chore: point lance dependency at jack branch

wait() (sync + async) only stopped on finished/stale/committed, so a job the
server already reported as state=failed was polled until the (default 3600s)
timeout, then raised a misleading TimeoutError instead of the real cause. A
doomed backfill -- e.g. a multi-column REFRESH COLUMN of a scalar UDF -- hung
the client even though get_job surfaced the failure within ~3s.

Add a terminal failed branch that raises JobFailedError carrying the server
error, exported from the package. Verified end-to-end against the cluster:
raises in 3.6s instead of hanging. Unit-tested with a mock conn (sync+async,
failure + success + committed paths).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.agents/skills/README.md

@@ -0,0 +1,7 @@
+# Agent Skills
+
+This directory contains repo-scoped code agent skills for the LanceDB project.
+
+Each skill is a folder that contains a required `SKILL.md` and optional bundled resources.
+
+Codex discovers skills from `.agents/skills` in the current working directory and parent directories.

.agents/skills/lancedb-column-metadata/SKILL.md

@@ -0,0 +1,178 @@
+---
+name: lancedb-column-metadata
+description: Column metadata authoring for LanceDB tables via the REST API. This skill is required for tasks like writing field descriptions, setting tags on columns (field_type, model, project_id, version), classifying columns as embeddings vs labels vs eval metrics, or grouping versioned columns into logical families — because it has the API integration needed to read the schema and persist metadata back. Invoke whenever someone wants to document, annotate, tag, or classify what their table columns ARE. Trigger even without an explicit "LanceDB" mention, as long as the context is column-level documentation or tagging for an ML or vector database table.
+metadata:
+  short-description: Write column descriptions, tags, and logical groupings to a LanceDB table
+---
+
+## Overview
+
+This skill authors column-level metadata for a LanceDB table. It connects to a LanceDB deployment over its REST API, inspects the table schema, generates appropriate metadata, and writes it back.
+
+## Step 0: Establish the connection
+
+Use the `lancedb-connect` skill (invoke it via the Skill tool) to resolve the base URL and auth headers (`x-api-key`, `x-lancedb-database`) for whichever deployment the user is working against — enterprise/self-hosted or a local dev server. Skip it only if the connection details are already established in the conversation.
+
+All examples below use `{base_url}` — substitute the resolved endpoint and include the resolved headers on every request.
+
+## Metadata keys
+
+All metadata uses namespaced keys:
+
+| Key | Purpose | Example value |
+|-----|---------|---------------|
+| `lancedb:description` | Human-readable explanation of what the column contains | `"CLIP ViT-L/14 image embedding, L2-normalized (768-dim)"` |
+| `lancedb:tag:<name>` | Flexible key-value tag; the suffix names the tag category | `lancedb:tag:field_type: "embedding"`, `lancedb:tag:model: "clip"`, `lancedb:tag:project_id: "foo"` |
+| `lancedb:logical-column` | Logical group/family this column belongs to | `"clip_features"` |
+
+Tags are open-ended — use whatever key suffix and value make sense given the user's intent. The tag suffix should describe *what is being classified* (e.g., `field_type`, `model`, `project_id`) and the value describes *how*.
+
+## Step 1: Resolve the table identifier
+
+You need:
+- **Table name** (required) — e.g., `my_table` or `my_namespace.my_table`
+- **Database name** — ask if not provided and not inferable from context; it goes in the `x-lancedb-database` header, never in the URL path
+
+The table identifier in the URL path is typically `table_name` for a top-level table, or `namespace$table_name` if the table lives in a namespace. The API accepts a `delimiter` query parameter to parse compound identifiers (default `$`).
+
+## Step 2: Describe the table
+
+```http
+POST {base_url}/v1/table/{table_id}/describe
+Content-Type: application/json
+
+{}
+```
+
+The response contains `schema.fields` — an array of field objects:
+
+```json
+{
+  "schema": {
+    "fields": [
+      {
+        "name": "clip_embedding_v3",
+        "type": { "type": "FixedSizeList", "fields": [...], "listSize": 768 },
+        "nullable": true,
+        "metadata": { "lancedb:description": "..." }
+      }
+    ]
+  }
+}
+```
+
+Each field has:
+- `name` — field name
+- `type` — Arrow data type (check `type.type` for the type string)
+- `nullable` — boolean
+- `metadata` — existing key-value metadata (read this before writing to avoid redundant updates)
+
+For struct/nested fields, recurse into `type.fields` and represent them as dot-notation paths (e.g., `parent.child`).
+
+If the user hasn't specified which columns to update, work with all columns.
+
+## Step 3: Generate metadata
+
+Decide what to generate based on the user's request.
+
+### Writing descriptions (`lancedb:description`)
+
+Base descriptions on:
+- The column name and Arrow type (e.g., `FixedSizeList` of floats → likely an embedding)
+- User-supplied context (upstream pipeline, sample values, domain knowledge)
+- Name patterns: `_embedding`/`_vec`/`_embed` → vector; `_label`/`_class` → label; `_score`/`_eval`/`_metric` → evaluation metric
+
+Be specific and concise. Good: `"Sentence-BERT embedding of the query text (768-dim)."` Not: `"An embedding column."`
+
+### Tagging columns (`lancedb:tag:<name>`)
+
+Choose tag key names that match what the user asked to annotate. Common patterns:
+
+- Semantic field type → `lancedb:tag:field_type: "embedding"` / `"text"` / `"image"` / `"label"` / `"eval"` / `"id"` / `"metadata"`
+- Model or source → `lancedb:tag:model: "clip"` / `"bert"` / `"vit"`
+- Project affiliation → `lancedb:tag:project_id: "<name>"`
+- Version → `lancedb:tag:version: "v3"` (and `lancedb:tag:latest: "true"` for the newest)
+
+Use Arrow type as a hint: `FixedSizeList` + float → embedding; `Utf8`/`LargeUtf8` → text; `Binary` → image or blob.
+
+Multiple tags on the same column are fine — each is a separate key.
+
+### Grouping into logical columns (`lancedb:logical-column`)
+
+Look for naming patterns across columns:
+- `clip_v1`, `clip_v2`, `clip_v3` → logical column `"clip"`, latest is `v3`
+- `text_embed_20240101`, `text_embed_20240601` → logical column `"text_embed"`, latest is the most recent date suffix
+
+Write `lancedb:logical-column` on all members of a group. Mark the newest with `lancedb:tag:latest: "true"` (in addition to its version tag).
+
+## Step 4: Write the metadata
+
+```http
+POST {base_url}/v1/table/{table_id}/update_field_metadata
+Content-Type: application/json
+
+{
+  "updates": [
+    {
+      "path": "clip_v3",
+      "metadata": {
+        "lancedb:description": "CLIP ViT-L/14 image embedding, L2-normalized (1024-dim).",
+        "lancedb:tag:field_type": "embedding",
+        "lancedb:tag:model": "clip",
+        "lancedb:tag:version": "v3",
+        "lancedb:tag:latest": "true",
+        "lancedb:logical-column": "clip"
+      },
+      "replace": false
+    },
+    {
+      "path": "clip_v2",
+      "metadata": {
+        "lancedb:description": "CLIP ViT-B/32 image embedding (768-dim), superseded by v3.",
+        "lancedb:tag:field_type": "embedding",
+        "lancedb:tag:model": "clip",
+        "lancedb:tag:version": "v2",
+        "lancedb:logical-column": "clip"
+      },
+      "replace": false
+    }
+  ]
+}
+```
+
+Rules:
+- **Use `"replace": false`** (merge) by default — this preserves existing metadata the user didn't ask to change
+- Use `"replace": true` only if the user explicitly asks to overwrite all existing metadata on a column
+- Set a value to `null` to delete a specific key
+- Batch all updates in a single request when possible
+
+The response includes `version` (new table version) and `fields` (the updated metadata per field).
+
+## Step 5: Confirm
+
+Report back:
+- Which columns were updated and what was written
+- The new table version number
+- Any columns skipped (e.g., already had up-to-date metadata)
+
+---
+
+## Quick examples
+
+**"Write descriptions for all columns in the `product_embeddings` table"**
+1. POST `/v1/table/product_embeddings/describe` → get all fields
+2. Generate a `lancedb:description` for each column based on name + type
+3. POST `update_field_metadata` with descriptions
+4. Report
+
+**"Tag the columns in `model_outputs` with their field type and model"**
+1. Describe `model_outputs`
+2. For each field, classify by name + Arrow type → set `lancedb:tag:field_type` and `lancedb:tag:model` where applicable
+3. POST `update_field_metadata`
+4. Report
+
+**"Group the feature columns in `training_features` into logical families and mark the latest version"**
+1. Describe the table
+2. Find version patterns → assign `lancedb:logical-column` and `lancedb:tag:version`; mark newest with `lancedb:tag:latest: "true"`
+3. POST `update_field_metadata`
+4. Show the grouping

