  Int32,
+  FixedSizeList,
+  Int64,
+  Float64,
+} from "apache-arrow";
+import { makeArrowTable } from "../dist/arrow";
+
+describe("Given a table", () => {
+  let tmpDir: tmp.DirResult;
+  let table: Table;
+  const schema = new Schema([new Field("id", new Float64(), true)]);
+  beforeEach(async () => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+    const conn = await connect(tmpDir.name);
+    table = await conn.createEmptyTable("some_table", schema);
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("be displayable", async () => {
+    expect(table.display()).toMatch(
+      /NativeTable\(some_table, uri=.*, read_consistency_interval=None\)/,
+    );
+    table.close();
+    expect(table.display()).toBe("ClosedTable(some_table)");
+  });
+
+  it("should let me add data", async () => {
+    await table.add([{ id: 1 }, { id: 2 }]);
+    await table.add([{ id: 1 }]);
+    await expect(table.countRows()).resolves.toBe(3);
+  });
+
+  it("should overwrite data if asked", async () => {
+    await table.add([{ id: 1 }, { id: 2 }]);
+    await table.add([{ id: 1 }], { mode: "overwrite" });
+    await expect(table.countRows()).resolves.toBe(1);
+  });
+
+  it("should let me close the table", async () => {
+    expect(table.isOpen()).toBe(true);
+    table.close();
+    expect(table.isOpen()).toBe(false);
+    expect(table.countRows()).rejects.toThrow("Table some_table is closed");
+  });
+});
+
+describe("Test creating index", () => {
+  let tmpDir: tmp.DirResult;
+  const schema = new Schema([
+    new Field("id", new Int32(), true),
+    new Field("vec", new FixedSizeList(32, new Field("item", new Float32()))),
+  ]);
+
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  test("create vector index with no column", async () => {
+    const db = await connect(tmpDir.name);
+    const data = makeArrowTable(
+      Array(300)
+        .fill(1)
+        .map((_, i) => ({
+          id: i,
+          vec: Array(32)
+            .fill(1)
+            .map(() => Math.random()),
+        })),
+      {
+        schema,
+      },
+    );
+    const tbl = await db.createTable("test", data);
+    await tbl.createIndex().build();
+
+    // check index directory
+    const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
+    expect(fs.readdirSync(indexDir)).toHaveLength(1);
+    // TODO: check index type.
+
+    // Search without specifying the column
+    const queryVector = data.toArray()[5].vec.toJSON();
+    const rst = await tbl.query().nearestTo(queryVector).limit(2).toArrow();
+    expect(rst.numRows).toBe(2);
+
+    // Search with specifying the column
+    const rst2 = await tbl.search(queryVector, "vec").limit(2).toArrow();
+    expect(rst2.numRows).toBe(2);
+    expect(rst.toString()).toEqual(rst2.toString());
+  });
+
+  test("no vector column available", async () => {
+    const db = await connect(tmpDir.name);
+    const tbl = await db.createTable(
+      "no_vec",
+      makeArrowTable([
+        { id: 1, val: 2 },
+        { id: 2, val: 3 },
+      ]),
+    );
+    await expect(tbl.createIndex().build()).rejects.toThrow(
+      "No vector column found",
+    );
+
+    await tbl.createIndex("val").build();
+    const indexDir = path.join(tmpDir.name, "no_vec.lance", "_indices");
+    expect(fs.readdirSync(indexDir)).toHaveLength(1);
+
+    for await (const r of tbl.query().filter("id > 1").select(["id"])) {
+      expect(r.numRows).toBe(1);
+    }
+  });
+
+  test("two columns with different dimensions", async () => {
+    const db = await connect(tmpDir.name);
+    const schema = new Schema([
+      new Field("id", new Int32(), true),
+      new Field("vec", new FixedSizeList(32, new Field("item", new Float32()))),
+      new Field(
+        "vec2",
+        new FixedSizeList(64, new Field("item", new Float32())),
+      ),
+    ]);
+    const tbl = await db.createTable(
+      "two_vectors",
+      makeArrowTable(
+        Array(300)
+          .fill(1)
+          .map((_, i) => ({
+            id: i,
+            vec: Array(32)
+              .fill(1)
+              .map(() => Math.random()),
+            vec2: Array(64) // different dimension
+              .fill(1)
+              .map(() => Math.random()),
+          })),
+        { schema },
+      ),
+    );
+
+    // Only build index over v1
+    await expect(tbl.createIndex().build()).rejects.toThrow(
+      /.*More than one vector columns found.*/,
+    );
+    tbl
+      .createIndex("vec")
+      // eslint-disable-next-line @typescript-eslint/naming-convention
+      .ivf_pq({ num_partitions: 2, num_sub_vectors: 2 })
+      .build();
+
+    const rst = await tbl
+      .query()
+      .nearestTo(
+        Array(32)
+          .fill(1)
+          .map(() => Math.random()),
+      )
+      .limit(2)
+      .toArrow();
+    expect(rst.numRows).toBe(2);
+
+    // Search with specifying the column
+    await expect(
+      tbl
+        .search(
+          Array(64)
+            .fill(1)
+            .map(() => Math.random()),
+          "vec",
+        )
+        .limit(2)
+        .toArrow(),
+    ).rejects.toThrow(/.*does not match the dimension.*/);
+
+    const query64 = Array(64)
+      .fill(1)
+      .map(() => Math.random());
+    const rst64Query = await tbl.query().nearestTo(query64).limit(2).toArrow();
+    const rst64Search = await tbl.search(query64, "vec2").limit(2).toArrow();
+    expect(rst64Query.toString()).toEqual(rst64Search.toString());
+    expect(rst64Query.numRows).toBe(2);
+  });
+
+  test("create scalar index", async () => {
+    const db = await connect(tmpDir.name);
+    const data = makeArrowTable(
+      Array(300)
+        .fill(1)
+        .map((_, i) => ({
+          id: i,
+          vec: Array(32)
+            .fill(1)
+            .map(() => Math.random()),
+        })),
+      {
+        schema,
+      },
+    );
+    const tbl = await db.createTable("test", data);
+    await tbl.createIndex("id").build();
+
+    // check index directory
+    const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
+    expect(fs.readdirSync(indexDir)).toHaveLength(1);
+    // TODO: check index type.
+  });
+});
+
+describe("Read consistency interval", () => {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  // const intervals = [undefined, 0, 0.1];
+  const intervals = [0];
+  test.each(intervals)("read consistency interval %p", async (interval) => {
+    const db = await connect(tmpDir.name);
+    const table = await db.createTable("my_table", [{ id: 1 }]);
+
+    const db2 = await connect(tmpDir.name, {
+      readConsistencyInterval: interval,
+    });
+    const table2 = await db2.openTable("my_table");
+    expect(await table2.countRows()).toEqual(await table.countRows());
+
+    await table.add([{ id: 2 }]);
+
+    if (interval === undefined) {
+      expect(await table2.countRows()).toEqual(1);
+      // TODO: once we implement time travel we can uncomment this part of the test.
+      // await table2.checkout_latest();
+      // expect(await table2.countRows()).toEqual(2);
+    } else if (interval === 0) {
+      expect(await table2.countRows()).toEqual(2);
+    } else {
+      // interval == 0.1
+      expect(await table2.countRows()).toEqual(1);
+      await new Promise((r) => setTimeout(r, 100));
+      expect(await table2.countRows()).toEqual(2);
+    }
+  });
+});
+
+describe("schema evolution", function () {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => {
+    tmpDir.removeCallback();
+  });
+
+  // Create a new sample table
+  it("can add a new column to the schema", async function () {
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+
+    await table.addColumns([
+      { name: "price", valueSql: "cast(10.0 as float)" },
+    ]);
+
+    const expectedSchema = new Schema([
+      new Field("id", new Int64(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float32(), false),
+    ]);
+    expect(await table.schema()).toEqual(expectedSchema);
+  });
+
+  it("can alter the columns in the schema", async function () {
+    const con = await connect(tmpDir.name);
+    const schema = new Schema([
+      new Field("id", new Int64(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float64(), false),
+    ]);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+    // Can create a non-nullable column only through addColumns at the moment.
+    await table.addColumns([
+      { name: "price", valueSql: "cast(10.0 as double)" },
+    ]);
+    expect(await table.schema()).toEqual(schema);
+
+    await table.alterColumns([
+      { path: "id", rename: "new_id" },
+      { path: "price", nullable: true },
+    ]);
+
+    const expectedSchema = new Schema([
+      new Field("new_id", new Int64(), true),
+      new Field(
+        "vector",
+        new FixedSizeList(2, new Field("item", new Float32(), true)),
+        true,
+      ),
+      new Field("price", new Float64(), true),
+    ]);
+    expect(await table.schema()).toEqual(expectedSchema);
+  });
+
+  it("can drop a column from the schema", async function () {
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+    await table.dropColumns(["vector"]);
+
+    const expectedSchema = new Schema([new Field("id", new Int64(), true)]);
+    expect(await table.schema()).toEqual(expectedSchema);
+  });
+});

nodejs/__test__/tsconfig.json

@@ -0,0 +1,10 @@
+{
+  "extends": "../tsconfig.json",
+  "compilerOptions": {
+    "outDir": "./dist/spec",
+    "module": "commonjs",
+    "target": "es2022",
+    "types": ["jest", "node"]
+  },
+  "include": ["**/*"]
+}

nodejs/build.rs

@@ -0,0 +1,5 @@
+extern crate napi_build;
+
+fn main() {
+    napi_build::setup();
+}

nodejs/eslint.config.js

@@ -0,0 +1,17 @@
+/* eslint-disable @typescript-eslint/naming-convention */
+// @ts-check
+
+const eslint = require("@eslint/js");
+const tseslint = require("typescript-eslint");
+const eslintConfigPrettier = require("eslint-config-prettier");
+
+module.exports = tseslint.config(
+  eslint.configs.recommended,
+  eslintConfigPrettier,
+  ...tseslint.configs.recommended,
+  {
+    rules: {
+      "@typescript-eslint/naming-convention": "error",
+    },
+  },
+);

