Bump version: 0.24.2-beta.2 → 0.24.2

## Summary
Fixes IndexError when creating tables with empty list data and a
provided schema. Previously, `_into_pyarrow_reader()` would attempt to
access `data[0]` on empty lists, causing an IndexError. Now properly
handles empty lists by using the provided schema.

Also adds regression tests for GitHub issues #1968 and #303 to prevent
future regressions with empty table scenarios.

## Changes
- Fix IndexError in `_into_pyarrow_reader()` for empty list + schema
case
- Add Optional[pa.Schema] parameter to handle empty data gracefully  
- Add `test_create_table_empty_list_with_schema` for the IndexError fix
- Add `test_create_empty_then_add_data` for issue #1968
- Add `test_search_empty_table` for issue #303

## Test plan
- [x] All new regression tests pass
- [x] Existing tests continue to pass
- [x] Code formatted with `make format`

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.19.1-beta.5"
+current_version = "0.21.2-beta.1"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

.github/workflows/cargo-publish.yml

@@ -5,8 +5,8 @@ on:
     tags-ignore:
       # We don't publish pre-releases for Rust. Crates.io is just a source
       # distribution, so we don't need to publish pre-releases.
-      - 'v*-beta*'
+      - "v*-beta*"
-      - '*-v*' # for example, python-vX.Y.Z
+      - "*-v*" # for example, python-vX.Y.Z
 
 env:
   # This env var is used by Swatinem/rust-cache@v2 for the cache
@@ -19,6 +19,8 @@ env:
 jobs:
   build:
     runs-on: ubuntu-22.04
+    permissions:
+      id-token: write
     timeout-minutes: 30
     # Only runs on tags that matches the make-release action
     if: startsWith(github.ref, 'refs/tags/v')
@@ -31,6 +33,8 @@ jobs:
         run: |
           sudo apt update
           sudo apt install -y protobuf-compiler libssl-dev
+      - uses: rust-lang/crates-io-auth-action@v1
+        id: auth
       - name: Publish the package
         run: |
-          cargo publish -p lancedb --all-features --token ${{ secrets.CARGO_REGISTRY_TOKEN }}
+          cargo publish -p lancedb --all-features --token ${{ steps.auth.outputs.token }}

.github/workflows/java.yml

@@ -35,6 +35,9 @@ jobs:
       - uses: Swatinem/rust-cache@v2
         with:
           workspaces: java/core/lancedb-jni
+      - uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          components: rustfmt
       - name: Run cargo fmt
         run: cargo fmt --check
         working-directory: ./java/core/lancedb-jni
@@ -68,6 +71,9 @@ jobs:
       - uses: Swatinem/rust-cache@v2
         with:
           workspaces: java/core/lancedb-jni
+      - uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          components: rustfmt
       - name: Run cargo fmt
         run: cargo fmt --check
         working-directory: ./java/core/lancedb-jni
@@ -110,4 +116,3 @@ jobs:
           -Djdk.reflect.useDirectMethodHandle=false \
           -Dio.netty.tryReflectionSetAccessible=true"
           JAVA_HOME=$JAVA_17 mvn clean test
-  

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

@@ -84,6 +84,7 @@ jobs:
         run: |
           pip install bump-my-version PyGithub packaging
           bash ci/bump_version.sh ${{ inputs.type }} ${{ inputs.bump-minor }} v $COMMIT_BEFORE_BUMP
+          bash ci/update_lockfiles.sh --amend
       - name: Push new version tag
         if: ${{ !inputs.dry_run }}
         uses: ad-m/github-push-action@master
@@ -92,11 +93,3 @@ jobs:
           github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
           branch: ${{ github.ref }}
           tags: true
-      - uses: ./.github/workflows/update_package_lock
-        if: ${{ !inputs.dry_run && inputs.other }}
-        with:
-          github_token: ${{ secrets.GITHUB_TOKEN }}
-      - uses: ./.github/workflows/update_package_lock_nodejs
-        if: ${{ !inputs.dry_run && inputs.other }}
-        with:
-          github_token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/nodejs.yml

@@ -47,6 +47,9 @@ jobs:
       run: |
         sudo apt update
         sudo apt install -y protobuf-compiler libssl-dev
+    - uses: actions-rust-lang/setup-rust-toolchain@v1
+      with:
+        components: rustfmt, clippy
     - name: Lint
       run: |
         cargo fmt --all -- --check
@@ -113,7 +116,7 @@ jobs:
         set -e
         npm ci
         npm run docs
-        if ! git diff --exit-code; then
+        if ! git diff --exit-code -- . ':(exclude)Cargo.lock'; then
           echo "Docs need to be updated"
           echo "Run 'npm run docs', fix any warnings, and commit the changes."
           exit 1

.github/workflows/npm-publish.yml

@@ -505,6 +505,8 @@ jobs:
     name: vectordb NPM Publish
     needs: [node, node-macos, node-linux-gnu, node-windows]
     runs-on: ubuntu-latest
+    permissions:
+      contents: write
     # Only runs on tags that matches the make-release action
     if: startsWith(github.ref, 'refs/tags/v')
     steps:
@@ -537,6 +539,20 @@ jobs:
         # We need to deprecate the old package to avoid confusion.
         # Each time we publish a new version, it gets undeprecated.
         run: npm deprecate vectordb "Use @lancedb/lancedb instead."
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          ref: main
+      - name: Update package-lock.json
+        run: |
+          git config user.name 'Lance Release'
+          git config user.email 'lance-dev@lancedb.com'
+          bash ci/update_lockfiles.sh
+      - name: Push new commit
+        uses: ad-m/github-push-action@master
+        with:
+          github_token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
+          branch: main
       - name: Notify Slack Action
         uses: ravsamhq/notify-slack-action@2.3.0
         if: ${{ always() }}
@@ -546,21 +562,3 @@ jobs:
           notification_title: "{workflow} is failing"
         env:
           SLACK_WEBHOOK_URL: ${{ secrets.ACTION_MONITORING_SLACK }}
-
-  update-package-lock:
-    if: startsWith(github.ref, 'refs/tags/v')
-    needs: [release]
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v4
-        with:
-          ref: main
-          token: ${{ secrets.LANCEDB_RELEASE_TOKEN }}
-          fetch-depth: 0
-          lfs: true
-      - uses: ./.github/workflows/update_package_lock
-        with:
-          github_token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/run_tests/action.yml

@@ -24,8 +24,8 @@ runs:
     - name: pytest (with integration)
       shell: bash
       if: ${{ inputs.integration == 'true' }}
-      run: pytest -m "not slow" -x -v --durations=30 python/python/tests
+      run: pytest -m "not slow" -vv --durations=30 python/python/tests
     - name: pytest (no integration tests)
       shell: bash
       if: ${{ inputs.integration != 'true' }}
-      run: pytest -m "not slow and not s3_test" -x -v --durations=30 python/python/tests
+      run: pytest -m "not slow and not s3_test" -vv --durations=30 python/python/tests

.github/workflows/update_package_lock/action.yml

@@ -1,33 +0,0 @@
-name: update_package_lock
-description: "Update node's package.lock"
-
-inputs:
-  github_token:
-    required: true
-    description: "github token for the repo"
-
-runs:
-  using: "composite"
-  steps:
-    - uses: actions/setup-node@v3
-      with:
-        node-version: 20
-    - name: Set git configs
-      shell: bash
-      run: |
-        git config user.name 'Lance Release'
-        git config user.email 'lance-dev@lancedb.com'
-    - name: Update package-lock.json file
-      working-directory: ./node
-      run: |
-        npm install
-        git add package-lock.json
-        git commit -m "Updating package-lock.json"
-      shell: bash
-    - name: Push changes
-      if: ${{ inputs.dry_run }} == "false"
-      uses: ad-m/github-push-action@master
-      with:
-        github_token: ${{ inputs.github_token }}
-        branch: main
-        tags: true

.github/workflows/update_package_lock_nodejs/action.yml

@@ -1,33 +0,0 @@
-name: update_package_lock_nodejs
-description: "Update nodejs's package.lock"
-
-inputs:
-  github_token:
-    required: true
-    description: "github token for the repo"
-
-runs:
-  using: "composite"
-  steps:
-    - uses: actions/setup-node@v3
-      with:
-        node-version: 20
-    - name: Set git configs
-      shell: bash
-      run: |
-        git config user.name 'Lance Release'
-        git config user.email 'lance-dev@lancedb.com'
-    - name: Update package-lock.json file
-      working-directory: ./nodejs
-      run: |
-        npm install
-        git add package-lock.json
-        git commit -m "Updating package-lock.json"
-      shell: bash
-    - name: Push changes
-      if: ${{ inputs.dry_run }} == "false"
-      uses: ad-m/github-push-action@master
-      with:
-        github_token: ${{ inputs.github_token }}
-        branch: main
-        tags: true

CLAUDE.md

@@ -0,0 +1,24 @@
+LanceDB is a database designed for retrieval, including vector, full-text, and hybrid search.
+It is a wrapper around Lance. There are two backends: local (in-process like SQLite) and
+remote (against LanceDB Cloud).
+
+The core of LanceDB is written in Rust. There are bindings in Python, Typescript, and Java.
+
+Project layout:
+
+* `rust/lancedb`: The LanceDB core Rust implementation.
+* `python`: The Python bindings, using PyO3.
+* `nodejs`: The Typescript bindings, using napi-rs
+* `java`: The Java bindings
+
+(`rust/ffi` and `node/` are for a deprecated package. You can ignore them.)
+
+Common commands:
+
+* Check for compiler errors: `cargo check --features remote --tests --examples`
+* Run tests: `cargo test --features remote --tests`
+* Run specific test: `cargo test --features remote -p <package_name> --test <test_name>`
+* Lint: `cargo clippy --features remote --tests --examples`
+* Format: `cargo fmt --all`
+
+Before committing changes, run formatting.

Cargo.toml

@@ -21,49 +21,49 @@ categories = ["database-implementations"]
 rust-version = "1.78.0"
 
 [workspace.dependencies]
-lance = { "version" = "=0.27.2", "features" = ["dynamodb"] }
+lance = { "version" = "=0.32.0", "features" = ["dynamodb"] }
-lance-io = { version = "=0.27.2" }
+lance-io = "=0.32.0"
-lance-index = { version = "=0.27.2" }
+lance-index = "=0.32.0"
-lance-linalg = { version = "=0.27.2" }
+lance-linalg = "=0.32.0"
-lance-table = { version = "=0.27.2" }
+lance-table = "=0.32.0"
-lance-testing = { version = "=0.27.2" }
+lance-testing = "=0.32.0"
-lance-datafusion = { version = "=0.27.2" }
+lance-datafusion = "=0.32.0"
-lance-encoding = { version = "=0.27.2" }
+lance-encoding = "=0.32.0"
 # Note that this one does not include pyarrow
-arrow = { version = "54.1", optional = false }
+arrow = { version = "55.1", optional = false }
-arrow-array = "54.1"
+arrow-array = "55.1"
-arrow-data = "54.1"
+arrow-data = "55.1"
-arrow-ipc = "54.1"
+arrow-ipc = "55.1"
-arrow-ord = "54.1"
+arrow-ord = "55.1"
-arrow-schema = "54.1"
+arrow-schema = "55.1"
-arrow-arith = "54.1"
+arrow-arith = "55.1"
-arrow-cast = "54.1"
+arrow-cast = "55.1"
 async-trait = "0"
-datafusion = { version = "46.0", default-features = false }
+datafusion = { version = "48.0", default-features = false }
-datafusion-catalog = "46.0"
+datafusion-catalog = "48.0"
-datafusion-common = { version = "46.0", default-features = false }
+datafusion-common = { version = "48.0", default-features = false }
-datafusion-execution = "46.0"
+datafusion-execution = "48.0"
-datafusion-expr = "46.0"
+datafusion-expr = "48.0"
-datafusion-physical-plan = "46.0"
+datafusion-physical-plan = "48.0"
 env_logger = "0.11"
-half = { "version" = "=2.4.1", default-features = false, features = [
+half = { "version" = "2.6.0", default-features = false, features = [
     "num-traits",
 ] }
 futures = "0"
 log = "0.4"
 moka = { version = "0.12", features = ["future"] }
-object_store = "0.11.0"
+object_store = "0.12.0"
 pin-project = "1.0.7"
 snafu = "0.8"
 url = "2"
 num-traits = "0.2"
-rand = "0.8"
+rand = "0.9"
 regex = "1.10"
 lazy_static = "1"
 semver = "1.0.25"
 # Temporary pins to work around downstream issues
 # https://github.com/apache/arrow-rs/commit/2fddf85afcd20110ce783ed5b4cdeb82293da30b
-chrono = "=0.4.39"
+chrono = "=0.4.41"
 # https://github.com/RustCrypto/formats/issues/1684
 base64ct = "=1.6.0"
 # Workaround for: https://github.com/eira-fransham/crunchy/issues/13

README.md

@@ -1,94 +1,97 @@
 <a href="https://cloud.lancedb.com" target="_blank">
   <img src="https://github.com/user-attachments/assets/92dad0a2-2a37-4ce1-b783-0d1b4f30a00c" alt="LanceDB Cloud Public Beta" width="100%" style="max-width: 100%;">
 </a>
-
 <div align="center">
-<p align="center">
 
-<picture>
+[![LanceDB](docs/src/assets/hero-header.png)](https://lancedb.com)
-  <source media="(prefers-color-scheme: dark)" srcset="https://github.com/user-attachments/assets/ac270358-333e-4bea-a132-acefaa94040e">
+[![Website](https://img.shields.io/badge/-Website-100000?style=for-the-badge&labelColor=645cfb&color=645cfb)](https://lancedb.com/)
-  <source media="(prefers-color-scheme: light)" srcset="https://github.com/user-attachments/assets/b864d814-0d29-4784-8fd9-807297c758c0">
+[![Blog](https://img.shields.io/badge/Blog-100000?style=for-the-badge&labelColor=645cfb&color=645cfb)](https://blog.lancedb.com/)
-  <img alt="LanceDB Logo" src="https://github.com/user-attachments/assets/b864d814-0d29-4784-8fd9-807297c758c0" width=300>
+[![Discord](https://img.shields.io/badge/-Discord-100000?style=for-the-badge&logo=discord&logoColor=white&labelColor=645cfb&color=645cfb)](https://discord.gg/zMM32dvNtd)
-</picture>
+[![Twitter](https://img.shields.io/badge/-Twitter-100000?style=for-the-badge&logo=x&logoColor=white&labelColor=645cfb&color=645cfb)](https://twitter.com/lancedb)
+[![LinkedIn](https://img.shields.io/badge/-LinkedIn-100000?style=for-the-badge&logo=linkedin&logoColor=white&labelColor=645cfb&color=645cfb)](https://www.linkedin.com/company/lancedb/)
 
-**Search More, Manage Less**
 
-<a href='https://github.com/lancedb/vectordb-recipes/tree/main' target="_blank"><img alt='LanceDB' src='https://img.shields.io/badge/VectorDB_Recipes-100000?style=for-the-badge&logo=LanceDB&logoColor=white&labelColor=645cfb&color=645cfb'/></a>
+<img src="docs/src/assets/lancedb.png" alt="LanceDB" width="50%">
-<a href='https://lancedb.github.io/lancedb/' target="_blank"><img alt='lancdb' src='https://img.shields.io/badge/DOCS-100000?style=for-the-badge&logo=lancdb&logoColor=white&labelColor=645cfb&color=645cfb'/></a>
-[![Blog](https://img.shields.io/badge/Blog-12100E?style=for-the-badge&logoColor=white)](https://blog.lancedb.com/)
-[![Discord](https://img.shields.io/badge/Discord-%235865F2.svg?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/zMM32dvNtd)
-[![Twitter](https://img.shields.io/badge/Twitter-%231DA1F2.svg?style=for-the-badge&logo=Twitter&logoColor=white)](https://twitter.com/lancedb)
-[![Gurubase](https://img.shields.io/badge/Gurubase-Ask%20LanceDB%20Guru-006BFF?style=for-the-badge)](https://gurubase.io/g/lancedb)
 
-</p>
+# **The Multimodal AI Lakehouse**
 
-<img max-width="750px" alt="LanceDB Multimodal Search" src="https://github.com/lancedb/lancedb/assets/917119/09c5afc5-7816-4687-bae4-f2ca194426ec">
+[**How to Install** ](#how-to-install) ✦ [**Detailed Documentation**](https://lancedb.github.io/lancedb/) ✦ [**Tutorials and Recipes**](https://github.com/lancedb/vectordb-recipes/tree/main) ✦  [**Contributors**](#contributors) 
+
+**The ultimate multimodal data platform for AI/ML applications.** 
+
+LanceDB is designed for fast, scalable, and production-ready vector search. It is built on top of the Lance columnar format. You can store, index, and search over petabytes of multimodal data and vectors with ease. 
+LanceDB is a central location where developers can build, train and analyze their AI workloads.
 
-</p>
 </div>
 
-<hr />
+<br>
 
-LanceDB is an open-source database for vector-search built with persistent storage, which greatly simplifies retrieval, filtering and management of embeddings.
+## **Demo: Multimodal Search by Keyword, Vector or with SQL**
+<img max-width="750px" alt="LanceDB Multimodal Search" src="https://github.com/lancedb/lancedb/assets/917119/09c5afc5-7816-4687-bae4-f2ca194426ec">
 
-The key features of LanceDB include:
+## **Star LanceDB to get updates!**
 
-* Production-scale vector search with no servers to manage.
+<details>
+<summary>⭐ Click here ⭐  to see how fast we're growing!</summary>
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="https://api.star-history.com/svg?repos=lancedb/lancedb&theme=dark&type=Date">
+  <img width="100%" src="https://api.star-history.com/svg?repos=lancedb/lancedb&theme=dark&type=Date">
+</picture>
+</details>
 
-* Store, query and filter vectors, metadata and multi-modal data (text, images, videos, point clouds, and more).
+## **Key Features**:
 
-* Support for vector similarity search, full-text search and SQL.
+- **Fast Vector Search**: Search billions of vectors in milliseconds with state-of-the-art indexing.
+- **Comprehensive Search**: Support for vector similarity search, full-text search and SQL.
+- **Multimodal Support**: Store, query and filter vectors, metadata and multimodal data (text, images, videos, point clouds, and more).
+- **Advanced Features**: Zero-copy, automatic versioning, manage versions of your data without needing extra infrastructure. GPU support in building vector index.
 
-* Native Python and Javascript/Typescript support.
+### **Products**:
+- **Open Source & Local**: 100% open source, runs locally or in your cloud. No vendor lock-in.
+- **Cloud and Enterprise**: Production-scale vector search with no servers to manage. Complete data sovereignty and security.
 
-* Zero-copy, automatic versioning, manage versions of your data without needing extra infrastructure.
+### **Ecosystem**:
+- **Columnar Storage**: Built on the Lance columnar format for efficient storage and analytics.
+- **Seamless Integration**: Python, Node.js, Rust, and REST APIs for easy integration. Native Python and Javascript/Typescript support.
+- **Rich Ecosystem**: Integrations with [**LangChain** 🦜️🔗](https://python.langchain.com/docs/integrations/vectorstores/lancedb/), [**LlamaIndex** 🦙](https://gpt-index.readthedocs.io/en/latest/examples/vector_stores/LanceDBIndexDemo.html), Apache-Arrow, Pandas, Polars, DuckDB and more on the way.
 
-* GPU support in building vector index(*).
+## **How to Install**:
 
-* Ecosystem integrations with [LangChain 🦜️🔗](https://python.langchain.com/docs/integrations/vectorstores/lancedb/), [LlamaIndex 🦙](https://gpt-index.readthedocs.io/en/latest/examples/vector_stores/LanceDBIndexDemo.html), Apache-Arrow, Pandas, Polars, DuckDB and more on the way.
+Follow the [Quickstart](https://lancedb.github.io/lancedb/basic/) doc to set up LanceDB locally. 
 
-LanceDB's core is written in Rust 🦀 and is built using <a href="https://github.com/lancedb/lance">Lance</a>, an open-source columnar format designed for performant ML workloads.
+**API & SDK:** We also support Python, Typescript and Rust SDKs
 
-## Quick Start
+| Interface | Documentation |
+|-----------|---------------|
+| Python SDK | https://lancedb.github.io/lancedb/python/python/ |
+| Typescript SDK | https://lancedb.github.io/lancedb/js/globals/ |
+| Rust SDK | https://docs.rs/lancedb/latest/lancedb/index.html |
+| REST API | https://docs.lancedb.com/api-reference/introduction |
 
-**Javascript**
+## **Join Us and Contribute**
-```shell
-npm install @lancedb/lancedb
-```
 
-```javascript
+We welcome contributions from everyone! Whether you're a developer, researcher, or just someone who wants to help out. 
-import * as lancedb from "@lancedb/lancedb";
 
-const db = await lancedb.connect("data/sample-lancedb");
+If you have any suggestions or feature requests, please feel free to open an issue on GitHub or discuss it on our [**Discord**](https://discord.gg/G5DcmnZWKB) server.
-const table = await db.createTable("vectors", [
+
-	{ id: 1, vector: [0.1, 0.2], item: "foo", price: 10 },
+[**Check out the GitHub Issues**](https://github.com/lancedb/lancedb/issues) if you would like to work on the features that are planned for the future. If you have any suggestions or feature requests, please feel free to open an issue on GitHub. 
-	{ id: 2, vector: [1.1, 1.2], item: "bar", price: 50 },
+
-], {mode: 'overwrite'});
+## **Contributors**
+
+<a href="https://github.com/lancedb/lancedb/graphs/contributors">
+  <img src="https://contrib.rocks/image?repo=lancedb/lancedb" />
+</a>
 
 
-const query = table.vectorSearch([0.1, 0.3]).limit(2);
+## **Stay in Touch With Us**
-const results = await query.toArray();
+<div align="center">
 
-// You can also search for rows by specific criteria without involving a vector search.
+</br>
-const rowsByCriteria = await table.query().where("price >= 10").toArray();
-```
 
-**Python**
+[![Website](https://img.shields.io/badge/-Website-100000?style=for-the-badge&labelColor=645cfb&color=645cfb)](https://lancedb.com/)
-```shell
+[![Blog](https://img.shields.io/badge/Blog-100000?style=for-the-badge&labelColor=645cfb&color=645cfb)](https://blog.lancedb.com/)
-pip install lancedb
+[![Discord](https://img.shields.io/badge/-Discord-100000?style=for-the-badge&logo=discord&logoColor=white&labelColor=645cfb&color=645cfb)](https://discord.gg/zMM32dvNtd)
-```
+[![Twitter](https://img.shields.io/badge/-Twitter-100000?style=for-the-badge&logo=x&logoColor=white&labelColor=645cfb&color=645cfb)](https://twitter.com/lancedb)
+[![LinkedIn](https://img.shields.io/badge/-LinkedIn-100000?style=for-the-badge&logo=linkedin&logoColor=white&labelColor=645cfb&color=645cfb)](https://www.linkedin.com/company/lancedb/)
 