.agents/skills/lancedb-connect/SKILL.md

@@ -0,0 +1,42 @@
+---
+name: lancedb-connect
+description: Resolve how to connect to a LanceDB deployment over the REST API — figure out the base URL, API key, and database header. Use this before making any REST requests to a LanceDB table, whenever the endpoint or auth setup is not already known. Also useful on its own when someone asks how to connect, authenticate, or curl their LanceDB instance.
+metadata:
+  short-description: Resolve the base URL and auth headers for a LanceDB deployment
+---
+
+## Goal
+
+Produce two things every REST request needs:
+
+1. **Base URL** — the endpoint
+2. **Headers** — `x-api-key`, and usually `x-lancedb-database`
+
+## Resolution steps
+
+1. If the user already gave a URL and API key (or said which environment they're working against), use that.
+2. Otherwise, look for credentials already available in the environment:
+   - Env vars like `LANCEDB_URI` / `LANCEDB_HOST` / `LANCEDB_API_KEY`
+   - A LanceDB endpoint already running or port-forwarded locally (the REST default port is 2333, i.e. `http://localhost:2333`)
+3. If you didn't find both pieces, ask the user directly: **"What's your LanceDB endpoint's URL, and what's your API key?"** Also ask which database to use if it isn't obvious. Don't guess or probe further — the user knows their deployment.
+
+## Validating the connection
+
+Make a cheap authenticated request and check the status:
+
+```bash
+curl -s -w "\n%{http_code}" "{base_url}/v1/table/?limit=1" \
+  -H "x-api-key: <key>" \
+  -H "x-lancedb-database: <database>"
+```
+
+- `200` — connection, key, and database header all good
+- `401` — API key missing or wrong
+- `400` mentioning a database header — this deployment expects `x-lancedb-database`
+
+## Non-REST equivalents
+
+If the caller would rather use the SDK or CLI than raw REST, the same credentials work:
+
+- Python SDK: `lancedb.connect("db://<database>", api_key="<key>", host_override="<base_url>")`
+- `lancedb` CLI: a `[profiles.<name>]` entry in `~/.lancedb/config.toml` with `http_server_url`, `api_key`, `database`

.agents/skills/lancedb-update-lance-dependency/SKILL.md

@@ -0,0 +1,98 @@
+---
+name: lancedb-update-lance-dependency
+description: Update LanceDB to a specific Lance release or tag. Use when bumping Lance dependencies in the lancedb repository, including Rust workspace Lance crates, Java lance-core, validation, branch creation, commit, push, and PR creation when requested.
+---
+
+# LanceDB Update Lance Dependency
+
+## Scope
+
+Use this skill in the `lancedb/lancedb` repository when updating the Lance dependency to a specific Lance version or tag.
+
+Inputs can be a version (`7.2.0-beta.1`), a tag (`v7.2.0-beta.1`), a tag ref (`refs/tags/v7.2.0-beta.1`), or `latest`.
+
+## Workflow
+
+1. Confirm the worktree status with `git status --short`.
+2. Resolve the target Lance version:
+
+   - If the input is `latest`, empty, or omitted, run:
+
+     ```bash
+     python3 ci/check_lance_release.py
+     ```
+
+     Parse the JSON output. If `needs_update` is not `true`, stop without creating a PR. Otherwise use `latest_tag`.
+
+   - If the input is explicit, use it directly.
+
+3. Compute update metadata without changing files:
+
+   ```bash
+   python3 ci/update_lance_dependency.py "$TAG_OR_VERSION" --metadata-only
+   ```
+
+   Before making changes, check for an existing open PR with the emitted `pr_title`:
+
+   ```bash
+   gh pr list --search "\"$PR_TITLE\" in:title" --state open --limit 1 --json number,url,title
+   ```
+
+   If a matching open PR exists, stop and report it instead of creating a duplicate.
+
+4. Run the deterministic update entrypoint:
+
+   ```bash
+   python3 ci/update_lance_dependency.py "$TAG_OR_VERSION"
+   ```
+
+   This updates the Rust workspace Lance dependencies through `ci/set_lance_version.py`, updates `java/pom.xml`, refreshes Cargo metadata, and prints JSON metadata containing `branch_name`, `commit_message`, and `pr_title`.
+
+5. Run validation:
+
+   ```bash
+   cargo clippy --quiet --workspace --tests --all-features -- -D warnings
+   cargo fmt --all --quiet
+   ```
+
+   Fix real diagnostics and rerun clippy until it succeeds. Do not skip warnings.
+
+6. Inspect `git status --short` and `git diff` to ensure only the Lance dependency update and required compatibility fixes are present.
+
+7. If the task only asks to prepare local changes, stop here and report the changed files and validation result.
+
+8. If the task asks to publish the update, create a branch using the printed `branch_name`, stage all relevant files, and commit using the printed `commit_message`. Do not amend or rewrite existing commits.
+
+9. Push to `origin`. Before creating the PR, check that the current token has push permission:
+
+   ```bash
+   gh api repos/lancedb/lancedb --jq .permissions.push
+   ```
+
+   If the remote branch already exists for the same generated branch name, delete the remote ref with `gh api -X DELETE repos/lancedb/lancedb/git/refs/heads/$BRANCH_NAME`, then push. Do not force-push.
+
+10. Create a PR targeting `main` with the printed `pr_title`. If there is no PR template, keep the body to two or three concise sentences: state the Lance dependency bump, note any required compatibility fixes, and link the triggering Lance tag or release.
+
+11. Read back the remote PR title after creation. If it is not a Conventional Commit title, fix it immediately.
+
+12. When running in GitHub Actions after creating the LanceDB PR, trigger the Sophon dependency update:
+
+    ```bash
+    gh workflow run codex-bump-lancedb-lance.yml \
+      --repo lancedb/sophon \
+      -f lance_ref="$LANCE_TAG" \
+      -f lancedb_ref="$BRANCH_NAME"
+    gh run list --repo lancedb/sophon --workflow codex-bump-lancedb-lance.yml --limit 1 --json databaseId,url,displayTitle
+    ```
+
+    Use the emitted metadata `tag` value as `LANCE_TAG`. Do this only after a new LanceDB PR has been created. If the update was skipped because no update is needed or an open PR already exists, do not trigger Sophon.
+
+## GitHub Actions
+
+When this skill is used from GitHub Actions, `TAG`, `GH_TOKEN`, and `GITHUB_TOKEN` may already be set. Resolve `latest` first when `TAG` is empty. Once an explicit tag or version is known, use:
+
+```bash
+python3 ci/update_lance_dependency.py "$TAG" --github-output "$GITHUB_OUTPUT"
+```
+
+Then use the emitted `branch_name`, `commit_message`, and `pr_title` values for branch, commit, and PR creation.

.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.29.1-beta.0"
+current_version = "0.31.0-beta.4"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.
@@ -23,6 +23,8 @@ allow_dirty = true
 commit = true
 message = "Bump version: {current_version} → {new_version}"
 commit_args = ""
+# bump-my-version >=1.4.0 rejects pre_commit_hooks containing shell syntax unless opted in.
+allow_shell_hooks = true
 
 # Java maven files
 pre_commit_hooks = [

.github/dependabot.yml

@@ -21,3 +21,14 @@ updates:
         update-types:
           - minor
           - patch
+
+  - package-ecosystem: pip
+    directory: /python
+    schedule:
+      interval: weekly
+    # Only update uv.lock, never widen version requirements in pyproject.toml.
+    versioning-strategy: lockfile-only
+    groups:
+      python-deps:
+        patterns:
+          - "*"

.github/workflows/build_windows_wheel/action.yml

@@ -29,7 +29,3 @@ runs:
         args: ${{ inputs.args }}
         docker-options: "-e PIP_EXTRA_INDEX_URL='https://pypi.fury.io/lance-format/ https://pypi.fury.io/lancedb/'"
         working-directory: python
-    - uses: actions/upload-artifact@v4
-      with:
-        name: windows-wheels
-        path: python\target\wheels

.github/workflows/codex-update-lance-dependency.yml

@@ -4,14 +4,16 @@ on:
   workflow_call:
     inputs:
       tag:
-        description: "Tag name from Lance"
-        required: true
+        description: "Tag name from Lance. If omitted, the skill will use the latest Lance release that needs an update."
+        required: false
+        default: ""
         type: string
   workflow_dispatch:
     inputs:
       tag:
-        description: "Tag name from Lance"
-        required: true
+        description: "Tag name from Lance. Leave empty to use the latest Lance release that needs an update."
+        required: false
+        default: ""
         type: string
 
 permissions:
@@ -25,7 +27,7 @@ jobs:
     steps:
       - name: Show inputs
         run: |
-          echo "tag = ${{ inputs.tag }}"
+          echo "tag = ${{ inputs.tag || 'latest' }}"
 
       - name: Checkout Repo LanceDB
         uses: actions/checkout@v4
@@ -71,65 +73,21 @@ jobs:
           OPENAI_API_KEY: ${{ secrets.CODEX_TOKEN }}
         run: |
           set -euo pipefail
-          VERSION="${TAG#refs/tags/}"
-          VERSION="${VERSION#v}"
-          BRANCH_NAME="codex/update-lance-${VERSION//[^a-zA-Z0-9]/-}"
-
-          # Use "chore" for beta/rc versions, "feat" for stable releases
-          if [[ "${VERSION}" == *beta* ]] || [[ "${VERSION}" == *rc* ]]; then
-            COMMIT_TYPE="chore"
-          else
-            COMMIT_TYPE="feat"
-          fi
+          TARGET_TAG="${TAG:-latest}"
 
           cat <<EOF >/tmp/codex-prompt.txt
-          You are running inside the lancedb repository on a GitHub Actions runner. Update the Lance dependency to version ${VERSION} and prepare a pull request for maintainers to review.
+          You are running inside the lancedb repository on a GitHub Actions runner.
 