nodejs/jest.config.js

@@ -0,0 +1,7 @@
+/** @type {import('ts-jest').JestConfigWithTsJest} */
+module.exports = {
+  preset: "ts-jest",
+  testEnvironment: "node",
+  moduleDirectories: ["node_modules", "./dist"],
+  moduleFileExtensions: ["js", "ts"],
+};

nodejs/lancedb/arrow.ts

@@ -0,0 +1,634 @@
+// Copyright 2023 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import {
+  Field,
+  makeBuilder,
+  RecordBatchFileWriter,
+  Utf8,
+  type Vector,
+  FixedSizeList,
+  vectorFromArray,
+  type Schema,
+  Table as ArrowTable,
+  RecordBatchStreamWriter,
+  List,
+  RecordBatch,
+  makeData,
+  Struct,
+  type Float,
+  DataType,
+  Binary,
+  Float32,
+} from "apache-arrow";
+import { type EmbeddingFunction } from "./embedding/embedding_function";
+import { sanitizeSchema } from "./sanitize";
+
+/** Data type accepted by NodeJS SDK */
+export type Data = Record<string, unknown>[] | ArrowTable;
+
+/*
+ * Options to control how a column should be converted to a vector array
+ */
+export class VectorColumnOptions {
+  /** Vector column type. */
+  type: Float = new Float32();
+
+  constructor(values?: Partial<VectorColumnOptions>) {
+    Object.assign(this, values);
+  }
+}
+
+/** Options to control the makeArrowTable call. */
+export class MakeArrowTableOptions {
+  /*
+   * Schema of the data.
+   *
+   * If this is not provided then the data type will be inferred from the
+   * JS type.  Integer numbers will become int64, floating point numbers
+   * will become float64 and arrays will become variable sized lists with
+   * the data type inferred from the first element in the array.
+   *
+   * The schema must be specified if there are no records (e.g. to make
+   * an empty table)
+   */
+  schema?: Schema;
+
+  /*
+   * Mapping from vector column name to expected type
+   *
+   * Lance expects vector columns to be fixed size list arrays (i.e. tensors)
+   * However, `makeArrowTable` will not infer this by default (it creates
+   * variable size list arrays).  This field can be used to indicate that a column
+   * should be treated as a vector column and converted to a fixed size list.
+   *
+   * The keys should be the names of the vector columns.  The value specifies the
+   * expected data type of the vector columns.
+   *
+   * If `schema` is provided then this field is ignored.
+   *
+   * By default, the column named "vector" will be assumed to be a float32
+   * vector column.
+   */
+  vectorColumns: Record<string, VectorColumnOptions> = {
+    vector: new VectorColumnOptions(),
+  };
+
+  /**
+   * If true then string columns will be encoded with dictionary encoding
+   *
+   * Set this to true if your string columns tend to repeat the same values
+   * often.  For more precise control use the `schema` property to specify the
+   * data type for individual columns.
+   *
+   * If `schema` is provided then this property is ignored.
+   */
+  dictionaryEncodeStrings: boolean = false;
+
+  constructor(values?: Partial<MakeArrowTableOptions>) {
+    Object.assign(this, values);
+  }
+}
+
+/**
+ * An enhanced version of the {@link makeTable} function from Apache Arrow
+ * that supports nested fields and embeddings columns.
+ *
+ * This function converts an array of Record<String, any> (row-major JS objects)
+ * to an Arrow Table (a columnar structure)
+ *
+ * Note that it currently does not support nulls.
+ *
+ * If a schema is provided then it will be used to determine the resulting array
+ * types.  Fields will also be reordered to fit the order defined by the schema.
+ *
+ * If a schema is not provided then the types will be inferred and the field order
+ * will be controlled by the order of properties in the first record.  If a type
+ * is inferred it will always be nullable.
+ *
+ * If the input is empty then a schema must be provided to create an empty table.
+ *
+ * When a schema is not specified then data types will be inferred.  The inference
+ * rules are as follows:
+ *
+ *  - boolean => Bool
+ *  - number => Float64
+ *  - String => Utf8
+ *  - Buffer => Binary
+ *  - Record<String, any> => Struct
+ *  - Array<any> => List
+ *
+ * @param data input data
+ * @param options options to control the makeArrowTable call.
+ *
+ * @example
+ *
+ * ```ts
+ *
+ * import { fromTableToBuffer, makeArrowTable } from "../arrow";
+ * import { Field, FixedSizeList, Float16, Float32, Int32, Schema } from "apache-arrow";
+ *
+ * const schema = new Schema([
+ *   new Field("a", new Int32()),
+ *   new Field("b", new Float32()),
+ *   new Field("c", new FixedSizeList(3, new Field("item", new Float16()))),
+ *  ]);
+ *  const table = makeArrowTable([
+ *    { a: 1, b: 2, c: [1, 2, 3] },
+ *    { a: 4, b: 5, c: [4, 5, 6] },
+ *    { a: 7, b: 8, c: [7, 8, 9] },
+ *  ], { schema });
+ * ```
+ *
+ * By default it assumes that the column named `vector` is a vector column
+ * and it will be converted into a fixed size list array of type float32.
+ * The `vectorColumns` option can be used to support other vector column
+ * names and data types.
+ *
+ * ```ts
+ *
+ * const schema = new Schema([
+    new Field("a", new Float64()),
+    new Field("b", new Float64()),
+    new Field(
+      "vector",
+      new FixedSizeList(3, new Field("item", new Float32()))
+    ),
+  ]);
+  const table = makeArrowTable([
+    { a: 1, b: 2, vector: [1, 2, 3] },
+    { a: 4, b: 5, vector: [4, 5, 6] },
+    { a: 7, b: 8, vector: [7, 8, 9] },
+  ]);
+  assert.deepEqual(table.schema, schema);
+ * ```
+ *
+ * You can specify the vector column types and names using the options as well
+ *
+ * ```typescript
+ *
+ * const schema = new Schema([
+    new Field('a', new Float64()),
+    new Field('b', new Float64()),
+    new Field('vec1', new FixedSizeList(3, new Field('item', new Float16()))),
+    new Field('vec2', new FixedSizeList(3, new Field('item', new Float16())))
+  ]);
+ * const table = makeArrowTable([
+    { a: 1, b: 2, vec1: [1, 2, 3], vec2: [2, 4, 6] },
+    { a: 4, b: 5, vec1: [4, 5, 6], vec2: [8, 10, 12] },
+    { a: 7, b: 8, vec1: [7, 8, 9], vec2: [14, 16, 18] }
+  ], {
+    vectorColumns: {
+      vec1: { type: new Float16() },
+      vec2: { type: new Float16() }
+    }
+  }
+ * assert.deepEqual(table.schema, schema)
+ * ```
+ */
+export function makeArrowTable(
+  data: Array<Record<string, unknown>>,
+  options?: Partial<MakeArrowTableOptions>,
+): ArrowTable {
+  if (
+    data.length === 0 &&
+    (options?.schema === undefined || options?.schema === null)
+  ) {
+    throw new Error("At least one record or a schema needs to be provided");
+  }
+
+  const opt = new MakeArrowTableOptions(options !== undefined ? options : {});
+  if (opt.schema !== undefined && opt.schema !== null) {
+    opt.schema = sanitizeSchema(opt.schema);
+  }
+  const columns: Record<string, Vector> = {};
+  // TODO: sample dataset to find missing columns
+  // Prefer the field ordering of the schema, if present
+  const columnNames =
+    opt.schema != null ? (opt.schema.names as string[]) : Object.keys(data[0]);
+  for (const colName of columnNames) {
+    if (
+      data.length !== 0 &&
+      !Object.prototype.hasOwnProperty.call(data[0], colName)
+    ) {
+      // The field is present in the schema, but not in the data, skip it
+      continue;
+    }
+    // Extract a single column from the records (transpose from row-major to col-major)
+    let values = data.map((datum) => datum[colName]);
+
+    // By default (type === undefined) arrow will infer the type from the JS type
+    let type;
+    if (opt.schema !== undefined) {
+      // If there is a schema provided, then use that for the type instead
+      type = opt.schema?.fields.filter((f) => f.name === colName)[0]?.type;
+      if (DataType.isInt(type) && type.bitWidth === 64) {
+        // wrap in BigInt to avoid bug: https://github.com/apache/arrow/issues/40051
+        values = values.map((v) => {
+          if (v === null) {
+            return v;
+          }
+          if (typeof v === "bigint") {
+            return v;
+          }
+          if (typeof v === "number") {
+            return BigInt(v);
+          }
+          throw new Error(
+            `Expected BigInt or number for column ${colName}, got ${typeof v}`,
+          );
+        });
+      }
+    } else {
+      // Otherwise, check to see if this column is one of the vector columns
+      // defined by opt.vectorColumns and, if so, use the fixed size list type
+      const vectorColumnOptions = opt.vectorColumns[colName];
+      if (vectorColumnOptions !== undefined) {
+        const firstNonNullValue = values.find((v) => v !== null);
+        if (Array.isArray(firstNonNullValue)) {
+          type = newVectorType(
+            firstNonNullValue.length,
+            vectorColumnOptions.type,
+          );
+        } else {
+          throw new Error(
+            `Column ${colName} is expected to be a vector column but first non-null value is not an array.  Could not determine size of vector column`,
+          );
+        }
+      }
+    }
+
+    try {
+      // Convert an Array of JS values to an arrow vector
+      columns[colName] = makeVector(values, type, opt.dictionaryEncodeStrings);
+    } catch (error: unknown) {
+      // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+      throw Error(`Could not convert column "${colName}" to Arrow: ${error}`);
+    }
+  }
+
+  if (opt.schema != null) {
+    // `new ArrowTable(columns)` infers a schema which may sometimes have
+    // incorrect nullability (it assumes nullable=true always)
+    //
+    // `new ArrowTable(schema, columns)` will also fail because it will create a
+    // batch with an inferred schema and then complain that the batch schema
+    // does not match the provided schema.
+    //
+    // To work around this we first create a table with the wrong schema and
+    // then patch the schema of the batches so we can use
+    // `new ArrowTable(schema, batches)` which does not do any schema inference
+    const firstTable = new ArrowTable(columns);
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const batchesFixed = firstTable.batches.map(
+      (batch) => new RecordBatch(opt.schema!, batch.data),
+    );
+    return new ArrowTable(opt.schema, batchesFixed);
+  } else {
+    return new ArrowTable(columns);
+  }
+}
+
+/**
+ * Create an empty Arrow table with the provided schema
+ */
+export function makeEmptyTable(schema: Schema): ArrowTable {
+  return makeArrowTable([], { schema });
+}
+
+// Helper function to convert Array<Array<any>> to a variable sized list array
+// @ts-expect-error (Vector<unknown> is not assignable to Vector<any>)
+function makeListVector(lists: unknown[][]): Vector<unknown> {