-```python
+</div>
-import lancedb
-
-uri = "data/sample-lancedb"
-db = lancedb.connect(uri)
-table = db.create_table("my_table",
-                         data=[{"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
-                               {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}])
-result = table.search([100, 100]).limit(2).to_pandas()
-```
-
-## Blogs, Tutorials & Videos
-* 📈 <a href="https://blog.lancedb.com/benchmarking-random-access-in-lance/">2000x better performance with Lance over Parquet</a>
-* 🤖 <a href="https://github.com/lancedb/vectordb-recipes/tree/main/examples/Youtube-Search-QA-Bot">Build a question and answer bot with LanceDB</a>

ci/set_lance_version.py

@@ -0,0 +1,188 @@
+import argparse
+import sys
+import json
+
+
+def run_command(command: str) -> str:
+    """
+    Run a shell command and return stdout as a string.
+    If exit code is not 0, raise an exception with the stderr output.
+    """
+    import subprocess
+
+    result = subprocess.run(command, shell=True, capture_output=True, text=True)
+    if result.returncode != 0:
+        raise Exception(f"Command failed with error: {result.stderr.strip()}")
+    return result.stdout.strip()
+
+
+def get_latest_stable_version() -> str:
+    version_line = run_command("cargo info lance | grep '^version:'")
+    version = version_line.split(" ")[1].strip()
+    return version
+
+
+def get_latest_preview_version() -> str:
+    lance_tags = run_command(
+        "git ls-remote --tags https://github.com/lancedb/lance.git | grep 'refs/tags/v[0-9beta.-]\\+$'"
+    ).splitlines()
+    lance_tags = (
+        tag.split("refs/tags/")[1]
+        for tag in lance_tags
+        if "refs/tags/" in tag and "beta" in tag
+    )
+    from packaging.version import Version
+
+    latest = max(
+        (tag[1:] for tag in lance_tags if tag.startswith("v")), key=lambda t: Version(t)
+    )
+    return str(latest)
+
+
+def extract_features(line: str) -> list:
+    """
+    Extracts the features from a line in Cargo.toml.
+    Example: 'lance = { "version" = "=0.29.0", "features" = ["dynamodb"] }'
+    Returns: ['dynamodb']
+    """
+    import re
+
+    match = re.search(r'"features"\s*=\s*\[\s*(.*?)\s*\]', line, re.DOTALL)
+    if match:
+        features_str = match.group(1)
+        return [f.strip('"') for f in features_str.split(",") if len(f) > 0]
+    return []
+
+
+def update_cargo_toml(line_updater):
+    """
+    Updates the Cargo.toml file by applying the line_updater function to each line.
+    The line_updater function should take a line as input and return the updated line.
+    """
+    with open("Cargo.toml", "r") as f:
+        lines = f.readlines()
+
+    new_lines = []
+    lance_line = ""
+    is_parsing_lance_line = False
+    for line in lines:
+        if line.startswith("lance"):
+            # Update the line using the provided function
+            if line.strip().endswith("}"):
+                new_lines.append(line_updater(line))
+            else:
+                lance_line = line
+                is_parsing_lance_line = True
+        elif is_parsing_lance_line:
+            lance_line += line
+            if line.strip().endswith("}"):
+                new_lines.append(line_updater(lance_line))
+                lance_line = ""
+                is_parsing_lance_line = False
+            else:
+                print("doesn't end with }:", line)
+        else:
+            # Keep the line unchanged
+            new_lines.append(line)
+
+    with open("Cargo.toml", "w") as f:
+        f.writelines(new_lines)
+
+
+def set_stable_version(version: str):
+    """
+    Sets lines to
+    lance = { "version" = "=0.29.0", "features" = ["dynamodb"] }
+    lance-io = "=0.29.0"
+    ...
+    """
+
+    def line_updater(line: str) -> str:
+        package_name = line.split("=", maxsplit=1)[0].strip()
+        features = extract_features(line)
+        if features:
+            return f'{package_name} = {{ "version" = "={version}", "features" = {json.dumps(features)} }}\n'
+        else:
+            return f'{package_name} = "={version}"\n'
+
+    update_cargo_toml(line_updater)
+
+
+def set_preview_version(version: str):
+    """
+    Sets lines to
+    lance = { "version" = "=0.29.0", "features" = ["dynamodb"], tag = "v0.29.0-beta.2", git="https://github.com/lancedb/lance.git" }
+    lance-io = { version = "=0.29.0", tag = "v0.29.0-beta.2", git="https://github.com/lancedb/lance.git" }
+    ...
+    """
+
+    def line_updater(line: str) -> str:
+        package_name = line.split("=", maxsplit=1)[0].strip()
+        features = extract_features(line)
+        base_version = version.split("-")[0]  # Get the base version without beta suffix
+        if features:
+            return f'{package_name} = {{ "version" = "={base_version}", "features" = {json.dumps(features)}, "tag" = "v{version}", "git" = "https://github.com/lancedb/lance.git" }}\n'
+        else:
+            return f'{package_name} = {{ "version" = "={base_version}", "tag" = "v{version}", "git" = "https://github.com/lancedb/lance.git" }}\n'
+
+    update_cargo_toml(line_updater)
+
+
+def set_local_version():
+    """
+    Sets lines to
+    lance = { path = "../lance/rust/lance", features = ["dynamodb"] }
+    lance-io = { path = "../lance/rust/lance-io" }
+    ...
+    """
+
+    def line_updater(line: str) -> str:
+        package_name = line.split("=", maxsplit=1)[0].strip()
+        features = extract_features(line)
+        if features:
+            return f'{package_name} = {{ "path" = "../lance/rust/{package_name}", "features" = {json.dumps(features)} }}\n'
+        else:
+            return f'{package_name} = {{ "path" = "../lance/rust/{package_name}" }}\n'
+
+    update_cargo_toml(line_updater)
+
+
+parser = argparse.ArgumentParser(description="Set the version of the Lance package.")
+parser.add_argument(
+    "version",
+    type=str,
+    help="The version to set for the Lance package. Use 'stable' for the latest stable version, 'preview' for latest preview version, or a specific version number (e.g., '0.1.0'). You can also specify 'local' to use a local path.",
+)
+args = parser.parse_args()
+
+if args.version == "stable":
+    latest_stable_version = get_latest_stable_version()
+    print(
+        f"Found latest stable version: \033[1mv{latest_stable_version}\033[0m",
+        file=sys.stderr,
+    )
+    set_stable_version(latest_stable_version)
+elif args.version == "preview":
+    latest_preview_version = get_latest_preview_version()
+    print(
+        f"Found latest preview version: \033[1mv{latest_preview_version}\033[0m",
+        file=sys.stderr,
+    )
+    set_preview_version(latest_preview_version)
+elif args.version == "local":
+    set_local_version()
+else:
+    # Parse the version number.
+    version = args.version
+    # Ignore initial v if present.
+    if version.startswith("v"):
+        version = version[1:]
+
+    if "beta" in version:
+        set_preview_version(version)
+    else:
+        set_stable_version(version)
+
+print("Updating lockfiles...", file=sys.stderr, end="")
+run_command("cargo metadata > /dev/null")
+print(" done.", file=sys.stderr)

ci/update_lockfiles.sh

@@ -0,0 +1,30 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+AMEND=false
+
+for arg in "$@"; do
+  if [[ "$arg" == "--amend" ]]; then
+    AMEND=true
+  fi
+done
+
+# This updates the lockfile without building
+cargo metadata --quiet > /dev/null
+
+pushd nodejs || exit 1
+npm install --package-lock-only --silent
+popd
+pushd node || exit 1
+npm install --package-lock-only --silent
+popd
+
+if git diff --quiet --exit-code; then
+  echo "No lockfile changes to commit; skipping amend."
+elif $AMEND; then
+  git add Cargo.lock nodejs/package-lock.json node/package-lock.json
+  git commit --amend --no-edit
+else
+  git add Cargo.lock nodejs/package-lock.json node/package-lock.json
+  git commit -m "Update lockfiles"
+fi

docs/mkdocs.yml

@@ -193,6 +193,7 @@ nav:
           - Pandas and PyArrow: python/pandas_and_pyarrow.md
           - Polars: python/polars_arrow.md
           - DuckDB: python/duckdb.md
+          - Datafusion: python/datafusion.md
           - LangChain:
               - LangChain 🔗: integrations/langchain.md
               - LangChain demo: notebooks/langchain_demo.ipynb
@@ -248,6 +249,7 @@ nav:
       - Data management: concepts/data_management.md
   - Guides:
       - Working with tables: guides/tables.md
+      - Working with SQL: guides/sql_querying.md
       - Building an ANN index: ann_indexes.md
       - Vector Search: search.md
       - Full-text search (native): fts.md
@@ -324,6 +326,7 @@ nav:
       - Pandas and PyArrow: python/pandas_and_pyarrow.md
       - Polars: python/polars_arrow.md
       - DuckDB: python/duckdb.md
+      - Datafusion: python/datafusion.md
       - LangChain 🦜️🔗↗: integrations/langchain.md
       - LangChain.js 🦜️🔗↗: https://js.langchain.com/docs/integrations/vectorstores/lancedb
       - LlamaIndex 🦙↗: integrations/llamaIndex.md

docs/overrides/partials/main.html

@@ -0,0 +1,5 @@
+{% extends "base.html" %}
+
+{% block announce %}
+  📚 Starting June 1st, 2025, please use <a href="https://lancedb.github.io/documentation" target="_blank" rel="noopener noreferrer">lancedb.github.io/documentation</a> for the latest docs.
+{% endblock %}

docs/package-lock.json

@@ -19,7 +19,7 @@
     },
     "../node": {
       "name": "vectordb",
-      "version": "0.12.0",
+      "version": "0.21.2-beta.0",
       "cpu": [
         "x64",
         "arm64"
@@ -65,11 +65,11 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.12.0",
+        "@lancedb/vectordb-darwin-arm64": "0.21.2-beta.0",
-        "@lancedb/vectordb-darwin-x64": "0.12.0",
+        "@lancedb/vectordb-darwin-x64": "0.21.2-beta.0",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.12.0",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.21.2-beta.0",
-        "@lancedb/vectordb-linux-x64-gnu": "0.12.0",
+        "@lancedb/vectordb-linux-x64-gnu": "0.21.2-beta.0",
-        "@lancedb/vectordb-win32-x64-msvc": "0.12.0"
+        "@lancedb/vectordb-win32-x64-msvc": "0.21.2-beta.0"
       },
       "peerDependencies": {
         "@apache-arrow/ts": "^14.0.2",

docs/src/ann_indexes.md

@@ -291,7 +291,7 @@ Product quantization can lead to approximately `16 * sizeof(float32) / 1 = 64` t
 
 `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.
+On `SIFT-1M` dataset, our benchmark shows that keeping each partition 4K-8K rows lead to a good latency / recall.
 
 `num_sub_vectors` specifies how many Product Quantization (PQ) short codes to generate on each vector. The number should be a factor of the vector dimension. Because
 PQ is a lossy compression of the original vector, a higher `num_sub_vectors` usually results in

docs/src/guides/sql_querying.md

@@ -0,0 +1,60 @@
+# SQL Querying
+
+You can use DuckDB and Apache Datafusion to query your LanceDB tables using SQL.
+This guide will show how to query Lance tables them using both.
+
+We will re-use the dataset [created previously](./tables.md):
+
+```python
+import lancedb
+
+db = lancedb.connect("data/sample-lancedb")
+data = [
+    {"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
+    {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}
+]
+table = db.create_table("pd_table", data=data)
+```
+
+## Querying a LanceDB Table with DuckDb
+
+The `to_lance` method converts the LanceDB table to a `LanceDataset`, which is accessible to DuckDB through the Arrow compatibility layer.
+To query the resulting Lance dataset in DuckDB, all you need to do is reference the dataset by the same name in your SQL query.
+
+```python
+import duckdb
+
+arrow_table = table.to_lance()
+
+duckdb.query("SELECT * FROM arrow_table")
+```
+
+| vector      | item | price |
+| ----------- | ---- | ----- |
+| [3.1, 4.1]  | foo  | 10.0  |
+| [5.9, 26.5] | bar  | 20.0  |
+
+## Querying a LanceDB Table with Apache Datafusion
+
+Have the required imports before doing any querying.
+
+=== "Python"
+
+    ```python
+    --8<-- "python/python/tests/docs/test_guide_tables.py:import-lancedb"
+    --8<-- "python/python/tests/docs/test_guide_tables.py:import-session-context"
+    --8<-- "python/python/tests/docs/test_guide_tables.py:import-ffi-dataset"
+    ```
+
+Register the table created with the Datafusion session context.
+
+=== "Python"
+
+    ```python
+    --8<-- "python/python/tests/docs/test_guide_tables.py:lance_sql_basic"
+    ```
+
+| vector      | item | price |
+| ----------- | ---- | ----- |
+| [3.1, 4.1]  | foo  | 10.0  |
+| [5.9, 26.5] | bar  | 20.0  |

docs/src/guides/tables.md

@@ -765,7 +765,7 @@ This can be used to update zero to all rows depending on how many rows match the
         ];
         const tbl = await db.createTable("my_table", data)
 
-        await tbl.update({ 
+        await tbl.update({
             values: { vector: [10, 10] },
             where: "x = 2"
         });
@@ -787,9 +787,9 @@ This can be used to update zero to all rows depending on how many rows match the
         ];
         const tbl = await db.createTable("my_table", data)
 
-        await tbl.update({ 
+        await tbl.update({
-            where: "x = 2", 
+            where: "x = 2",
-            values: { vector: [10, 10] } 
+            values: { vector: [10, 10] }
         });
         ```
 

docs/src/js/classes/BooleanQuery.md

@@ -0,0 +1,53 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / BooleanQuery
+
+# Class: BooleanQuery
+
+Represents a full-text query interface.
+This interface defines the structure and behavior for full-text queries,
+including methods to retrieve the query type and convert the query to a dictionary format.
+
+## Implements
+
+- [`FullTextQuery`](../interfaces/FullTextQuery.md)
+
+## Constructors
+
+### new BooleanQuery()
+
+```ts
+new BooleanQuery(queries): BooleanQuery
+```
+
+Creates an instance of BooleanQuery.
+
+#### Parameters
+
+* **queries**: [[`Occur`](../enumerations/Occur.md), [`FullTextQuery`](../interfaces/FullTextQuery.md)][]
+    An array of (Occur, FullTextQuery objects) to combine.
+    Occur specifies whether the query must match, or should match.
+
+#### Returns
+
+[`BooleanQuery`](BooleanQuery.md)
+
+## Methods
+
+### queryType()
+
+```ts
+queryType(): FullTextQueryType
+```
+
+The type of the full-text query.
+
+#### Returns
+
+[`FullTextQueryType`](../enumerations/FullTextQueryType.md)
+
+#### Implementation of
+
+[`FullTextQuery`](../interfaces/FullTextQuery.md).[`queryType`](../interfaces/FullTextQuery.md#querytype)

docs/src/js/classes/MatchQuery.md

@@ -40,6 +40,8 @@ Creates an instance of MatchQuery.
     - `boost`: The boost factor for the query (default is 1.0).
     - `fuzziness`: The fuzziness level for the query (default is 0).
     - `maxExpansions`: The maximum number of terms to consider for fuzzy matching (default is 50).
+    - `operator`: The logical operator to use for combining terms in the query (default is "OR").
+    - `prefixLength`: The number of beginning characters being unchanged for fuzzy matching.
 
 * **options.boost?**: `number`
 
@@ -47,6 +49,10 @@ Creates an instance of MatchQuery.
 
 * **options.maxExpansions?**: `number`
 
+* **options.operator?**: [`Operator`](../enumerations/Operator.md)
+
+* **options.prefixLength?**: `number`
+
 #### Returns
 
 [`MatchQuery`](MatchQuery.md)

docs/src/js/classes/MultiMatchQuery.md

@@ -38,9 +38,12 @@ Creates an instance of MultiMatchQuery.
 * **options?**
     Optional parameters for the multi-match query.
     - `boosts`: An array of boost factors for each column (default is 1.0 for all).
+    - `operator`: The logical operator to use for combining terms in the query (default is "OR").
 
 * **options.boosts?**: `number`[]
 
+* **options.operator?**: [`Operator`](../enumerations/Operator.md)
+
 #### Returns
 
 [`MultiMatchQuery`](MultiMatchQuery.md)

docs/src/js/classes/PhraseQuery.md

@@ -19,7 +19,10 @@ including methods to retrieve the query type and convert the query to a dictiona
 ### new PhraseQuery()
 
 ```ts
-new PhraseQuery(query, column): PhraseQuery
+new PhraseQuery(
+   query,
+   column,
+   options?): PhraseQuery
 ```
 
 Creates an instance of `PhraseQuery`.
@@ -32,6 +35,12 @@ Creates an instance of `PhraseQuery`.
 * **column**: `string`
     The name of the column to search within.
 
+* **options?**
+    Optional parameters for the phrase query.
+    - `slop`: The maximum number of intervening unmatched positions allowed between words in the phrase (default is 0).
+
+* **options.slop?**: `number`
+
 #### Returns
 
 [`PhraseQuery`](PhraseQuery.md)

docs/src/js/classes/Session.md

@@ -0,0 +1,84 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / Session
+
+# Class: Session
+
+A session for managing caches and object stores across LanceDB operations.
+
+Sessions allow you to configure cache sizes for index and metadata caches,
+which can significantly impact performance for large datasets.
+
+## Constructors
+
+### new Session()
+
+```ts
+new Session(indexCacheSizeBytes?, metadataCacheSizeBytes?): Session
+```
+
+Create a new session with custom cache sizes.
+
+# Parameters
+
+- `index_cache_size_bytes`: The size of the index cache in bytes.
+  Defaults to 6GB if not specified.
+- `metadata_cache_size_bytes`: The size of the metadata cache in bytes.
+  Defaults to 1GB if not specified.
+
+#### Parameters
+
+* **indexCacheSizeBytes?**: `null` \| `bigint`
+
+* **metadataCacheSizeBytes?**: `null` \| `bigint`
+
+#### Returns
+
+[`Session`](Session.md)
+
+## Methods
+
+### approxNumItems()
+
+```ts
+approxNumItems(): number
+```
+
+Get the approximate number of items cached in the session.
+
+#### Returns
+
+`number`
+
+***
+
+### sizeBytes()
+
+```ts
+sizeBytes(): bigint
+```
+
+Get the current size of the session caches in bytes.
+
+#### Returns
+
+`bigint`
+
+***
+
+### default()
+
+```ts
+static default(): Session
+```
+
+Create a session with default cache sizes.
+
+This is equivalent to creating a session with 6GB index cache
+and 1GB metadata cache.
+
+#### Returns
+
+[`Session`](Session.md)

docs/src/js/classes/Table.md

@@ -612,7 +612,7 @@ of the given query
 
 #### Parameters
 
-* **query**: `string` \| [`IntoVector`](../type-aliases/IntoVector.md) \| [`FullTextQuery`](../interfaces/FullTextQuery.md)
+* **query**: `string` \| [`IntoVector`](../type-aliases/IntoVector.md) \| [`MultiVector`](../type-aliases/MultiVector.md) \| [`FullTextQuery`](../interfaces/FullTextQuery.md)
     the query, a vector or string
 
 * **queryType?**: `string`
@@ -799,7 +799,7 @@ by `query`.
 
 #### Parameters
 
-* **vector**: [`IntoVector`](../type-aliases/IntoVector.md)
+* **vector**: [`IntoVector`](../type-aliases/IntoVector.md) \| [`MultiVector`](../type-aliases/MultiVector.md)
 
 #### Returns
 

docs/src/js/classes/VectorQuery.md

@@ -386,6 +386,53 @@ called then every valid row from the table will be returned.
 
 ***
 
+### maximumNprobes()
+
+```ts
+maximumNprobes(maximumNprobes): VectorQuery
+```
+
+Set the maximum number of probes used.
+
+This controls the maximum number of partitions that will be searched.  If this
+number is greater than minimumNprobes then the excess partitions will _only_ be
+searched if we have not found enough results.  This can be useful when there is
+a narrow filter to allow these queries to spend more time searching and avoid
+potential false negatives.
+
+#### Parameters
+
+* **maximumNprobes**: `number`
+
+#### Returns
+
+[`VectorQuery`](VectorQuery.md)
+
+***
+
+### minimumNprobes()
+
+```ts
+minimumNprobes(minimumNprobes): VectorQuery
+```
+
+Set the minimum number of probes used.
+
+This controls the minimum number of partitions that will be searched.  This
+parameter will impact every query against a vector index, regardless of the
+filter.  See `nprobes` for more details.  Higher values will increase recall
+but will also increase latency.
+
+#### Parameters
+
+* **minimumNprobes**: `number`
+
+#### Returns
+
+[`VectorQuery`](VectorQuery.md)
+
+***
+
 ### nprobes()
 
 ```ts
@@ -413,6 +460,10 @@ For best results we recommend tuning this parameter with a benchmark against
 your actual data to find the smallest possible value that will still give
 you the desired recall.
 
+For more fine grained control over behavior when you have a very narrow filter
+you can use `minimumNprobes` and `maximumNprobes`.  This method sets both
+the minimum and maximum to the same value.
+
 #### Parameters
 
 * **nprobes**: `number`

docs/src/js/enumerations/FullTextQueryType.md

@@ -15,6 +15,14 @@ Enum representing the types of full-text queries supported.
 
 ## Enumeration Members
 
+### Boolean
+
+```ts
+Boolean: "boolean";
+```
+
+***
+
 ### Boost
 
 ```ts

docs/src/js/enumerations/Occur.md

@@ -0,0 +1,37 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / Occur
+
+# Enumeration: Occur
+
+Enum representing the occurrence of terms in full-text queries.
+
+- `Must`: The term must be present in the document.
+- `Should`: The term should contribute to the document score, but is not required.
+- `MustNot`: The term must not be present in the document.
+
+## Enumeration Members
+
+### Must
+
+```ts
+Must: "MUST";
+```
+
+***
+
+### MustNot
+
+```ts
+MustNot: "MUST_NOT";
+```
+
+***
+
+### Should
+
+```ts
+Should: "SHOULD";
+```

docs/src/js/enumerations/Operator.md

@@ -0,0 +1,28 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / Operator
+
+# Enumeration: Operator
+
+Enum representing the logical operators used in full-text queries.
+
+- `And`: All terms must match.
+- `Or`: At least one term must match.
+
+## Enumeration Members
+
+### And
+
+```ts
+And: "AND";
+```
+
+***
+
+### Or
+
+```ts
+Or: "OR";
+```

docs/src/js/functions/connect.md

@@ -6,10 +6,13 @@
 
 # Function: connect()
 
-## connect(uri, options)
+## connect(uri, options, session)
 
 ```ts
-function connect(uri, options?): Promise<Connection>
+function connect(
+   uri,
+   options?,
+   session?): Promise<Connection>
 ```
 
 Connect to a LanceDB instance at the given URI.
@@ -29,6 +32,8 @@ Accepted formats:
 * **options?**: `Partial`&lt;[`ConnectionOptions`](../interfaces/ConnectionOptions.md)&gt;
     The options to use when connecting to the database
 
+* **session?**: [`Session`](../classes/Session.md)
+
 ### Returns
 
 `Promise`&lt;[`Connection`](../classes/Connection.md)&gt;
@@ -77,7 +82,7 @@ Accepted formats:
 
 [ConnectionOptions](../interfaces/ConnectionOptions.md) for more details on the URI format.
 
-### Example
+### Examples
 
 ```ts
 const conn = await connect({
@@ -85,3 +90,11 @@ const conn = await connect({
   storageOptions: {timeout: "60s"}
 });
 ```
+
+```ts
+const session = Session.default();
+const conn = await connect({
+  uri: "/path/to/database",
+  session: session
+});
+```

docs/src/js/globals.md

@@ -12,9 +12,12 @@
 ## Enumerations
 
 - [FullTextQueryType](enumerations/FullTextQueryType.md)
+- [Occur](enumerations/Occur.md)
+- [Operator](enumerations/Operator.md)
 
 ## Classes
 
+- [BooleanQuery](classes/BooleanQuery.md)
 - [BoostQuery](classes/BoostQuery.md)
 - [Connection](classes/Connection.md)
 - [Index](classes/Index.md)
@@ -26,6 +29,7 @@
 - [Query](classes/Query.md)
 - [QueryBase](classes/QueryBase.md)
 - [RecordBatchIterator](classes/RecordBatchIterator.md)
+- [Session](classes/Session.md)
 - [Table](classes/Table.md)
 - [TagContents](classes/TagContents.md)
 - [Tags](classes/Tags.md)
@@ -81,6 +85,7 @@
 - [FieldLike](type-aliases/FieldLike.md)
 - [IntoSql](type-aliases/IntoSql.md)
 - [IntoVector](type-aliases/IntoVector.md)
+- [MultiVector](type-aliases/MultiVector.md)
 - [RecordBatchLike](type-aliases/RecordBatchLike.md)
 - [SchemaLike](type-aliases/SchemaLike.md)
 - [TableLike](type-aliases/TableLike.md)

docs/src/js/interfaces/ConnectionOptions.md

@@ -70,6 +70,17 @@ Defaults to 'us-east-1'.
 
 ***
 
+### session?
+
+```ts
+optional session: Session;
+```
+
+(For LanceDB OSS only): the session to use for this connection. Holds
+shared caches and other session-specific state.
+
+***
+
 ### storageOptions?
 
 ```ts

docs/src/js/interfaces/FtsOptions.md

@@ -23,7 +23,7 @@ whether to remove punctuation
 ### baseTokenizer?
 
 ```ts
-optional baseTokenizer: "raw" | "simple" | "whitespace";
+optional baseTokenizer: "raw" | "simple" | "whitespace" | "ngram";
 ```
 
 The tokenizer to use when building the index.
@@ -71,6 +71,36 @@ tokens longer than this length will be ignored
 
 ***
 
+### ngramMaxLength?
+
+```ts
+optional ngramMaxLength: number;
+```
+
+ngram max length
+
+***
+
+### ngramMinLength?
+
+```ts
+optional ngramMinLength: number;
+```
+
+ngram min length
+
+***
+
+### prefixOnly?
+
+```ts
+optional prefixOnly: boolean;
+```
+
+whether to only index the prefix of the token for ngram tokenizer
+
+***
+
 ### removeStopWords?
 