-          Follow these steps exactly:
-          1. Use script "ci/set_lance_version.py" to update Lance Rust dependencies. The script already refreshes Cargo metadata, so allow it to finish even if it takes time.
-          2. Update the Java lance-core dependency version in "java/pom.xml": change the "<lance-core.version>...</lance-core.version>" property to "${VERSION}".
-          3. Run "cargo clippy --workspace --tests --all-features -- -D warnings". If diagnostics appear, fix them yourself and rerun clippy until it exits cleanly. Do not skip any warnings.
-          4. After clippy succeeds, run "cargo fmt --all" to format the workspace.
-          5. Ensure the repository is clean except for intentional changes. Inspect "git status --short" and "git diff" to confirm the dependency update and any required fixes.
-          6. Create and switch to a new branch named "${BRANCH_NAME}" (replace any duplicated hyphens if necessary).
-          7. Stage all relevant files with "git add -A". Commit using the message "${COMMIT_TYPE}: update lance dependency to v${VERSION}".
-          8. Push the branch to origin. If the remote branch already exists, delete it first with "gh api -X DELETE repos/lancedb/lancedb/git/refs/heads/${BRANCH_NAME}" then push with "git push origin ${BRANCH_NAME}". Do NOT use "git push --force" or "git push -f".
-          9. env "GH_TOKEN" is available, use "gh" tools for github related operations like creating pull request.
-          10. Create a pull request targeting "main" with title "${COMMIT_TYPE}: update lance dependency to v${VERSION}". First, write the PR body to /tmp/pr-body.md using a heredoc (cat <<'EOF' > /tmp/pr-body.md). The body should summarize the dependency bump, clippy/fmt verification, and link the triggering tag (${TAG}). Then run "gh pr create --body-file /tmp/pr-body.md".
-          11. After creating the PR, display the PR URL, "git status --short", and a concise summary of the commands run and their results.
+          Use \$lancedb-update-lance-dependency with target "${TARGET_TAG}".
 
           Constraints:
-          - Use bash commands; avoid modifying GitHub workflow files other than through the scripted task above.
-          - Do not merge the PR.
-          - If any command fails, diagnose and fix the issue instead of aborting.
+          - Use env "GH_TOKEN" for GitHub operations.
+          - Do not merge the pull request.
+          - Do not force-push.
+          - Do not create a duplicate pull request if an open PR already exists for the target Lance version.
+          - If any command fails, diagnose and fix the root cause instead of aborting.
+          - After creating the PR, display the PR URL, "git status --short", and a concise summary of the commands run and their results.
           EOF
 
           printenv OPENAI_API_KEY | codex login --with-api-key
           codex --config shell_environment_policy.ignore_default_excludes=true exec --dangerously-bypass-approvals-and-sandbox "$(cat /tmp/codex-prompt.txt)"
-
-      - name: Trigger sophon dependency update
-        env:
-          TAG: ${{ inputs.tag }}
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          VERSION="${TAG#refs/tags/}"
-          VERSION="${VERSION#v}"
-          LANCEDB_BRANCH="codex/update-lance-${VERSION//[^a-zA-Z0-9]/-}"
-
-          echo "Triggering sophon workflow with:"
-          echo "  lance_ref: ${TAG#refs/tags/}"
-          echo "  lancedb_ref: ${LANCEDB_BRANCH}"
-
-          gh workflow run codex-bump-lancedb-lance.yml \
-            --repo lancedb/sophon \
-            -f lance_ref="${TAG#refs/tags/}" \
-            -f lancedb_ref="${LANCEDB_BRANCH}"
-
-      - name: Show latest sophon workflow run
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          echo "Latest sophon workflow run:"
-          gh run list --repo lancedb/sophon --workflow codex-bump-lancedb-lance.yml --limit 1 --json databaseId,url,displayTitle

.github/workflows/lance-release-timer.yml

@@ -1,62 +0,0 @@
-name: Lance Release Timer
-
-on:
-  schedule:
-    - cron: "*/10 * * * *"
-  workflow_dispatch:
-
-permissions:
-  contents: read
-  actions: write
-
-concurrency:
-  group: lance-release-timer
-  cancel-in-progress: false
-
-jobs:
-  trigger-update:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Check for new Lance tag
-        id: check
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          python3 ci/check_lance_release.py --github-output "$GITHUB_OUTPUT"
-
-      - name: Look for existing PR
-        if: steps.check.outputs.needs_update == 'true'
-        id: pr
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          TITLE="chore: update lance dependency to v${{ steps.check.outputs.latest_version }}"
-          COUNT=$(gh pr list --search "\"$TITLE\" in:title" --state open --limit 1 --json number --jq 'length')
-          if [ "$COUNT" -gt 0 ]; then
-            echo "Open PR already exists for $TITLE"
-            echo "pr_exists=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "No existing PR for $TITLE"
-            echo "pr_exists=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Trigger codex update workflow
-        if: steps.check.outputs.needs_update == 'true' && steps.pr.outputs.pr_exists != 'true'
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          TAG=${{ steps.check.outputs.latest_tag }}
-          gh workflow run codex-update-lance-dependency.yml -f tag=refs/tags/$TAG
-
-      - name: Show latest codex workflow run
-        if: steps.check.outputs.needs_update == 'true' && steps.pr.outputs.pr_exists != 'true'
-        env:
-          GH_TOKEN: ${{ secrets.ROBOT_TOKEN }}
-        run: |
-          set -euo pipefail
-          gh run list --workflow codex-update-lance-dependency.yml --limit 1 --json databaseId,url,displayTitle

.github/workflows/pypi-publish.yml

@@ -8,6 +8,9 @@ on:
     # This should trigger a dry run (we skip the final publish step)
     paths:
       - .github/workflows/pypi-publish.yml
+      - .github/workflows/build_linux_wheel/action.yml
+      - .github/workflows/build_mac_wheel/action.yml
+      - .github/workflows/build_windows_wheel/action.yml
       - Cargo.toml # Change in dependency frequently breaks builds
       - Cargo.lock
 
@@ -21,32 +24,21 @@ jobs:
   linux:
     name: Python ${{ matrix.config.platform }} manylinux${{ matrix.config.manylinux }}
     timeout-minutes: 60
-    permissions:
-      id-token: write
-      contents: read
     strategy:
       matrix:
         config:
-          - platform: x86_64
-            manylinux: "2_17"
-            extra_args: ""
-            runner: ubuntu-22.04
           - platform: x86_64
             manylinux: "2_28"
             extra_args: "--features fp16kernels"
             runner: ubuntu-22.04
-          - platform: aarch64
-            manylinux: "2_17"
-            extra_args: ""
-            # For successful fat LTO builds, we need a large runner to avoid OOM errors.
-            runner: ubuntu-2404-8x-arm64
+          # For successful fat LTO builds, we need a large runner to avoid OOM errors.
           - platform: aarch64
             manylinux: "2_28"
             extra_args: "--features fp16kernels"
             runner: ubuntu-2404-8x-arm64
     runs-on: ${{ matrix.config.runner }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -60,15 +52,14 @@ jobs:
           args: "--release --strip ${{ matrix.config.extra_args }}"
           arm-build: ${{ matrix.config.platform == 'aarch64' }}
           manylinux: ${{ matrix.config.manylinux }}
-      - uses: ./.github/workflows/upload_wheel
+      - uses: actions/upload-artifact@v7
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          fury_token: ${{ secrets.FURY_TOKEN }}
+          name: wheels-linux-${{ matrix.config.platform }}-${{ matrix.config.manylinux }}
+          path: target/wheels/lancedb-*.whl
+          if-no-files-found: error
   mac:
     timeout-minutes: 90
-    permissions:
-      id-token: write
-      contents: read
     runs-on: ${{ matrix.config.runner }}
     strategy:
       matrix:
@@ -78,7 +69,7 @@ jobs:
     env:
       MACOSX_DEPLOYMENT_TARGET: 10.15
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -90,18 +81,21 @@ jobs:
         with:
           python-minor-version: 10
           args: "--release --strip --target ${{ matrix.config.target }} --features fp16kernels"
-      - uses: ./.github/workflows/upload_wheel
+      - uses: actions/upload-artifact@v7
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          fury_token: ${{ secrets.FURY_TOKEN }}
+          name: wheels-mac-${{ matrix.config.target }}
+          path: target/wheels/lancedb-*.whl
+          if-no-files-found: error
   windows:
-    timeout-minutes: 60
-    permissions:
-      id-token: write
-      contents: read
+    timeout-minutes: 90
     runs-on: windows-latest
+    env:
+      # link.exe is single-threaded and the long pole on Windows builds. Use
+      # rustc's bundled lld-link instead.
+      CARGO_TARGET_X86_64_PC_WINDOWS_MSVC_LINKER: rust-lld
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -113,18 +107,70 @@ jobs:
         with:
           python-minor-version: 10
           args: "--release --strip"
-          vcpkg_token: ${{ secrets.VCPKG_GITHUB_PACKAGES }}
-      - uses: ./.github/workflows/upload_wheel
+      - uses: actions/upload-artifact@v7
         if: startsWith(github.ref, 'refs/tags/python-v')
         with:
-          fury_token: ${{ secrets.FURY_TOKEN }}
+          name: wheels-windows
+          path: target/wheels/lancedb-*.whl
+          if-no-files-found: error
+  publish:
+    name: Publish wheels
+    if: startsWith(github.ref, 'refs/tags/python-v')
+    needs: [linux, mac, windows]
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write
+      contents: read
+    steps:
+      - uses: actions/checkout@v6
+      - name: Download wheel artifacts
+        uses: actions/download-artifact@v8
+        with:
+          pattern: wheels-*
+          path: target/wheels
+          merge-multiple: true
+      - name: List wheels
+        run: ls -la target/wheels
+      - name: Choose repo
+        id: choose_repo
+        run: |
+          if [[ ${{ github.ref }} == *beta* ]]; then
+            echo "repo=fury" >> $GITHUB_OUTPUT
+          else
+            echo "repo=pypi" >> $GITHUB_OUTPUT
+          fi
+      - name: Publish to Fury
+        if: steps.choose_repo.outputs.repo == 'fury'
+        env:
+          FURY_TOKEN: ${{ secrets.FURY_TOKEN }}
+        run: |
+          shopt -s nullglob
+          WHEELS=(target/wheels/lancedb-*.whl)
+          if [[ ${#WHEELS[@]} -eq 0 ]]; then
+            echo "No wheels found in target/wheels/" >&2
+            exit 1
+          fi
+          for WHEEL in "${WHEELS[@]}"; do
+            echo "Uploading $WHEEL to Fury"
+            curl -f -F package=@"$WHEEL" "https://$FURY_TOKEN@push.fury.io/lancedb/"
+          done
+      # NOTE: pypa/gh-action-pypi-publish must be invoked directly from a
+      # workflow file, not from inside a composite action. When called from a
+      # composite, `github.action_repository` is empty (actions/runner#2473)
+      # and the action falls back to `github.repository`, producing a bogus
+      # `docker://ghcr.io/<repo>:<ref>` image reference that GHA tries to pull.
+      - name: Publish to PyPI
+        if: steps.choose_repo.outputs.repo == 'pypi'
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          packages-dir: target/wheels/
   gh-release:
     if: startsWith(github.ref, 'refs/tags/python-v')
     runs-on: ubuntu-latest
     permissions:
       contents: write
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
           lfs: true
@@ -187,13 +233,13 @@ jobs:
   report-failure:
     name: Report Workflow Failure
     runs-on: ubuntu-latest
-    needs: [linux, mac, windows]
+    needs: [linux, mac, windows, publish]
     permissions:
       contents: read
       issues: write
     if: always() && failure() && startsWith(github.ref, 'refs/tags/python-v')
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
       - uses: ./.github/actions/create-failure-issue
         with:
           job-results: ${{ toJSON(needs) }}