+  if (lists.length === 0 || lists[0].length === 0) {
+    throw Error("Cannot infer list vector from empty array or empty list");
+  }
+  const sampleList = lists[0];
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  let inferredType: any;
+  try {
+    const sampleVector = makeVector(sampleList);
+    inferredType = sampleVector.type;
+  } catch (error: unknown) {
+    // eslint-disable-next-line @typescript-eslint/restrict-template-expressions
+    throw Error(`Cannot infer list vector.  Cannot infer inner type: ${error}`);
+  }
+
+  const listBuilder = makeBuilder({
+    type: new List(new Field("item", inferredType, true)),
+  });
+  for (const list of lists) {
+    listBuilder.append(list);
+  }
+  return listBuilder.finish().toVector();
+}
+
+// Helper function to convert an Array of JS values to an Arrow Vector
+function makeVector(
+  values: unknown[],
+  type?: DataType,
+  stringAsDictionary?: boolean,
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+): Vector<any> {
+  if (type !== undefined) {
+    // No need for inference, let Arrow create it
+    return vectorFromArray(values, type);
+  }
+  if (values.length === 0) {
+    throw Error(
+      "makeVector requires at least one value or the type must be specfied",
+    );
+  }
+  const sampleValue = values.find((val) => val !== null && val !== undefined);
+  if (sampleValue === undefined) {
+    throw Error(
+      "makeVector cannot infer the type if all values are null or undefined",
+    );
+  }
+  if (Array.isArray(sampleValue)) {
+    // Default Arrow inference doesn't handle list types
+    return makeListVector(values as unknown[][]);
+  } else if (Buffer.isBuffer(sampleValue)) {
+    // Default Arrow inference doesn't handle Buffer
+    return vectorFromArray(values, new Binary());
+  } else if (
+    !(stringAsDictionary ?? false) &&
+    (typeof sampleValue === "string" || sampleValue instanceof String)
+  ) {
+    // If the type is string then don't use Arrow's default inference unless dictionaries are requested
+    // because it will always use dictionary encoding for strings
+    return vectorFromArray(values, new Utf8());
+  } else {
+    // Convert a JS array of values to an arrow vector
+    return vectorFromArray(values);
+  }
+}
+
+async function applyEmbeddings<T>(
+  table: ArrowTable,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<ArrowTable> {
+  if (embeddings == null) {
+    return table;
+  }
+
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+
+  // Convert from ArrowTable to Record<String, Vector>
+  const colEntries = [...Array(table.numCols).keys()].map((_, idx) => {
+    const name = table.schema.fields[idx].name;
+    // eslint-disable-next-line @typescript-eslint/no-non-null-assertion
+    const vec = table.getChildAt(idx)!;
+    return [name, vec];
+  });
+  const newColumns = Object.fromEntries(colEntries);
+
+  const sourceColumn = newColumns[embeddings.sourceColumn];
+  const destColumn = embeddings.destColumn ?? "vector";
+  const innerDestType = embeddings.embeddingDataType ?? new Float32();
+  if (sourceColumn === undefined) {
+    throw new Error(
+      `Cannot apply embedding function because the source column '${embeddings.sourceColumn}' was not present in the data`,
+    );
+  }
+
+  if (table.numRows === 0) {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      // We have an empty table and it already has the embedding column so no work needs to be done
+      // Note: we don't return an error like we did below because this is a common occurrence.  For example,
+      // if we call convertToTable with 0 records and a schema that includes the embedding
+      return table;
+    }
+    if (embeddings.embeddingDimension !== undefined) {
+      const destType = newVectorType(
+        embeddings.embeddingDimension,
+        innerDestType,
+      );
+      newColumns[destColumn] = makeVector([], destType);
+    } else if (schema != null) {
+      const destField = schema.fields.find((f) => f.name === destColumn);
+      if (destField != null) {
+        newColumns[destColumn] = makeVector([], destField.type);
+      } else {
+        throw new Error(
+          `Attempt to apply embeddings to an empty table failed because schema was missing embedding column '${destColumn}'`,
+        );
+      }
+    } else {
+      throw new Error(
+        "Attempt to apply embeddings to an empty table when the embeddings function does not specify `embeddingDimension`",
+      );
+    }
+  } else {
+    if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
+      throw new Error(
+        `Attempt to apply embeddings to table failed because column ${destColumn} already existed`,
+      );
+    }
+    if (table.batches.length > 1) {
+      throw new Error(
+        "Internal error: `makeArrowTable` unexpectedly created a table with more than one batch",
+      );
+    }
+    const values = sourceColumn.toArray();
+    const vectors = await embeddings.embed(values as T[]);
+    if (vectors.length !== values.length) {
+      throw new Error(
+        "Embedding function did not return an embedding for each input element",
+      );
+    }
+    const destType = newVectorType(vectors[0].length, innerDestType);
+    newColumns[destColumn] = makeVector(vectors, destType);
+  }
+
+  const newTable = new ArrowTable(newColumns);
+  if (schema != null) {
+    if (schema.fields.find((f) => f.name === destColumn) === undefined) {
+      throw new Error(
+        `When using embedding functions and specifying a schema the schema should include the embedding column but the column ${destColumn} was missing`,
+      );
+    }
+    return alignTable(newTable, schema);
+  }
+  return newTable;
+}
+
+/*
+ * Convert an Array of records into an Arrow Table, optionally applying an
+ * embeddings function to it.
+ *
+ * This function calls `makeArrowTable` first to create the Arrow Table.
+ * Any provided `makeTableOptions` (e.g. a schema) will be passed on to
+ * that call.
+ *
+ * The embedding function will be passed a column of values (based on the
+ * `sourceColumn` of the embedding function) and expects to receive back
+ * number[][] which will be converted into a fixed size list column.  By
+ * default this will be a fixed size list of Float32 but that can be
+ * customized by the `embeddingDataType` property of the embedding function.
+ *
+ * If a schema is provided in `makeTableOptions` then it should include the
+ * embedding columns.  If no schema is provded then embedding columns will
+ * be placed at the end of the table, after all of the input columns.
+ */
+export async function convertToTable<T>(
+  data: Array<Record<string, unknown>>,
+  embeddings?: EmbeddingFunction<T>,
+  makeTableOptions?: Partial<MakeArrowTableOptions>,
+): Promise<ArrowTable> {
+  const table = makeArrowTable(data, makeTableOptions);
+  return await applyEmbeddings(table, embeddings, makeTableOptions?.schema);
+}
+
+// Creates the Arrow Type for a Vector column with dimension `dim`
+function newVectorType<T extends Float>(
+  dim: number,
+  innerType: T,
+): FixedSizeList<T> {
+  // in Lance we always default to have the elements nullable, so we need to set it to true
+  // otherwise we often get schema mismatches because the stored data always has schema with nullable elements
+  const children = new Field<T>("item", innerType, true);
+  return new FixedSizeList(dim, children);
+}
+
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC File serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
+export async function fromRecordsToBuffer<T>(
+  data: Array<Record<string, unknown>>,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  const table = await convertToTable(data, embeddings, { schema });
+  const writer = RecordBatchFileWriter.writeAll(table);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+/**
+ * Serialize an Array of records into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will call `convertToTable` and pass on `embeddings` and `schema`
+ *
+ * `schema` is required if data is empty
+ */
+export async function fromRecordsToStreamBuffer<T>(
+  data: Array<Record<string, unknown>>,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  const table = await convertToTable(data, embeddings, { schema });
+  const writer = RecordBatchStreamWriter.writeAll(table);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC File serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
+export async function fromTableToBuffer<T>(
+  table: ArrowTable,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema);
+  const writer = RecordBatchFileWriter.writeAll(tableWithEmbeddings);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+export async function fromDataToBuffer<T>(
+  data: Data,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  if (schema !== undefined && schema !== null) {
+    schema = sanitizeSchema(schema);
+  }
+  if (data instanceof ArrowTable) {
+    return fromTableToBuffer(data, embeddings, schema);
+  } else {
+    const table = await convertToTable(data);
+    return fromTableToBuffer(table, embeddings, schema);
+  }
+}
+
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC Stream serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
+export async function fromTableToStreamBuffer<T>(
+  table: ArrowTable,
+  embeddings?: EmbeddingFunction<T>,
+  schema?: Schema,
+): Promise<Buffer> {
+  const tableWithEmbeddings = await applyEmbeddings(table, embeddings, schema);
+  const writer = RecordBatchStreamWriter.writeAll(tableWithEmbeddings);
+  return Buffer.from(await writer.toUint8Array());
+}
+
+function alignBatch(batch: RecordBatch, schema: Schema): RecordBatch {
+  const alignedChildren = [];
+  for (const field of schema.fields) {
+    const indexInBatch = batch.schema.fields?.findIndex(
+      (f) => f.name === field.name,
+    );
+    if (indexInBatch < 0) {
+      throw new Error(
+        `The column ${field.name} was not found in the Arrow Table`,
+      );
+    }
+    alignedChildren.push(batch.data.children[indexInBatch]);
+  }
+  const newData = makeData({
+    type: new Struct(schema.fields),
+    length: batch.numRows,
+    nullCount: batch.nullCount,
+    children: alignedChildren,
+  });
+  return new RecordBatch(schema, newData);
+}
+
+function alignTable(table: ArrowTable, schema: Schema): ArrowTable {
+  const alignedBatches = table.batches.map((batch) =>
+    alignBatch(batch, schema),
+  );
+  return new ArrowTable(schema, alignedBatches);
+}
+
+// Creates an empty Arrow Table
+export function createEmptyTable(schema: Schema): ArrowTable {
+  return new ArrowTable(sanitizeSchema(schema));
+}