 ```ts

docs/src/js/interfaces/OpenTableOptions.md

@@ -8,7 +8,7 @@
 
 ## Properties
 
-### indexCacheSize?
+### ~~indexCacheSize?~~
 
 ```ts
 optional indexCacheSize: number;
@@ -16,6 +16,11 @@ optional indexCacheSize: number;
 
 Set the size of the index cache, specified as a number of entries
 
+#### Deprecated
+
+Use session-level cache configuration instead.
+Create a Session with custom cache sizes and pass it to the connect() function.
+
 The exact meaning of an "entry" will depend on the type of index:
 - IVF: there is one entry for each IVF partition
 - BTREE: there is one entry for the entire index

docs/src/js/interfaces/OptimizeOptions.md

@@ -24,10 +24,10 @@ The default is 7 days
 // Delete all versions older than 1 day
 const olderThan = new Date();
 olderThan.setDate(olderThan.getDate() - 1));
-tbl.cleanupOlderVersions(olderThan);
+tbl.optimize({cleanupOlderThan: olderThan});
 
 // Delete all versions except the current version
-tbl.cleanupOlderVersions(new Date());
+tbl.optimize({cleanupOlderThan: new Date()});
 ```
 
 ***

docs/src/js/type-aliases/MultiVector.md

@@ -0,0 +1,11 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / MultiVector
+
+# Type Alias: MultiVector
+
+```ts
+type MultiVector: IntoVector[];
+```

docs/src/notebooks/Multivector_on_LanceDB.ipynb

@@ -428,7 +428,7 @@
         "\n",
         "**Why?**  \n",
         "Embedding the UFO dataset and ingesting it into LanceDB takes **~2 hours on a T4 GPU**. To save time:  \n",
-        "- **Use the pre-prepared table with index created ** (provided below) to proceed directly to step7: search.  \n",
+        "- **Use the pre-prepared table with index created** (provided below) to proceed directly to **Step 7**: search.  \n",
         "- **Step 5a** contains the full ingestion code for reference (run it only if necessary).  \n",
         "- **Step 6** contains the details on creating the index on the multivector column"
       ]

docs/src/python/datafusion.md

@@ -0,0 +1,53 @@
+# Apache Datafusion
+
+In Python, LanceDB tables can also be queried with [Apache Datafusion](https://datafusion.apache.org/), an extensible query engine written in Rust that uses Apache Arrow as its in-memory format. This means you can write complex SQL queries to analyze your data in LanceDB.
+
+This integration is done via [Datafusion FFI](https://docs.rs/datafusion-ffi/latest/datafusion_ffi/), which provides a native integration between LanceDB and Datafusion.
+The Datafusion FFI allows to pass down column selections and basic filters to LanceDB, reducing the amount of scanned data when executing your query. Additionally, the integration allows streaming data from LanceDB tables which allows to do aggregation larger-than-memory.
+
+We can demonstrate this by first installing `datafusion` and `lancedb`.
+
+```shell
+pip install datafusion lancedb
+```
+
+We will re-use the dataset [created previously](./pandas_and_pyarrow.md):
+
+```python
+import lancedb
+
+from datafusion import SessionContext
+from lance import FFILanceTableProvider
+
+db = lancedb.connect("data/sample-lancedb")
+data = [
+    {"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
+    {"vector": [5.9, 26.5], "item": "bar", "price": 20.0}
+]
+lance_table = db.create_table("lance_table", data)
+
+ctx = SessionContext()
+
+ffi_lance_table = FFILanceTableProvider(
+    lance_table.to_lance(), with_row_id=True, with_row_addr=True
+)
+ctx.register_table_provider("ffi_lance_table", ffi_lance_table)
+```
+
+The `to_lance` method converts the LanceDB table to a `LanceDataset`, which is accessible to Datafusion through the Datafusion FFI integration layer.
+To query the resulting Lance dataset in Datafusion, you first need to register the dataset with Datafusion and then just reference it by the same name in your SQL query.
+
+```python
+ctx.table("ffi_lance_table")
+ctx.sql("SELECT * FROM ffi_lance_table")
+```
+
+```
+┌─────────────┬─────────┬────────┬─────────────────┬─────────────────┐
+│   vector    │  item   │ price  │ _rowid          │ _rowaddr        │
+│   float[]   │ varchar │ double │ bigint unsigned │ bigint unsigned │
+├─────────────┼─────────┼────────┼─────────────────┼─────────────────┤
+│ [3.1, 4.1]  │ foo     │   10.0 │               0 │               0 │
+│ [5.9, 26.5] │ bar     │   20.0 │               1 │               1 │
+└─────────────┴─────────┴────────┴─────────────────┴─────────────────┘
+```

docs/test/md_testing.py

@@ -30,7 +30,8 @@ excluded_globs = [
     "../src/rag/advanced_techniques/*.md",
     "../src/guides/scalar_index.md",
     "../src/guides/storage.md",
-    "../src/search.md"
+    "../src/search.md",
+    "../src/guides/sql_querying.md",
 ]
 
 python_prefix = "py"

docs/test/requirements.txt

@@ -7,3 +7,4 @@ tantivy==0.20.1
 --extra-index-url https://download.pytorch.org/whl/cpu
 torch
 polars>=0.19, <=1.3.0
+datafusion

java/.mvn/wrapper/maven-wrapper.properties

@@ -0,0 +1,19 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you 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.
+wrapperVersion=3.3.2
+distributionType=only-script
+distributionUrl=https://repo.maven.apache.org/maven2/org/apache/maven/apache-maven/3.9.9/apache-maven-3.9.9-bin.zip

java/README.md

@@ -0,0 +1,37 @@
+# LanceDB Java SDK
+
+## Configuration and Initialization
+
+### LanceDB Cloud
+
+For LanceDB Cloud, use the simplified builder API:
+
+```java
+import com.lancedb.lance.namespace.LanceRestNamespace;
+
+// If your DB url is db://example-db, then your database here is example-db
+LanceRestNamespace namespace = LanceDBRestNamespaces.builder()
+    .apiKey("your_lancedb_cloud_api_key")
+    .database("your_database_name")
+    .build();
+```
+
+### LanceDB Enterprise
+
+For Enterprise deployments, use your VPC endpoint:
+
+```java
+LanceRestNamespace namespace = LanceDBRestNamespaces.builder()
+    .apiKey("your_lancedb_enterprise_api_key")
+    .database("your-top-dir") // Your top level folder under your cloud bucket, e.g. s3://your-bucket/your-top-dir/
+    .hostOverride("http://<vpc_endpoint_dns_name>:80")
+    .build();
+```
+
+## Development
+
+Build:
+
+```shell
+./mvnw install
+```

java/core/lancedb-jni/Cargo.toml

@@ -19,7 +19,7 @@ lancedb = { path = "../../../rust/lancedb" }
 lance = { workspace = true }
 arrow = { workspace = true, features = ["ffi"] }
 arrow-schema.workspace = true
-tokio = "1.23"
+tokio = "1.46"
 jni = "0.21.1"
 snafu.workspace = true
 lazy_static.workspace = true

java/core/pom.xml

@@ -8,18 +8,24 @@
     <parent>
         <groupId>com.lancedb</groupId>
         <artifactId>lancedb-parent</artifactId>
-        <version>0.19.1-beta.5</version>
+        <version>0.21.2-beta.1</version>
         <relativePath>../pom.xml</relativePath>
     </parent>
 
     <artifactId>lancedb-core</artifactId>
-    <name>LanceDB Core</name>
+    <name>${project.artifactId}</name>
+    <description>LanceDB Core</description>
     <packaging>jar</packaging>
     <properties>
         <rust.release.build>false</rust.release.build>
     </properties>
 
     <dependencies>
+        <dependency>
+            <groupId>com.lancedb</groupId>
+            <artifactId>lance-namespace-core</artifactId>
+            <version>0.0.1</version>
+        </dependency>
         <dependency>
             <groupId>org.apache.arrow</groupId>
             <artifactId>arrow-vector</artifactId>

java/lance-namespace/pom.xml

@@ -0,0 +1,26 @@
+<?xml version="1.0" encoding="UTF-8"?>
+
+<project xmlns="http://maven.apache.org/POM/4.0.0"
+    xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
+    xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
+    <modelVersion>4.0.0</modelVersion>
+
+    <parent>
+        <groupId>com.lancedb</groupId>
+        <artifactId>lancedb-parent</artifactId>
+        <version>0.21.2-beta.1</version>
+        <relativePath>../pom.xml</relativePath>
+    </parent>
+
+    <artifactId>lancedb-lance-namespace</artifactId>
+    <name>${project.artifactId}</name>
+    <description>LanceDB Java Integration with Lance Namespace</description>
+    <packaging>jar</packaging>
+
+    <dependencies>
+        <dependency>
+            <groupId>com.lancedb</groupId>
+            <artifactId>lance-namespace-core</artifactId>
+        </dependency>
+    </dependencies>
+</project>

java/lance-namespace/src/main/java/com/lancedb/lancedb/LanceDbRestNamespaces.java

@@ -0,0 +1,146 @@
+/*
+ * 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.
+ */
+package com.lancedb.lancedb;
+
+import com.lancedb.lance.namespace.LanceRestNamespace;
+import com.lancedb.lance.namespace.client.apache.ApiClient;
+
+import java.util.HashMap;
+import java.util.Map;
+import java.util.Optional;
+
+/** Util class to help construct a {@link LanceRestNamespace} for LanceDB. */
+public class LanceDbRestNamespaces {
+  private static final String DEFAULT_REGION = "us-east-1";
+  private static final String CLOUD_URL_PATTERN = "https://%s.%s.api.lancedb.com";
+
+  private String apiKey;
+  private String database;
+  private Optional<String> hostOverride = Optional.empty();
+  private Optional<String> region = Optional.empty();
+  private Map<String, String> additionalConfig = new HashMap<>();
+
+  private LanceDbRestNamespaces() {}
+
+  /**
+   * Create a new builder instance.
+   *
+   * @return A new LanceRestNamespaceBuilder
+   */
+  public static LanceDbRestNamespaces builder() {
+    return new LanceDbRestNamespaces();
+  }
+
+  /**
+   * Set the API key (required).
+   *
+   * @param apiKey The LanceDB API key
+   * @return This builder
+   */
+  public LanceDbRestNamespaces apiKey(String apiKey) {
+    if (apiKey == null || apiKey.trim().isEmpty()) {
+      throw new IllegalArgumentException("API key cannot be null or empty");
+    }
+    this.apiKey = apiKey;
+    return this;
+  }
+
+  /**
+   * Set the database name (required).
+   *
+   * @param database The database name
+   * @return This builder
+   */
+  public LanceDbRestNamespaces database(String database) {
+    if (database == null || database.trim().isEmpty()) {
+      throw new IllegalArgumentException("Database cannot be null or empty");
+    }
+    this.database = database;
+    return this;
+  }
+
+  /**
+   * Set a custom host override (optional). When set, this overrides the default LanceDB Cloud URL
+   * construction. Use this for LanceDB Enterprise deployments.
+   *
+   * @param hostOverride The complete base URL (e.g., "http://your-vpc-endpoint:80")
+   * @return This builder
+   */
+  public LanceDbRestNamespaces hostOverride(String hostOverride) {
+    this.hostOverride = Optional.ofNullable(hostOverride);
+    return this;
+  }
+
+  /**
+   * Set the region for LanceDB Cloud (optional). Defaults to "us-east-1" if not specified. This is
+   * ignored when hostOverride is set.
+   *
+   * @param region The AWS region (e.g., "us-east-1", "eu-west-1")
+   * @return This builder
+   */
+  public LanceDbRestNamespaces region(String region) {
+    this.region = Optional.ofNullable(region);
+    return this;
+  }
+
+  /**
+   * Add additional configuration parameters.
+   *
+   * @param key The configuration key
+   * @param value The configuration value
+   * @return This builder
+   */
+  public LanceDbRestNamespaces config(String key, String value) {
+    this.additionalConfig.put(key, value);
+    return this;
+  }
+
+  /**
+   * Build the LanceRestNamespace instance.
+   *
+   * @return A configured LanceRestNamespace
+   * @throws IllegalStateException if required parameters are missing
+   */
+  public LanceRestNamespace build() {
+    // Validate required fields
+    if (apiKey == null) {
+      throw new IllegalStateException("API key is required");
+    }
+    if (database == null) {
+      throw new IllegalStateException("Database is required");
+    }
+
+    // Build configuration map
+    Map<String, String> config = new HashMap<>(additionalConfig);
+    config.put("headers.x-lancedb-database", database);
+    config.put("headers.x-api-key", apiKey);
+
+    // Determine base URL
+    String baseUrl;
+    if (hostOverride.isPresent()) {
+      baseUrl = hostOverride.get();
+      config.put("host_override", hostOverride.get());
+    } else {
+      String effectiveRegion = region.orElse(DEFAULT_REGION);
+      baseUrl = String.format(CLOUD_URL_PATTERN, database, effectiveRegion);
+      config.put("region", effectiveRegion);
+    }
+
+    // Create and configure ApiClient
+    ApiClient apiClient = new ApiClient();
+    apiClient.setBasePath(baseUrl);
+
+    return new LanceRestNamespace(apiClient, config);
+  }
+}

java/mvnw

@@ -0,0 +1,259 @@
+#!/bin/sh
+# ----------------------------------------------------------------------------
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you 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.
+# ----------------------------------------------------------------------------
+
+# ----------------------------------------------------------------------------
+# Apache Maven Wrapper startup batch script, version 3.3.2
+#
+# Optional ENV vars
+# -----------------
+#   JAVA_HOME - location of a JDK home dir, required when download maven via java source
+#   MVNW_REPOURL - repo url base for downloading maven distribution
+#   MVNW_USERNAME/MVNW_PASSWORD - user and password for downloading maven
+#   MVNW_VERBOSE - true: enable verbose log; debug: trace the mvnw script; others: silence the output
+# ----------------------------------------------------------------------------
+
+set -euf
+[ "${MVNW_VERBOSE-}" != debug ] || set -x
+
+# OS specific support.
+native_path() { printf %s\\n "$1"; }
+case "$(uname)" in
+CYGWIN* | MINGW*)
+  [ -z "${JAVA_HOME-}" ] || JAVA_HOME="$(cygpath --unix "$JAVA_HOME")"
+  native_path() { cygpath --path --windows "$1"; }
+  ;;
+esac
+
+# set JAVACMD and JAVACCMD
+set_java_home() {
+  # For Cygwin and MinGW, ensure paths are in Unix format before anything is touched
+  if [ -n "${JAVA_HOME-}" ]; then
+    if [ -x "$JAVA_HOME/jre/sh/java" ]; then
+      # IBM's JDK on AIX uses strange locations for the executables
+      JAVACMD="$JAVA_HOME/jre/sh/java"
+      JAVACCMD="$JAVA_HOME/jre/sh/javac"
+    else
+      JAVACMD="$JAVA_HOME/bin/java"
+      JAVACCMD="$JAVA_HOME/bin/javac"
+
+      if [ ! -x "$JAVACMD" ] || [ ! -x "$JAVACCMD" ]; then
+        echo "The JAVA_HOME environment variable is not defined correctly, so mvnw cannot run." >&2
+        echo "JAVA_HOME is set to \"$JAVA_HOME\", but \"\$JAVA_HOME/bin/java\" or \"\$JAVA_HOME/bin/javac\" does not exist." >&2
+        return 1
+      fi
+    fi
+  else
+    JAVACMD="$(
+      'set' +e
+      'unset' -f command 2>/dev/null
+      'command' -v java
+    )" || :
+    JAVACCMD="$(
+      'set' +e
+      'unset' -f command 2>/dev/null
+      'command' -v javac
+    )" || :
+
+    if [ ! -x "${JAVACMD-}" ] || [ ! -x "${JAVACCMD-}" ]; then
+      echo "The java/javac command does not exist in PATH nor is JAVA_HOME set, so mvnw cannot run." >&2
+      return 1
+    fi
+  fi
+}
+
+# hash string like Java String::hashCode
+hash_string() {
+  str="${1:-}" h=0
+  while [ -n "$str" ]; do
+    char="${str%"${str#?}"}"
+    h=$(((h * 31 + $(LC_CTYPE=C printf %d "'$char")) % 4294967296))
+    str="${str#?}"
+  done
+  printf %x\\n $h
+}
+
+verbose() { :; }
+[ "${MVNW_VERBOSE-}" != true ] || verbose() { printf %s\\n "${1-}"; }
+
+die() {
+  printf %s\\n "$1" >&2
+  exit 1
+}
+
+trim() {
+  # MWRAPPER-139:
+  #   Trims trailing and leading whitespace, carriage returns, tabs, and linefeeds.
+  #   Needed for removing poorly interpreted newline sequences when running in more
+  #   exotic environments such as mingw bash on Windows.
+  printf "%s" "${1}" | tr -d '[:space:]'
+}
+
+# parse distributionUrl and optional distributionSha256Sum, requires .mvn/wrapper/maven-wrapper.properties
+while IFS="=" read -r key value; do
+  case "${key-}" in
+  distributionUrl) distributionUrl=$(trim "${value-}") ;;
+  distributionSha256Sum) distributionSha256Sum=$(trim "${value-}") ;;
+  esac
+done <"${0%/*}/.mvn/wrapper/maven-wrapper.properties"
+[ -n "${distributionUrl-}" ] || die "cannot read distributionUrl property in ${0%/*}/.mvn/wrapper/maven-wrapper.properties"
+
+case "${distributionUrl##*/}" in
+maven-mvnd-*bin.*)
+  MVN_CMD=mvnd.sh _MVNW_REPO_PATTERN=/maven/mvnd/
+  case "${PROCESSOR_ARCHITECTURE-}${PROCESSOR_ARCHITEW6432-}:$(uname -a)" in
+  *AMD64:CYGWIN* | *AMD64:MINGW*) distributionPlatform=windows-amd64 ;;
+  :Darwin*x86_64) distributionPlatform=darwin-amd64 ;;
+  :Darwin*arm64) distributionPlatform=darwin-aarch64 ;;
+  :Linux*x86_64*) distributionPlatform=linux-amd64 ;;
+  *)
+    echo "Cannot detect native platform for mvnd on $(uname)-$(uname -m), use pure java version" >&2
+    distributionPlatform=linux-amd64
+    ;;
+  esac
+  distributionUrl="${distributionUrl%-bin.*}-$distributionPlatform.zip"
+  ;;
+maven-mvnd-*) MVN_CMD=mvnd.sh _MVNW_REPO_PATTERN=/maven/mvnd/ ;;
+*) MVN_CMD="mvn${0##*/mvnw}" _MVNW_REPO_PATTERN=/org/apache/maven/ ;;
+esac
+
+# apply MVNW_REPOURL and calculate MAVEN_HOME
+# maven home pattern: ~/.m2/wrapper/dists/{apache-maven-<version>,maven-mvnd-<version>-<platform>}/<hash>
+[ -z "${MVNW_REPOURL-}" ] || distributionUrl="$MVNW_REPOURL$_MVNW_REPO_PATTERN${distributionUrl#*"$_MVNW_REPO_PATTERN"}"
+distributionUrlName="${distributionUrl##*/}"
+distributionUrlNameMain="${distributionUrlName%.*}"
+distributionUrlNameMain="${distributionUrlNameMain%-bin}"
+MAVEN_USER_HOME="${MAVEN_USER_HOME:-${HOME}/.m2}"
+MAVEN_HOME="${MAVEN_USER_HOME}/wrapper/dists/${distributionUrlNameMain-}/$(hash_string "$distributionUrl")"
+
+exec_maven() {
+  unset MVNW_VERBOSE MVNW_USERNAME MVNW_PASSWORD MVNW_REPOURL || :
+  exec "$MAVEN_HOME/bin/$MVN_CMD" "$@" || die "cannot exec $MAVEN_HOME/bin/$MVN_CMD"
+}
+
+if [ -d "$MAVEN_HOME" ]; then
+  verbose "found existing MAVEN_HOME at $MAVEN_HOME"
+  exec_maven "$@"
+fi
+
+case "${distributionUrl-}" in
+*?-bin.zip | *?maven-mvnd-?*-?*.zip) ;;
+*) die "distributionUrl is not valid, must match *-bin.zip or maven-mvnd-*.zip, but found '${distributionUrl-}'" ;;
+esac
+
+# prepare tmp dir
+if TMP_DOWNLOAD_DIR="$(mktemp -d)" && [ -d "$TMP_DOWNLOAD_DIR" ]; then
+  clean() { rm -rf -- "$TMP_DOWNLOAD_DIR"; }
+  trap clean HUP INT TERM EXIT
+else
+  die "cannot create temp dir"
+fi
+
+mkdir -p -- "${MAVEN_HOME%/*}"
+
+# Download and Install Apache Maven
+verbose "Couldn't find MAVEN_HOME, downloading and installing it ..."
+verbose "Downloading from: $distributionUrl"
+verbose "Downloading to: $TMP_DOWNLOAD_DIR/$distributionUrlName"
+
+# select .zip or .tar.gz
+if ! command -v unzip >/dev/null; then
+  distributionUrl="${distributionUrl%.zip}.tar.gz"
+  distributionUrlName="${distributionUrl##*/}"
+fi
+
+# verbose opt
+__MVNW_QUIET_WGET=--quiet __MVNW_QUIET_CURL=--silent __MVNW_QUIET_UNZIP=-q __MVNW_QUIET_TAR=''
+[ "${MVNW_VERBOSE-}" != true ] || __MVNW_QUIET_WGET='' __MVNW_QUIET_CURL='' __MVNW_QUIET_UNZIP='' __MVNW_QUIET_TAR=v
+
+# normalize http auth
+case "${MVNW_PASSWORD:+has-password}" in
+'') MVNW_USERNAME='' MVNW_PASSWORD='' ;;
+has-password) [ -n "${MVNW_USERNAME-}" ] || MVNW_USERNAME='' MVNW_PASSWORD='' ;;
+esac
+
+if [ -z "${MVNW_USERNAME-}" ] && command -v wget >/dev/null; then
+  verbose "Found wget ... using wget"
+  wget ${__MVNW_QUIET_WGET:+"$__MVNW_QUIET_WGET"} "$distributionUrl" -O "$TMP_DOWNLOAD_DIR/$distributionUrlName" || die "wget: Failed to fetch $distributionUrl"
+elif [ -z "${MVNW_USERNAME-}" ] && command -v curl >/dev/null; then
+  verbose "Found curl ... using curl"
+  curl ${__MVNW_QUIET_CURL:+"$__MVNW_QUIET_CURL"} -f -L -o "$TMP_DOWNLOAD_DIR/$distributionUrlName" "$distributionUrl" || die "curl: Failed to fetch $distributionUrl"
+elif set_java_home; then
+  verbose "Falling back to use Java to download"
+  javaSource="$TMP_DOWNLOAD_DIR/Downloader.java"
+  targetZip="$TMP_DOWNLOAD_DIR/$distributionUrlName"
+  cat >"$javaSource" <<-END
+	public class Downloader extends java.net.Authenticator
+	{
+	  protected java.net.PasswordAuthentication getPasswordAuthentication()
+	  {
+	    return new java.net.PasswordAuthentication( System.getenv( "MVNW_USERNAME" ), System.getenv( "MVNW_PASSWORD" ).toCharArray() );
+	  }
+	  public static void main( String[] args ) throws Exception
+	  {
+	    setDefault( new Downloader() );
+	    java.nio.file.Files.copy( java.net.URI.create( args[0] ).toURL().openStream(), java.nio.file.Paths.get( args[1] ).toAbsolutePath().normalize() );
+	  }
+	}
+	END
+  # For Cygwin/MinGW, switch paths to Windows format before running javac and java
+  verbose " - Compiling Downloader.java ..."
+  "$(native_path "$JAVACCMD")" "$(native_path "$javaSource")" || die "Failed to compile Downloader.java"
+  verbose " - Running Downloader.java ..."
+  "$(native_path "$JAVACMD")" -cp "$(native_path "$TMP_DOWNLOAD_DIR")" Downloader "$distributionUrl" "$(native_path "$targetZip")"
+fi
+
+# If specified, validate the SHA-256 sum of the Maven distribution zip file
+if [ -n "${distributionSha256Sum-}" ]; then
+  distributionSha256Result=false
+  if [ "$MVN_CMD" = mvnd.sh ]; then
+    echo "Checksum validation is not supported for maven-mvnd." >&2
+    echo "Please disable validation by removing 'distributionSha256Sum' from your maven-wrapper.properties." >&2
+    exit 1
+  elif command -v sha256sum >/dev/null; then
+    if echo "$distributionSha256Sum  $TMP_DOWNLOAD_DIR/$distributionUrlName" | sha256sum -c >/dev/null 2>&1; then
+      distributionSha256Result=true
+    fi
+  elif command -v shasum >/dev/null; then
+    if echo "$distributionSha256Sum  $TMP_DOWNLOAD_DIR/$distributionUrlName" | shasum -a 256 -c >/dev/null 2>&1; then
+      distributionSha256Result=true
+    fi
+  else
+    echo "Checksum validation was requested but neither 'sha256sum' or 'shasum' are available." >&2
+    echo "Please install either command, or disable validation by removing 'distributionSha256Sum' from your maven-wrapper.properties." >&2
+    exit 1
+  fi
+  if [ $distributionSha256Result = false ]; then
+    echo "Error: Failed to validate Maven distribution SHA-256, your Maven distribution might be compromised." >&2
+    echo "If you updated your Maven version, you need to update the specified distributionSha256Sum property." >&2
+    exit 1
+  fi
+fi
+
+# unzip and move
+if command -v unzip >/dev/null; then
+  unzip ${__MVNW_QUIET_UNZIP:+"$__MVNW_QUIET_UNZIP"} "$TMP_DOWNLOAD_DIR/$distributionUrlName" -d "$TMP_DOWNLOAD_DIR" || die "failed to unzip"
+else
+  tar xzf${__MVNW_QUIET_TAR:+"$__MVNW_QUIET_TAR"} "$TMP_DOWNLOAD_DIR/$distributionUrlName" -C "$TMP_DOWNLOAD_DIR" || die "failed to untar"
+fi
+printf %s\\n "$distributionUrl" >"$TMP_DOWNLOAD_DIR/$distributionUrlNameMain/mvnw.url"
+mv -- "$TMP_DOWNLOAD_DIR/$distributionUrlNameMain" "$MAVEN_HOME" || [ -d "$MAVEN_HOME" ] || die "fail to move MAVEN_HOME"
+
+clean || :
+exec_maven "$@"

java/pom.xml

@@ -6,11 +6,10 @@
 
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-parent</artifactId>
-    <version>0.19.1-beta.5</version>
+    <version>0.21.2-beta.1</version>
     <packaging>pom</packaging>
-
+    <name>${project.artifactId}</name>
-    <name>LanceDB Parent</name>
+    <description>LanceDB Java SDK Parent POM</description>
-    <description>LanceDB vector database Java API</description>
     <url>http://lancedb.com/</url>
 