.github/workflows/upload_wheel/action.yml

@@ -1,34 +0,0 @@
-name: upload-wheel
-
-description: "Upload wheels to Pypi"
-inputs:
-  fury_token:
-    required: true
-    description: "release token for the fury repo"
-
-runs:
-  using: "composite"
-  steps:
-  - name: Choose repo
-    shell: bash
-    id: choose_repo
-    run: |
-      if [[ ${{ github.ref }} == *beta* ]]; then
-        echo "repo=fury" >> $GITHUB_OUTPUT
-      else
-        echo "repo=pypi" >> $GITHUB_OUTPUT
-      fi
-  - name: Publish to Fury
-    if: steps.choose_repo.outputs.repo == 'fury'
-    shell: bash
-    env:
-      FURY_TOKEN: ${{ inputs.fury_token }}
-    run: |
-      WHEEL=$(ls target/wheels/lancedb-*.whl 2> /dev/null | head -n 1)
-      echo "Uploading $WHEEL to Fury"
-      curl -f -F package=@$WHEEL https://$FURY_TOKEN@push.fury.io/lancedb/
-  - name: Publish to PyPI
-    if: steps.choose_repo.outputs.repo == 'pypi'
-    uses: pypa/gh-action-pypi-publish@release/v1
-    with:
-      packages-dir: target/wheels/

.gitignore

@@ -27,6 +27,7 @@ python/dist
 *.so
 *.dylib
 *.dll
+*.pdb
 
 ## Javascript
 *.node

AGENTS.md

@@ -37,10 +37,13 @@ Before committing changes, run formatting for every language you touched. At min
   and run targeted tests through `cd python && uv run ...`.
 * TypeScript changes: run the relevant `npm`/`pnpm` lint, format, build, and docs commands in `nodejs`.
 
-Before creating a PR, make sure the PR title follows Conventional Commits, such as
-`fix: support nested field paths in native index creation` or
-`feat(python): add dataset multiprocessing support`. The semantic-release check uses the
-PR title and body as the merge commit message, so a non-conventional PR title will fail CI.
+Before creating a PR, the exact value passed to `gh pr create --title` must follow
+Conventional Commits, such as `fix: support nested field paths in native index creation`
+or `feat(python): add dataset multiprocessing support`. Do not use a plain natural
+language summary like `Support nested field paths in native index creation` as the PR
+title. The semantic-release check uses the PR title and body as the merge commit message,
+so a non-conventional PR title will fail CI. After creating a PR, read the remote PR title
+back and fix it immediately if it is not conventional.
 
 ## Coding tips
 

Cargo.toml

@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"
 
 [workspace.dependencies]
-lance = { "version" = "=7.0.0-beta.13", default-features = false, "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=7.0.0-beta.13", default-features = false, "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=7.0.0-beta.13", default-features = false, "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=7.0.0-beta.13", "tag" = "v7.0.0-beta.13", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=9.0.0-beta.10", default-features = false, "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-core = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-datagen = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-file = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-io = { "version" = "=9.0.0-beta.10", default-features = false, "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-index = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-linalg = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-namespace = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-namespace-impls = { "version" = "=9.0.0-beta.10", default-features = false, "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-table = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-testing = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-datafusion = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-encoding = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
+lance-arrow = { "version" = "=9.0.0-beta.10", "branch" = "jack/sophon-pr-6325", "git" = "https://github.com/jackye1995/lance.git" }
 ahash = "0.8"
 # Note that this one does not include pyarrow
 arrow = { version = "58.0.0", optional = false }

REVIEW.md

@@ -0,0 +1,26 @@
+# Code review guidelines
+
+Repo-specific guidance for automated PR reviews.
+
+## Cross-SDK parity
+
+LanceDB exposes the same core (`rust/lancedb`) through Python, TypeScript (`nodejs`),
+and Java bindings. Behavioral drift between SDKs is a recurring problem, so watch for
+parity gaps when reviewing — but only flag real ones:
+
+* If the change adds or modifies user-facing API or behavior in the shared core
+  (`rust/lancedb`), check whether each binding that should expose it (`python`,
+  `nodejs`) does. A core change with no corresponding binding update is worth a note.
+* If the change adds or modifies a public API in one SDK but not the other, open the
+  sibling SDK's corresponding module and state whether an equivalent exists. If not,
+  note it as a possible parity gap and suggest a follow-up issue.
+* For bug fixes, first read the sibling SDK's analogous code path to check whether the
+  same bug exists there. Only raise parity if it actually does. Do not ask to "port" a
+  fix for a bug that only ever existed in one binding.
+* Stay silent on internal-only refactors, tests, docs, and changes with no cross-SDK
+  surface.
+* Parity expectations apply to the Python and TypeScript (`nodejs`) SDKs. Java currently
+  implements only the remote table, not the local/embedded backend, so it is expected to
+  be partial — do not flag Java for missing local-only functionality.
+* Keep parity feedback to a short, clearly-labeled note (e.g. "Possible SDK parity
+  gap: …"). It is advisory, not a merge blocker.

ci/check_lance_release.py

@@ -112,25 +112,25 @@ def fetch_remote_tags() -> List[TagInfo]:
             "api",
             "-X",
             "GET",
-            f"repos/{LANCE_REPO}/git/refs/tags",
-            "--paginate",
+            f"repos/{LANCE_REPO}/releases",
             "--jq",
-            ".[].ref",
+            ".[].tag_name",
+            "-F",
+            "per_page=20",
         ]
     )
     tags: List[TagInfo] = []
     for line in output.splitlines():
-        ref = line.strip()
-        if not ref.startswith("refs/tags/v"):
+        tag = line.strip()
+        if not tag.startswith("v"):
             continue
-        tag = ref.split("refs/tags/")[-1]
         version = tag.lstrip("v")
         try:
             tags.append(TagInfo(tag=tag, version=version, semver=parse_semver(version)))
         except ValueError:
             continue
     if not tags:
-        raise RuntimeError("No Lance tags could be parsed from GitHub API output")
+        raise RuntimeError("No Lance releases could be parsed from GitHub API output")
     return tags
 
 

ci/update_lance_dependency.py

@@ -0,0 +1,126 @@
+#!/usr/bin/env python3
+"""Prepare a Lance dependency update for LanceDB."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import subprocess
+import sys
+from pathlib import Path
+from typing import Sequence
+
+try:
+    from check_lance_release import parse_semver
+except ModuleNotFoundError:
+    # Supports importing as ci.update_lance_dependency from tests or ad hoc checks.
+    from ci.check_lance_release import parse_semver  # type: ignore
+
+
+def normalize_version(raw: str) -> str:
+    value = raw.strip()
+    value = value.removeprefix("refs/tags/")
+    value = value.removeprefix("v")
+    try:
+        parse_semver(value)
+    except ValueError:
+        raise ValueError(f"Unsupported Lance version or tag: {raw}")
+    return value
+
+
+def normalized_tag(version: str) -> str:
+    return f"v{version}"
+
+
+def branch_name(version: str) -> str:
+    suffix = re.sub(r"[^a-zA-Z0-9]+", "-", version).strip("-")
+    suffix = re.sub(r"-+", "-", suffix)
+    return f"codex/update-lance-{suffix}"
+
+
+def commit_type(version: str) -> str:
+    prerelease = version.split("-", maxsplit=1)[1] if "-" in version else ""
+    return "chore" if "beta" in prerelease or "rc" in prerelease else "feat"
+
+
+def metadata_for(version: str) -> dict[str, str]:
+    kind = commit_type(version)
+    message = f"{kind}: update lance dependency to v{version}"
+    return {
+        "version": version,
+        "tag": normalized_tag(version),
+        "branch_name": branch_name(version),
+        "commit_type": kind,
+        "commit_message": message,
+        "pr_title": message,
+    }
+
+
+def run_command(cmd: Sequence[str], *, cwd: Path) -> None:
+    subprocess.run(cmd, cwd=cwd, check=True)
+
+
+def update_java_lance_core_version(repo_root: Path, version: str) -> None:
+    pom_path = repo_root / "java" / "pom.xml"
+    contents = pom_path.read_text(encoding="utf-8")
+    updated, count = re.subn(
+        r"(<lance-core\.version>)[^<]+(</lance-core\.version>)",
+        rf"\g<1>{version}\g<2>",
+        contents,
+        count=1,
+    )
+    if count != 1:
+        raise RuntimeError(
+            "Expected exactly one <lance-core.version> entry in java/pom.xml"
+        )
+    pom_path.write_text(updated, encoding="utf-8")
+
+
+def write_github_outputs(path: str | None, payload: dict[str, str]) -> None:
+    if not path:
+        return
+    with open(path, "a", encoding="utf-8") as output:
+        for key, value in payload.items():
+            output.write(f"{key}={value}\n")
+
+
+def main(argv: Sequence[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "tag_or_version",
+        help="Lance tag or version, for example refs/tags/v7.2.0-beta.1 or 7.2.0",
+    )
+    parser.add_argument(
+        "--repo-root",
+        type=Path,
+        default=Path(__file__).resolve().parents[1],
+        help="Path to the lancedb repository root",
+    )
+    parser.add_argument(
+        "--github-output",
+        default=None,
+        help="Optional GitHub Actions output file to receive metadata fields",
+    )
+    parser.add_argument(
+        "--metadata-only",
+        action="store_true",
+        help="Only print derived metadata; do not modify dependency files",
+    )
+    args = parser.parse_args(argv)
+
+    repo_root = args.repo_root.resolve()
+    version = normalize_version(args.tag_or_version)
+    payload = metadata_for(version)
+
+    if not args.metadata_only:
+        run_command([sys.executable, "ci/set_lance_version.py", version], cwd=repo_root)
+        update_java_lance_core_version(repo_root, version)
+
+    write_github_outputs(args.github_output, payload)
+    print(json.dumps(payload, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