nodejs/lancedb/connection.ts

@@ -0,0 +1,177 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { fromTableToBuffer, makeArrowTable, makeEmptyTable } from "./arrow";
+import { Connection as LanceDbConnection } from "./native";
+import { Table } from "./table";
+import { Table as ArrowTable, Schema } from "apache-arrow";
+
+export interface CreateTableOptions {
+  /**
+   * The mode to use when creating the table.
+   *
+   * If this is set to "create" and the table already exists then either
+   * an error will be thrown or, if existOk is true, then nothing will
+   * happen.  Any provided data will be ignored.
+   *
+   * If this is set to "overwrite" then any existing table will be replaced.
+   */
+  mode: "create" | "overwrite";
+  /**
+   * If this is true and the table already exists and the mode is "create"
+   * then no error will be raised.
+   */
+  existOk: boolean;
+}
+
+export interface TableNamesOptions {
+  /**
+   * If present, only return names that come lexicographically after the
+   * supplied value.
+   *
+   * This can be combined with limit to implement pagination by setting this to
+   * the last table name from the previous page.
+   */
+  startAfter?: string;
+  /** An optional limit to the number of results to return. */
+  limit?: number;
+}
+
+/**
+ * A LanceDB Connection that allows you to open tables and create new ones.
+ *
+ * Connection could be local against filesystem or remote against a server.
+ *
+ * A Connection is intended to be a long lived object and may hold open
+ * resources such as HTTP connection pools.  This is generally fine and
+ * a single connection should be shared if it is going to be used many
+ * times. However, if you are finished with a connection, you may call
+ * close to eagerly free these resources.  Any call to a Connection
+ * method after it has been closed will result in an error.
+ *
+ * Closing a connection is optional.  Connections will automatically
+ * be closed when they are garbage collected.
+ *
+ * Any created tables are independent and will continue to work even if
+ * the underlying connection has been closed.
+ */
+export class Connection {
+  readonly inner: LanceDbConnection;
+
+  constructor(inner: LanceDbConnection) {
+    this.inner = inner;
+  }
+
+  /** Return true if the connection has not been closed */
+  isOpen(): boolean {
+    return this.inner.isOpen();
+  }
+
+  /** Close the connection, releasing any underlying resources.
+   *
+   * It is safe to call this method multiple times.
+   *
+   * Any attempt to use the connection after it is closed will result in an error.
+   */
+  close(): void {
+    this.inner.close();
+  }
+
+  /** Return a brief description of the connection */
+  display(): string {
+    return this.inner.display();
+  }
+
+  /** List all the table names in this database.
+   *
+   * Tables will be returned in lexicographical order.
+   *
+   * @param options Optional parameters to control the listing.
+   */
+  async tableNames(options?: Partial<TableNamesOptions>): Promise<string[]> {
+    return this.inner.tableNames(options?.startAfter, options?.limit);
+  }
+
+  /**
+   * Open a table in the database.
+   *
+   * @param name The name of the table.
+   * @param embeddings An embedding function to use on this table
+   */
+  async openTable(name: string): Promise<Table> {
+    const innerTable = await this.inner.openTable(name);
+    return new Table(innerTable);
+  }
+
+  /**
+   * Creates a new Table and initialize it with new data.
+   *
+   * @param {string} name - The name of the table.
+   * @param data - Non-empty Array of Records to be inserted into the table
+   */
+  async createTable(
+    name: string,
+    data: Record<string, unknown>[] | ArrowTable,
+    options?: Partial<CreateTableOptions>,
+  ): Promise<Table> {
+    let mode: string = options?.mode ?? "create";
+    const existOk = options?.existOk ?? false;
+
+    if (mode === "create" && existOk) {
+      mode = "exist_ok";
+    }
+
+    let table: ArrowTable;
+    if (data instanceof ArrowTable) {
+      table = data;
+    } else {
+      table = makeArrowTable(data);
+    }
+    const buf = await fromTableToBuffer(table);
+    const innerTable = await this.inner.createTable(name, buf, mode);
+    return new Table(innerTable);
+  }
+
+  /**
+   * Creates a new empty Table
+   *
+   * @param {string} name - The name of the table.
+   * @param schema - The schema of the table
+   */
+  async createEmptyTable(
+    name: string,
+    schema: Schema,
+    options?: Partial<CreateTableOptions>,
+  ): Promise<Table> {
+    let mode: string = options?.mode ?? "create";
+    const existOk = options?.existOk ?? false;
+
+    if (mode === "create" && existOk) {
+      mode = "exist_ok";
+    }
+
+    const table = makeEmptyTable(schema);
+    const buf = await fromTableToBuffer(table);
+    const innerTable = await this.inner.createEmptyTable(name, buf, mode);
+    return new Table(innerTable);
+  }
+
+  /**
+   * Drop an existing table.
+   * @param name The name of the table to drop.
+   */
+  async dropTable(name: string): Promise<void> {
+    return this.inner.dropTable(name);
+  }
+}

nodejs/lancedb/embedding/embedding_function.ts

@@ -0,0 +1,77 @@
+// Copyright 2023 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { type Float } from "apache-arrow";
+
+/**
+ * An embedding function that automatically creates vector representation for a given column.
+ */
+export interface EmbeddingFunction<T> {
+  /**
+   * The name of the column that will be used as input for the Embedding Function.
+   */
+  sourceColumn: string;
+
+  /**
+   * The data type of the embedding
+   *
+   * The embedding function should return `number`.  This will be converted into
+   * an Arrow float array.  By default this will be Float32 but this property can
+   * be used to control the conversion.
+   */
+  embeddingDataType?: Float;
+
+  /**
+   * The dimension of the embedding
+   *
+   * This is optional, normally this can be determined by looking at the results of
+   * `embed`.  If this is not specified, and there is an attempt to apply the embedding
+   * to an empty table, then that process will fail.
+   */
+  embeddingDimension?: number;
+
+  /**
+   * The name of the column that will contain the embedding
+   *
+   * By default this is "vector"
+   */
+  destColumn?: string;
+
+  /**
+   * Should the source column be excluded from the resulting table
+   *
+   * By default the source column is included.  Set this to true and
+   * only the embedding will be stored.
+   */
+  excludeSource?: boolean;
+
+  /**
+   * Creates a vector representation for the given values.
+   */
+  embed: (data: T[]) => Promise<number[][]>;
+}
+
+export function isEmbeddingFunction<T>(
+  value: unknown,
+): value is EmbeddingFunction<T> {
+  if (typeof value !== "object" || value === null) {
+    return false;
+  }
+  if (!("sourceColumn" in value) || !("embed" in value)) {
+    return false;
+  }
+  return (
+    typeof value.sourceColumn === "string" && typeof value.embed === "function"
+  );
+}

nodejs/lancedb/embedding/openai.ts

@@ -0,0 +1,62 @@
+// Copyright 2023 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { type EmbeddingFunction } from "./embedding_function";
+import type OpenAI from "openai";
+
+export class OpenAIEmbeddingFunction implements EmbeddingFunction<string> {
+  private readonly _openai: OpenAI;
+  private readonly _modelName: string;
+
+  constructor(
+    sourceColumn: string,
+    openAIKey: string,
+    modelName: string = "text-embedding-ada-002",
+  ) {
+    /**
+     * @type {import("openai").default}
+     */
+    // eslint-disable-next-line @typescript-eslint/naming-convention
+    let Openai;
+    try {
+      // eslint-disable-next-line @typescript-eslint/no-var-requires
+      Openai = require("openai");
+    } catch {
+      throw new Error("please install openai@^4.24.1 using npm install openai");
+    }
+
+    this.sourceColumn = sourceColumn;
+    const configuration = {
+      apiKey: openAIKey,
+    };
+
+    this._openai = new Openai(configuration);
+    this._modelName = modelName;
+  }
+
+  async embed(data: string[]): Promise<number[][]> {
+    const response = await this._openai.embeddings.create({
+      model: this._modelName,
+      input: data,
+    });
+
+    const embeddings: number[][] = [];
+    for (let i = 0; i < response.data.length; i++) {
+      embeddings.push(response.data[i].embedding);
+    }
+    return embeddings;
+  }
+
+  sourceColumn: string;
+}

nodejs/lancedb/index.ts

@@ -0,0 +1,51 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { Connection } from "./connection";
+import {
+  Connection as LanceDbConnection,
+  ConnectionOptions,
+} from "./native.js";
+
+export {
+  ConnectionOptions,
+  WriteOptions,
+  Query,
+  MetricType,
+} from "./native.js";
+export { Connection } from "./connection";
+export { Table } from "./table";
+export { IvfPQOptions, IndexBuilder } from "./indexer";
+
+/**
+ * Connect to a LanceDB instance at the given URI.
+ *
+ * Accpeted formats:
+ *
+ * - `/path/to/database` - local database
+ * - `s3://bucket/path/to/database` or `gs://bucket/path/to/database` - database on cloud storage
+ * - `db://host:port` - remote database (LanceDB cloud)
+ *
+ * @param uri The uri of the database. If the database uri starts with `db://` then it connects to a remote database.
+ *
+ * @see {@link ConnectionOptions} for more details on the URI format.
+ */
+export async function connect(
+  uri: string,
+  opts?: Partial<ConnectionOptions>,
+): Promise<Connection> {
+  opts = opts ?? {};
+  const nativeConn = await LanceDbConnection.new(uri, opts);
+  return new Connection(nativeConn);
+}