     <developers>
@@ -29,6 +28,7 @@
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <arrow.version>15.0.0</arrow.version>
+        <lance-namespace.verison>0.0.1</lance-namespace.verison>
         <spotless.skip>false</spotless.skip>
         <spotless.version>2.30.0</spotless.version>
         <spotless.java.googlejavaformat.version>1.7</spotless.java.googlejavaformat.version>
@@ -52,6 +52,7 @@
 
     <modules>
         <module>core</module>
+        <module>lance-namespace</module>
     </modules>
 
     <scm>
@@ -62,6 +63,11 @@
 
     <dependencyManagement>
         <dependencies>
+            <dependency>
+                <groupId>com.lancedb</groupId>
+                <artifactId>lance-namespace-core</artifactId>
+                <version>${lance-namespace.verison}</version>
+            </dependency>
             <dependency>
                 <groupId>org.apache.arrow</groupId>
                 <artifactId>arrow-vector</artifactId>

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.19.1-beta.5",
+  "version": "0.21.2-beta.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
       "cpu": [
         "x64",
         "arm64"
@@ -52,11 +52,11 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.19.1-beta.5",
+        "@lancedb/vectordb-darwin-arm64": "0.21.2-beta.1",
-        "@lancedb/vectordb-darwin-x64": "0.19.1-beta.5",
+        "@lancedb/vectordb-darwin-x64": "0.21.2-beta.1",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.19.1-beta.5",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.21.2-beta.1",
-        "@lancedb/vectordb-linux-x64-gnu": "0.19.1-beta.5",
+        "@lancedb/vectordb-linux-x64-gnu": "0.21.2-beta.1",
-        "@lancedb/vectordb-win32-x64-msvc": "0.19.1-beta.5"
+        "@lancedb/vectordb-win32-x64-msvc": "0.21.2-beta.1"
       },
       "peerDependencies": {
         "@apache-arrow/ts": "^14.0.2",
@@ -327,65 +327,60 @@
       }
     },
     "node_modules/@lancedb/vectordb-darwin-arm64": {
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.19.1-beta.5.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.21.2-beta.1.tgz",
-      "integrity": "sha512-9WcTw67We5HYGayDt5jFquGoyAVzFSt/I65ag8+q7H9q4ZYKxeDhgNyQZJ8BmXEvbJtnYtYBSAtTEdFKYMce6w==",
+      "integrity": "sha512-7QXVJNTei7PMuXRyyc+F3WGiudRNq9HfeOaMmMOJJpuCAO0zLq1pM9DCl5aPF5MddrodPHJxi+IWV+iAFH7zcg==",
       "cpu": [
         "arm64"
       ],
-      "license": "Apache-2.0",
       "optional": true,
       "os": [
         "darwin"
       ]
     },
     "node_modules/@lancedb/vectordb-darwin-x64": {
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.19.1-beta.5.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.21.2-beta.1.tgz",
-      "integrity": "sha512-6Pe3PxEMi0VKGsu5R7IhOxTijUM3b5olRAqhxfcu5ti34gXIPNtu7g+T9lS78LKe+0D0v2BjZEY/JQakIFBNRw==",
+      "integrity": "sha512-M/TWcJ3WVc6DNFgG/lWI7L5tQ05IF3WoWuZfRfbbimGhRvY7xf1O3uOt+jMcNJCa5mHFGCg2SZDA8mebd/mL7g==",
       "cpu": [
         "x64"
       ],
-      "license": "Apache-2.0",
       "optional": true,
       "os": [
         "darwin"
       ]
     },
     "node_modules/@lancedb/vectordb-linux-arm64-gnu": {
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.19.1-beta.5.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.21.2-beta.1.tgz",
-      "integrity": "sha512-VJbBd+Y+6L2SREaOO1OzuUfTPHXyHE4AcsZuM6VMyoeX8k7lPnaA+vNk96o0w4V2KFEAI6o4QPgrRAXmMAzmbg==",
+      "integrity": "sha512-OEsM9znf9DDmdwGuTg2EVu+ebwuWQ1lCx0cYy4+hNy3ntolwMC39ePg2H9WD9SsEnQ2vcGJgBJTQLPKgXww+iQ==",
       "cpu": [
         "arm64"
       ],
-      "license": "Apache-2.0",
       "optional": true,
       "os": [
         "linux"
       ]
     },
     "node_modules/@lancedb/vectordb-linux-x64-gnu": {
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.19.1-beta.5.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.21.2-beta.1.tgz",
-      "integrity": "sha512-3wS8Zn5NmHoszXfrY4JzMimHoh5LAmVi3pTX4gD+C9kVGoUJcDBP7/CrAbjnAz7VzzAIPmz8kvBuPz8l9X4hjw==",
+      "integrity": "sha512-7FTq/O1zNzD71rgX2PEVmkct4jk2wc+ADU3rss+0VqoBSO9XeMqZEVD2WgZWuSTg6bYai//FHGDHSaknHBNsdw==",
       "cpu": [
         "x64"
       ],
-      "license": "Apache-2.0",
       "optional": true,
       "os": [
         "linux"
       ]
     },
     "node_modules/@lancedb/vectordb-win32-x64-msvc": {
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.19.1-beta.5.tgz",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.21.2-beta.1.tgz",
-      "integrity": "sha512-TemM9cvrPa2jFCjvYmKnrL0DTHegi/+LOQ3No9nPDHie2ka2fM9O2q60fAbYsYz+Mo9aV7MvL49ATbNCyl9MLA==",
+      "integrity": "sha512-mN1p/J0kdqy6MrlKtmA8set/PibqFPyytQJFAuxSLXC/rwD7vgqUCt0SI0zVWPGG7J5Y65kvdc99l7Yl7lJtwQ==",
       "cpu": [
         "x64"
       ],
-      "license": "Apache-2.0",
       "optional": true,
       "os": [
         "win32"

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.19.1-beta.5",
+  "version": "0.21.2-beta.1",
   "description": " Serverless, low-latency vector database for AI applications",
   "private": false,
   "main": "dist/index.js",
@@ -89,10 +89,10 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-x64": "0.19.1-beta.5",
+    "@lancedb/vectordb-darwin-x64": "0.21.2-beta.1",
-    "@lancedb/vectordb-darwin-arm64": "0.19.1-beta.5",
+    "@lancedb/vectordb-darwin-arm64": "0.21.2-beta.1",
-    "@lancedb/vectordb-linux-x64-gnu": "0.19.1-beta.5",
+    "@lancedb/vectordb-linux-x64-gnu": "0.21.2-beta.1",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.19.1-beta.5",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.21.2-beta.1",
-    "@lancedb/vectordb-win32-x64-msvc": "0.19.1-beta.5"
+    "@lancedb/vectordb-win32-x64-msvc": "0.21.2-beta.1"
   }
 }

node/src/integration_test/test.ts

@@ -49,7 +49,7 @@ describe('LanceDB Mirrored Store Integration test', function () {
   it('s3://...?mirroredStore=... param is processed correctly', async function () {
     this.timeout(600000)
 
-    const dir = tmpdir()
+    const dir = await fs.promises.mkdtemp(path.join(tmpdir(), 'lancedb-mirror-'))
     console.log(dir)
     const conn = await lancedb.connect({ uri: `s3://lancedb-integtest?mirroredStore=${dir}`, storageOptions: { allowHttp: 'true' } })
     const data = Array(200).fill({ vector: Array(128).fill(1.0), id: 0 })
@@ -63,118 +63,93 @@ describe('LanceDB Mirrored Store Integration test', function () {
     const t = await conn.createTable(tableName, data, { writeMode: lancedb.WriteMode.Overwrite })
 
     const mirroredPath = path.join(dir, `${tableName}.lance`)
-    fs.readdir(mirroredPath, { withFileTypes: true }, (err, files) => {
-      if (err != null) throw err
-      // there should be three dirs
-      assert.equal(files.length, 3)
-      assert.isTrue(files[0].isDirectory())
-      assert.isTrue(files[1].isDirectory())
 
-      fs.readdir(path.join(mirroredPath, '_transactions'), { withFileTypes: true }, (err, files) => {
+    const files = await fs.promises.readdir(mirroredPath, { withFileTypes: true })
-        if (err != null) throw err
+    // there should be three dirs
-        assert.equal(files.length, 1)
+    assert.equal(files.length, 3, 'files after table creation')
-        assert.isTrue(files[0].name.endsWith('.txn'))
+    assert.isTrue(files[0].isDirectory())
-      })
+    assert.isTrue(files[1].isDirectory())
 
-      fs.readdir(path.join(mirroredPath, '_versions'), { withFileTypes: true }, (err, files) => {
+    const transactionFiles = await fs.promises.readdir(path.join(mirroredPath, '_transactions'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(transactionFiles.length, 1, 'transactionFiles after table creation')
-        assert.equal(files.length, 1)
+    assert.isTrue(transactionFiles[0].name.endsWith('.txn'))
-        assert.isTrue(files[0].name.endsWith('.manifest'))
-      })
 
-      fs.readdir(path.join(mirroredPath, 'data'), { withFileTypes: true }, (err, files) => {
+    const versionFiles = await fs.promises.readdir(path.join(mirroredPath, '_versions'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(versionFiles.length, 1, 'versionFiles after table creation')
-        assert.equal(files.length, 1)
+    assert.isTrue(versionFiles[0].name.endsWith('.manifest'))
-        assert.isTrue(files[0].name.endsWith('.lance'))
+
-      })
+    const dataFiles = await fs.promises.readdir(path.join(mirroredPath, 'data'), { withFileTypes: true })
-    })
+    assert.equal(dataFiles.length, 1, 'dataFiles after table creation')
+    assert.isTrue(dataFiles[0].name.endsWith('.lance'))
 
     // try create index and check if it's mirrored
     await t.createIndex({ column: 'vector', type: 'ivf_pq' })
 
-    fs.readdir(mirroredPath, { withFileTypes: true }, (err, files) => {
+    const filesAfterIndex = await fs.promises.readdir(mirroredPath, { withFileTypes: true })
-      if (err != null) throw err
+    // there should be four dirs
-      // there should be four dirs
+    assert.equal(filesAfterIndex.length, 4, 'filesAfterIndex')
-      assert.equal(files.length, 4)
+    assert.isTrue(filesAfterIndex[0].isDirectory())
-      assert.isTrue(files[0].isDirectory())
+    assert.isTrue(filesAfterIndex[1].isDirectory())
-      assert.isTrue(files[1].isDirectory())
+    assert.isTrue(filesAfterIndex[2].isDirectory())
-      assert.isTrue(files[2].isDirectory())
 
-      // Two TXs now
+    // Two TXs now
-      fs.readdir(path.join(mirroredPath, '_transactions'), { withFileTypes: true }, (err, files) => {
+    const transactionFilesAfterIndex = await fs.promises.readdir(path.join(mirroredPath, '_transactions'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(transactionFilesAfterIndex.length, 2, 'transactionFilesAfterIndex')
-        assert.equal(files.length, 2)
+    assert.isTrue(transactionFilesAfterIndex[0].name.endsWith('.txn'))
-        assert.isTrue(files[0].name.endsWith('.txn'))
+    assert.isTrue(transactionFilesAfterIndex[1].name.endsWith('.txn'))
-        assert.isTrue(files[1].name.endsWith('.txn'))
-      })
 
-      fs.readdir(path.join(mirroredPath, 'data'), { withFileTypes: true }, (err, files) => {
+    const dataFilesAfterIndex = await fs.promises.readdir(path.join(mirroredPath, 'data'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(dataFilesAfterIndex.length, 1, 'dataFilesAfterIndex')
-        assert.equal(files.length, 1)
+    assert.isTrue(dataFilesAfterIndex[0].name.endsWith('.lance'))
-        assert.isTrue(files[0].name.endsWith('.lance'))
-      })
 
-      fs.readdir(path.join(mirroredPath, '_indices'), { withFileTypes: true }, (err, files) => {
+    const indicesFiles = await fs.promises.readdir(path.join(mirroredPath, '_indices'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(indicesFiles.length, 1, 'indicesFiles')
-        assert.equal(files.length, 1)
+    assert.isTrue(indicesFiles[0].isDirectory())
-        assert.isTrue(files[0].isDirectory())
 
-        fs.readdir(path.join(mirroredPath, '_indices', files[0].name), { withFileTypes: true }, (err, files) => {
+    const indexFiles = await fs.promises.readdir(path.join(mirroredPath, '_indices', indicesFiles[0].name), { withFileTypes: true })
-          if (err != null) throw err
+    console.log(`DEBUG indexFiles in ${indicesFiles[0].name}:`, indexFiles.map(f => `${f.name} (${f.isFile() ? 'file' : 'dir'})`))
-
+    assert.equal(indexFiles.length, 2, 'indexFiles')
-          assert.equal(files.length, 1)
+    const fileNames = indexFiles.map(f => f.name).sort()
-          assert.isTrue(files[0].isFile())
+    assert.isTrue(fileNames.includes('auxiliary.idx'), 'auxiliary.idx should be present')
-          assert.isTrue(files[0].name.endsWith('.idx'))
+    assert.isTrue(fileNames.includes('index.idx'), 'index.idx should be present')
-        })
+    assert.isTrue(indexFiles.every(f => f.isFile()), 'all index files should be files')
-      })
-    })
 
     // try delete and check if it's mirrored
     await t.delete('id = 0')
 
-    fs.readdir(mirroredPath, { withFileTypes: true }, (err, files) => {
+    const filesAfterDelete = await fs.promises.readdir(mirroredPath, { withFileTypes: true })
-      if (err != null) throw err
+    // there should be five dirs
-      // there should be five dirs
+    assert.equal(filesAfterDelete.length, 5, 'filesAfterDelete')
-      assert.equal(files.length, 5)
+    assert.isTrue(filesAfterDelete[0].isDirectory())
-      assert.isTrue(files[0].isDirectory())
+    assert.isTrue(filesAfterDelete[1].isDirectory())
-      assert.isTrue(files[1].isDirectory())
+    assert.isTrue(filesAfterDelete[2].isDirectory())
-      assert.isTrue(files[2].isDirectory())
+    assert.isTrue(filesAfterDelete[3].isDirectory())
-      assert.isTrue(files[3].isDirectory())
+    assert.isTrue(filesAfterDelete[4].isDirectory())
-      assert.isTrue(files[4].isDirectory())
 
-      // Three TXs now
+    // Three TXs now
-      fs.readdir(path.join(mirroredPath, '_transactions'), { withFileTypes: true }, (err, files) => {
+    const transactionFilesAfterDelete = await fs.promises.readdir(path.join(mirroredPath, '_transactions'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(transactionFilesAfterDelete.length, 3, 'transactionFilesAfterDelete')
-        assert.equal(files.length, 3)
+    assert.isTrue(transactionFilesAfterDelete[0].name.endsWith('.txn'))
-        assert.isTrue(files[0].name.endsWith('.txn'))
+    assert.isTrue(transactionFilesAfterDelete[1].name.endsWith('.txn'))
-        assert.isTrue(files[1].name.endsWith('.txn'))
-      })
 
-      fs.readdir(path.join(mirroredPath, 'data'), { withFileTypes: true }, (err, files) => {
+    const dataFilesAfterDelete = await fs.promises.readdir(path.join(mirroredPath, 'data'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(dataFilesAfterDelete.length, 1, 'dataFilesAfterDelete')
-        assert.equal(files.length, 1)
+    assert.isTrue(dataFilesAfterDelete[0].name.endsWith('.lance'))
-        assert.isTrue(files[0].name.endsWith('.lance'))
-      })
 
-      fs.readdir(path.join(mirroredPath, '_indices'), { withFileTypes: true }, (err, files) => {
+    const indicesFilesAfterDelete = await fs.promises.readdir(path.join(mirroredPath, '_indices'), { withFileTypes: true })
-        if (err != null) throw err
+    assert.equal(indicesFilesAfterDelete.length, 1, 'indicesFilesAfterDelete')
-        assert.equal(files.length, 1)
+    assert.isTrue(indicesFilesAfterDelete[0].isDirectory())
-        assert.isTrue(files[0].isDirectory())
 
-        fs.readdir(path.join(mirroredPath, '_indices', files[0].name), { withFileTypes: true }, (err, files) => {
+    const indexFilesAfterDelete = await fs.promises.readdir(path.join(mirroredPath, '_indices', indicesFilesAfterDelete[0].name), { withFileTypes: true })
-          if (err != null) throw err
+    console.log(`DEBUG indexFilesAfterDelete in ${indicesFilesAfterDelete[0].name}:`, indexFilesAfterDelete.map(f => `${f.name} (${f.isFile() ? 'file' : 'dir'})`))
+    assert.equal(indexFilesAfterDelete.length, 2, 'indexFilesAfterDelete')
+    const fileNamesAfterDelete = indexFilesAfterDelete.map(f => f.name).sort()
+    assert.isTrue(fileNamesAfterDelete.includes('auxiliary.idx'), 'auxiliary.idx should be present after delete')
+    assert.isTrue(fileNamesAfterDelete.includes('index.idx'), 'index.idx should be present after delete')
+    assert.isTrue(indexFilesAfterDelete.every(f => f.isFile()), 'all index files should be files after delete')
 
-          assert.equal(files.length, 1)
+    const deletionFiles = await fs.promises.readdir(path.join(mirroredPath, '_deletions'), { withFileTypes: true })
-          assert.isTrue(files[0].isFile())
+    assert.equal(deletionFiles.length, 1, 'deletionFiles')
-          assert.isTrue(files[0].name.endsWith('.idx'))
+    assert.isTrue(deletionFiles[0].name.endsWith('.arrow'))
-        })
-      })
-
-      fs.readdir(path.join(mirroredPath, '_deletions'), { withFileTypes: true }, (err, files) => {
-        if (err != null) throw err
-        assert.equal(files.length, 1)
-        assert.isTrue(files[0].name.endsWith('.arrow'))
-      })
-    })
   })
 })

nodejs/CLAUDE.md

@@ -0,0 +1,13 @@
+These are the typescript bindings of LanceDB.
+The core Rust library is in the `../rust/lancedb` directory, the rust binding
+code is in the `src/` directory and the typescript bindings are in
+the `lancedb/` directory.
+
+Whenever you change the Rust code, you will need to recompile: `npm run build`.
+
+Common commands:
+* Build: `npm run build`
+* Lint: `npm run lint`
+* Fix lints: `npm run lint-fix`
+* Test: `npm test`
+* Run single test file: `npm test __test__/arrow.test.ts`

nodejs/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.19.1-beta.5"
+version = "0.21.2-beta.1"
 license.workspace = true
 description.workspace = true
 repository.workspace = true
@@ -30,6 +30,7 @@ log.workspace = true
 
 # Workaround for build failure until we can fix it.
 aws-lc-sys = "=0.28.0"
+aws-lc-rs = "=1.13.0"
 
 [build-dependencies]
 napi-build = "2.1"

nodejs/__test__/arrow.test.ts

@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-import { Schema } from "apache-arrow";
+import { Bool, Field, Int32, List, Schema, Struct, Utf8 } from "apache-arrow";
 
 import * as arrow15 from "apache-arrow-15";
 import * as arrow16 from "apache-arrow-16";
@@ -11,10 +11,12 @@ import * as arrow18 from "apache-arrow-18";
 import {
   convertToTable,
   fromBufferToRecordBatch,
+  fromDataToBuffer,
   fromRecordBatchToBuffer,
   fromTableToBuffer,
   makeArrowTable,
   makeEmptyTable,
+  tableFromIPC,
 } from "../lancedb/arrow";
 import {
   EmbeddingFunction,
@@ -375,8 +377,221 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
         expect(table2.schema).toEqual(schema);
       });
 
+      it("will handle missing columns in schema alignment when using embeddings", async function () {
+        const schema = new Schema(
+          [
+            new Field("domain", new Utf8(), true),
+            new Field("name", new Utf8(), true),
+            new Field("description", new Utf8(), true),
+          ],
+          new Map([["embedding_functions", JSON.stringify([])]]),
+        );
+
+        const data = [
+          { domain: "google.com", name: "Google" },
+          { domain: "facebook.com", name: "Facebook" },
+        ];
+
+        const table = await convertToTable(data, undefined, { schema });
+
+        expect(table.numCols).toBe(3);
+        expect(table.numRows).toBe(2);
+
+        const descriptionColumn = table.getChild("description");
+        expect(descriptionColumn).toBeDefined();
+        expect(descriptionColumn?.nullCount).toBe(2);
+        expect(descriptionColumn?.toArray()).toEqual([null, null]);
+
+        expect(table.getChild("domain")?.toArray()).toEqual([
+          "google.com",
+          "facebook.com",
+        ]);
+        expect(table.getChild("name")?.toArray()).toEqual([
+          "Google",
+          "Facebook",
+        ]);
+      });
+
+      it("will handle completely missing nested struct columns", async function () {
+        const schema = new Schema(
+          [
+            new Field("id", new Utf8(), true),
+            new Field("name", new Utf8(), true),
+            new Field(
+              "metadata",
+              new Struct([
+                new Field("version", new Int32(), true),
+                new Field("author", new Utf8(), true),
+                new Field(
+                  "tags",
+                  new List(new Field("item", new Utf8(), true)),
+                  true,
+                ),
+              ]),
+              true,
+            ),
+          ],
+          new Map([["embedding_functions", JSON.stringify([])]]),
+        );
+
+        const data = [
+          { id: "doc1", name: "Document 1" },
+          { id: "doc2", name: "Document 2" },
+        ];
+
+        const table = await convertToTable(data, undefined, { schema });
+
+        expect(table.numCols).toBe(3);
+        expect(table.numRows).toBe(2);
+
+        const buf = await fromTableToBuffer(table);
+        const retrievedTable = tableFromIPC(buf);
+
+        const rows = [];
+        for (let i = 0; i < retrievedTable.numRows; i++) {
+          rows.push(retrievedTable.get(i));
+        }
+
+        expect(rows[0].metadata.version).toBe(null);
+        expect(rows[0].metadata.author).toBe(null);
+        expect(rows[0].metadata.tags).toBe(null);
+        expect(rows[0].id).toBe("doc1");
+        expect(rows[0].name).toBe("Document 1");
+      });
+
+      it("will handle partially missing nested struct fields", async function () {
+        const schema = new Schema(
+          [
+            new Field("id", new Utf8(), true),
+            new Field(
+              "metadata",
+              new Struct([
+                new Field("version", new Int32(), true),
+                new Field("author", new Utf8(), true),
+                new Field("created_at", new Utf8(), true),
+              ]),
+              true,
+            ),
+          ],
+          new Map([["embedding_functions", JSON.stringify([])]]),
+        );
+
+        const data = [
+          { id: "doc1", metadata: { version: 1, author: "Alice" } },
+          { id: "doc2", metadata: { version: 2 } },
+        ];
+
+        const table = await convertToTable(data, undefined, { schema });
+
+        expect(table.numCols).toBe(2);
+        expect(table.numRows).toBe(2);
+
+        const metadataColumn = table.getChild("metadata");
+        expect(metadataColumn).toBeDefined();
+        expect(metadataColumn?.type.toString()).toBe(
+          "Struct<{version:Int32, author:Utf8, created_at:Utf8}>",
+        );
+      });
+
+      it("will handle multiple levels of nested structures", async function () {
+        const schema = new Schema(
+          [
+            new Field("id", new Utf8(), true),
+            new Field(
+              "config",
+              new Struct([
+                new Field("database", new Utf8(), true),
+                new Field(
+                  "connection",
+                  new Struct([
+                    new Field("host", new Utf8(), true),
+                    new Field("port", new Int32(), true),
+                    new Field(
+                      "ssl",
+                      new Struct([
+                        new Field("enabled", new Bool(), true),
+                        new Field("cert_path", new Utf8(), true),
+                      ]),
+                      true,
+                    ),
+                  ]),
+                  true,
+                ),
+              ]),
+              true,
+            ),
+          ],
+          new Map([["embedding_functions", JSON.stringify([])]]),
+        );
+
+        const data = [
+          {
+            id: "config1",
+            config: {
+              database: "postgres",
+              connection: { host: "localhost" },
+            },
+          },
+          {
+            id: "config2",
+            config: { database: "mysql" },
+          },
+          {
+            id: "config3",
+          },
+        ];
+
+        const table = await convertToTable(data, undefined, { schema });
+
+        expect(table.numCols).toBe(2);
+        expect(table.numRows).toBe(3);
+
+        const configColumn = table.getChild("config");
+        expect(configColumn).toBeDefined();
+        expect(configColumn?.type.toString()).toBe(
+          "Struct<{database:Utf8, connection:Struct<{host:Utf8, port:Int32, ssl:Struct<{enabled:Bool, cert_path:Utf8}>}>}>",
+        );
+      });
+
+      it("will handle missing columns in Arrow table input when using embeddings", async function () {
+        const incompleteTable = makeArrowTable([
+          { domain: "google.com", name: "Google" },
+          { domain: "facebook.com", name: "Facebook" },
+        ]);
+
+        const schema = new Schema(
+          [
+            new Field("domain", new Utf8(), true),
+            new Field("name", new Utf8(), true),
+            new Field("description", new Utf8(), true),
+          ],
+          new Map([["embedding_functions", JSON.stringify([])]]),
+        );
+
+        const buf = await fromDataToBuffer(incompleteTable, undefined, schema);
+
+        expect(buf.byteLength).toBeGreaterThan(0);
+
+        const retrievedTable = tableFromIPC(buf);
+        expect(retrievedTable.numCols).toBe(3);
+        expect(retrievedTable.numRows).toBe(2);
+
+        const descriptionColumn = retrievedTable.getChild("description");
+        expect(descriptionColumn).toBeDefined();
+        expect(descriptionColumn?.nullCount).toBe(2);
+        expect(descriptionColumn?.toArray()).toEqual([null, null]);
+
+        expect(retrievedTable.getChild("domain")?.toArray()).toEqual([
+          "google.com",
+          "facebook.com",
+        ]);
+        expect(retrievedTable.getChild("name")?.toArray()).toEqual([
+          "Google",
+          "Facebook",
+        ]);
+      });
+
       it("should correctly retain values in nested struct fields", async function () {
-        // Define test data with nested struct
         const testData = [
           {
             id: "doc1",
@@ -400,10 +615,8 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
           },
         ];
 