deny.toml

@@ -113,6 +113,12 @@ ignore = [
     # rand from a custom logger; upgrade once all pinned chains accept 0.8.6+.
     # https://rustsec.org/advisories/RUSTSEC-2026-0097
     { id = "RUSTSEC-2026-0097", reason = "transitive rand 0.8.5; LanceDB does not call ThreadRng from custom logging" },
+
+    # pyo3 advisories in the Python bindings; tracked pending a patched pyo3 release.
+    # https://rustsec.org/advisories/RUSTSEC-2026-0176
+    # https://rustsec.org/advisories/RUSTSEC-2026-0177
+    { id = "RUSTSEC-2026-0176", reason = "pyo3 in Python bindings; awaiting patched pyo3 release" },
+    { id = "RUSTSEC-2026-0177", reason = "pyo3 in Python bindings; awaiting patched pyo3 release" },
 ]
 
 # ---------------------------------------------------------------------------
@@ -147,6 +153,14 @@ allow = [
     "CDLA-Permissive-2.0",
 ]
 confidence-threshold = 0.8
+# Per-crate license exceptions: allow a license for a specific crate only,
+# rather than globally via the `allow` list above.
+exceptions = [
+    # CDDL-1.0 (copyleft) is pulled in only as a dev/profiling dependency via
+    # `inferno` -> `pprof` -> `lance-testing`; it is a test dependency that we
+    # do not distribute, so scope the allowance to `inferno` alone.
+    { allow = ["CDDL-1.0"], crate = "inferno" },
+]
 # Crates whose license cannot be determined from Cargo metadata but whose
 # license we've manually confirmed from upstream. Keep this list minimal.
 [[licenses.clarify]]

docs/src/java/java.md

@@ -14,7 +14,7 @@ Add the following dependency to your `pom.xml`:
 <dependency>
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-core</artifactId>
-    <version>0.29.1-beta.0</version>
+    <version>0.31.0-beta.4</version>
 </dependency>
 ```
 

docs/src/js/classes/BranchContents.md

@@ -0,0 +1,43 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / BranchContents
+
+# Class: BranchContents
+
+## Constructors
+
+### new BranchContents()
+
+```ts
+new BranchContents(): BranchContents
+```
+
+#### Returns
+
+[`BranchContents`](BranchContents.md)
+
+## Properties
+
+### manifestSize
+
+```ts
+manifestSize: number;
+```
+
+***
+
+### parentBranch?
+
+```ts
+optional parentBranch: string;
+```
+
+***
+
+### parentVersion
+
+```ts
+parentVersion: number;
+```

docs/src/js/classes/Branches.md

@@ -0,0 +1,96 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / Branches
+
+# Class: Branches
+
+Branch manager for a [Table](Table.md).
+
+Unlike tags, `create` and `checkout` return a new [Table](Table.md) handle scoped
+to the branch; writes on it do not affect `main`.
+
+## Methods
+
+### checkout()
+
+```ts
+checkout(name, version?): Promise<Table>
+```
+
+Check out an existing branch and return a handle scoped to it.
+
+With `version` set, the returned handle is pinned to that version of the
+branch (a read-only, detached view); otherwise it tracks the branch's
+latest and stays writable.
+
+#### Parameters
+
+* **name**: `string`
+
+* **version?**: `number`
+
+#### Returns
+
+`Promise`&lt;[`Table`](Table.md)&gt;
+
+***
+
+### create()
+
+```ts
+create(
+   name,
+   fromRef?,
+   fromVersion?): Promise<Table>
+```
+
+Create a branch and return a handle scoped to it.
+
+#### Parameters
+
+* **name**: `string`
+    Name of the new branch.
+
+* **fromRef?**: `string`
+    Source branch to fork from. Defaults to `main`.
+
+* **fromVersion?**: `number`
+    A specific version on `fromRef`. Defaults to latest.
+
+#### Returns
+
+`Promise`&lt;[`Table`](Table.md)&gt;
+
+***
+
+### delete()
+
+```ts
+delete(name): Promise<void>
+```
+
+Delete a branch.
+
+#### Parameters
+
+* **name**: `string`
+
+#### Returns
+
+`Promise`&lt;`void`&gt;
+
+***
+
+### list()
+
+```ts
+list(): Promise<Record<string, BranchContents>>
+```
+
+List all branches, mapping name to branch metadata.
+
+#### Returns
+
+`Promise`&lt;`Record`&lt;`string`, [`BranchContents`](BranchContents.md)&gt;&gt;

docs/src/js/classes/Index.md

@@ -57,6 +57,24 @@ block size may be added in the future.
 
 ***
 
+### fm()
+
+```ts
+static fm(): Index
+```
+
+Create an FM-Index.
+
+An FM-Index is a scalar index on string or binary columns that accelerates
+substring search, i.e. `contains(col, 'needle')`. Unlike the tokenized
+full-text-search index, it matches arbitrary substrings of the raw bytes.
+
+#### Returns
+
+[`Index`](Index.md)
+
+***
+
 ### fts()
 
 ```ts

docs/src/js/classes/MergeInsertBuilder.md

@@ -76,6 +76,57 @@ the query optimizer chooses a suboptimal path.
 
 ***
 
+### useLsmWrite()
+
+```ts
+useLsmWrite(useLsmWrite): MergeInsertBuilder
+```
+
+Controls whether the merge uses the MemWAL LSM write path.
+
+By default (unset), a `mergeInsert` on a table with an LSM write spec is
+routed through Lance's MemWAL shard writer, and a table without one uses
+the standard path. Pass `false` to force the standard path even when a
+spec is set. Pass `true` to require a spec — `mergeInsert` rejects if none
+is installed.
+
+#### Parameters
+
+* **useLsmWrite**: `boolean`
+    Whether to use the LSM write path.
+
+#### Returns
+
+[`MergeInsertBuilder`](MergeInsertBuilder.md)
+
+***
+
+### validateSingleShard()
+
+```ts
+validateSingleShard(validateSingleShard): MergeInsertBuilder
+```
+
+Controls how an LSM merge checks that its input targets a single shard.
+
+When a table has an LSM write spec, every row in a `mergeInsert` call must
+route to the same shard. When `true` (the default), every row is inspected
+to verify this. When `false`, only the first row is inspected and the
+shard it routes to is used for the whole input — a faster path for callers
+that have already pre-sharded their input. Has no effect on tables without
+an LSM write spec.
+
+#### Parameters
+
+* **validateSingleShard**: `boolean`
+    Whether to check every row routes to one shard. Defaults to `true`.
+
+#### Returns
+
+[`MergeInsertBuilder`](MergeInsertBuilder.md)
+
+***
+
 ### whenMatchedUpdateAll()
 
 ```ts

docs/src/js/classes/Table.md

@@ -110,6 +110,23 @@ containing the new version number of the table after altering the columns.
 
 ***
 
+### branches()
+
+```ts
+abstract branches(): Promise<Branches>
+```
+
+Get the branch manager for this table.
+
+Branches are isolated, writable lines of history forked from another
+branch (or version). Writes on a branch do not affect `main`.
+
+#### Returns
+
+`Promise`&lt;[`Branches`](Branches.md)&gt;
+
+***
+
 ### checkout()
 
 ```ts
@@ -187,6 +204,25 @@ Any attempt to use the table after it is closed will result in an error.
 