nodejs/lancedb/indexer.ts

@@ -0,0 +1,105 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// TODO: Re-enable this as part of https://github.com/lancedb/lancedb/pull/1052
+/* eslint-disable @typescript-eslint/naming-convention */
+
+import {
+  MetricType,
+  IndexBuilder as NativeBuilder,
+  Table as NativeTable,
+} from "./native";
+
+/** Options to create `IVF_PQ` index */
+export interface IvfPQOptions {
+  /** Number of IVF partitions. */
+  num_partitions?: number;
+
+  /** Number of sub-vectors in PQ coding. */
+  num_sub_vectors?: number;
+
+  /** Number of bits used for each PQ code.
+   */
+  num_bits?: number;
+
+  /** Metric type to calculate the distance between vectors.
+   *
+   * Supported metrics: `L2`, `Cosine` and `Dot`.
+   */
+  metric_type?: MetricType;
+
+  /** Number of iterations to train K-means.
+   *
+   * Default is 50. The more iterations it usually yield better results,
+   * but it takes longer to train.
+   */
+  max_iterations?: number;
+
+  sample_rate?: number;
+}
+
+/**
+ * Building an index on LanceDB {@link Table}
+ *
+ * @see {@link Table.createIndex} for detailed usage.
+ */
+export class IndexBuilder {
+  private inner: NativeBuilder;
+
+  constructor(tbl: NativeTable) {
+    this.inner = tbl.createIndex();
+  }
+
+  /** Instruct the builder to build an `IVF_PQ` index */
+  ivf_pq(options?: IvfPQOptions): IndexBuilder {
+    this.inner.ivfPq(
+      options?.metric_type,
+      options?.num_partitions,
+      options?.num_sub_vectors,
+      options?.num_bits,
+      options?.max_iterations,
+      options?.sample_rate,
+    );
+    return this;
+  }
+
+  /** Instruct the builder to build a Scalar index. */
+  scalar(): IndexBuilder {
+    this.scalar();
+    return this;
+  }
+
+  /** Set the column(s) to create index on top of. */
+  column(col: string): IndexBuilder {
+    this.inner.column(col);
+    return this;
+  }
+
+  /** Set to true to replace existing index. */
+  replace(val: boolean): IndexBuilder {
+    this.inner.replace(val);
+    return this;
+  }
+
+  /** Specify the name of the index. Optional */
+  name(n: string): IndexBuilder {
+    this.inner.name(n);
+    return this;
+  }
+
+  /** Building the index. */
+  async build() {
+    await this.inner.build();
+  }
+}

nodejs/lancedb/native.d.ts

@@ -0,0 +1,133 @@
+/* tslint:disable */
+/* eslint-disable */
+
+/* auto-generated by NAPI-RS */
+
+export const enum IndexType {
+  Scalar = 0,
+  IvfPq = 1
+}
+export const enum MetricType {
+  L2 = 0,
+  Cosine = 1,
+  Dot = 2
+}
+/**
+ *  A definition of a column alteration. The alteration changes the column at
+ * `path` to have the new name `name`, to be nullable if `nullable` is true,
+ * and to have the data type `data_type`. At least one of `rename` or `nullable`
+ * must be provided.
+ */
+export interface ColumnAlteration {
+  /**
+   * The path to the column to alter. This is a dot-separated path to the column.
+   * If it is a top-level column then it is just the name of the column. If it is
+   * a nested column then it is the path to the column, e.g. "a.b.c" for a column
+   * `c` nested inside a column `b` nested inside a column `a`.
+   */
+  path: string
+  /**
+   * The new name of the column. If not provided then the name will not be changed.
+   * This must be distinct from the names of all other columns in the table.
+   */
+  rename?: string
+  /** Set the new nullability. Note that a nullable column cannot be made non-nullable. */
+  nullable?: boolean
+}
+/** A definition of a new column to add to a table. */
+export interface AddColumnsSql {
+  /** The name of the new column. */
+  name: string
+  /**
+   * The values to populate the new column with, as a SQL expression.
+   * The expression can reference other columns in the table.
+   */
+  valueSql: string
+}
+export interface ConnectionOptions {
+  apiKey?: string
+  hostOverride?: string
+  /**
+   * (For LanceDB OSS only): The interval, in seconds, at which to check for
+   * updates to the table from other processes. If None, then consistency is not
+   * checked. For performance reasons, this is the default. For strong
+   * consistency, set this to zero seconds. Then every read will check for
+   * updates from other processes. As a compromise, you can set this to a
+   * non-zero value for eventual consistency. If more than that interval
+   * has passed since the last check, then the table will be checked for updates.
+   * Note: this consistency only applies to read operations. Write operations are
+   * always consistent.
+   */
+  readConsistencyInterval?: number
+}
+/** Write mode for writing a table. */
+export const enum WriteMode {
+  Create = 'Create',
+  Append = 'Append',
+  Overwrite = 'Overwrite'
+}
+/** Write options when creating a Table. */
+export interface WriteOptions {
+  mode?: WriteMode
+}
+export function connect(uri: string, options: ConnectionOptions): Promise<Connection>
+export class Connection {
+  /** Create a new Connection instance from the given URI. */
+  static new(uri: string, options: ConnectionOptions): Promise<Connection>
+  display(): string
+  isOpen(): boolean
+  close(): void
+  /** List all tables in the dataset. */
+  tableNames(startAfter?: string | undefined | null, limit?: number | undefined | null): Promise<Array<string>>
+  /**
+   * Create table from a Apache Arrow IPC (file) buffer.
+   *
+   * Parameters:
+   * - name: The name of the table.
+   * - buf: The buffer containing the IPC file.
+   *
+   */
+  createTable(name: string, buf: Buffer, mode: string): Promise<Table>
+  createEmptyTable(name: string, schemaBuf: Buffer, mode: string): Promise<Table>
+  openTable(name: string): Promise<Table>
+  /** Drop table with the name. Or raise an error if the table does not exist. */
+  dropTable(name: string): Promise<void>
+}
+export class IndexBuilder {
+  replace(v: boolean): void
+  column(c: string): void
+  name(name: string): void
+  ivfPq(metricType?: MetricType | undefined | null, numPartitions?: number | undefined | null, numSubVectors?: number | undefined | null, numBits?: number | undefined | null, maxIterations?: number | undefined | null, sampleRate?: number | undefined | null): void
+  scalar(): void
+  build(): Promise<void>
+}
+/** Typescript-style Async Iterator over RecordBatches  */
+export class RecordBatchIterator {
+  next(): Promise<Buffer | null>
+}
+export class Query {
+  column(column: string): void
+  filter(filter: string): void
+  select(columns: Array<string>): void
+  limit(limit: number): void
+  prefilter(prefilter: boolean): void
+  nearestTo(vector: Float32Array): void
+  refineFactor(refineFactor: number): void
+  nprobes(nprobe: number): void
+  executeStream(): Promise<RecordBatchIterator>
+}
+export class Table {
+  display(): string
+  isOpen(): boolean
+  close(): void
+  /** Return Schema as empty Arrow IPC file. */
+  schema(): Promise<Buffer>
+  add(buf: Buffer, mode: string): Promise<void>
+  countRows(filter?: string | undefined | null): Promise<number>
+  delete(predicate: string): Promise<void>
+  createIndex(): IndexBuilder
+  query(): Query
+  addColumns(transforms: Array<AddColumnsSql>): Promise<void>
+  alterColumns(alterations: Array<ColumnAlteration>): Promise<void>
+  dropColumns(columns: Array<string>): Promise<void>
+}