-        // Create Arrow table from the data
         const table = makeArrowTable(testData);
 
-        // Verify schema has the nested struct fields
         const metadataField = table.schema.fields.find(
           (f) => f.name === "metadata",
         );
@@ -417,23 +630,17 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
           "text",
         ]);
 
-        // Convert to buffer and back (simulating storage and retrieval)
         const buf = await fromTableToBuffer(table);
         const retrievedTable = tableFromIPC(buf);
 
-        // Verify the retrieved table has the same structure
         const rows = [];
         for (let i = 0; i < retrievedTable.numRows; i++) {
           rows.push(retrievedTable.get(i));
         }
 
-        // Check values in the first row
         const firstRow = rows[0];
         expect(firstRow.id).toBe("doc1");
         expect(firstRow.vector.toJSON()).toEqual([1, 2, 3]);
-
-        // Verify metadata values are preserved (this is where the bug is)
-        expect(firstRow.metadata).toBeDefined();
         expect(firstRow.metadata.filePath).toBe("/path/to/file1.ts");
         expect(firstRow.metadata.startLine).toBe(10);
         expect(firstRow.metadata.endLine).toBe(20);
@@ -592,14 +799,14 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
         ).rejects.toThrow("column vector was missing");
       });
 
-      it("will provide a nice error if run twice", async function () {
+      it("will skip embedding application if already applied", async function () {
         const records = sampleRecords();
         const table = await convertToTable(records, dummyEmbeddingConfig);
 
         // fromTableToBuffer will try and apply the embeddings again
-        await expect(
+        // but should skip since the column already has non-null values
-          fromTableToBuffer(table, dummyEmbeddingConfig),
+        const result = await fromTableToBuffer(table, dummyEmbeddingConfig);
-        ).rejects.toThrow("already existed");
+        expect(result.byteLength).toBeGreaterThan(0);
       });
     });
 

nodejs/__test__/session.test.ts

@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+import * as tmp from "tmp";
+import { Session, connect } from "../lancedb";
+
+describe("Session", () => {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should configure cache sizes and work with database operations", async () => {
+    // Create session with small cache limits for testing
+    const indexCacheSize = BigInt(1024 * 1024); // 1MB
+    const metadataCacheSize = BigInt(512 * 1024); // 512KB
+
+    const session = new Session(indexCacheSize, metadataCacheSize);
+
+    // Record initial cache state
+    const initialCacheSize = session.sizeBytes();
+    const initialCacheItems = session.approxNumItems();
+
+    // Test session works with database connection
+    const db = await connect({ uri: tmpDir.name, session: session });
+
+    // Create and use a table to exercise the session
+    const data = Array.from({ length: 100 }, (_, i) => ({
+      id: i,
+      text: `item ${i}`,
+    }));
+    const table = await db.createTable("test", data);
+    const results = await table.query().limit(5).toArray();
+
+    expect(results).toHaveLength(5);
+
+    // Verify cache usage increased after operations
+    const finalCacheSize = session.sizeBytes();
+    const finalCacheItems = session.approxNumItems();
+
+    expect(finalCacheSize).toBeGreaterThan(initialCacheSize); // Cache should have grown
+    expect(finalCacheItems).toBeGreaterThanOrEqual(initialCacheItems); // Items should not decrease
+    expect(initialCacheSize).toBeLessThan(indexCacheSize + metadataCacheSize); // Within limits
+  });
+});

nodejs/__test__/table.test.ts

@@ -33,7 +33,12 @@ import {
   register,
 } from "../lancedb/embedding";
 import { Index } from "../lancedb/indices";
-import { instanceOfFullTextQuery } from "../lancedb/query";
+import {
+  BooleanQuery,
+  Occur,
+  Operator,
+  instanceOfFullTextQuery,
+} from "../lancedb/query";
 import exp = require("constants");
 
 describe.each([arrow15, arrow16, arrow17, arrow18])(
@@ -363,9 +368,9 @@ describe("merge insert", () => {
       { a: 4, b: "z" },
     ];
 
-    expect(
+    const result = (await table.toArrow()).toArray().sort((a, b) => a.a - b.a);
-      JSON.parse(JSON.stringify((await table.toArrow()).toArray())),
+
-    ).toEqual(expected);
+    expect(result.map((row) => ({ ...row }))).toEqual(expected);
   });
   test("conditional update", async () => {
     const newData = [
@@ -554,6 +559,32 @@ describe("When creating an index", () => {
     rst = await tbl.query().limit(2).offset(1).nearestTo(queryVec).toArrow();
     expect(rst.numRows).toBe(1);
 
+    // test nprobes
+    rst = await tbl.query().nearestTo(queryVec).limit(2).nprobes(50).toArrow();
+    expect(rst.numRows).toBe(2);
+    rst = await tbl
+      .query()
+      .nearestTo(queryVec)
+      .limit(2)
+      .minimumNprobes(15)
+      .toArrow();
+    expect(rst.numRows).toBe(2);
+    rst = await tbl
+      .query()
+      .nearestTo(queryVec)
+      .limit(2)
+      .minimumNprobes(10)
+      .maximumNprobes(20)
+      .toArrow();
+    expect(rst.numRows).toBe(2);
+
+    expect(() => tbl.query().nearestTo(queryVec).minimumNprobes(0)).toThrow(
+      "Invalid input, minimum_nprobes must be greater than 0",
+    );
+    expect(() => tbl.query().nearestTo(queryVec).maximumNprobes(5)).toThrow(
+      "Invalid input, maximum_nprobes must be greater than minimum_nprobes",
+    );
+
     await tbl.dropIndex("vec_idx");
     const indices2 = await tbl.listIndices();
     expect(indices2.length).toBe(0);
@@ -1506,7 +1537,9 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
       ];
       const table = await db.createTable("test", data);
       await table.createIndex("text", {
-        config: Index.fts(),
+        config: Index.fts({
+          withPosition: true,
+        }),
       });
 
       const results = await table.search("lance").toArray();
@@ -1529,6 +1562,18 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
 
       const results = await table.search("hello").toArray();
       expect(results[0].text).toBe(data[0].text);
+
+      const results2 = await table
+        .search(new MatchQuery("hello world", "text"))
+        .toArray();
+      expect(results2.length).toBe(2);
+
+      const results3 = await table
+        .search(
+          new MatchQuery("hello world", "text", { operator: Operator.And }),
+        )
+        .toArray();
+      expect(results3.length).toBe(1);
     });
 
     test("full text search without lowercase", async () => {
@@ -1559,7 +1604,9 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
       ];
       const table = await db.createTable("test", data);
       await table.createIndex("text", {
-        config: Index.fts(),
+        config: Index.fts({
+          withPosition: true,
+        }),
       });
 
       const results = await table.search("world").toArray();
@@ -1603,6 +1650,114 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
       expect(resultSet.has("fob")).toBe(true);
       expect(resultSet.has("fo")).toBe(true);
       expect(resultSet.has("food")).toBe(true);
+
+      const prefixResults = await table
+        .search(
+          new MatchQuery("foo", "text", { fuzziness: 3, prefixLength: 3 }),
+        )
+        .toArray();
+      expect(prefixResults.length).toBe(2);
+      const resultSet2 = new Set(prefixResults.map((r) => r.text));
+      expect(resultSet2.has("foo")).toBe(true);
+      expect(resultSet2.has("food")).toBe(true);
+    });
+
+    test("full text search boolean query", async () => {
+      const db = await connect(tmpDir.name);
+      const data = [
+        { text: "The cat and dog are playing" },
+        { text: "The cat is sleeping" },
+        { text: "The dog is barking" },
+        { text: "The dog chases the cat" },
+      ];
+      const table = await db.createTable("test", data);
+      await table.createIndex("text", {
+        config: Index.fts({ withPosition: false }),
+      });
+
+      const shouldResults = await table
+        .search(
+          new BooleanQuery([
+            [Occur.Should, new MatchQuery("cat", "text")],
+            [Occur.Should, new MatchQuery("dog", "text")],
+          ]),
+        )
+        .toArray();
+      expect(shouldResults.length).toBe(4);
+
+      const mustResults = await table
+        .search(
+          new BooleanQuery([
+            [Occur.Must, new MatchQuery("cat", "text")],
+            [Occur.Must, new MatchQuery("dog", "text")],
+          ]),
+        )
+        .toArray();
+      expect(mustResults.length).toBe(2);
+
+      const mustNotResults = await table
+        .search(
+          new BooleanQuery([
+            [Occur.Must, new MatchQuery("cat", "text")],
+            [Occur.MustNot, new MatchQuery("dog", "text")],
+          ]),
+        )
+        .toArray();
+      expect(mustNotResults.length).toBe(1);
+    });
+
+    test("full text search ngram", async () => {
+      const db = await connect(tmpDir.name);
+      const data = [
+        { text: "hello world", vector: [0.1, 0.2, 0.3] },
+        { text: "lance database", vector: [0.4, 0.5, 0.6] },
+        { text: "lance is cool", vector: [0.7, 0.8, 0.9] },
+      ];
+      const table = await db.createTable("test", data);
+      await table.createIndex("text", {
+        config: Index.fts({ baseTokenizer: "ngram" }),
+      });
+
+      const results = await table.search("lan").toArray();
+      expect(results.length).toBe(2);
+      const resultSet = new Set(results.map((r) => r.text));
+      expect(resultSet.has("lance database")).toBe(true);
+      expect(resultSet.has("lance is cool")).toBe(true);
+
+      const results2 = await table.search("nce").toArray(); // spellchecker:disable-line
+      expect(results2.length).toBe(2);
+      const resultSet2 = new Set(results2.map((r) => r.text));
+      expect(resultSet2.has("lance database")).toBe(true);
+      expect(resultSet2.has("lance is cool")).toBe(true);
+
+      // the default min_ngram_length is 3, so "la" should not match
+      const results3 = await table.search("la").toArray();
+      expect(results3.length).toBe(0);
+
+      // test setting min_ngram_length and prefix_only
+      await table.createIndex("text", {
+        config: Index.fts({
+          baseTokenizer: "ngram",
+          ngramMinLength: 2,
+          prefixOnly: true,
+        }),
+        replace: true,
+      });
+
+      const results4 = await table.search("lan").toArray();
+      expect(results4.length).toBe(2);
+      const resultSet4 = new Set(results4.map((r) => r.text));
+      expect(resultSet4.has("lance database")).toBe(true);
+      expect(resultSet4.has("lance is cool")).toBe(true);
+
+      const results5 = await table.search("nce").toArray(); // spellchecker:disable-line
+      expect(results5.length).toBe(0);
+
+      const results6 = await table.search("la").toArray();
+      expect(results6.length).toBe(2);
+      const resultSet6 = new Set(results6.map((r) => r.text));
+      expect(resultSet6.has("lance database")).toBe(true);
+      expect(resultSet6.has("lance is cool")).toBe(true);
     });
 
     test.each([
@@ -1708,4 +1863,43 @@ describe("column name options", () => {
     expect(results[0].query_index).toBe(0);
     expect(results[1].query_index).toBe(1);
   });
+
+  test("index and search multivectors", async () => {
+    const db = await connect(tmpDir.name);
+    const data = [];
+    // generate 512 random multivectors
+    for (let i = 0; i < 256; i++) {
+      data.push({
+        multivector: Array.from({ length: 10 }, () =>
+          Array(2).fill(Math.random()),
+        ),
+      });
+    }
+    const table = await db.createTable("multivectors", data, {
+      schema: new Schema([
+        new Field(
+          "multivector",
+          new List(
+            new Field(
+              "item",
+              new FixedSizeList(2, new Field("item", new Float32())),
+            ),
+          ),
+        ),
+      ]),
+    });
+
+    const results = await table.search(data[0].multivector).limit(10).toArray();
+    expect(results.length).toBe(10);
+
+    await table.createIndex("multivector", {
+      config: Index.ivfPq({ numPartitions: 2, distanceType: "cosine" }),
+    });
+
+    const results2 = await table
+      .search(data[0].multivector)
+      .limit(10)
+      .toArray();
+    expect(results2.length).toBe(10);
+  });
 });

nodejs/lancedb/arrow.ts

@@ -107,6 +107,20 @@ export type IntoVector =
   | number[]
   | Promise<Float32Array | Float64Array | number[]>;
 
+export type MultiVector = IntoVector[];
+
+export function isMultiVector(value: unknown): value is MultiVector {
+  return Array.isArray(value) && isIntoVector(value[0]);
+}
+
+export function isIntoVector(value: unknown): value is IntoVector {
+  return (
+    value instanceof Float32Array ||
+    value instanceof Float64Array ||
+    (Array.isArray(value) && !Array.isArray(value[0]))
+  );
+}
+
 export function isArrowTable(value: object): value is TableLike {
   if (value instanceof ArrowTable) return true;
   return "schema" in value && "batches" in value;
@@ -417,7 +431,9 @@ function inferSchema(
         } else {
           const inferredType = inferType(value, path, opts);
           if (inferredType === undefined) {
-            throw new Error(`Failed to infer data type for field ${path.join(".")} at row ${rowI}. \
+            throw new Error(`Failed to infer data type for field ${path.join(
+              ".",
+            )} at row ${rowI}. \
                              Consider providing an explicit schema.`);
           }
           pathTree.set(path, inferredType);
@@ -799,11 +815,17 @@ async function applyEmbeddingsFromMetadata(
         `Cannot apply embedding function because the source column '${functionEntry.sourceColumn}' was not present in the data`,
       );
     }
+
+    // Check if destination column exists and handle accordingly
     if (columns[destColumn] !== undefined) {
-      throw new Error(
+      const existingColumn = columns[destColumn];
-        `Attempt to apply embeddings to table failed because column ${destColumn} already existed`,
+      // If the column exists but is all null, we can fill it with embeddings
-      );
+      if (existingColumn.nullCount !== existingColumn.length) {
+        // Column has non-null values, skip embedding application
+        continue;
+      }
     }
+
     if (table.batches.length > 1) {
       throw new Error(
         "Internal error: `makeArrowTable` unexpectedly created a table with more than one batch",
@@ -831,6 +853,15 @@ async function applyEmbeddingsFromMetadata(
     const vector = makeVector(vectors, destType);
     columns[destColumn] = vector;
   }
+
+  // Add any missing columns from the schema as null vectors
+  for (const field of schema.fields) {
+    if (!(field.name in columns)) {
+      const nullValues = new Array(table.numRows).fill(null);
+      columns[field.name] = makeVector(nullValues, field.type);
+    }
+  }
+
   const newTable = new ArrowTable(columns);
   return alignTable(newTable, schema);
 }
@@ -903,11 +934,23 @@ async function applyEmbeddings<T>(
       );
     }
   } else {
+    // Check if destination column exists and handle accordingly
     if (Object.prototype.hasOwnProperty.call(newColumns, destColumn)) {
-      throw new Error(
+      const existingColumn = newColumns[destColumn];
-        `Attempt to apply embeddings to table failed because column ${destColumn} already existed`,
+      // If the column exists but is all null, we can fill it with embeddings
-      );
+      if (existingColumn.nullCount !== existingColumn.length) {
+        // Column has non-null values, skip embedding application and return table as-is
+        let newTable = new ArrowTable(newColumns);
+        if (schema != null) {
+          newTable = alignTable(newTable, schema as Schema);
+        }
+        return new ArrowTable(
+          new Schema(newTable.schema.fields, schemaMetadata),
+          newTable.batches,
+        );
+      }
     }
+
     if (table.batches.length > 1) {
       throw new Error(
         "Internal error: `makeArrowTable` unexpectedly created a table with more than one batch",
@@ -967,7 +1010,21 @@ export async function convertToTable(
   embeddings?: EmbeddingFunctionConfig,
   makeTableOptions?: Partial<MakeArrowTableOptions>,
 ): Promise<ArrowTable> {
-  const table = makeArrowTable(data, makeTableOptions);
+  let processedData = data;
+
+  // If we have a schema with embedding metadata, we need to preprocess the data
+  // to ensure all nested fields are present
+  if (
+    makeTableOptions?.schema &&
+    makeTableOptions.schema.metadata?.has("embedding_functions")
+  ) {
+    processedData = ensureNestedFieldsExist(
+      data,
+      makeTableOptions.schema as Schema,
+    );
+  }
+
+  const table = makeArrowTable(processedData, makeTableOptions);
   return await applyEmbeddings(table, embeddings, makeTableOptions?.schema);
 }
 
@@ -1060,7 +1117,16 @@ export async function fromDataToBuffer(
     schema = sanitizeSchema(schema);
   }
   if (isArrowTable(data)) {
-    return fromTableToBuffer(sanitizeTable(data), embeddings, schema);
+    const table = sanitizeTable(data);
+    // If we have a schema with embedding functions, we need to ensure all columns exist
+    // before applying embeddings, since applyEmbeddingsFromMetadata expects all columns
+    // to be present in the table
+    if (schema && schema.metadata?.has("embedding_functions")) {
+      const alignedTable = alignTableToSchema(table, schema);
+      return fromTableToBuffer(alignedTable, embeddings, schema);
+    } else {
+      return fromTableToBuffer(table, embeddings, schema);
+    }
   } else {
     const table = await convertToTable(data, embeddings, { schema });
     return fromTableToBuffer(table);
@@ -1129,7 +1195,7 @@ function alignBatch(batch: RecordBatch, schema: Schema): RecordBatch {
     type: new Struct(schema.fields),
     length: batch.numRows,
     nullCount: batch.nullCount,
-    children: alignedChildren,
+    children: alignedChildren as unknown as ArrowData<DataType>[],
   });
   return new RecordBatch(schema, newData);
 }
@@ -1201,6 +1267,79 @@ function validateSchemaEmbeddings(
   return new Schema(fields, schema.metadata);
 }
 
+/**
+ * Ensures that all nested fields defined in the schema exist in the data,
+ * filling missing fields with null values.
+ */
+export function ensureNestedFieldsExist(
+  data: Array<Record<string, unknown>>,
+  schema: Schema,
+): Array<Record<string, unknown>> {
+  return data.map((row) => {
+    const completeRow: Record<string, unknown> = {};
+
+    for (const field of schema.fields) {
+      if (field.name in row) {
+        if (
+          field.type.constructor.name === "Struct" &&
+          row[field.name] !== null &&
+          row[field.name] !== undefined
+        ) {
+          // Handle nested struct
+          const nestedValue = row[field.name] as Record<string, unknown>;
+          completeRow[field.name] = ensureStructFieldsExist(
+            nestedValue,
+            field.type,
+          );
+        } else {
+          // Non-struct field or null struct value
+          completeRow[field.name] = row[field.name];
+        }
+      } else {
+        // Field is missing from the data - set to null
+        completeRow[field.name] = null;
+      }
+    }
+
+    return completeRow;
+  });
+}
+
+/**
+ * Recursively ensures that all fields in a struct type exist in the data,
+ * filling missing fields with null values.
+ */
+function ensureStructFieldsExist(
+  data: Record<string, unknown>,
+  structType: Struct,
+): Record<string, unknown> {
+  const completeStruct: Record<string, unknown> = {};
+
+  for (const childField of structType.children) {
+    if (childField.name in data) {
+      if (
+        childField.type.constructor.name === "Struct" &&
+        data[childField.name] !== null &&
+        data[childField.name] !== undefined
+      ) {
+        // Recursively handle nested struct
+        completeStruct[childField.name] = ensureStructFieldsExist(
+          data[childField.name] as Record<string, unknown>,
+          childField.type,
+        );
+      } else {
+        // Non-struct field or null struct value
+        completeStruct[childField.name] = data[childField.name];
+      }
+    } else {
+      // Field is missing - set to null
+      completeStruct[childField.name] = null;
+    }
+  }
+
+  return completeStruct;
+}
+
 interface JsonDataType {
   type: string;
   fields?: JsonField[];
@@ -1334,3 +1473,64 @@ function fieldToJson(field: Field): JsonField {
     metadata: field.metadata,
   };
 }
+
+function alignTableToSchema(
+  table: ArrowTable,
+  targetSchema: Schema,
+): ArrowTable {
+  const existingColumns = new Map<string, Vector>();
+
+  // Map existing columns
+  for (const field of table.schema.fields) {
+    existingColumns.set(field.name, table.getChild(field.name)!);
+  }
+
+  // Create vectors for all fields in target schema
+  const alignedColumns: Record<string, Vector> = {};
+
+  for (const field of targetSchema.fields) {
+    if (existingColumns.has(field.name)) {
+      // Column exists, use it
+      alignedColumns[field.name] = existingColumns.get(field.name)!;
+    } else {
+      // Column missing, create null vector
+      alignedColumns[field.name] = createNullVector(field, table.numRows);
+    }
+  }
+
+  // Create new table with aligned schema and columns
+  return new ArrowTable(targetSchema, alignedColumns);
+}
+
+function createNullVector(field: Field, numRows: number): Vector {
+  if (field.type.constructor.name === "Struct") {
+    // For struct types, create a struct with null fields
+    const structType = field.type as Struct;
+    const childVectors = structType.children.map((childField) =>
+      createNullVector(childField, numRows),
+    );
+
+    // Create struct data
+    const structData = makeData({
+      type: structType,
+      length: numRows,
+      nullCount: 0,
+      children: childVectors.map((v) => v.data[0]),
+    });
+
+    return arrowMakeVector(structData);
+  } else {
+    // For other types, create a vector of nulls
+    const nullBitmap = new Uint8Array(Math.ceil(numRows / 8));
+    // All bits are 0, meaning all values are null
+
+    const data = makeData({
+      type: field.type,
+      length: numRows,
+      nullCount: numRows,
+      nullBitmap,
+    });
+
+    return arrowMakeVector(data);
+  }
+}

nodejs/lancedb/connection.ts

@@ -85,6 +85,9 @@ export interface OpenTableOptions {
   /**
    * Set the size of the index cache, specified as a number of entries
    *
+   * @deprecated Use session-level cache configuration instead.
+   * Create a Session with custom cache sizes and pass it to the connect() function.
+   *
    * The exact meaning of an "entry" will depend on the type of index:
    * - IVF: there is one entry for each IVF partition
    * - BTREE: there is one entry for the entire index

nodejs/lancedb/index.ts

@@ -10,6 +10,7 @@ import {
 import {
   ConnectionOptions,
   Connection as LanceDbConnection,
+  Session,
 } from "./native.js";
 
 export {
@@ -51,6 +52,8 @@ export {
   OpenTableOptions,
 } from "./connection";
 