 ***
 
+### closeLsmWriters()
+
+```ts
+abstract closeLsmWriters(): Promise<void>
+```
+
+Drain and close any cached MemWAL shard writers held for this table.
+
+When an [LsmWriteSpec](../interfaces/LsmWriteSpec.md) is installed, `mergeInsert` opens MemWAL
+shard writers and caches them for reuse across calls. This closes them,
+flushing pending data; writers reopen lazily on the next `mergeInsert`.
+It is a no-op when no writers are cached.
+
+#### Returns
+
+`Promise`&lt;`void`&gt;
+
+***
+
 ### countRows()
 
 ```ts
@@ -259,6 +295,23 @@ await table.createIndex("my_float_col");
 
 ***
 
+### currentBranch()
+
+```ts
+abstract currentBranch(): null | string
+```
+
+The branch this table handle is scoped to, or `null` for the main branch.
+
+A handle returned by [Branches.create](Branches.md#create) or [Branches.checkout](Branches.md#checkout)
+reports the branch it targets; a handle opened normally reports `null`.
+
+#### Returns
+
+`null` \| `string`
+
+***
+
 ### delete()
 
 ```ts
@@ -975,6 +1028,29 @@ based on the row being updated (e.g. "my_col + 1")
 
 ***
 
+### updateFieldMetadata()
+
+```ts
+abstract updateFieldMetadata(updates): Promise<UpdateFieldMetadataResult>
+```
+
+Update per-field (column) metadata.
+
+#### Parameters
+
+* **updates**: [`FieldMetadataUpdate`](../interfaces/FieldMetadataUpdate.md)[]
+    One or more per-field updates. Each
+    update's metadata is merged into the field's existing metadata by default;
+    a value of `null` deletes that key, and `replace: true` swaps the whole map.
+
+#### Returns
+
+`Promise`&lt;[`UpdateFieldMetadataResult`](../interfaces/UpdateFieldMetadataResult.md)&gt;
+
+resolves to the new table version.
+
+***
+
 ### vectorSearch()
 
 ```ts

docs/src/js/enumerations/OAuthFlowType.md

@@ -0,0 +1,29 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / OAuthFlowType
+
+# Enumeration: OAuthFlowType
+
+OAuth authentication flow types.
+
+## Enumeration Members
+
+### AzureManagedIdentity
+
+```ts
+AzureManagedIdentity: "azure_managed_identity";
+```
+
+Azure Managed Identity via IMDS.
+
+***
+
+### ClientCredentials
+
+```ts
+ClientCredentials: "client_credentials";
+```
+
+Client Credentials grant (service-to-service / M2M).

docs/src/js/globals.md

@@ -12,6 +12,7 @@
 ## Enumerations
 
 - [FullTextQueryType](enumerations/FullTextQueryType.md)
+- [OAuthFlowType](enumerations/OAuthFlowType.md)
 - [Occur](enumerations/Occur.md)
 - [Operator](enumerations/Operator.md)
 
@@ -19,6 +20,8 @@
 
 - [BooleanQuery](classes/BooleanQuery.md)
 - [BoostQuery](classes/BoostQuery.md)
+- [BranchContents](classes/BranchContents.md)
+- [Branches](classes/Branches.md)
 - [Connection](classes/Connection.md)
 - [HeaderProvider](classes/HeaderProvider.md)
 - [Index](classes/Index.md)
@@ -65,6 +68,7 @@
 - [DropNamespaceOptions](interfaces/DropNamespaceOptions.md)
 - [DropNamespaceResponse](interfaces/DropNamespaceResponse.md)
 - [ExecutableQuery](interfaces/ExecutableQuery.md)
+- [FieldMetadataUpdate](interfaces/FieldMetadataUpdate.md)
 - [FragmentStatistics](interfaces/FragmentStatistics.md)
 - [FragmentSummaryStats](interfaces/FragmentSummaryStats.md)
 - [FtsOptions](interfaces/FtsOptions.md)
@@ -82,6 +86,8 @@
 - [ListNamespacesResponse](interfaces/ListNamespacesResponse.md)
 - [LsmWriteSpec](interfaces/LsmWriteSpec.md)
 - [MergeResult](interfaces/MergeResult.md)
+- [NativeOAuthConfig](interfaces/NativeOAuthConfig.md)
+- [OAuthConfig](interfaces/OAuthConfig.md)
 - [OpenTableOptions](interfaces/OpenTableOptions.md)
 - [OptimizeOptions](interfaces/OptimizeOptions.md)
 - [OptimizeStats](interfaces/OptimizeStats.md)
@@ -101,6 +107,7 @@
 - [TimeoutConfig](interfaces/TimeoutConfig.md)
 - [TlsConfig](interfaces/TlsConfig.md)
 - [TokenResponse](interfaces/TokenResponse.md)
+- [UpdateFieldMetadataResult](interfaces/UpdateFieldMetadataResult.md)
 - [UpdateOptions](interfaces/UpdateOptions.md)
 - [UpdateResult](interfaces/UpdateResult.md)
 - [Version](interfaces/Version.md)

docs/src/js/interfaces/ConnectionOptions.md

@@ -64,22 +64,39 @@ client used by manifest-enabled native connections.
 
 ***
 
+### oauthConfig?
+
+```ts
+optional oauthConfig: NativeOAuthConfig;
+```
+
+(For LanceDB cloud only): OAuth configuration for IdP-based
+authentication (e.g., Azure Entra ID). When set, token acquisition
+and refresh are handled entirely in Rust. TypeScript users should pass
+the public `OAuthConfig` type exported from `@lancedb/lancedb`.
+
+***
+
 ### readConsistencyInterval?
 
 ```ts
 optional readConsistencyInterval: number;
 ```
 
-(For LanceDB OSS only): The interval, in seconds, at which to check for
-updates to the table from other processes. If None, then consistency is not
-checked. For performance reasons, this is the default. For strong
-consistency, set this to zero seconds. Then every read will check for
-updates from other processes. As a compromise, you can set this to a
-non-zero value for eventual consistency. If more than that interval
-has passed since the last check, then the table will be checked for updates.
-Note: this consistency only applies to read operations. Write operations are
+The interval, in seconds, at which to check for updates to the table
+from other processes. If None, then consistency is not checked. For
+performance reasons, this is the default. For strong consistency, set
+this to zero seconds. Then every read will check for updates from other
+processes. As a compromise, you can set this to a non-zero value for
+eventual consistency. If more than that interval has passed since the
+last check, then the table will be checked for updates. Note: this
+consistency only applies to read operations. Write operations are
 always consistent.
 
+Stronger consistency is not free. The smaller the interval, the more
+often each read pays the cost of checking for updates against object
+storage, raising per-read latency and cost.
+
 ***
 
 ### region?

docs/src/js/interfaces/FieldMetadataUpdate.md

@@ -0,0 +1,41 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / FieldMetadataUpdate
+
+# Interface: FieldMetadataUpdate
+
+A per-field metadata update, addressed by dot-path.
+
+## Properties
+
+### metadata
+
+```ts
+metadata: Record<string, null | string>;
+```
+
+Metadata key/value pairs. Merged into the field's existing metadata by
+default; a value of `null` deletes that key.
+
+***
+
+### path
+
+```ts
+path: string;
+```
+
+Dot-separated path to the field. For a top-level column this is just its
+name; for a nested field it's the path, e.g. "a.b.c".
+
+***
+
+### replace?
+
+```ts
+optional replace: boolean;
+```
+
+If true, replace the field's entire metadata map instead of merging.

docs/src/js/interfaces/IndexConfig.md

@@ -23,6 +23,31 @@ be more columns to represent composite indices.
 
 ***
 
+### createdAt?
+
+```ts
+optional createdAt: Date;
+```
+
+When the index was created.
+
+`undefined` for remote tables or indices created before timestamps were tracked.
+
+***
+
+### indexDetails?
+
+```ts
+optional indexDetails: any;
+```
+
+Index-type-specific details parsed as a JavaScript object.
+
+Falls back to a raw string if JSON parsing fails. `undefined` for
+remote tables or when details are unavailable.
+
+***
+
 ### indexType
 
 ```ts
@@ -33,6 +58,30 @@ The type of the index
 
 ***
 
+### indexUuid?
+
+```ts
+optional indexUuid: string;
+```
+
+The UUID of the first segment of the index.
+
+`undefined` for remote tables, which do not yet surface this.
+
+***
+
+### indexVersion?
+
+```ts
+optional indexVersion: number;
+```
+
+The on-disk index format version.
+
+`undefined` for remote tables.
+
+***
+
 ### name
 
 ```ts
@@ -40,3 +89,63 @@ name: string;
 ```
 
 The name of the index
+
+***
+
+### numIndexedRows?
+
+```ts
+optional numIndexedRows: number;
+```
+
+The number of rows indexed, across all segments.
+
+`undefined` for remote tables.
+
+***
+
+### numSegments?
+
+```ts
+optional numSegments: number;
+```
+
+The number of segments that make up the index.
+
+`undefined` for remote tables.
+
+***
+
+### numUnindexedRows?
+
+```ts
+optional numUnindexedRows: number;
+```
+
+The number of rows not yet covered by this index.
+
+`undefined` for remote tables.
+
+***
+
+### sizeBytes?
+
+```ts
+optional sizeBytes: number;
+```
+
+The total size in bytes of all index files across all segments.
+
+`undefined` for remote tables or indices without size tracking.
+
+***
+
+### typeUrl?
+
+```ts
+optional typeUrl: string;
+```
+
+The protobuf type URL, a precise type identifier for the index.
+
+`undefined` for remote tables.

docs/src/js/interfaces/IndexStatistics.md

@@ -30,17 +30,6 @@ The type of the index
 
 ***
 
-### loss?
-
-```ts
-optional loss: number;
-```
-
-The KMeans loss value of the index,
-it is only present for vector indices.
-
-***
-
 ### numIndexedRows
 
 ```ts

docs/src/js/interfaces/LsmWriteSpec.md

@@ -11,7 +11,10 @@ Specification selecting Lance's MemWAL LSM-style write path for
 
 `specType` is `"bucket"`, `"identity"`, or `"unsharded"`. For `"bucket"`,
 `column` and `numBuckets` are required; for `"identity"`, `column` is
-required.
+required and must be a deterministic function of the unenforced primary
+key (every row with a given primary key must always produce the same
+`column` value, or upserts of that key can land in different shards and a
+stale version can win).
 
 ## Properties
 

docs/src/js/interfaces/MergeResult.md

@@ -32,6 +32,14 @@ numInsertedRows: number;
 
 ***
 
+### numRows
+
+```ts
+numRows: number;
+```
+
+***
+
 ### numUpdatedRows
 