nodejs/lancedb/native.js

@@ -0,0 +1,308 @@
+/* tslint:disable */
+/* eslint-disable */
+/* prettier-ignore */
+
+/* auto-generated by NAPI-RS */
+
+const { existsSync, readFileSync } = require('fs')
+const { join } = require('path')
+
+const { platform, arch } = process
+
+let nativeBinding = null
+let localFileExisted = false
+let loadError = null
+
+function isMusl() {
+  // For Node 10
+  if (!process.report || typeof process.report.getReport !== 'function') {
+    try {
+      const lddPath = require('child_process').execSync('which ldd').toString().trim()
+      return readFileSync(lddPath, 'utf8').includes('musl')
+    } catch (e) {
+      return true
+    }
+  } else {
+    const { glibcVersionRuntime } = process.report.getReport().header
+    return !glibcVersionRuntime
+  }
+}
+
+switch (platform) {
+  case 'android':
+    switch (arch) {
+      case 'arm64':
+        localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.android-arm64.node'))
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.android-arm64.node')
+          } else {
+            nativeBinding = require('lancedb-android-arm64')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      case 'arm':
+        localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.android-arm-eabi.node'))
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.android-arm-eabi.node')
+          } else {
+            nativeBinding = require('lancedb-android-arm-eabi')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      default:
+        throw new Error(`Unsupported architecture on Android ${arch}`)
+    }
+    break
+  case 'win32':
+    switch (arch) {
+      case 'x64':
+        localFileExisted = existsSync(
+          join(__dirname, 'lancedb-nodejs.win32-x64-msvc.node')
+        )
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.win32-x64-msvc.node')
+          } else {
+            nativeBinding = require('lancedb-win32-x64-msvc')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      case 'ia32':
+        localFileExisted = existsSync(
+          join(__dirname, 'lancedb-nodejs.win32-ia32-msvc.node')
+        )
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.win32-ia32-msvc.node')
+          } else {
+            nativeBinding = require('lancedb-win32-ia32-msvc')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      case 'arm64':
+        localFileExisted = existsSync(
+          join(__dirname, 'lancedb-nodejs.win32-arm64-msvc.node')
+        )
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.win32-arm64-msvc.node')
+          } else {
+            nativeBinding = require('lancedb-win32-arm64-msvc')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      default:
+        throw new Error(`Unsupported architecture on Windows: ${arch}`)
+    }
+    break
+  case 'darwin':
+    localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.darwin-universal.node'))
+    try {
+      if (localFileExisted) {
+        nativeBinding = require('./lancedb-nodejs.darwin-universal.node')
+      } else {
+        nativeBinding = require('lancedb-darwin-universal')
+      }
+      break
+    } catch {}
+    switch (arch) {
+      case 'x64':
+        localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.darwin-x64.node'))
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.darwin-x64.node')
+          } else {
+            nativeBinding = require('lancedb-darwin-x64')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      case 'arm64':
+        localFileExisted = existsSync(
+          join(__dirname, 'lancedb-nodejs.darwin-arm64.node')
+        )
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.darwin-arm64.node')
+          } else {
+            nativeBinding = require('lancedb-darwin-arm64')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      default:
+        throw new Error(`Unsupported architecture on macOS: ${arch}`)
+    }
+    break
+  case 'freebsd':
+    if (arch !== 'x64') {
+      throw new Error(`Unsupported architecture on FreeBSD: ${arch}`)
+    }
+    localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.freebsd-x64.node'))
+    try {
+      if (localFileExisted) {
+        nativeBinding = require('./lancedb-nodejs.freebsd-x64.node')
+      } else {
+        nativeBinding = require('lancedb-freebsd-x64')
+      }
+    } catch (e) {
+      loadError = e
+    }
+    break
+  case 'linux':
+    switch (arch) {
+      case 'x64':
+        if (isMusl()) {
+          localFileExisted = existsSync(
+            join(__dirname, 'lancedb-nodejs.linux-x64-musl.node')
+          )
+          try {
+            if (localFileExisted) {
+              nativeBinding = require('./lancedb-nodejs.linux-x64-musl.node')
+            } else {
+              nativeBinding = require('lancedb-linux-x64-musl')
+            }
+          } catch (e) {
+            loadError = e
+          }
+        } else {
+          localFileExisted = existsSync(
+            join(__dirname, 'lancedb-nodejs.linux-x64-gnu.node')
+          )
+          try {
+            if (localFileExisted) {
+              nativeBinding = require('./lancedb-nodejs.linux-x64-gnu.node')
+            } else {
+              nativeBinding = require('lancedb-linux-x64-gnu')
+            }
+          } catch (e) {
+            loadError = e
+          }
+        }
+        break
+      case 'arm64':
+        if (isMusl()) {
+          localFileExisted = existsSync(
+            join(__dirname, 'lancedb-nodejs.linux-arm64-musl.node')
+          )
+          try {
+            if (localFileExisted) {
+              nativeBinding = require('./lancedb-nodejs.linux-arm64-musl.node')
+            } else {
+              nativeBinding = require('lancedb-linux-arm64-musl')
+            }
+          } catch (e) {
+            loadError = e
+          }
+        } else {
+          localFileExisted = existsSync(
+            join(__dirname, 'lancedb-nodejs.linux-arm64-gnu.node')
+          )
+          try {
+            if (localFileExisted) {
+              nativeBinding = require('./lancedb-nodejs.linux-arm64-gnu.node')
+            } else {
+              nativeBinding = require('lancedb-linux-arm64-gnu')
+            }
+          } catch (e) {
+            loadError = e
+          }
+        }
+        break
+      case 'arm':
+        localFileExisted = existsSync(
+          join(__dirname, 'lancedb-nodejs.linux-arm-gnueabihf.node')
+        )
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.linux-arm-gnueabihf.node')
+          } else {
+            nativeBinding = require('lancedb-linux-arm-gnueabihf')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      case 'riscv64':
+        if (isMusl()) {
+          localFileExisted = existsSync(
+            join(__dirname, 'lancedb-nodejs.linux-riscv64-musl.node')
+          )
+          try {
+            if (localFileExisted) {
+              nativeBinding = require('./lancedb-nodejs.linux-riscv64-musl.node')
+            } else {
+              nativeBinding = require('lancedb-linux-riscv64-musl')
+            }
+          } catch (e) {
+            loadError = e
+          }
+        } else {
+          localFileExisted = existsSync(
+            join(__dirname, 'lancedb-nodejs.linux-riscv64-gnu.node')
+          )
+          try {
+            if (localFileExisted) {
+              nativeBinding = require('./lancedb-nodejs.linux-riscv64-gnu.node')
+            } else {
+              nativeBinding = require('lancedb-linux-riscv64-gnu')
+            }
+          } catch (e) {
+            loadError = e
+          }
+        }
+        break
+      case 's390x':
+        localFileExisted = existsSync(
+          join(__dirname, 'lancedb-nodejs.linux-s390x-gnu.node')
+        )
+        try {
+          if (localFileExisted) {
+            nativeBinding = require('./lancedb-nodejs.linux-s390x-gnu.node')
+          } else {
+            nativeBinding = require('lancedb-linux-s390x-gnu')
+          }
+        } catch (e) {
+          loadError = e
+        }
+        break
+      default:
+        throw new Error(`Unsupported architecture on Linux: ${arch}`)
+    }
+    break
+  default:
+    throw new Error(`Unsupported OS: ${platform}, architecture: ${arch}`)
+}
+
+if (!nativeBinding) {
+  if (loadError) {
+    throw loadError
+  }
+  throw new Error(`Failed to load native binding`)
+}
+
+const { Connection, IndexType, MetricType, IndexBuilder, RecordBatchIterator, Query, Table, WriteMode, connect } = nativeBinding
+
+module.exports.Connection = Connection
+module.exports.IndexType = IndexType
+module.exports.MetricType = MetricType
+module.exports.IndexBuilder = IndexBuilder
+module.exports.RecordBatchIterator = RecordBatchIterator
+module.exports.Query = Query
+module.exports.Table = Table
+module.exports.WriteMode = WriteMode
+module.exports.connect = connect

nodejs/lancedb/query.ts

@@ -0,0 +1,154 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { RecordBatch, tableFromIPC, Table as ArrowTable } from "apache-arrow";
+import {
+  RecordBatchIterator as NativeBatchIterator,
+  Query as NativeQuery,
+  Table as NativeTable,
+} from "./native";
+
+class RecordBatchIterator implements AsyncIterator<RecordBatch> {
+  private promisedInner?: Promise<NativeBatchIterator>;
+  private inner?: NativeBatchIterator;
+
+  constructor(
+    inner?: NativeBatchIterator,
+    promise?: Promise<NativeBatchIterator>,
+  ) {
+    // TODO: check promise reliably so we dont need to pass two arguments.
+    this.inner = inner;
+    this.promisedInner = promise;
+  }
+
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  async next(): Promise<IteratorResult<RecordBatch<any>>> {
+    if (this.inner === undefined) {
+      this.inner = await this.promisedInner;
+    }
+    if (this.inner === undefined) {
+      throw new Error("Invalid iterator state state");
+    }
+    const n = await this.inner.next();
+    if (n == null) {
+      return Promise.resolve({ done: true, value: null });
+    }
+    const tbl = tableFromIPC(n);
+    if (tbl.batches.length != 1) {
+      throw new Error("Expected only one batch");
+    }
+    return Promise.resolve({ done: false, value: tbl.batches[0] });
+  }
+}
+/* eslint-enable */
+
+/** Query executor */
+export class Query implements AsyncIterable<RecordBatch> {
+  private readonly inner: NativeQuery;
+
+  constructor(tbl: NativeTable) {
+    this.inner = tbl.query();
+  }
+
+  /** Set the column to run query. */
+  column(column: string): Query {
+    this.inner.column(column);
+    return this;
+  }
+
+  /** Set the filter predicate, only returns the results that satisfy the filter.
+   *
+   */
+  filter(predicate: string): Query {
+    this.inner.filter(predicate);
+    return this;
+  }
+
+  /**
+   * Select the columns to return. If not set, all columns are returned.
+   */
+  select(columns: string[]): Query {
+    this.inner.select(columns);
+    return this;
+  }
+
+  /**
+   * Set the limit of rows to return.
+   */
+  limit(limit: number): Query {
+    this.inner.limit(limit);
+    return this;
+  }
+
+  prefilter(prefilter: boolean): Query {
+    this.inner.prefilter(prefilter);
+    return this;
+  }
+
+  /**
+   * Set the query vector.
+   */
+  nearestTo(vector: number[]): Query {
+    this.inner.nearestTo(Float32Array.from(vector));
+    return this;
+  }
+
+  /**
+   * Set the number of IVF partitions to use for the query.
+   */
+  nprobes(nprobes: number): Query {
+    this.inner.nprobes(nprobes);
+    return this;
+  }
+
+  /**
+   * Set the refine factor for the query.
+   */
+  refineFactor(refineFactor: number): Query {
+    this.inner.refineFactor(refineFactor);
+    return this;
+  }
+
+  /**
+   * Execute the query and return the results as an AsyncIterator.
+   */
+  async executeStream(): Promise<RecordBatchIterator> {
+    const inner = await this.inner.executeStream();
+    return new RecordBatchIterator(inner);
+  }
+
+  /** Collect the results as an Arrow Table. */
+  async toArrow(): Promise<ArrowTable> {
+    const batches = [];
+    for await (const batch of this) {
+      batches.push(batch);
+    }
+    return new ArrowTable(batches);
+  }
+
+  /** Returns a JSON Array of All results.
+   *
+   */
+  async toArray(): Promise<unknown[]> {
+    const tbl = await this.toArrow();
+    // eslint-disable-next-line @typescript-eslint/no-unsafe-return
+    return tbl.toArray();
+  }
+
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  [Symbol.asyncIterator](): AsyncIterator<RecordBatch<any>> {
+    const promise = this.inner.executeStream();
+    return new RecordBatchIterator(undefined, promise);
+  }
+}