+export { Session } from "./native.js";
+
 export {
   ExecutableQuery,
   Query,
@@ -64,7 +67,10 @@ export {
   PhraseQuery,
   BoostQuery,
   MultiMatchQuery,
+  BooleanQuery,
   FullTextQueryType,
+  Operator,
+  Occur,
 } from "./query";
 
 export {
@@ -97,6 +103,7 @@ export {
   RecordBatchLike,
   DataLike,
   IntoVector,
+  MultiVector,
 } from "./arrow";
 export { IntoSql, packBits } from "./util";
 
@@ -127,6 +134,7 @@ export { IntoSql, packBits } from "./util";
 export async function connect(
   uri: string,
   options?: Partial<ConnectionOptions>,
+  session?: Session,
 ): Promise<Connection>;
 /**
  * Connect to a LanceDB instance at the given URI.
@@ -145,31 +153,43 @@ export async function connect(
  *   storageOptions: {timeout: "60s"}
  * });
  * ```
+ *
+ * @example
+ * ```ts
+ * const session = Session.default();
+ * const conn = await connect({
+ *   uri: "/path/to/database",
+ *   session: session
+ * });
+ * ```
  */
 export async function connect(
   options: Partial<ConnectionOptions> & { uri: string },
 ): Promise<Connection>;
 export async function connect(
   uriOrOptions: string | (Partial<ConnectionOptions> & { uri: string }),
-  options: Partial<ConnectionOptions> = {},
+  options?: Partial<ConnectionOptions>,
 ): Promise<Connection> {
   let uri: string | undefined;
+  let finalOptions: Partial<ConnectionOptions> = {};
+
   if (typeof uriOrOptions !== "string") {
     const { uri: uri_, ...opts } = uriOrOptions;
     uri = uri_;
-    options = opts;
+    finalOptions = opts;
   } else {
     uri = uriOrOptions;
+    finalOptions = options || {};
   }
 
   if (!uri) {
     throw new Error("uri is required");
   }
 
-  options = (options as ConnectionOptions) ?? {};
+  finalOptions = (finalOptions as ConnectionOptions) ?? {};
-  (<ConnectionOptions>options).storageOptions = cleanseStorageOptions(
+  (<ConnectionOptions>finalOptions).storageOptions = cleanseStorageOptions(
-    (<ConnectionOptions>options).storageOptions,
+    (<ConnectionOptions>finalOptions).storageOptions,
   );
-  const nativeConn = await LanceDbConnection.new(uri, options);
+  const nativeConn = await LanceDbConnection.new(uri, finalOptions);
   return new LocalConnection(nativeConn);
 }

nodejs/lancedb/indices.ts

@@ -439,7 +439,7 @@ export interface FtsOptions {
    *
    * "raw" - Raw tokenizer. This tokenizer does not split the text into tokens and indexes the entire text as a single token.
    */
-  baseTokenizer?: "simple" | "whitespace" | "raw";
+  baseTokenizer?: "simple" | "whitespace" | "raw" | "ngram";
 
   /**
    * language for stemming and stop words
@@ -472,6 +472,21 @@ export interface FtsOptions {
    * whether to remove punctuation
    */
   asciiFolding?: boolean;
+
+  /**
+   * ngram min length
+   */
+  ngramMinLength?: number;
+
+  /**
+   * ngram max length
+   */
+  ngramMaxLength?: number;
+
+  /**
+   * whether to only index the prefix of the token for ngram tokenizer
+   */
+  prefixOnly?: boolean;
 }
 
 export class Index {
@@ -608,6 +623,9 @@ export class Index {
         options?.stem,
         options?.removeStopWords,
         options?.asciiFolding,
+        options?.ngramMinLength,
+        options?.ngramMaxLength,
+        options?.prefixOnly,
       ),
     );
   }

nodejs/lancedb/query.ts

@@ -448,6 +448,10 @@ export class VectorQuery extends QueryBase<NativeVectorQuery> {
    * For best results we recommend tuning this parameter with a benchmark against
    * your actual data to find the smallest possible value that will still give
    * you the desired recall.
+   *
+   * For more fine grained control over behavior when you have a very narrow filter
+   * you can use `minimumNprobes` and `maximumNprobes`.  This method sets both
+   * the minimum and maximum to the same value.
    */
   nprobes(nprobes: number): VectorQuery {
     super.doCall((inner) => inner.nprobes(nprobes));
@@ -455,6 +459,33 @@ export class VectorQuery extends QueryBase<NativeVectorQuery> {
     return this;
   }
 
+  /**
+   * Set the minimum number of probes used.
+   *
+   * This controls the minimum number of partitions that will be searched.  This
+   * parameter will impact every query against a vector index, regardless of the
+   * filter.  See `nprobes` for more details.  Higher values will increase recall
+   * but will also increase latency.
+   */
+  minimumNprobes(minimumNprobes: number): VectorQuery {
+    super.doCall((inner) => inner.minimumNprobes(minimumNprobes));
+    return this;
+  }
+
+  /**
+   * Set the maximum number of probes used.
+   *
+   * This controls the maximum number of partitions that will be searched.  If this
+   * number is greater than minimumNprobes then the excess partitions will _only_ be
+   * searched if we have not found enough results.  This can be useful when there is
+   * a narrow filter to allow these queries to spend more time searching and avoid
+   * potential false negatives.
+   */
+  maximumNprobes(maximumNprobes: number): VectorQuery {
+    super.doCall((inner) => inner.maximumNprobes(maximumNprobes));
+    return this;
+  }
+
   /*
    * Set the distance range to use
    *
@@ -762,6 +793,31 @@ export enum FullTextQueryType {
   MatchPhrase = "match_phrase",
   Boost = "boost",
   MultiMatch = "multi_match",
+  Boolean = "boolean",
+}
+
+/**
+ * Enum representing the logical operators used in full-text queries.
+ *
+ * - `And`: All terms must match.
+ * - `Or`: At least one term must match.
+ */
+export enum Operator {
+  And = "AND",
+  Or = "OR",
+}
+
+/**
+ * Enum representing the occurrence of terms in full-text queries.
+ *
+ * - `Must`: The term must be present in the document.
+ * - `Should`: The term should contribute to the document score, but is not required.
+ * - `MustNot`: The term must not be present in the document.
+ */
+export enum Occur {
+  Should = "SHOULD",
+  Must = "MUST",
+  MustNot = "MUST_NOT",
 }
 
 /**
@@ -791,6 +847,7 @@ export function instanceOfFullTextQuery(obj: any): obj is FullTextQuery {
 export class MatchQuery implements FullTextQuery {
   /** @ignore */
   public readonly inner: JsFullTextQuery;
+
   /**
    * Creates an instance of MatchQuery.
    *
@@ -800,6 +857,8 @@ export class MatchQuery implements FullTextQuery {
    *   - `boost`: The boost factor for the query (default is 1.0).
    *   - `fuzziness`: The fuzziness level for the query (default is 0).
    *   - `maxExpansions`: The maximum number of terms to consider for fuzzy matching (default is 50).
+   *   - `operator`: The logical operator to use for combining terms in the query (default is "OR").
+   *   - `prefixLength`: The number of beginning characters being unchanged for fuzzy matching.
    */
   constructor(
     query: string,
@@ -808,6 +867,8 @@ export class MatchQuery implements FullTextQuery {
       boost?: number;
       fuzziness?: number;
       maxExpansions?: number;
+      operator?: Operator;
+      prefixLength?: number;
     },
   ) {
     let fuzziness = options?.fuzziness;
@@ -820,6 +881,8 @@ export class MatchQuery implements FullTextQuery {
       options?.boost ?? 1.0,
       fuzziness,
       options?.maxExpansions ?? 50,
+      options?.operator ?? Operator.Or,
+      options?.prefixLength ?? 0,
     );
   }
 
@@ -836,9 +899,11 @@ export class PhraseQuery implements FullTextQuery {
    *
    * @param query - The phrase to search for in the specified column.
    * @param column - The name of the column to search within.
+   * @param options - Optional parameters for the phrase query.
+   *   - `slop`: The maximum number of intervening unmatched positions allowed between words in the phrase (default is 0).
    */
-  constructor(query: string, column: string) {
+  constructor(query: string, column: string, options?: { slop?: number }) {
-    this.inner = JsFullTextQuery.phraseQuery(query, column);
+    this.inner = JsFullTextQuery.phraseQuery(query, column, options?.slop ?? 0);
   }
 
   queryType(): FullTextQueryType {
@@ -889,18 +954,21 @@ export class MultiMatchQuery implements FullTextQuery {
    * @param columns - An array of column names to search within.
    * @param options - Optional parameters for the multi-match query.
    *  - `boosts`: An array of boost factors for each column (default is 1.0 for all).
+   *  - `operator`: The logical operator to use for combining terms in the query (default is "OR").
    */
   constructor(
     query: string,
     columns: string[],
     options?: {
       boosts?: number[];
+      operator?: Operator;
     },
   ) {
     this.inner = JsFullTextQuery.multiMatchQuery(
       query,
       columns,
       options?.boosts,
+      options?.operator ?? Operator.Or,
     );
   }
 
@@ -908,3 +976,23 @@ export class MultiMatchQuery implements FullTextQuery {
     return FullTextQueryType.MultiMatch;
   }
 }
+
+export class BooleanQuery implements FullTextQuery {
+  /** @ignore */
+  public readonly inner: JsFullTextQuery;
+  /**
+   * Creates an instance of BooleanQuery.
+   *
+   * @param queries - An array of (Occur, FullTextQuery objects) to combine.
+   * Occur specifies whether the query must match, or should match.
+   */
+  constructor(queries: [Occur, FullTextQuery][]) {
+    this.inner = JsFullTextQuery.booleanQuery(
+      queries.map(([occur, query]) => [occur, query.inner]),
+    );
+  }
+
+  queryType(): FullTextQueryType {
+    return FullTextQueryType.Boolean;
+  }
+}

nodejs/lancedb/table.ts

@@ -6,9 +6,11 @@ import {
   Data,
   DataType,
   IntoVector,
+  MultiVector,
   Schema,
   dataTypeToJson,
   fromDataToBuffer,
+  isMultiVector,
   tableFromIPC,
 } from "./arrow";
 
@@ -75,10 +77,10 @@ export interface OptimizeOptions {
    * // Delete all versions older than 1 day
    * const olderThan = new Date();
    * olderThan.setDate(olderThan.getDate() - 1));
-   * tbl.cleanupOlderVersions(olderThan);
+   * tbl.optimize({cleanupOlderThan: olderThan});
    *
    * // Delete all versions except the current version
-   * tbl.cleanupOlderVersions(new Date());
+   * tbl.optimize({cleanupOlderThan: new Date()});
    */
   cleanupOlderThan: Date;
   deleteUnverified: boolean;
@@ -346,7 +348,7 @@ export abstract class Table {
    * if the query is a string and no embedding function is defined, it will be treated as a full text search query
    */
   abstract search(
-    query: string | IntoVector | FullTextQuery,
+    query: string | IntoVector | MultiVector | FullTextQuery,
     queryType?: string,
     ftsColumns?: string | string[],
   ): VectorQuery | Query;
@@ -357,7 +359,7 @@ export abstract class Table {
    * is the same thing as calling `nearestTo` on the builder returned
    * by `query`.  @see {@link Query#nearestTo} for more details.
    */
-  abstract vectorSearch(vector: IntoVector): VectorQuery;
+  abstract vectorSearch(vector: IntoVector | MultiVector): VectorQuery;
   /**
    * Add new columns with defined values.
    * @param {AddColumnsSql[]} newColumnTransforms pairs of column names and
@@ -668,7 +670,7 @@ export class LocalTable extends Table {
   }
 
   search(
-    query: string | IntoVector | FullTextQuery,
+    query: string | IntoVector | MultiVector | FullTextQuery,
     queryType: string = "auto",
     ftsColumns?: string | string[],
   ): VectorQuery | Query {
@@ -715,7 +717,15 @@ export class LocalTable extends Table {
     return this.query().nearestTo(queryPromise);
   }
 
-  vectorSearch(vector: IntoVector): VectorQuery {
+  vectorSearch(vector: IntoVector | MultiVector): VectorQuery {
+    if (isMultiVector(vector)) {
+      const query = this.query().nearestTo(vector[0]);
+      for (const v of vector.slice(1)) {
+        query.addQueryVector(v);
+      }
+      return query;
+    }
+
     return this.query().nearestTo(vector);
   }
 

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",

nodejs/npm/darwin-x64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-x64",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["darwin"],
 	"cpu": ["x64"],
 	"main": "lancedb.darwin-x64.node",

nodejs/npm/linux-arm64-gnu/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",

nodejs/npm/linux-arm64-musl/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-musl",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-musl.node",

nodejs/npm/linux-x64-gnu/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",

nodejs/npm/linux-x64-musl/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-musl",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-musl.node",

nodejs/npm/win32-arm64-msvc/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lancedb/lancedb-win32-arm64-msvc",
-  "version": "0.19.1-beta.5",
+  "version": "0.21.2-beta.1",
   "os": [
     "win32"
   ],

nodejs/npm/win32-x64-msvc/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.19.1-beta.5",
+	"version": "0.21.2-beta.1",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.19.1-beta.5",
+  "version": "0.21.2-beta.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.19.1-beta.5",
+      "version": "0.21.2-beta.1",
       "cpu": [
         "x64",
         "arm64"

nodejs/package.json

@@ -11,7 +11,7 @@
     "ann"
   ],
   "private": false,
-  "version": "0.19.1-beta.5",
+  "version": "0.21.2-beta.1",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",

nodejs/src/connection.rs

@@ -74,6 +74,10 @@ impl Connection {
             builder = builder.host_override(&host_override);
         }
 
+        if let Some(session) = options.session {
+            builder = builder.session(session.inner.clone());
+        }
+
         Ok(Self::inner_new(builder.execute().await.default_error()?))
     }
 

nodejs/src/index.rs

@@ -123,34 +123,44 @@ impl Index {
         stem: Option<bool>,
         remove_stop_words: Option<bool>,
         ascii_folding: Option<bool>,
+        ngram_min_length: Option<u32>,
+        ngram_max_length: Option<u32>,
+        prefix_only: Option<bool>,
     ) -> Self {
         let mut opts = FtsIndexBuilder::default();
-        let mut tokenizer_configs = opts.tokenizer_configs.clone();
         if let Some(with_position) = with_position {
             opts = opts.with_position(with_position);
         }
         if let Some(base_tokenizer) = base_tokenizer {
-            tokenizer_configs = tokenizer_configs.base_tokenizer(base_tokenizer);
+            opts = opts.base_tokenizer(base_tokenizer);
         }
         if let Some(language) = language {
-            tokenizer_configs = tokenizer_configs.language(&language).unwrap();
+            opts = opts.language(&language).unwrap();
         }
         if let Some(max_token_length) = max_token_length {
-            tokenizer_configs = tokenizer_configs.max_token_length(Some(max_token_length as usize));
+            opts = opts.max_token_length(Some(max_token_length as usize));
         }
         if let Some(lower_case) = lower_case {
-            tokenizer_configs = tokenizer_configs.lower_case(lower_case);
+            opts = opts.lower_case(lower_case);
         }
         if let Some(stem) = stem {
-            tokenizer_configs = tokenizer_configs.stem(stem);
+            opts = opts.stem(stem);
         }
         if let Some(remove_stop_words) = remove_stop_words {
-            tokenizer_configs = tokenizer_configs.remove_stop_words(remove_stop_words);
+            opts = opts.remove_stop_words(remove_stop_words);
         }
         if let Some(ascii_folding) = ascii_folding {
-            tokenizer_configs = tokenizer_configs.ascii_folding(ascii_folding);
+            opts = opts.ascii_folding(ascii_folding);
+        }
+        if let Some(ngram_min_length) = ngram_min_length {
+            opts = opts.ngram_min_length(ngram_min_length);
+        }
+        if let Some(ngram_max_length) = ngram_max_length {
+            opts = opts.ngram_max_length(ngram_max_length);
+        }
+        if let Some(prefix_only) = prefix_only {
+            opts = opts.ngram_prefix_only(prefix_only);
         }
-        opts.tokenizer_configs = tokenizer_configs;
 
         Self {
             inner: Mutex::new(Some(LanceDbIndex::FTS(opts))),

nodejs/src/lib.rs

@@ -14,6 +14,7 @@ pub mod merge;
 mod query;
 pub mod remote;
 mod rerankers;
+mod session;
 mod table;
 mod util;
 
@@ -34,6 +35,9 @@ pub struct ConnectionOptions {
     ///
     /// The available options are described at https://lancedb.github.io/lancedb/guides/storage/
     pub storage_options: Option<HashMap<String, String>>,
+    /// (For LanceDB OSS only): the session to use for this connection. Holds
+    /// shared caches and other session-specific state.
+    pub session: Option<session::Session>,
 
     /// (For LanceDB cloud only): configuration for the remote HTTP client.
     pub client_config: Option<remote::ClientConfig>,

nodejs/src/query.rs

@@ -4,7 +4,8 @@
 use std::sync::Arc;
 
 use lancedb::index::scalar::{
-    BoostQuery, FtsQuery, FullTextSearchQuery, MatchQuery, MultiMatchQuery, PhraseQuery,
+    BooleanQuery, BoostQuery, FtsQuery, FullTextSearchQuery, MatchQuery, MultiMatchQuery, Occur,
+    Operator, PhraseQuery,
 };
 use lancedb::query::ExecutableQuery;
 use lancedb::query::Query as LanceDbQuery;
@@ -177,6 +178,31 @@ impl VectorQuery {
         self.inner = self.inner.clone().nprobes(nprobe as usize);
     }
 
+    #[napi]
+    pub fn minimum_nprobes(&mut self, minimum_nprobe: u32) -> napi::Result<()> {
+        self.inner = self
+            .inner
+            .clone()
+            .minimum_nprobes(minimum_nprobe as usize)
+            .default_error()?;
+        Ok(())
+    }
+
+    #[napi]
+    pub fn maximum_nprobes(&mut self, maximum_nprobes: u32) -> napi::Result<()> {
+        let maximum_nprobes = if maximum_nprobes == 0 {
+            None
+        } else {
+            Some(maximum_nprobes as usize)
+        };
+        self.inner = self
+            .inner
+            .clone()
+            .maximum_nprobes(maximum_nprobes)
+            .default_error()?;
+        Ok(())
+    }
+
     #[napi]
     pub fn distance_range(&mut self, lower_bound: Option<f64>, upper_bound: Option<f64>) {
         // napi doesn't support f32, so we have to convert to f32
@@ -308,6 +334,8 @@ impl JsFullTextQuery {
         boost: f64,
         fuzziness: Option<u32>,
         max_expansions: u32,
+        operator: String,
+        prefix_length: u32,
     ) -> napi::Result<Self> {
         Ok(Self {
             inner: MatchQuery::new(query)
@@ -315,14 +343,23 @@ impl JsFullTextQuery {
                 .with_boost(boost as f32)
                 .with_fuzziness(fuzziness)
                 .with_max_expansions(max_expansions as usize)
+                .with_operator(
+                    Operator::try_from(operator.as_str()).map_err(|e| {
+                        napi::Error::from_reason(format!("Invalid operator: {}", e))
+                    })?,
+                )
+                .with_prefix_length(prefix_length)
                 .into(),
         })
     }
 
     #[napi(factory)]
-    pub fn phrase_query(query: String, column: String) -> napi::Result<Self> {
+    pub fn phrase_query(query: String, column: String, slop: u32) -> napi::Result<Self> {
         Ok(Self {
-            inner: PhraseQuery::new(query).with_column(Some(column)).into(),
+            inner: PhraseQuery::new(query)
+                .with_column(Some(column))
+                .with_slop(slop)
+                .into(),
         })
     }
 
@@ -348,6 +385,7 @@ impl JsFullTextQuery {
         query: String,
         columns: Vec<String>,
         boosts: Option<Vec<f64>>,
+        operator: String,
     ) -> napi::Result<Self> {
         let q = match boosts {
             Some(boosts) => MultiMatchQuery::try_new(query, columns)
@@ -358,7 +396,37 @@ impl JsFullTextQuery {
             napi::Error::from_reason(format!("Failed to create multi match query: {}", e))
         })?;
 
-        Ok(Self { inner: q.into() })
+        let operator = Operator::try_from(operator.as_str()).map_err(|e| {
+            napi::Error::from_reason(format!("Invalid operator for multi match query: {}", e))
+        })?;
+
+        Ok(Self {
+            inner: q.with_operator(operator).into(),
+        })
+    }
+
+    #[napi(factory)]
+    pub fn boolean_query(queries: Vec<(String, &JsFullTextQuery)>) -> napi::Result<Self> {
+        let mut sub_queries = Vec::with_capacity(queries.len());
+        for (occur, q) in queries {
+            let occur = Occur::try_from(occur.as_str())
+                .map_err(|e| napi::Error::from_reason(e.to_string()))?;
+            sub_queries.push((occur, q.inner.clone()));
+        }
+        Ok(Self {
+            inner: BooleanQuery::new(sub_queries).into(),
+        })
+    }
+
+    #[napi(getter)]
+    pub fn query_type(&self) -> String {
+        match self.inner {
+            FtsQuery::Match(_) => "match".to_string(),
+            FtsQuery::Phrase(_) => "phrase".to_string(),
+            FtsQuery::Boost(_) => "boost".to_string(),
+            FtsQuery::MultiMatch(_) => "multi_match".to_string(),
+            FtsQuery::Boolean(_) => "boolean".to_string(),
+        }
     }
 }
 

nodejs/src/session.rs

@@ -0,0 +1,102 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+use std::sync::Arc;
+
+use lancedb::{ObjectStoreRegistry, Session as LanceSession};
+use napi::bindgen_prelude::*;
+use napi_derive::*;
+
+/// A session for managing caches and object stores across LanceDB operations.
+///
+/// Sessions allow you to configure cache sizes for index and metadata caches,
+/// which can significantly impact memory use and performance. They can
+/// also be re-used across multiple connections to share the same cache state.
+#[napi]
+#[derive(Clone)]
+pub struct Session {
+    pub(crate) inner: Arc<LanceSession>,
+}
+
+impl std::fmt::Debug for Session {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("Session")
+            .field("size_bytes", &self.inner.size_bytes())
+            .field("approx_num_items", &self.inner.approx_num_items())
+            .finish()
+    }
+}
+
+#[napi]
+impl Session {
+    /// Create a new session with custom cache sizes.
+    ///
+    /// # Parameters
+    ///
+    /// - `index_cache_size_bytes`: The size of the index cache in bytes.
+    ///   Index data is stored in memory in this cache to speed up queries.
+    ///   Defaults to 6GB if not specified.
+    /// - `metadata_cache_size_bytes`: The size of the metadata cache in bytes.
+    ///   The metadata cache stores file metadata and schema information in memory.
+    ///   This cache improves scan and write performance.
+    ///   Defaults to 1GB if not specified.
+    #[napi(constructor)]
+    pub fn new(
+        index_cache_size_bytes: Option<BigInt>,
+        metadata_cache_size_bytes: Option<BigInt>,
+    ) -> napi::Result<Self> {
+        let index_cache_size = index_cache_size_bytes
+            .map(|size| size.get_u64().1 as usize)
+            .unwrap_or(6 * 1024 * 1024 * 1024); // 6GB default
+
+        let metadata_cache_size = metadata_cache_size_bytes
+            .map(|size| size.get_u64().1 as usize)
+            .unwrap_or(1024 * 1024 * 1024); // 1GB default
+
+        let session = LanceSession::new(
+            index_cache_size,
+            metadata_cache_size,
+            Arc::new(ObjectStoreRegistry::default()),
+        );
+
+        Ok(Self {
+            inner: Arc::new(session),
+        })
+    }
+
+    /// Create a session with default cache sizes.
+    ///
+    /// This is equivalent to creating a session with 6GB index cache
+    /// and 1GB metadata cache.
+    #[napi(factory)]
+    pub fn default() -> Self {
+        Self {
+            inner: Arc::new(LanceSession::default()),
+        }
+    }
+
+    /// Get the current size of the session caches in bytes.
+    #[napi]
+    pub fn size_bytes(&self) -> BigInt {
+        BigInt::from(self.inner.size_bytes())
+    }
+
+    /// Get the approximate number of items cached in the session.
+    #[napi]
+    pub fn approx_num_items(&self) -> u32 {
+        self.inner.approx_num_items() as u32
+    }
+}
+
+// Implement FromNapiValue for Session to work with napi(object)
+impl napi::bindgen_prelude::FromNapiValue for Session {
+    unsafe fn from_napi_value(
+        env: napi::sys::napi_env,
+        napi_val: napi::sys::napi_value,
+    ) -> napi::Result<Self> {
+        let object: napi::bindgen_prelude::ClassInstance<Session> =
+            napi::bindgen_prelude::ClassInstance::from_napi_value(env, napi_val)?;
+        let copy = object.clone();
+        Ok(copy)
+    }
+}

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.22.1"
+current_version = "0.24.2"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.

python/CLAUDE.md

@@ -0,0 +1,19 @@
+These are the Python bindings of LanceDB.
+The core Rust library is in the `../rust/lancedb` directory, the rust binding
+code is in the `src/` directory and the Python bindings are in the `lancedb/` directory.
+
+Common commands:
+
+* Build: `make develop`
+* Format: `make format`
+* Lint: `make check`
+* Fix lints: `make fix`
+* Test: `make test`
+* Doc test: `make doctest`
+
+Before committing changes, run lints and then formatting.
+
+When you change the Rust code, you will need to recompile the Python bindings: `make develop`.
+
+When you export new types from Rust to Python, you must manually update `python/lancedb/_lancedb.pyi`
+with the corresponding type hints. You can run `pyright` to check for type errors in the Python code.

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.22.1"
+version = "0.24.2"
 edition.workspace = true
 description = "Python bindings for LanceDB"
 license.workspace = true
@@ -14,11 +14,11 @@ name = "_lancedb"
 crate-type = ["cdylib"]
 
 [dependencies]
-arrow = { version = "54.1", features = ["pyarrow"] }
+arrow = { version = "55.1", features = ["pyarrow"] }
 lancedb = { path = "../rust/lancedb", default-features = false }
 env_logger.workspace = true
-pyo3 = { version = "0.23", features = ["extension-module", "abi3-py39"] }
+pyo3 = { version = "0.24", features = ["extension-module", "abi3-py39"] }