 ```ts

docs/src/js/interfaces/NativeOAuthConfig.md

@@ -0,0 +1,88 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / NativeOAuthConfig
+
+# Interface: NativeOAuthConfig
+
+OAuth configuration for LanceDB authentication.
+
+This is the generated napi-rs binding shape. TypeScript users should prefer
+the public `OAuthConfig` type exported from `@lancedb/lancedb`.
+
+All token acquisition and refresh is handled in the Rust layer.
+
+## Properties
+
+### clientId
+
+```ts
+clientId: string;
+```
+
+Application / Client ID.
+
+***
+
+### clientSecret?
+
+```ts
+optional clientSecret: string;
+```
+
+Client secret (required for client_credentials).
+
+***
+
+### flow?
+
+```ts
+optional flow: string;
+```
+
+Authentication flow: "client_credentials" or "azure_managed_identity"
+
+***
+
+### issuerUrl
+
+```ts
+issuerUrl: string;
+```
+
+OIDC issuer URL or OAuth authority URL.
+For Azure: `https://login.microsoftonline.com/{tenant_id}/v2.0`
+
+***
+
+### managedIdentityClientId?
+
+```ts
+optional managedIdentityClientId: string;
+```
+
+Client ID for user-assigned managed identity (azure_managed_identity).
+
+***
+
+### refreshBufferSecs?
+
+```ts
+optional refreshBufferSecs: number;
+```
+
+Seconds before expiry to trigger proactive refresh (default: 300).
+Keep this well below the token TTL; if it is greater than or equal to
+the TTL, each request refreshes the token.
+
+***
+
+### scopes
+
+```ts
+scopes: string[];
+```
+
+OAuth scopes to request. For Azure managed identity, exactly one scope
+or resource is required. For example: `["api://{app_id}/.default"]`

docs/src/js/interfaces/OAuthConfig.md

@@ -0,0 +1,111 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / OAuthConfig
+
+# Interface: OAuthConfig
+
+OAuth configuration for LanceDB authentication.
+
+This is the public TypeScript OAuth configuration type. The generated
+`NativeOAuthConfig` type has the same runtime shape but is an implementation
+detail of the napi-rs binding.
+
+All token acquisition and refresh is handled in the Rust layer.
+This config is passed through to Rust via napi-rs.
+
+## Examples
+
+```typescript
+const config: OAuthConfig = {
+  issuerUrl: "https://login.microsoftonline.com/{tenant}/v2.0",
+  clientId: "app-id",
+  clientSecret: "secret",
+  scopes: ["api://lancedb-api/.default"],
+};
+```
+
+```typescript
+const config: OAuthConfig = {
+  issuerUrl: "https://login.microsoftonline.com/{tenant}/v2.0",
+  clientId: "app-id",
+  scopes: ["api://lancedb-api/.default"],
+  flow: OAuthFlowType.AzureManagedIdentity,
+};
+```
+
+## Properties
+
+### clientId
+
+```ts
+clientId: string;
+```
+
+Application / Client ID.
+
+***
+
+### clientSecret?
+
+```ts
+optional clientSecret: string;
+```
+
+Client secret (required for ClientCredentials).
+
+***
+
+### flow?
+
+```ts
+optional flow: OAuthFlowType;
+```
+
+Authentication flow (default: ClientCredentials).
+
+***
+
+### issuerUrl
+
+```ts
+issuerUrl: string;
+```
+
+OIDC issuer URL or OAuth authority URL.
+For Azure: `https://login.microsoftonline.com/{tenant_id}/v2.0`
+
+***
+
+### managedIdentityClientId?
+
+```ts
+optional managedIdentityClientId: string;
+```
+
+Client ID for user-assigned managed identity (AzureManagedIdentity).
+
+***
+
+### refreshBufferSecs?
+
+```ts
+optional refreshBufferSecs: number;
+```
+
+Seconds before expiry to trigger proactive refresh (default: 300).
+Keep this well below the token TTL; if it is greater than or equal to
+the TTL, each request refreshes the token.
+
+***
+
+### scopes
+
+```ts
+scopes: string[];
+```
+
+OAuth scopes to request.
+For Azure managed identity, exactly one scope or resource is required.
+For example: `["api://{app_id}/.default"]`