nodejs/lancedb/sanitize.ts

@@ -0,0 +1,509 @@
+// Copyright 2023 LanceDB Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+// The utilities in this file help sanitize data from the user's arrow
+// library into the types expected by vectordb's arrow library.  Node
+// generally allows for mulitple versions of the same library (and sometimes
+// even multiple copies of the same version) to be installed at the same
+// time.  However, arrow-js uses instanceof which expected that the input
+// comes from the exact same library instance.  This is not always the case
+// and so we must sanitize the input to ensure that it is compatible.
+
+import {
+  Field,
+  Utf8,
+  FixedSizeBinary,
+  FixedSizeList,
+  Schema,
+  List,
+  Struct,
+  Float,
+  Bool,
+  Date_,
+  Decimal,
+  DataType,
+  Dictionary,
+  Binary,
+  Float32,
+  Interval,
+  Map_,
+  Duration,
+  Union,
+  Time,
+  Timestamp,
+  Type,
+  Null,
+  Int,
+  type Precision,
+  type DateUnit,
+  Int8,
+  Int16,
+  Int32,
+  Int64,
+  Uint8,
+  Uint16,
+  Uint32,
+  Uint64,
+  Float16,
+  Float64,
+  DateDay,
+  DateMillisecond,
+  DenseUnion,
+  SparseUnion,
+  TimeNanosecond,
+  TimeMicrosecond,
+  TimeMillisecond,
+  TimeSecond,
+  TimestampNanosecond,
+  TimestampMicrosecond,
+  TimestampMillisecond,
+  TimestampSecond,
+  IntervalDayTime,
+  IntervalYearMonth,
+  DurationNanosecond,
+  DurationMicrosecond,
+  DurationMillisecond,
+  DurationSecond,
+} from "apache-arrow";
+import type { IntBitWidth, TKeys, TimeBitWidth } from "apache-arrow/type";
+
+function sanitizeMetadata(
+  metadataLike?: unknown,
+): Map<string, string> | undefined {
+  if (metadataLike === undefined || metadataLike === null) {
+    return undefined;
+  }
+  if (!(metadataLike instanceof Map)) {
+    throw Error("Expected metadata, if present, to be a Map<string, string>");
+  }
+  for (const item of metadataLike) {
+    if (!(typeof item[0] === "string" || !(typeof item[1] === "string"))) {
+      throw Error(
+        "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values",
+      );
+    }
+  }
+  return metadataLike as Map<string, string>;
+}
+
+function sanitizeInt(typeLike: object) {
+  if (
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number" ||
+    !("isSigned" in typeLike) ||
+    typeof typeLike.isSigned !== "boolean"
+  ) {
+    throw Error(
+      "Expected an Int Type to have a `bitWidth` and `isSigned` property",
+    );
+  }
+  return new Int(typeLike.isSigned, typeLike.bitWidth as IntBitWidth);
+}
+
+function sanitizeFloat(typeLike: object) {
+  if (!("precision" in typeLike) || typeof typeLike.precision !== "number") {
+    throw Error("Expected a Float Type to have a `precision` property");
+  }
+  return new Float(typeLike.precision as Precision);
+}
+
+function sanitizeDecimal(typeLike: object) {
+  if (
+    !("scale" in typeLike) ||
+    typeof typeLike.scale !== "number" ||
+    !("precision" in typeLike) ||
+    typeof typeLike.precision !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Decimal Type to have `scale`, `precision`, and `bitWidth` properties",
+    );
+  }
+  return new Decimal(typeLike.scale, typeLike.precision, typeLike.bitWidth);
+}
+
+function sanitizeDate(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Date type to have a `unit` property");
+  }
+  return new Date_(typeLike.unit as DateUnit);
+}
+
+function sanitizeTime(typeLike: object) {
+  if (
+    !("unit" in typeLike) ||
+    typeof typeLike.unit !== "number" ||
+    !("bitWidth" in typeLike) ||
+    typeof typeLike.bitWidth !== "number"
+  ) {
+    throw Error(
+      "Expected a Time type to have `unit` and `bitWidth` properties",
+    );
+  }
+  return new Time(typeLike.unit, typeLike.bitWidth as TimeBitWidth);
+}
+
+function sanitizeTimestamp(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Timestamp type to have a `unit` property");
+  }
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Timestamp(typeLike.unit, timezone);
+}
+
+function sanitizeTypedTimestamp(
+  typeLike: object,
+  // eslint-disable-next-line @typescript-eslint/naming-convention
+  Datatype:
+    | typeof TimestampNanosecond
+    | typeof TimestampMicrosecond
+    | typeof TimestampMillisecond
+    | typeof TimestampSecond,
+) {
+  let timezone = null;
+  if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
+    timezone = typeLike.timezone;
+  }
+  return new Datatype(timezone);
+}
+
+function sanitizeInterval(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected an Interval type to have a `unit` property");
+  }
+  return new Interval(typeLike.unit);
+}
+
+function sanitizeList(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a List type to have an array-like `children` property",
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a List type to have exactly one child");
+  }
+  return new List(sanitizeField(typeLike.children[0]));
+}
+
+function sanitizeStruct(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Struct type to have an array-like `children` property",
+    );
+  }
+  return new Struct(typeLike.children.map((child) => sanitizeField(child)));
+}
+
+function sanitizeUnion(typeLike: object) {
+  if (
+    !("typeIds" in typeLike) ||
+    !("mode" in typeLike) ||
+    typeof typeLike.mode !== "number"
+  ) {
+    throw Error(
+      "Expected a Union type to have `typeIds` and `mode` properties",
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Union type to have an array-like `children` property",
+    );
+  }
+
+  return new Union(
+    typeLike.mode,
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    typeLike.typeIds as any,
+    typeLike.children.map((child) => sanitizeField(child)),
+  );
+}
+
+function sanitizeTypedUnion(
+  typeLike: object,
+  // eslint-disable-next-line @typescript-eslint/naming-convention
+  UnionType: typeof DenseUnion | typeof SparseUnion,
+) {
+  if (!("typeIds" in typeLike)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have a `typeIds` property",
+    );
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a DenseUnion/SparseUnion type to have an array-like `children` property",
+    );
+  }
+
+  return new UnionType(
+    typeLike.typeIds as Int32Array | number[],
+    typeLike.children.map((child) => sanitizeField(child)),
+  );
+}
+
+function sanitizeFixedSizeBinary(typeLike: object) {
+  if (!("byteWidth" in typeLike) || typeof typeLike.byteWidth !== "number") {
+    throw Error(
+      "Expected a FixedSizeBinary type to have a `byteWidth` property",
+    );
+  }
+  return new FixedSizeBinary(typeLike.byteWidth);
+}
+
+function sanitizeFixedSizeList(typeLike: object) {
+  if (!("listSize" in typeLike) || typeof typeLike.listSize !== "number") {
+    throw Error("Expected a FixedSizeList type to have a `listSize` property");
+  }
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a FixedSizeList type to have an array-like `children` property",
+    );
+  }
+  if (typeLike.children.length !== 1) {
+    throw Error("Expected a FixedSizeList type to have exactly one child");
+  }
+  return new FixedSizeList(
+    typeLike.listSize,
+    sanitizeField(typeLike.children[0]),
+  );
+}
+
+function sanitizeMap(typeLike: object) {
+  if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
+    throw Error(
+      "Expected a Map type to have an array-like `children` property",
+    );
+  }
+  if (!("keysSorted" in typeLike) || typeof typeLike.keysSorted !== "boolean") {
+    throw Error("Expected a Map type to have a `keysSorted` property");
+  }
+
+  return new Map_(
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    typeLike.children.map((field) => sanitizeField(field)) as any,
+    typeLike.keysSorted,
+  );
+}
+
+function sanitizeDuration(typeLike: object) {
+  if (!("unit" in typeLike) || typeof typeLike.unit !== "number") {
+    throw Error("Expected a Duration type to have a `unit` property");
+  }
+  return new Duration(typeLike.unit);
+}
+
+function sanitizeDictionary(typeLike: object) {
+  if (!("id" in typeLike) || typeof typeLike.id !== "number") {
+    throw Error("Expected a Dictionary type to have an `id` property");
+  }
+  if (!("indices" in typeLike) || typeof typeLike.indices !== "object") {
+    throw Error("Expected a Dictionary type to have an `indices` property");
+  }
+  if (!("dictionary" in typeLike) || typeof typeLike.dictionary !== "object") {
+    throw Error("Expected a Dictionary type to have an `dictionary` property");
+  }
+  if (!("isOrdered" in typeLike) || typeof typeLike.isOrdered !== "boolean") {
+    throw Error("Expected a Dictionary type to have an `isOrdered` property");
+  }
+  return new Dictionary(
+    sanitizeType(typeLike.dictionary),
+    sanitizeType(typeLike.indices) as TKeys,
+    typeLike.id,
+    typeLike.isOrdered,
+  );
+}
+
+// eslint-disable-next-line @typescript-eslint/no-explicit-any
+function sanitizeType(typeLike: unknown): DataType<any> {
+  if (typeof typeLike !== "object" || typeLike === null) {
+    throw Error("Expected a Type but object was null/undefined");
+  }
+  if (!("typeId" in typeLike) || !(typeof typeLike.typeId !== "function")) {
+    throw Error("Expected a Type to have a typeId function");
+  }
+  let typeId: Type;
+  if (typeof typeLike.typeId === "function") {
+    typeId = (typeLike.typeId as () => unknown)() as Type;
+  } else if (typeof typeLike.typeId === "number") {
+    typeId = typeLike.typeId as Type;
+  } else {
+    throw Error("Type's typeId property was not a function or number");
+  }
+
+  switch (typeId) {
+    case Type.NONE:
+      throw Error("Received a Type with a typeId of NONE");
+    case Type.Null:
+      return new Null();
+    case Type.Int:
+      return sanitizeInt(typeLike);
+    case Type.Float:
+      return sanitizeFloat(typeLike);
+    case Type.Binary:
+      return new Binary();
+    case Type.Utf8:
+      return new Utf8();
+    case Type.Bool:
+      return new Bool();
+    case Type.Decimal:
+      return sanitizeDecimal(typeLike);
+    case Type.Date:
+      return sanitizeDate(typeLike);
+    case Type.Time:
+      return sanitizeTime(typeLike);
+    case Type.Timestamp:
+      return sanitizeTimestamp(typeLike);
+    case Type.Interval:
+      return sanitizeInterval(typeLike);
+    case Type.List:
+      return sanitizeList(typeLike);
+    case Type.Struct:
+      return sanitizeStruct(typeLike);
+    case Type.Union:
+      return sanitizeUnion(typeLike);
+    case Type.FixedSizeBinary:
+      return sanitizeFixedSizeBinary(typeLike);
+    case Type.FixedSizeList:
+      return sanitizeFixedSizeList(typeLike);
+    case Type.Map:
+      return sanitizeMap(typeLike);
+    case Type.Duration:
+      return sanitizeDuration(typeLike);
+    case Type.Dictionary:
+      return sanitizeDictionary(typeLike);
+    case Type.Int8:
+      return new Int8();
+    case Type.Int16:
+      return new Int16();
+    case Type.Int32:
+      return new Int32();
+    case Type.Int64:
+      return new Int64();
+    case Type.Uint8:
+      return new Uint8();
+    case Type.Uint16:
+      return new Uint16();
+    case Type.Uint32:
+      return new Uint32();
+    case Type.Uint64:
+      return new Uint64();
+    case Type.Float16:
+      return new Float16();
+    case Type.Float32:
+      return new Float32();
+    case Type.Float64:
+      return new Float64();
+    case Type.DateMillisecond:
+      return new DateMillisecond();
+    case Type.DateDay:
+      return new DateDay();
+    case Type.TimeNanosecond:
+      return new TimeNanosecond();
+    case Type.TimeMicrosecond:
+      return new TimeMicrosecond();
+    case Type.TimeMillisecond:
+      return new TimeMillisecond();
+    case Type.TimeSecond:
+      return new TimeSecond();
+    case Type.TimestampNanosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampNanosecond);
+    case Type.TimestampMicrosecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMicrosecond);
+    case Type.TimestampMillisecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampMillisecond);
+    case Type.TimestampSecond:
+      return sanitizeTypedTimestamp(typeLike, TimestampSecond);
+    case Type.DenseUnion:
+      return sanitizeTypedUnion(typeLike, DenseUnion);
+    case Type.SparseUnion:
+      return sanitizeTypedUnion(typeLike, SparseUnion);
+    case Type.IntervalDayTime:
+      return new IntervalDayTime();
+    case Type.IntervalYearMonth:
+      return new IntervalYearMonth();
+    case Type.DurationNanosecond:
+      return new DurationNanosecond();
+    case Type.DurationMicrosecond:
+      return new DurationMicrosecond();
+    case Type.DurationMillisecond:
+      return new DurationMillisecond();
+    case Type.DurationSecond:
+      return new DurationSecond();
+    default:
+      throw new Error("Unrecoginized type id in schema: " + typeId);
+  }
+}
+
+function sanitizeField(fieldLike: unknown): Field {
+  if (fieldLike instanceof Field) {
+    return fieldLike;
+  }
+  if (typeof fieldLike !== "object" || fieldLike === null) {
+    throw Error("Expected a Field but object was null/undefined");
+  }
+  if (
+    !("type" in fieldLike) ||
+    !("name" in fieldLike) ||
+    !("nullable" in fieldLike)
+  ) {
+    throw Error(
+      "The field passed in is missing a `type`/`name`/`nullable` property",
+    );
+  }
+  const type = sanitizeType(fieldLike.type);
+  const name = fieldLike.name;
+  if (!(typeof name === "string")) {
+    throw Error("The field passed in had a non-string `name` property");
+  }
+  const nullable = fieldLike.nullable;
+  if (!(typeof nullable === "boolean")) {
+    throw Error("The field passed in had a non-boolean `nullable` property");
+  }
+  let metadata;
+  if ("metadata" in fieldLike) {
+    metadata = sanitizeMetadata(fieldLike.metadata);
+  }
+  return new Field(name, type, nullable, metadata);
+}
+
+export function sanitizeSchema(schemaLike: unknown): Schema {
+  if (schemaLike instanceof Schema) {
+    return schemaLike;
+  }
+  if (typeof schemaLike !== "object" || schemaLike === null) {
+    throw Error("Expected a Schema but object was null/undefined");
+  }
+  if (!("fields" in schemaLike)) {
+    throw Error(
+      "The schema passed in does not appear to be a schema (no 'fields' property)",
+    );
+  }
+  let metadata;
+  if ("metadata" in schemaLike) {
+    metadata = sanitizeMetadata(schemaLike.metadata);
+  }
+  if (!Array.isArray(schemaLike.fields)) {
+    throw Error(
+      "The schema passed in had a 'fields' property but it was not an array",
+    );
+  }
+  const sanitizedFields = schemaLike.fields.map((field) =>
+    sanitizeField(field),
+  );
+  return new Schema(sanitizedFields, metadata);
+}