-pyo3-async-runtimes = { version = "0.23", features = [
+pyo3-async-runtimes = { version = "0.24", features = [
     "attributes",
     "tokio-runtime",
 ] }
@@ -27,7 +27,7 @@ futures.workspace = true
 tokio = { version = "1.40", features = ["sync"] }
 
 [build-dependencies]
-pyo3-build-config = { version = "0.23", features = [
+pyo3-build-config = { version = "0.24", features = [
     "extension-module",
     "abi3-py39",
 ] }

python/pyproject.toml

@@ -60,6 +60,7 @@ tests = [
     "pyarrow-stubs",
     "pylance>=0.25",
     "requests",
+    "datafusion",
 ]
 dev = [
     "ruff",
@@ -84,8 +85,8 @@ embeddings = [
     "boto3>=1.28.57",
     "awscli>=1.29.57",
     "botocore>=1.31.57",
-    "ollama",
+    'ibm-watsonx-ai>=1.1.2; python_version >= "3.10"',
-    "ibm-watsonx-ai>=1.1.2",
+    "ollama>=0.3.0",
 ]
 azure = ["adlfs>=2024.2.0"]
 

python/python/lancedb/__init__.py

@@ -18,6 +18,7 @@ from .remote import ClientConfig
 from .remote.db import RemoteDBConnection
 from .schema import vector
 from .table import AsyncTable
+from ._lancedb import Session
 
 
 def connect(
@@ -30,6 +31,7 @@ def connect(
     request_thread_pool: Optional[Union[int, ThreadPoolExecutor]] = None,
     client_config: Union[ClientConfig, Dict[str, Any], None] = None,
     storage_options: Optional[Dict[str, str]] = None,
+    session: Optional[Session] = None,
     **kwargs: Any,
 ) -> DBConnection:
     """Connect to a LanceDB database.
@@ -64,6 +66,12 @@ def connect(
     storage_options: dict, optional
         Additional options for the storage backend. See available options at
         <https://lancedb.github.io/lancedb/guides/storage/>
+    session: Session, optional
+        (For LanceDB OSS only)
+        A session to use for this connection. Sessions allow you to configure
+        cache sizes for index and metadata caches, which can significantly
+        impact memory use and performance. They can also be re-used across
+        multiple connections to share the same cache state.
 
     Examples
     --------
@@ -92,7 +100,7 @@ def connect(
         if api_key is None:
             api_key = os.environ.get("LANCEDB_API_KEY")
         if api_key is None:
-            raise ValueError(f"api_key is required to connected LanceDB cloud: {uri}")
+            raise ValueError(f"api_key is required to connect to LanceDB cloud: {uri}")
         if isinstance(request_thread_pool, int):
             request_thread_pool = ThreadPoolExecutor(request_thread_pool)
         return RemoteDBConnection(
@@ -113,6 +121,7 @@ def connect(
         uri,
         read_consistency_interval=read_consistency_interval,
         storage_options=storage_options,
+        session=session,
     )
 
 
@@ -125,6 +134,7 @@ async def connect_async(
     read_consistency_interval: Optional[timedelta] = None,
     client_config: Optional[Union[ClientConfig, Dict[str, Any]]] = None,
     storage_options: Optional[Dict[str, str]] = None,
+    session: Optional[Session] = None,
 ) -> AsyncConnection:
     """Connect to a LanceDB database.
 
@@ -158,6 +168,12 @@ async def connect_async(
     storage_options: dict, optional
         Additional options for the storage backend. See available options at
         <https://lancedb.github.io/lancedb/guides/storage/>
+    session: Session, optional
+        (For LanceDB OSS only)
+        A session to use for this connection. Sessions allow you to configure
+        cache sizes for index and metadata caches, which can significantly
+        impact memory use and performance. They can also be re-used across
+        multiple connections to share the same cache state.
 
     Examples
     --------
@@ -197,6 +213,7 @@ async def connect_async(
             read_consistency_interval_secs,
             client_config,
             storage_options,
+            session,
         )
     )
 
@@ -212,6 +229,7 @@ __all__ = [
     "DBConnection",
     "LanceDBConnection",
     "RemoteDBConnection",
+    "Session",
     "__version__",
 ]
 

python/python/lancedb/_lancedb.pyi

@@ -6,6 +6,19 @@ import pyarrow as pa
 from .index import BTree, IvfFlat, IvfPq, Bitmap, LabelList, HnswPq, HnswSq, FTS
 from .remote import ClientConfig
 
+class Session:
+    def __init__(
+        self,
+        index_cache_size_bytes: Optional[int] = None,
+        metadata_cache_size_bytes: Optional[int] = None,
+    ): ...
+    @staticmethod
+    def default() -> "Session": ...
+    @property
+    def size_bytes(self) -> int: ...
+    @property
+    def approx_num_items(self) -> int: ...
+
 class Connection(object):
     uri: str
     async def table_names(
@@ -89,6 +102,7 @@ async def connect(
     read_consistency_interval: Optional[float],
     client_config: Optional[Union[ClientConfig, Dict[str, Any]]],
     storage_options: Optional[Dict[str, str]],
+    session: Optional[Session],
 ) -> Connection: ...
 
 class RecordBatchStream:
@@ -143,6 +157,8 @@ class VectorQuery:
     def postfilter(self): ...
     def refine_factor(self, refine_factor: int): ...
     def nprobes(self, nprobes: int): ...
+    def minimum_nprobes(self, minimum_nprobes: int): ...
+    def maximum_nprobes(self, maximum_nprobes: int): ...
     def bypass_vector_index(self): ...
     def nearest_to_text(self, query: dict) -> HybridQuery: ...
     def to_query_request(self) -> PyQueryRequest: ...
@@ -158,6 +174,8 @@ class HybridQuery:
     def distance_type(self, distance_type: str): ...
     def refine_factor(self, refine_factor: int): ...
     def nprobes(self, nprobes: int): ...
+    def minimum_nprobes(self, minimum_nprobes: int): ...
+    def maximum_nprobes(self, maximum_nprobes: int): ...
     def bypass_vector_index(self): ...
     def to_vector_query(self) -> VectorQuery: ...
     def to_fts_query(self) -> FTSQuery: ...
@@ -165,23 +183,21 @@ class HybridQuery:
     def get_with_row_id(self) -> bool: ...
     def to_query_request(self) -> PyQueryRequest: ...
 
-class PyFullTextSearchQuery:
+class FullTextQuery:
-    columns: Optional[List[str]]
+    pass
-    query: str
-    limit: Optional[int]
-    wand_factor: Optional[float]
 
 class PyQueryRequest:
     limit: Optional[int]
     offset: Optional[int]
     filter: Optional[Union[str, bytes]]
-    full_text_search: Optional[PyFullTextSearchQuery]
+    full_text_search: Optional[FullTextQuery]
     select: Optional[Union[str, List[str]]]
     fast_search: Optional[bool]
     with_row_id: Optional[bool]
     column: Optional[str]
     query_vector: Optional[List[pa.Array]]
-    nprobes: Optional[int]
+    minimum_nprobes: Optional[int]
+    maximum_nprobes: Optional[int]
     lower_bound: Optional[float]
     upper_bound: Optional[float]
     ef: Optional[int]

python/python/lancedb/common.py

@@ -94,9 +94,9 @@ def data_to_reader(
     else:
         raise TypeError(
             f"Unknown data type {type(data)}. "
-            "Please check "
+            "Supported types: list of dicts, pandas DataFrame, polars DataFrame, "
-            "https://lancedb.github.io/lance/read_and_write.html "
+            "pyarrow Table/RecordBatch, or Pydantic models. "
-            "to see supported types."
+            "See https://lancedb.github.io/lancedb/guides/tables/ for examples."
         )
 
 

python/python/lancedb/db.py

@@ -37,6 +37,7 @@ if TYPE_CHECKING:
     from ._lancedb import Connection as LanceDbConnection
     from .common import DATA, URI
     from .embeddings import EmbeddingFunctionConfig
+    from ._lancedb import Session
 
 
 class DBConnection(EnforceOverrides):
@@ -247,6 +248,9 @@ class DBConnection(EnforceOverrides):
         name: str
             The name of the table.
         index_cache_size: int, default 256
+            **Deprecated**: Use session-level cache configuration instead.
+            Create a Session with custom cache sizes and pass it to lancedb.connect().
+
             Set the size of the index cache, specified as a number of entries
 
             The exact meaning of an "entry" will depend on the type of index:
@@ -354,6 +358,7 @@ class LanceDBConnection(DBConnection):
         *,
         read_consistency_interval: Optional[timedelta] = None,
         storage_options: Optional[Dict[str, str]] = None,
+        session: Optional[Session] = None,
     ):
         if not isinstance(uri, Path):
             scheme = get_uri_scheme(uri)
@@ -367,6 +372,7 @@ class LanceDBConnection(DBConnection):
         self._entered = False
         self.read_consistency_interval = read_consistency_interval
         self.storage_options = storage_options
+        self.session = session
 
         if read_consistency_interval is not None:
             read_consistency_interval_secs = read_consistency_interval.total_seconds()
@@ -382,6 +388,7 @@ class LanceDBConnection(DBConnection):
                 read_consistency_interval_secs,
                 None,
                 storage_options,
+                session,
             )
 
         self._conn = AsyncConnection(LOOP.run(do_connect()))
@@ -475,6 +482,17 @@ class LanceDBConnection(DBConnection):
         -------
         A LanceTable object representing the table.
         """
+        if index_cache_size is not None:
+            import warnings
+
+            warnings.warn(
+                "index_cache_size is deprecated. Use session-level cache "
+                "configuration instead. Create a Session with custom cache sizes "
+                "and pass it to lancedb.connect().",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+
         return LanceTable.open(
             self,
             name,
@@ -820,6 +838,9 @@ class AsyncConnection(object):
             See available options at
             <https://lancedb.github.io/lancedb/guides/storage/>
         index_cache_size: int, default 256
+            **Deprecated**: Use session-level cache configuration instead.
+            Create a Session with custom cache sizes and pass it to lancedb.connect().
+
             Set the size of the index cache, specified as a number of entries
 
             The exact meaning of an "entry" will depend on the type of index:

python/python/lancedb/embeddings/__init__.py

@@ -11,7 +11,7 @@ from .instructor import InstructorEmbeddingFunction
 from .ollama import OllamaEmbeddings
 from .open_clip import OpenClipEmbeddings
 from .openai import OpenAIEmbeddings
-from .registry import EmbeddingFunctionRegistry, get_registry
+from .registry import EmbeddingFunctionRegistry, get_registry, register
 from .sentence_transformers import SentenceTransformerEmbeddings
 from .gte import GteEmbeddings
 from .transformers import TransformersEmbeddingFunction, ColbertEmbeddings

python/python/lancedb/embeddings/gte_mlx_model.py

@@ -9,11 +9,14 @@ from huggingface_hub import snapshot_download
 from pydantic import BaseModel
 from transformers import BertTokenizer
 
+from .utils import create_import_stub
+
 try:
     import mlx.core as mx
     import mlx.nn as nn
 except ImportError:
-    raise ImportError("You need to install MLX to use this model use - pip install mlx")
+    mx = create_import_stub("mlx.core", "mlx")
+    nn = create_import_stub("mlx.nn", "mlx")
 
 
 def average_pool(last_hidden_state: mx.array, attention_mask: mx.array) -> mx.array:
@@ -72,7 +75,7 @@ class TransformerEncoder(nn.Module):
         super().__init__()
         self.layers = [
             TransformerEncoderLayer(dims, num_heads, mlp_dims)
-            for i in range(num_layers)
+            for _ in range(num_layers)
         ]
 
     def __call__(self, x, mask):

python/python/lancedb/embeddings/ollama.py

@@ -2,14 +2,15 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 from functools import cached_property
-from typing import TYPE_CHECKING, List, Optional, Union
+from typing import TYPE_CHECKING, List, Optional, Sequence, Union
+
+import numpy as np
 
 from ..util import attempt_import_or_raise
 from .base import TextEmbeddingFunction
 from .registry import register
 
 if TYPE_CHECKING:
-    import numpy as np
     import ollama
 
 
@@ -28,23 +29,21 @@ class OllamaEmbeddings(TextEmbeddingFunction):
     keep_alive: Optional[Union[float, str]] = None
     ollama_client_kwargs: Optional[dict] = {}
 
-    def ndims(self):
+    def ndims(self) -> int:
         return len(self.generate_embeddings(["foo"])[0])
 
-    def _compute_embedding(self, text) -> Union["np.array", None]:
+    def _compute_embedding(self, text: Sequence[str]) -> Sequence[Sequence[float]]:
-        return (
+        response = self._ollama_client.embed(
-            self._ollama_client.embeddings(
+            model=self.name,
-                model=self.name,
+            input=text,
-                prompt=text,
+            options=self.options,
-                options=self.options,
+            keep_alive=self.keep_alive,
-                keep_alive=self.keep_alive,
-            )["embedding"]
-            or None
         )
+        return response.embeddings
 
     def generate_embeddings(
-        self, texts: Union[List[str], "np.ndarray"]
+        self, texts: Union[List[str], np.ndarray]
-    ) -> list[Union["np.array", None]]:
+    ) -> list[Union[np.array, None]]:
         """
         Get the embeddings for the given texts
 
@@ -54,8 +53,8 @@ class OllamaEmbeddings(TextEmbeddingFunction):
             The texts to embed
         """
         # TODO retry, rate limit, token limit
-        embeddings = [self._compute_embedding(text) for text in texts]
+        embeddings = self._compute_embedding(texts)
-        return embeddings
+        return list(embeddings)
 
     @cached_property
     def _ollama_client(self) -> "ollama.Client":

python/python/lancedb/embeddings/registry.py

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 import json
-from typing import Dict, Optional
+from typing import Dict, Optional, Type
 
 from .base import EmbeddingFunction, EmbeddingFunctionConfig
 
@@ -43,7 +43,7 @@ class EmbeddingFunctionRegistry:
         self._functions = {}
         self._variables = {}
 
-    def register(self, alias: str = None):
+    def register(self, alias: Optional[str] = None):
         """
         This creates a decorator that can be used to register
         an EmbeddingFunction.
@@ -75,7 +75,7 @@ class EmbeddingFunctionRegistry:
         """
         self._functions = {}
 
-    def get(self, name: str):
+    def get(self, name: str) -> Type[EmbeddingFunction]:
         """
         Fetch an embedding function class by name
 

python/python/lancedb/embeddings/utils.py

@@ -21,6 +21,36 @@ from ..dependencies import pandas as pd
 from ..util import attempt_import_or_raise
 
 
+def create_import_stub(module_name: str, package_name: str = None):
+    """
+    Create a stub module that allows class definition but fails when used.
+    This allows modules to be imported for doctest collection even when
+    optional dependencies are not available.
+
+    Parameters
+    ----------
+    module_name : str
+        The name of the module to create a stub for
+    package_name : str, optional
+        The package name to suggest in the error message
+
+    Returns
+    -------
+    object
+        A stub object that can be used in place of the module
+    """
+
+    class _ImportStub:
+        def __getattr__(self, name):
+            return _ImportStub  # Return stub for chained access like nn.Module
+
+        def __call__(self, *args, **kwargs):
+            pkg = package_name or module_name
+            raise ImportError(f"You need to install {pkg} to use this functionality")
+
+    return _ImportStub()
+
+
 # ruff: noqa: PERF203
 def retry(tries=10, delay=1, max_delay=30, backoff=3, jitter=1):
     def wrapper(fn):

python/python/lancedb/index.py

@@ -102,7 +102,7 @@ class FTS:
 
     Attributes
     ----------
-    with_position : bool, default True
+    with_position : bool, default False
         Whether to store the position of the token in the document. Setting this
         to False can reduce the size of the index and improve indexing speed,
         but it will disable support for phrase queries.
@@ -118,25 +118,28 @@ class FTS:
         ignored.
     lower_case : bool, default True
         Whether to convert the token to lower case. This makes queries case-insensitive.
-    stem : bool, default False
+    stem : bool, default True
         Whether to stem the token. Stemming reduces words to their root form.
         For example, in English "running" and "runs" would both be reduced to "run".
-    remove_stop_words : bool, default False
+    remove_stop_words : bool, default True
         Whether to remove stop words. Stop words are common words that are often
         removed from text before indexing. For example, in English "the" and "and".
-    ascii_folding : bool, default False
+    ascii_folding : bool, default True
         Whether to fold ASCII characters. This converts accented characters to
         their ASCII equivalent. For example, "café" would be converted to "cafe".
     """
 
-    with_position: bool = True
+    with_position: bool = False
     base_tokenizer: Literal["simple", "raw", "whitespace"] = "simple"
     language: str = "English"
     max_token_length: Optional[int] = 40
     lower_case: bool = True
-    stem: bool = False
+    stem: bool = True
-    remove_stop_words: bool = False
+    remove_stop_words: bool = True
-    ascii_folding: bool = False
+    ascii_folding: bool = True
+    ngram_min_length: int = 3
+    ngram_max_length: int = 3
+    prefix_only: bool = False
 
 
 @dataclass

python/python/lancedb/query.py

@@ -4,7 +4,6 @@
 from __future__ import annotations
 
 from abc import ABC, abstractmethod
-import abc
 from concurrent.futures import ThreadPoolExecutor
 from enum import Enum
 from datetime import timedelta
@@ -15,7 +14,7 @@ from typing import (
     Literal,
     Optional,
     Tuple,
-    Type,
+    TypeVar,
     Union,
     Any,
 )
@@ -59,6 +58,8 @@ if TYPE_CHECKING:
     else:
         from typing_extensions import Self
 
+T = TypeVar("T", bound="LanceModel")
+
 
 # Pydantic validation function for vector queries
 def ensure_vector_query(
@@ -88,15 +89,28 @@ def ensure_vector_query(
         return val
 
 
-class FullTextQueryType(Enum):
+class FullTextQueryType(str, Enum):
     MATCH = "match"
     MATCH_PHRASE = "match_phrase"
     BOOST = "boost"
     MULTI_MATCH = "multi_match"
+    BOOLEAN = "boolean"
 
 
-class FullTextQuery(abc.ABC, pydantic.BaseModel):
+class FullTextOperator(str, Enum):
-    @abc.abstractmethod
+    AND = "AND"
+    OR = "OR"
+
+
+class Occur(str, Enum):
+    SHOULD = "SHOULD"
+    MUST = "MUST"
+    MUST_NOT = "MUST_NOT"
+
+
+@pydantic.dataclasses.dataclass
+class FullTextQuery(ABC):
+    @abstractmethod
     def query_type(self) -> FullTextQueryType:
         """
         Get the query type of the query.
@@ -106,193 +120,178 @@ class FullTextQuery(abc.ABC, pydantic.BaseModel):
         str
             The type of the query.
         """
+        pass
 
-    @abc.abstractmethod
+    def __and__(self, other: "FullTextQuery") -> "FullTextQuery":
-    def to_dict(self) -> dict:
         """
-        Convert the query to a dictionary.
+        Combine two queries with a logical AND operation.
-
-        Returns
-        -------
-        dict
-            The query as a dictionary.
-        """
-
-
-class MatchQuery(FullTextQuery):
-    query: str
-    column: str
-    boost: float = 1.0
-    fuzziness: int = 0
-    max_expansions: int = 50
-
-    def __init__(
-        self,
-        query: str,
-        column: str,
-        *,
-        boost: float = 1.0,
-        fuzziness: int = 0,
-        max_expansions: int = 50,
-    ):
-        """
-        Match query for full-text search.
 
         Parameters
         ----------
-        query : str
+        other : FullTextQuery
-            The query string to match against.
+            The other query to combine with.
-        column : str
+
-            The name of the column to match against.
+        Returns
-        boost : float, default 1.0
+        -------
-            The boost factor for the query.
+        FullTextQuery
-            The score of each matching document is multiplied by this value.
+            A new query that combines both queries with AND.
-        fuzziness : int, optional
-            The maximum edit distance for each term in the match query.
-            Defaults to 0 (exact match).
-            If None, fuzziness is applied automatically by the rules:
-                - 0 for terms with length <= 2
-                - 1 for terms with length <= 5
-                - 2 for terms with length > 5
-        max_expansions : int, optional
-            The maximum number of terms to consider for fuzzy matching.
-            Defaults to 50.
         """
-        super().__init__(
+        return BooleanQuery([(Occur.MUST, self), (Occur.MUST, other)])
-            query=query,
+
-            column=column,
+    def __or__(self, other: "FullTextQuery") -> "FullTextQuery":
-            boost=boost,
+        """
-            fuzziness=fuzziness,
+        Combine two queries with a logical OR operation.
-            max_expansions=max_expansions,
+
-        )
+        Parameters
+        ----------
+        other : FullTextQuery
+            The other query to combine with.
+
+        Returns
+        -------
+        FullTextQuery
+            A new query that combines both queries with OR.
+        """
+        return BooleanQuery([(Occur.SHOULD, self), (Occur.SHOULD, other)])
+
+
+@pydantic.dataclasses.dataclass
+class MatchQuery(FullTextQuery):
+    """
+    Match query for full-text search.
+
+    Parameters
+    ----------
+    query : str
+        The query string to match against.
+    column : str
+        The name of the column to match against.
+    boost : float, default 1.0
+        The boost factor for the query.
+        The score of each matching document is multiplied by this value.
+    fuzziness : int, optional
+        The maximum edit distance for each term in the match query.
+        Defaults to 0 (exact match).
+        If None, fuzziness is applied automatically by the rules:
+            - 0 for terms with length <= 2
+            - 1 for terms with length <= 5
+            - 2 for terms with length > 5
+    max_expansions : int, optional
+        The maximum number of terms to consider for fuzzy matching.
+        Defaults to 50.
+    operator : FullTextOperator, default OR
+        The operator to use for combining the query results.
+        Can be either `AND` or `OR`.
+        If `AND`, all terms in the query must match.
+        If `OR`, at least one term in the query must match.
+    prefix_length : int, optional
+        The number of beginning characters being unchanged for fuzzy matching.
+        This is useful to achieve prefix matching.
+    """
+
+    query: str
+    column: str
+    boost: float = pydantic.Field(1.0, kw_only=True)
+    fuzziness: int = pydantic.Field(0, kw_only=True)
+    max_expansions: int = pydantic.Field(50, kw_only=True)
+    operator: FullTextOperator = pydantic.Field(FullTextOperator.OR, kw_only=True)
+    prefix_length: int = pydantic.Field(0, kw_only=True)
 
     def query_type(self) -> FullTextQueryType:
         return FullTextQueryType.MATCH
 
-    def to_dict(self) -> dict:
-        return {
-            "match": {
-                self.column: {
-                    "query": self.query,
-                    "boost": self.boost,
-                    "fuzziness": self.fuzziness,
-                    "max_expansions": self.max_expansions,
-                }
-            }
-        }
-
 
+@pydantic.dataclasses.dataclass
 class PhraseQuery(FullTextQuery):
+    """
+    Phrase query for full-text search.
+
+    Parameters
+    ----------
+    query : str
+        The query string to match against.
+    column : str
+        The name of the column to match against.
+    """
+
     query: str
     column: str
-
+    slop: int = pydantic.Field(0, kw_only=True)
-    def __init__(self, query: str, column: str):
-        """
-        Phrase query for full-text search.
-
-        Parameters
-        ----------
-        query : str
-            The query string to match against.
-        column : str
-            The name of the column to match against.
-        """
-        super().__init__(query=query, column=column)
 
     def query_type(self) -> FullTextQueryType:
         return FullTextQueryType.MATCH_PHRASE
 
-    def to_dict(self) -> dict:
-        return {
-            "match_phrase": {
-                self.column: self.query,
-            }
-        }
-
 
+@pydantic.dataclasses.dataclass
 class BoostQuery(FullTextQuery):
+    """
+    Boost query for full-text search.
+
+    Parameters
+    ----------
+    positive : dict
+        The positive query object.
+    negative : dict
+        The negative query object.
+    negative_boost : float, default 0.5
+        The boost factor for the negative query.
+    """
+
     positive: FullTextQuery
     negative: FullTextQuery
-    negative_boost: float = 0.5
+    negative_boost: float = pydantic.Field(0.5, kw_only=True)
-
-    def __init__(
-        self,
-        positive: FullTextQuery,
-        negative: FullTextQuery,
-        *,
-        negative_boost: float = 0.5,
-    ):
-        """
-        Boost query for full-text search.
-
-        Parameters
-        ----------
-        positive : dict
-            The positive query object.
-        negative : dict
-            The negative query object.
-        negative_boost : float
-            The boost factor for the negative query.
-        """
-        super().__init__(
-            positive=positive, negative=negative, negative_boost=negative_boost
-        )
 
     def query_type(self) -> FullTextQueryType:
         return FullTextQueryType.BOOST
 
-    def to_dict(self) -> dict:
-        return {
-            "boost": {
-                "positive": self.positive.to_dict(),
-                "negative": self.negative.to_dict(),
-                "negative_boost": self.negative_boost,