nodejs/lancedb/table.ts

@@ -0,0 +1,235 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { Schema, tableFromIPC } from "apache-arrow";
+import {
+  AddColumnsSql,
+  ColumnAlteration,
+  Table as _NativeTable,
+} from "./native";
+import { Query } from "./query";
+import { IndexBuilder } from "./indexer";
+import { Data, fromDataToBuffer } from "./arrow";
+
+/**
+ * Options for adding data to a table.
+ */
+export interface AddDataOptions {
+  /** If "append" (the default) then the new data will be added to the table
+   *
+   * If "overwrite" then the new data will replace the existing data in the table.
+   */
+  mode: "append" | "overwrite";
+}
+
+/**
+ * A Table is a collection of Records in a LanceDB Database.
+ *
+ * A Table object is expected to be long lived and reused for multiple operations.
+ * Table objects will cache a certain amount of index data in memory.  This cache
+ * will be freed when the Table is garbage collected.  To eagerly free the cache you
+ * can call the `close` method.  Once the Table is closed, it cannot be used for any
+ * further operations.
+ *
+ * Closing a table is optional.  It not closed, it will be closed when it is garbage
+ * collected.
+ */
+export class Table {
+  private readonly inner: _NativeTable;
+
+  /** Construct a Table. Internal use only. */
+  constructor(inner: _NativeTable) {
+    this.inner = inner;
+  }
+
+  /** Return true if the table has not been closed */
+  isOpen(): boolean {
+    return this.inner.isOpen();
+  }
+
+  /** Close the table, releasing any underlying resources.
+   *
+   * It is safe to call this method multiple times.
+   *
+   * Any attempt to use the table after it is closed will result in an error.
+   */
+  close(): void {
+    this.inner.close();
+  }
+
+  /** Return a brief description of the table */
+  display(): string {
+    return this.inner.display();
+  }
+
+  /** Get the schema of the table. */
+  async schema(): Promise<Schema> {
+    const schemaBuf = await this.inner.schema();
+    const tbl = tableFromIPC(schemaBuf);
+    return tbl.schema;
+  }
+
+  /**
+   * Insert records into this Table.
+   *
+   * @param {Data} data Records to be inserted into the Table
+   * @return The number of rows added to the table
+   */
+  async add(data: Data, options?: Partial<AddDataOptions>): Promise<void> {
+    const mode = options?.mode ?? "append";
+
+    const buffer = await fromDataToBuffer(data);
+    await this.inner.add(buffer, mode);
+  }
+
+  /** Count the total number of rows in the dataset. */
+  async countRows(filter?: string): Promise<number> {
+    return await this.inner.countRows(filter);
+  }
+
+  /** Delete the rows that satisfy the predicate. */
+  async delete(predicate: string): Promise<void> {
+    await this.inner.delete(predicate);
+  }
+
+  /** Create an index over the columns.
+   *
+   * @param {string} column The column to create the index on. If not specified,
+   *                        it will create an index on vector field.
+   *
+   * @example
+   *
+   * By default, it creates vector idnex on one vector column.
+   *
+   * ```typescript
+   * const table = await conn.openTable("my_table");
+   * await table.createIndex().build();
+   * ```
+   *
+   * You can specify `IVF_PQ` parameters via `ivf_pq({})` call.
+   * ```typescript
+   * const table = await conn.openTable("my_table");
+   * await table.createIndex("my_vec_col")
+   *   .ivf_pq({ num_partitions: 128, num_sub_vectors: 16 })
+   *   .build();
+   * ```
+   *
+   * Or create a Scalar index
+   *
+   * ```typescript
+   * await table.createIndex("my_float_col").build();
+   * ```
+   */
+  createIndex(column?: string): IndexBuilder {
+    let builder = new IndexBuilder(this.inner);
+    if (column !== undefined) {
+      builder = builder.column(column);
+    }
+    return builder;
+  }
+
+  /**
+   * Create a generic {@link Query} Builder.
+   *
+   * When appropriate, various indices and statistics based pruning will be used to
+   * accelerate the query.
+   *
+   * @example
+   *
+   * ### Run a SQL-style query
+   * ```typescript
+   * for await (const batch of table.query()
+   *                          .filter("id > 1").select(["id"]).limit(20)) {
+   *  console.log(batch);
+   * }
+   * ```
+   *
+   * ### Run Top-10 vector similarity search
+   * ```typescript
+   * for await (const batch of table.query()
+   *                    .nearestTo([1, 2, 3])
+   *                    .refineFactor(5).nprobe(10)
+   *                    .limit(10)) {
+   *  console.log(batch);
+   * }
+   *```
+   *
+   * ### Scan the full dataset
+   * ```typescript
+   * for await (const batch of table.query()) {
+   *   console.log(batch);
+   * }
+   *
+   * ### Return the full dataset as Arrow Table
+   * ```typescript
+   * let arrowTbl = await table.query().nearestTo([1.0, 2.0, 0.5, 6.7]).toArrow();
+   * ```
+   *
+   * @returns {@link Query}
+   */
+  query(): Query {
+    return new Query(this.inner);
+  }
+
+  /** Search the table with a given query vector.
+   *
+   * This is a convenience method for preparing an ANN {@link Query}.
+   */
+  search(vector: number[], column?: string): Query {
+    const q = this.query();
+    q.nearestTo(vector);
+    if (column !== undefined) {
+      q.column(column);
+    }
+    return q;
+  }
+
+  // TODO: Support BatchUDF
+  /**
+   * Add new columns with defined values.
+   *
+   * @param newColumnTransforms pairs of column names and the SQL expression to use
+   *                            to calculate the value of the new column. These
+   *                            expressions will be evaluated for each row in the
+   *                            table, and can reference existing columns in the table.
+   */
+  async addColumns(newColumnTransforms: AddColumnsSql[]): Promise<void> {
+    await this.inner.addColumns(newColumnTransforms);
+  }
+
+  /**
+   * Alter the name or nullability of columns.
+   *
+   * @param columnAlterations One or more alterations to apply to columns.
+   */
+  async alterColumns(columnAlterations: ColumnAlteration[]): Promise<void> {
+    await this.inner.alterColumns(columnAlterations);
+  }
+
+  /**
+   * Drop one or more columns from the dataset
+   *
+   * This is a metadata-only operation and does not remove the data from the
+   * underlying storage. In order to remove the data, you must subsequently
+   * call ``compact_files`` to rewrite the data without the removed columns and
+   * then call ``cleanup_files`` to remove the old files.
+   *
+   * @param columnNames The names of the columns to drop. These can be nested
+   *                    column references (e.g. "a.b.c") or top-level column
+   *                    names (e.g. "a").
+   */
+  async dropColumns(columnNames: string[]): Promise<void> {
+    await this.inner.dropColumns(columnNames);
+  }
+}