-            }
-        }
-
 
+@pydantic.dataclasses.dataclass
 class MultiMatchQuery(FullTextQuery):
+    """
+    Multi-match query for full-text search.
+
+    Parameters
+    ----------
+    query : str | list[Query]
+        If a string, the query string to match against.
+    columns : list[str]
+        The list of columns to match against.
+    boosts : list[float], optional
+        The list of boost factors for each column. If not provided,
+        all columns will have the same boost factor.
+    operator : FullTextOperator, default OR
+        The operator to use for combining the query results.
+        Can be either `AND` or `OR`.
+        It would be applied to all columns individually.
+        For example, if the operator is `AND`,
+        then the query "hello world" is equal to
+        `match("hello AND world", column1) OR match("hello AND world", column2)`.
+    """
+
     query: str
     columns: list[str]
-    boosts: list[float]
+    boosts: Optional[list[float]] = pydantic.Field(None, kw_only=True)
-
+    operator: FullTextOperator = pydantic.Field(FullTextOperator.OR, kw_only=True)
-    def __init__(
-        self,
-        query: str,
-        columns: list[str],
-        *,
-        boosts: Optional[list[float]] = None,
-    ):
-        """
-        Multi-match query for full-text search.
-
-        Parameters
-        ----------
-        query : str
-            The query string to match against.
-
-        columns : list[str]
-            The list of columns to match against.
-
-        boosts : list[float], optional
-            The list of boost factors for each column. If not provided,
-            all columns will have the same boost factor.
-        """
-        if boosts is None:
-            boosts = [1.0] * len(columns)
-        super().__init__(query=query, columns=columns, boosts=boosts)
 
     def query_type(self) -> FullTextQueryType:
         return FullTextQueryType.MULTI_MATCH
 
-    def to_dict(self) -> dict:
+
-        return {
+@pydantic.dataclasses.dataclass
-            "multi_match": {
+class BooleanQuery(FullTextQuery):
-                "query": self.query,
+    """
-                "columns": self.columns,
+    Boolean query for full-text search.
-                "boost": self.boosts,
+
-            }
+    Parameters
-        }
+    ----------
+    queries : list[tuple(Occur, FullTextQuery)]
+        The list of queries with their occurrence requirements.
+    """
+
+    queries: list[tuple[Occur, FullTextQuery]]
+
+    def query_type(self) -> FullTextQueryType:
+        return FullTextQueryType.BOOLEAN
 
 
 class FullTextSearchQuery(pydantic.BaseModel):
@@ -445,8 +444,18 @@ class Query(pydantic.BaseModel):
     # which columns to return in the results
     columns: Optional[Union[List[str], Dict[str, str]]] = None
 
-    # number of IVF partitions to search
+    # minimum number of IVF partitions to search
-    nprobes: Optional[int] = None
+    #
+    # If None then a default value (20) will be used.
+    minimum_nprobes: Optional[int] = None
+
+    # maximum number of IVF partitions to search
+    #
+    # If None then a default value (20) will be used.
+    #
+    # If 0 then no limit will be applied and all partitions could be searched
+    # if needed to satisfy the limit.
+    maximum_nprobes: Optional[int] = None
 
     # lower bound for distance search
     lower_bound: Optional[float] = None
@@ -484,7 +493,8 @@ class Query(pydantic.BaseModel):
         query.vector_column = req.column
         query.vector = req.query_vector
         query.distance_type = req.distance_type
-        query.nprobes = req.nprobes
+        query.minimum_nprobes = req.minimum_nprobes
+        query.maximum_nprobes = req.maximum_nprobes
         query.lower_bound = req.lower_bound
         query.upper_bound = req.upper_bound
         query.ef = req.ef
@@ -493,10 +503,8 @@ class Query(pydantic.BaseModel):
         query.postfilter = req.postfilter
         if req.full_text_search is not None:
             query.full_text_query = FullTextSearchQuery(
-                columns=req.full_text_search.columns,
+                columns=None,
-                query=req.full_text_search.query,
+                query=req.full_text_search,
-                limit=req.full_text_search.limit,
-                wand_factor=req.full_text_search.wand_factor,
             )
         return query
 
@@ -740,8 +748,8 @@ class LanceQueryBuilder(ABC):
         return self.to_arrow(timeout=timeout).to_pylist()
 
     def to_pydantic(
-        self, model: Type[LanceModel], *, timeout: Optional[timedelta] = None
+        self, model: type[T], *, timeout: Optional[timedelta] = None
-    ) -> List[LanceModel]:
+    ) -> list[T]:
         """Return the table as a list of pydantic models.
 
         Parameters
@@ -900,11 +908,11 @@ class LanceQueryBuilder(ABC):
         >>> plan = table.search(query).explain_plan(True)
         >>> print(plan) # doctest: +ELLIPSIS, +NORMALIZE_WHITESPACE
         ProjectionExec: expr=[vector@0 as vector, _distance@2 as _distance]
-        GlobalLimitExec: skip=0, fetch=10
+          GlobalLimitExec: skip=0, fetch=10
-          FilterExec: _distance@2 IS NOT NULL
+            FilterExec: _distance@2 IS NOT NULL
-            SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], preserve_partitioning=[false]
+              SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], preserve_partitioning=[false]
-              KNNVectorDistance: metric=l2
+                KNNVectorDistance: metric=l2
-                LanceScan: uri=..., projection=[vector], row_id=true, row_addr=false, ordered=false
+                  LanceRead: uri=..., projection=[vector], ...
 
         Parameters
         ----------
@@ -934,19 +942,19 @@ class LanceQueryBuilder(ABC):
         >>> plan = table.search(query).analyze_plan()
         >>> print(plan)  # doctest: +ELLIPSIS, +NORMALIZE_WHITESPACE
         AnalyzeExec verbose=true, metrics=[]
-          ProjectionExec: expr=[...], metrics=[...]
+          TracedExec, metrics=[]
-            GlobalLimitExec: skip=0, fetch=10, metrics=[...]
+            ProjectionExec: expr=[...], metrics=[...]
-              FilterExec: _distance@2 IS NOT NULL,
+              GlobalLimitExec: skip=0, fetch=10, metrics=[...]
-              metrics=[output_rows=..., elapsed_compute=...]
+                FilterExec: _distance@2 IS NOT NULL,
-                SortExec: TopK(fetch=10), expr=[...],
+                metrics=[output_rows=..., elapsed_compute=...]
-                preserve_partitioning=[...],
+                  SortExec: TopK(fetch=10), expr=[...],
-                metrics=[output_rows=..., elapsed_compute=..., row_replacements=...]
+                  preserve_partitioning=[...],
-                  KNNVectorDistance: metric=l2,
+                  metrics=[output_rows=..., elapsed_compute=..., row_replacements=...]
-                  metrics=[output_rows=..., elapsed_compute=..., output_batches=...]
+                    KNNVectorDistance: metric=l2,
-                    LanceScan: uri=..., projection=[vector], row_id=true,
+                    metrics=[output_rows=..., elapsed_compute=..., output_batches=...]
-                    row_addr=false, ordered=false,
+                      LanceRead: uri=..., projection=[vector], ...
-                    metrics=[output_rows=..., elapsed_compute=...,
+                      metrics=[output_rows=..., elapsed_compute=...,
-                    bytes_read=..., iops=..., requests=...]
+                      bytes_read=..., iops=..., requests=...]
 
         Returns
         -------
@@ -1047,7 +1055,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         super().__init__(table)
         self._query = query
         self._distance_type = None
-        self._nprobes = None
+        self._minimum_nprobes = None
+        self._maximum_nprobes = None
         self._lower_bound = None
         self._upper_bound = None
         self._refine_factor = None
@@ -1110,6 +1119,10 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         See discussion in [Querying an ANN Index][querying-an-ann-index] for
         tuning advice.
 
+        This method sets both the minimum and maximum number of probes to the same
+        value. See `minimum_nprobes` and `maximum_nprobes` for more fine-grained
+        control.
+
         Parameters
         ----------
         nprobes: int
@@ -1120,7 +1133,36 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         LanceVectorQueryBuilder
             The LanceQueryBuilder object.
         """
-        self._nprobes = nprobes
+        self._minimum_nprobes = nprobes
+        self._maximum_nprobes = nprobes
+        return self
+
+    def minimum_nprobes(self, minimum_nprobes: int) -> LanceVectorQueryBuilder:
+        """Set the minimum number of probes to use.
+
+        See `nprobes` for more details.
+
+        These partitions will be searched on every vector query and will increase recall
+        at the expense of latency.
+        """
+        self._minimum_nprobes = minimum_nprobes
+        return self
+
+    def maximum_nprobes(self, maximum_nprobes: int) -> LanceVectorQueryBuilder:
+        """Set the maximum number of probes to use.
+
+        See `nprobes` for more details.
+
+        If this value is greater than `minimum_nprobes` then the excess partitions
+        will be searched only if we have not found enough results.
+
+        This can be useful when there is a narrow filter to allow these queries to
+        spend more time searching and avoid potential false negatives.
+
+        If this value is 0 then no limit will be applied and all partitions could be
+        searched if needed to satisfy the limit.
+        """
+        self._maximum_nprobes = maximum_nprobes
         return self
 
     def distance_range(
@@ -1224,7 +1266,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
             limit=self._limit,
             distance_type=self._distance_type,
             columns=self._columns,
-            nprobes=self._nprobes,
+            minimum_nprobes=self._minimum_nprobes,
+            maximum_nprobes=self._maximum_nprobes,
             lower_bound=self._lower_bound,
             upper_bound=self._upper_bound,
             refine_factor=self._refine_factor,
@@ -1333,6 +1376,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         if query_string is not None and not isinstance(query_string, str):
             raise ValueError("Reranking currently only supports string queries")
         self._str_query = query_string if query_string is not None else self._str_query
+        if reranker.score == "all":
+            self.with_row_id(True)
         return self
 
     def bypass_vector_index(self) -> LanceVectorQueryBuilder:
@@ -1410,10 +1455,13 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
 
         query = self._query
         if self._phrase_query:
-            raise NotImplementedError(
+            if isinstance(query, str):
-                "Phrase query is not yet supported in Lance FTS. "
+                if not query.startswith('"') or not query.endswith('"'):
-                "Use tantivy-based index instead for now."
+                    query = f'"{query}"'
-            )
+            elif isinstance(query, FullTextQuery) and not isinstance(
+                query, PhraseQuery
+            ):
+                raise TypeError("Please use PhraseQuery for phrase queries.")
         query = self.to_query_object()
         results = self._table._execute_query(query, timeout=timeout)
         results = results.read_all()
@@ -1525,6 +1573,8 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
             The LanceQueryBuilder object.
         """
         self._reranker = reranker
+        if reranker.score == "all":
+            self.with_row_id(True)
         return self
 
 
@@ -1588,7 +1638,8 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
         self._fts_columns = fts_columns
         self._norm = None
         self._reranker = None
-        self._nprobes = None
+        self._minimum_nprobes = None
+        self._maximum_nprobes = None
         self._refine_factor = None
         self._distance_type = None
         self._phrase_query = None
@@ -1800,6 +1851,8 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
 
         self._norm = normalize
         self._reranker = reranker
+        if reranker.score == "all":
+            self.with_row_id(True)
 
         return self
 
@@ -1820,7 +1873,24 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
         LanceHybridQueryBuilder
             The LanceHybridQueryBuilder object.
         """
-        self._nprobes = nprobes
+        self._minimum_nprobes = nprobes
+        self._maximum_nprobes = nprobes
+        return self
+
+    def minimum_nprobes(self, minimum_nprobes: int) -> LanceHybridQueryBuilder:
+        """Set the minimum number of probes to use.
+
+        See `nprobes` for more details.
+        """
+        self._minimum_nprobes = minimum_nprobes
+        return self
+
+    def maximum_nprobes(self, maximum_nprobes: int) -> LanceHybridQueryBuilder:
+        """Set the maximum number of probes to use.
+
+        See `nprobes` for more details.
+        """
+        self._maximum_nprobes = maximum_nprobes
         return self
 
     def distance_range(
@@ -1975,7 +2045,7 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
           FilterExec: _distance@2 IS NOT NULL
             SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], preserve_partitioning=[false]
               KNNVectorDistance: metric=l2
-                LanceScan: uri=..., projection=[vector], row_id=true, row_addr=false, ordered=false
+                LanceRead: uri=..., projection=[vector], ...
 
         Parameters
         ----------
@@ -2049,8 +2119,10 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
             self._fts_query.phrase_query(True)
         if self._distance_type:
             self._vector_query.metric(self._distance_type)
-        if self._nprobes:
+        if self._minimum_nprobes:
-            self._vector_query.nprobes(self._nprobes)
+            self._vector_query.minimum_nprobes(self._minimum_nprobes)
+        if self._maximum_nprobes is not None:
+            self._vector_query.maximum_nprobes(self._maximum_nprobes)
         if self._refine_factor:
             self._vector_query.refine_factor(self._refine_factor)
         if self._ef:
@@ -2359,7 +2431,7 @@ class AsyncQueryBase(object):
             FilterExec: _distance@2 IS NOT NULL
               SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], preserve_partitioning=[false]
                 KNNVectorDistance: metric=l2
-                  LanceScan: uri=..., projection=[vector], row_id=true, row_addr=false, ordered=false
+                  LanceRead: uri=..., projection=[vector], ...
 
         Parameters
         ----------
@@ -2513,7 +2585,7 @@ class AsyncQuery(AsyncQueryBase):
                 self._inner.nearest_to_text({"query": query, "columns": columns})
             )
         # FullTextQuery object
-        return AsyncFTSQuery(self._inner.nearest_to_text({"query": query.to_dict()}))
+        return AsyncFTSQuery(self._inner.nearest_to_text({"query": query}))
 
 
 class AsyncFTSQuery(AsyncQueryBase):
@@ -2661,6 +2733,34 @@ class AsyncVectorQueryBase:
         self._inner.nprobes(nprobes)
         return self
 
+    def minimum_nprobes(self, minimum_nprobes: int) -> Self:
+        """Set the minimum number of probes to use.
+
+        See `nprobes` for more details.
+
+        These partitions will be searched on every indexed vector query and will
+        increase recall at the expense of latency.
+        """
+        self._inner.minimum_nprobes(minimum_nprobes)
+        return self
+
+    def maximum_nprobes(self, maximum_nprobes: int) -> Self:
+        """Set the maximum number of probes to use.
+
+        See `nprobes` for more details.
+
+        If this value is greater than `minimum_nprobes` then the excess partitions
+        will be searched only if we have not found enough results.
+
+        This can be useful when there is a narrow filter to allow these queries to
+        spend more time searching and avoid potential false negatives.
+
+        If this value is 0 then no limit will be applied and all partitions could be
+        searched if needed to satisfy the limit.
+        """
+        self._inner.maximum_nprobes(maximum_nprobes)
+        return self
+
     def distance_range(
         self, lower_bound: Optional[float] = None, upper_bound: Optional[float] = None
     ) -> Self:
@@ -2835,7 +2935,7 @@ class AsyncVectorQuery(AsyncQueryBase, AsyncVectorQueryBase):
                 self._inner.nearest_to_text({"query": query, "columns": columns})
             )
         # FullTextQuery object
-        return AsyncHybridQuery(self._inner.nearest_to_text({"query": query.to_dict()}))
+        return AsyncHybridQuery(self._inner.nearest_to_text({"query": query}))
 
     async def to_batches(
         self,
@@ -2950,15 +3050,21 @@ class AsyncHybridQuery(AsyncQueryBase, AsyncVectorQueryBase):
         >>> asyncio.run(doctest_example()) # doctest: +ELLIPSIS, +NORMALIZE_WHITESPACE
         Vector Search Plan:
         ProjectionExec: expr=[vector@0 as vector, text@3 as text, _distance@2 as _distance]
-            Take: columns="vector, _rowid, _distance, (text)"
+          Take: columns="vector, _rowid, _distance, (text)"
-                CoalesceBatchesExec: target_batch_size=1024
+            CoalesceBatchesExec: target_batch_size=1024
-                GlobalLimitExec: skip=0, fetch=10
+              GlobalLimitExec: skip=0, fetch=10
-                    FilterExec: _distance@2 IS NOT NULL
+                FilterExec: _distance@2 IS NOT NULL
-                    SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], preserve_partitioning=[false]
+                  SortExec: TopK(fetch=10), expr=[_distance@2 ASC NULLS LAST], preserve_partitioning=[false]
-                        KNNVectorDistance: metric=l2
+                    KNNVectorDistance: metric=l2
-                        LanceScan: uri=..., projection=[vector], row_id=true, row_addr=false, ordered=false
+                      LanceRead: uri=..., projection=[vector], ...
+        <BLANKLINE>
         FTS Search Plan:
-        LanceScan: uri=..., projection=[vector, text], row_id=false, row_addr=false, ordered=true
+        ProjectionExec: expr=[vector@2 as vector, text@3 as text, _score@1 as _score]
+          Take: columns="_rowid, _score, (vector), (text)"
+            CoalesceBatchesExec: target_batch_size=1024
+              GlobalLimitExec: skip=0, fetch=10
+                MatchQuery: query=hello
+        <BLANKLINE>
 
         Parameters
         ----------

python/python/lancedb/remote/table.py

@@ -18,7 +18,7 @@ from lancedb._lancedb import (
     UpdateResult,
 )
 from lancedb.embeddings.base import EmbeddingFunctionConfig
-from lancedb.index import FTS, BTree, Bitmap, HnswPq, HnswSq, IvfFlat, IvfPq, LabelList
+from lancedb.index import FTS, BTree, Bitmap, HnswSq, IvfFlat, IvfPq, LabelList
 from lancedb.remote.db import LOOP
 import pyarrow as pa
 
@@ -89,7 +89,7 @@ class RemoteTable(Table):
 
     def to_pandas(self):
         """to_pandas() is not yet supported on LanceDB cloud."""
-        return NotImplementedError("to_pandas() is not yet supported on LanceDB cloud.")
+        raise NotImplementedError("to_pandas() is not yet supported on LanceDB cloud.")
 
     def checkout(self, version: Union[int, str]):
         return LOOP.run(self._table.checkout(version))
@@ -149,15 +149,18 @@ class RemoteTable(Table):
         *,
         replace: bool = False,
         wait_timeout: timedelta = None,
-        with_position: bool = True,
+        with_position: bool = False,
         # tokenizer configs:
         base_tokenizer: str = "simple",
         language: str = "English",
         max_token_length: Optional[int] = 40,
         lower_case: bool = True,
-        stem: bool = False,
+        stem: bool = True,
-        remove_stop_words: bool = False,
+        remove_stop_words: bool = True,
-        ascii_folding: bool = False,
+        ascii_folding: bool = True,
+        ngram_min_length: int = 3,
+        ngram_max_length: int = 3,
+        prefix_only: bool = False,
     ):
         config = FTS(
             with_position=with_position,
@@ -168,6 +171,9 @@ class RemoteTable(Table):
             stem=stem,
             remove_stop_words=remove_stop_words,
             ascii_folding=ascii_folding,
+            ngram_min_length=ngram_min_length,
+            ngram_max_length=ngram_max_length,
+            prefix_only=prefix_only,
         )
         LOOP.run(
             self._table.create_index(
@@ -186,6 +192,8 @@ class RemoteTable(Table):
         accelerator: Optional[str] = None,
         index_type="vector",
         wait_timeout: Optional[timedelta] = None,
+        *,
+        num_bits: int = 8,
     ):
         """Create an index on the table.
         Currently, the only parameters that matter are
@@ -220,11 +228,6 @@ class RemoteTable(Table):
         >>> table.create_index("l2", "vector") # doctest: +SKIP
         """
 
-        if num_partitions is not None:
-            logging.warning(
-                "num_partitions is not supported on LanceDB cloud."
-                "This parameter will be tuned automatically."
-            )
         if num_sub_vectors is not None:
             logging.warning(
                 "num_sub_vectors is not supported on LanceDB cloud."
@@ -244,13 +247,21 @@ class RemoteTable(Table):
 
         index_type = index_type.upper()
         if index_type == "VECTOR" or index_type == "IVF_PQ":
-            config = IvfPq(distance_type=metric)
+            config = IvfPq(
+                distance_type=metric,
+                num_partitions=num_partitions,
+                num_sub_vectors=num_sub_vectors,
+                num_bits=num_bits,
+            )
         elif index_type == "IVF_HNSW_PQ":
-            config = HnswPq(distance_type=metric)
+            raise ValueError(
+                "IVF_HNSW_PQ is not supported on LanceDB cloud."
+                "Please use IVF_HNSW_SQ instead."
+            )
         elif index_type == "IVF_HNSW_SQ":
-            config = HnswSq(distance_type=metric)
+            config = HnswSq(distance_type=metric, num_partitions=num_partitions)
         elif index_type == "IVF_FLAT":
-            config = IvfFlat(distance_type=metric)
+            config = IvfFlat(distance_type=metric, num_partitions=num_partitions)
         else:
             raise ValueError(
                 f"Unknown vector index type: {index_type}. Valid options are"

python/python/lancedb/rerankers/answerdotai.py

@@ -74,9 +74,7 @@ class AnswerdotaiRerankers(Reranker):
         if self.score == "relevance":
             combined_results = self._keep_relevance_score(combined_results)
         elif self.score == "all":
-            raise NotImplementedError(
+            combined_results = self._merge_and_keep_scores(vector_results, fts_results)
-                "Answerdotai Reranker does not support score='all' yet"
-            )
         combined_results = combined_results.sort_by(
             [("_relevance_score", "descending")]
         )

python/python/lancedb/rerankers/base.py

@@ -232,6 +232,39 @@ class Reranker(ABC):
 
         return deduped_table
 
+    def _merge_and_keep_scores(self, vector_results: pa.Table, fts_results: pa.Table):
+        """
+        Merge the results from the vector and FTS search and keep the scores.
+        This op is slower than just keeping relevance score but can be useful
+        for debugging.
+        """
+        # add nulls to fts results for _distance
+        if "_distance" not in fts_results.column_names:
+            fts_results = fts_results.append_column(
+                "_distance",
+                pa.array([None] * len(fts_results), type=pa.float32()),
+            )
+        # add nulls to vector results for _score
+        if "_score" not in vector_results.column_names:
+            vector_results = vector_results.append_column(
+                "_score",
+                pa.array([None] * len(vector_results), type=pa.float32()),
+            )
+
+        # combine them and fill the scores
+        vector_results_dict = {row["_rowid"]: row for row in vector_results.to_pylist()}
+        fts_results_dict = {row["_rowid"]: row for row in fts_results.to_pylist()}
+
+        # merge them into vector_results
+        for key, value in fts_results_dict.items():
+            if key in vector_results_dict:
+                vector_results_dict[key]["_score"] = value["_score"]
+            else:
+                vector_results_dict[key] = value
+
+        combined = pa.Table.from_pylist(list(vector_results_dict.values()))
+        return combined
+
     def _keep_relevance_score(self, combined_results: pa.Table):
         if self.score == "relevance":
             if "_score" in combined_results.column_names:

python/python/lancedb/rerankers/cohere.py

@@ -92,14 +92,14 @@ class CohereReranker(Reranker):
         vector_results: pa.Table,
         fts_results: pa.Table,
     ):
-        combined_results = self.merge_results(vector_results, fts_results)
+        if self.score == "all":
+            combined_results = self._merge_and_keep_scores(vector_results, fts_results)
+        else:
+            combined_results = self.merge_results(vector_results, fts_results)
         combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
             combined_results = self._keep_relevance_score(combined_results)
-        elif self.score == "all":
+
-            raise NotImplementedError(
-                "return_score='all' not implemented for cohere reranker"
-            )
         return combined_results
 
     def rerank_vector(self, query: str, vector_results: pa.Table):