docs/src/js/interfaces/OpenTableOptions.md

@@ -8,6 +8,18 @@
 
 ## Properties
 
+### branch?
+
+```ts
+optional branch: string;
+```
+
+Open the table scoped to this branch instead of the default branch.
+
+Reads and writes on the returned table operate in the branch's context.
+
+***
+
 ### ~~indexCacheSize?~~
 
 ```ts
@@ -43,3 +55,17 @@ Options already set on the connection will be inherited by the table,
 but can be overridden here.
 
 The available options are described at https://docs.lancedb.com/storage/
+
+***
+
+### version?
+
+```ts
+optional version: number;
+```
+
+Open the table pinned to this version, producing a read-only view.
+
+Composes with [OpenTableOptions.branch](OpenTableOptions.md#branch): when both are set, opens
+that branch at the version; otherwise opens `main` at the version. Call
+`checkoutLatest` to return to a writable state.

docs/src/js/interfaces/UpdateFieldMetadataResult.md

@@ -0,0 +1,15 @@
+[**@lancedb/lancedb**](../README.md) • **Docs**
+
+***
+
+[@lancedb/lancedb](../globals.md) / UpdateFieldMetadataResult
+
+# Interface: UpdateFieldMetadataResult
+
+## Properties
+
+### version
+
+```ts
+version: number;
+```

java/lancedb-core/pom.xml

@@ -8,7 +8,7 @@
     <parent>
       <groupId>com.lancedb</groupId>
       <artifactId>lancedb-parent</artifactId>
-      <version>0.29.1-beta.0</version>
+      <version>0.31.0-beta.4</version>
       <relativePath>../pom.xml</relativePath>
     </parent>
 

java/pom.xml

@@ -6,7 +6,7 @@
 
     <groupId>com.lancedb</groupId>
     <artifactId>lancedb-parent</artifactId>
-    <version>0.29.1-beta.0</version>
+    <version>0.31.0-beta.4</version>
     <packaging>pom</packaging>
     <name>${project.artifactId}</name>
     <description>LanceDB Java SDK Parent POM</description>
@@ -28,7 +28,7 @@
     <properties>
         <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
         <arrow.version>15.0.0</arrow.version>
-        <lance-core.version>7.0.0-beta.13</lance-core.version>
+        <lance-core.version>9.0.0-beta.10</lance-core.version>
         <spotless.skip>false</spotless.skip>
         <spotless.version>2.30.0</spotless.version>
         <spotless.java.googlejavaformat.version>1.7</spotless.java.googlejavaformat.version>

nodejs/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.29.1-beta.0"
+version = "0.31.0-beta.4"
 publish = false
 license.workspace = true
 description.workspace = true
@@ -25,8 +25,12 @@ lancedb = { path = "../rust/lancedb", default-features = false }
 lance-namespace.workspace = true
 napi = { version = "3.8.3", default-features = false, features = [
     "napi9",
-    "async"
+    "async",
+    "chrono_date",
+    "serde-json",
 ] }
+chrono = { version = "0.4", default-features = false, features = ["clock"] }
+serde_json = "1"
 napi-derive = "3.5.2"
 # Prevent dynamic linking of lzma, which comes from datafusion
 lzma-sys = { version = "0.1", features = ["static"] }

nodejs/__test__/connection.test.ts

@@ -171,18 +171,22 @@ describe("given a connection", () => {
 
     let manifestDir =
       tmpDir.name + "/test_manifest_paths_v2_empty.lance/_versions";
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d{20}\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d{20}\.manifest$/);
+      });
 
     table = (await db.createTable("test_manifest_paths_v2", [{ id: 1 }], {
       enableV2ManifestPaths: true,
     })) as LocalTable;
     expect(await table.usesV2ManifestPaths()).toBe(true);
     manifestDir = tmpDir.name + "/test_manifest_paths_v2.lance/_versions";
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d{20}\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d{20}\.manifest$/);
+      });
   });
 
   it("should be able to migrate tables to the V2 manifest paths", async () => {
@@ -199,16 +203,20 @@ describe("given a connection", () => {
 
     const manifestDir =
       tmpDir.name + "/test_manifest_path_migration.lance/_versions";
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d\.manifest$/);
+      });
 
     await table.migrateManifestPathsV2();
     expect(await table.usesV2ManifestPaths()).toBe(true);
 
-    readdirSync(manifestDir).forEach((file) => {
-      expect(file).toMatch(/^\d{20}\.manifest$/);
-    });
+    readdirSync(manifestDir)
+      .filter((f) => f.endsWith(".manifest"))
+      .forEach((file) => {
+        expect(file).toMatch(/^\d{20}\.manifest$/);
+      });
   });
 });
 

nodejs/__test__/remote.test.ts

@@ -191,6 +191,40 @@ describe("remote connection", () => {
     );
   });
 
+  it("supports version time-travel and branches on remote", async () => {
+    await withMockDatabase(
+      (req, res) => {
+        const body = req.url?.includes("/branches/list")
+          ? JSON.stringify({
+              branches: {
+                exp: { parentVersion: 1, createAt: 1, manifestSize: 1 },
+              },
+            })
+          : JSON.stringify({ name: "t", version: 2, schema: { fields: [] } });
+        res.writeHead(200, { "Content-Type": "application/json" }).end(body);
+      },
+      async (db) => {
+        // version-only (and "main" + version) time-travel the main chain
+        const v2 = await db.openTable("t", undefined, { version: 2 });
+        expect(v2.currentBranch()).toBeNull();
+        const mainV2 = await db.openTable("t", undefined, {
+          branch: "main",
+          version: 2,
+        });
+        expect(mainV2.currentBranch()).toBeNull();
+
+        // a non-main branch opens a handle scoped to that branch
+        const exp = await db.openTable("t", undefined, { branch: "exp" });
+        expect(exp.currentBranch()).toBe("exp");
+        const expV2 = await db.openTable("t", undefined, {
+          branch: "exp",
+          version: 2,
+        });
+        expect(expV2.currentBranch()).toBe("exp");
+      },
+    );
+  });
+
   describe("TlsConfig", () => {
     it("should create TlsConfig with all fields", () => {
       const tlsConfig: TlsConfig = {

nodejs/__test__/table.test.ts

@@ -85,6 +85,140 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
       await expect(table.countRows()).resolves.toBe(3);
     });
 
+    it("should support branches", async () => {
+      await table.add([{ id: 1 }]);
+      expect(await table.countRows()).toBe(1);
+
+      expect(table.currentBranch()).toBeNull();
+
+      // fork an isolated, writable branch from main
+      const branch = await (await table.branches()).create("exp");
+      expect(branch.currentBranch()).toBe("exp");
+      expect(await branch.countRows()).toBe(1);
+      await branch.add([{ id: 2 }]);
+      expect(await branch.countRows()).toBe(2);
+      // main is untouched by branch writes
+      expect(await table.countRows()).toBe(1);
+
+      // listed, with main (null) as the parent
+      const list = await (await table.branches()).list();
+      expect(Object.keys(list)).toContain("exp");
+      expect(list["exp"].parentBranch).toBeNull();
+
+      // fromRef="main" is equivalent to the default
+      await (await table.branches()).create("exp2", "main");
+      const list2 = await (await table.branches()).list();
+      expect(list2["exp2"].parentBranch).toBeNull();
+
+      // checkout returns a handle scoped to the branch's latest
+      const checkedOut = await (await table.branches()).checkout("exp");
+      expect(checkedOut.currentBranch()).toBe("exp");
+      expect(await checkedOut.countRows()).toBe(2);
+
+      // delete removes it
+      await (await table.branches()).delete("exp");
+      await (await table.branches()).delete("exp2");
+      const after = await (await table.branches()).list();
+      expect(Object.keys(after)).not.toContain("exp");
+    });
+
+    it("should open a branch via open_table", async () => {
+      const db = await connect(tmpDir.name);
+      await table.add([{ id: 1 }]);
+      const branch = await (await table.branches()).create("exp");
+      await branch.add([{ id: 2 }]);
+
+      // open_table(..., { branch }) returns a handle scoped to the branch
+      const opened = await db.openTable("some_table", undefined, {
+        branch: "exp",
+      });
+      expect(await opened.countRows()).toBe(2);
+      // opening without branch still tracks main
+      expect(await (await db.openTable("some_table")).countRows()).toBe(1);
+    });
+
+    it("should open a branch at a version isolated from main and HEAD", async () => {
+      const db = await connect(tmpDir.name);
+      // main: a single fork-point row
+      const t = await db.createTable("bv_table", [{ id: 0 }]);
+      const mainV1 = await t.version();
+
+      // fork "exp", then advance exp AND main independently past the fork so
+      // they diverge while sharing version numbers
+      const exp = await (await t.branches()).create("exp");
+      await exp.add([{ id: 1 }]); // exp: {0, 1}
+      const expV2 = await exp.version();
+      await exp.add([{ id: 2 }]); // exp HEAD: {0, 1, 2}
+      await t.add([{ id: 100 }, { id: 101 }, { id: 102 }]); // main HEAD: {0,100,101,102}
+      expect(await t.version()).toBe(expV2);
+
+      // open exp at the shared version: the data must be exp's, not main's.
+      // count alone cannot prove this (main@v2 also exists), so assert
+      // provenance by content.
+      const pinned = await db.openTable("bv_table", undefined, {
+        branch: "exp",
+        version: expV2,
+      });
+      expect(await pinned.countRows()).toBe(2); // not exp HEAD (3), not main@v2 (4)
+      expect(await pinned.countRows("id = 1")).toBe(1); // exp's post-fork row
+      expect(await pinned.countRows("id = 100")).toBe(0); // main's rows invisible
+
+      // the same coordinate is reachable directly via branches().checkout(name, version)
+      const pinnedDirect = await (await t.branches()).checkout("exp", expV2);
+      expect(await pinnedDirect.countRows()).toBe(2);
+
+      // the HEADs are unaffected
+      expect(
+        await (
+          await db.openTable("bv_table", undefined, { branch: "exp" })
+        ).countRows(),
+      ).toBe(3);
+      expect(await (await db.openTable("bv_table")).countRows()).toBe(4);
+
+      // version-only (no branch) time-travels main itself: its fork-point
+      // version holds only main's first row, and the shared version number
+      // resolves to main's data, not the branch's ("opens main at the version")
+      const oldMain = await db.openTable("bv_table", undefined, {
+        version: mainV1,
+      });
+      expect(await oldMain.countRows()).toBe(1);
+      const sharedOnMain = await db.openTable("bv_table", undefined, {
+        version: expV2,
+      });
+      expect(await sharedOnMain.countRows()).toBe(4); // main@v2, not exp@v2 (2)
+
+      // detached head: writing to a pinned version is rejected
+      await expect(pinned.add([{ id: 9 }])).rejects.toThrow(
+        /cannot be modified/,
+      );
+
+      // a nonexistent version is rejected -- on main, and on a branch (a
+      // distinct resolution path, on the branch's manifests)
+      await expect(
+        db.openTable("bv_table", undefined, { version: 9999 }),
+      ).rejects.toThrow();
+      await expect(
+        db.openTable("bv_table", undefined, { branch: "exp", version: 9999 }),
+      ).rejects.toThrow();
+
+      // checkoutLatest re-attaches the pinned handle to the BRANCH's HEAD
+      // (writable again), not main's HEAD (4), and not staying pinned (2)
+      await pinned.checkoutLatest();
+      expect(await pinned.countRows()).toBe(3); // exp HEAD
+      await pinned.add([{ id: 3 }]);
+      expect(await pinned.countRows()).toBe(4); // writable again
+    });
+
+    it("rejects invalid branch inputs", async () => {
+      const branches = await table.branches();
+      await expect(branches.create("")).rejects.toThrow("non-empty");
+      await expect(branches.checkout("")).rejects.toThrow("non-empty");
+      await expect(branches.delete("")).rejects.toThrow("non-empty");
+      await expect(branches.create("bad", "main", -1)).rejects.toThrow(
+        "non-negative",
+      );
+    });
+
     it("should show table stats", async () => {
       await table.add([{ id: 1 }, { id: 2 }]);
       await table.add([{ id: 1 }]);
@@ -715,13 +849,15 @@ describe("When creating an index", () => {
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
     const indices = await tbl.listIndices();
     expect(indices.length).toBe(1);
-    expect(indices[0]).toEqual({
-      name: "vec_idx",
-      indexType: "IvfPq",
-      columns: ["vec"],
-    });
+    expect(indices[0]).toEqual(
+      expect.objectContaining({
+        name: "vec_idx",
+        indexType: "IvfPq",
+        columns: ["vec"],
+      }),
+    );
     const stats = await tbl.indexStats("vec_idx");
-    expect(stats?.loss).toBeDefined();
+    expect(stats).toBeDefined();
 
     // Search without specifying the column
     let rst = await tbl
@@ -781,10 +917,22 @@ describe("When creating an index", () => {
     expect(indices2.length).toBe(0);
   });
 
-  it("should create and search a nested vector index", async () => {
+  it("should preserve canonical nested field paths across index lifecycle", async () => {
     const db = await connect(tmpDir.name);
     const nestedSchema = new Schema([
-      new Field("id", new Int32(), true),
+      new Field("rowId", new Int32(), true),
+      new Field("row-id", new Int32(), true),
+      new Field("userId", new Int32(), true),
+      new Field(
+        "metadata",
+        new Struct([new Field("user_id", new Int32(), true)]),
+        true,
+      ),
+      new Field(
+        "MetaData",
+        new Struct([new Field("userId", new Int32(), true)]),
+        true,
+      ),
       new Field(
         "image",
         new Struct([
@@ -796,28 +944,147 @@ describe("When creating an index", () => {
         ]),
         true,
       ),
+      new Field(
+        "payload",
+        new Struct([new Field("text", new Utf8(), true)]),
+        true,
+      ),
+      new Field(
+        "meta-data",
+        new Struct([new Field("user-id", new Int32(), true)]),
+        true,
+      ),
+      new Field(
+        "literal",
+        new Struct([new Field("a.b", new Int32(), true)]),
+        true,
+      ),
     ]);
     const nestedTable = await db.createTable(
-      "nested_vector",
+      "nested_field_index_lifecycle",
       makeArrowTable(
-        Array.from({ length: 300 }, (_, id) => ({
-          id,
-          image: { embedding: [id, id + 1] },
+        Array.from({ length: 300 }, (_, rowId) => ({
+          rowId,
+          "row-id": rowId,
+          userId: rowId,
+          metadata: { ["user_id"]: rowId },
+          ["MetaData"]: { userId: rowId },
+          image: { embedding: [rowId, rowId + 1] },
+          payload: { text: `document ${rowId}` },
+          "meta-data": { "user-id": rowId },
+          literal: { "a.b": rowId },
         })),
         { schema: nestedSchema },
       ),
     );
 
+    await nestedTable.createIndex("rowId", {
+      config: Index.btree(),
+      name: "row_id_idx",
+    });
+    await nestedTable.createIndex("`row-id`", {
+      config: Index.btree(),
+      name: "row_dash_id_idx",
+    });
+    await nestedTable.createIndex("userId", {
+      config: Index.btree(),
+      name: "top_user_id_idx",
+    });
+    await nestedTable.createIndex("metadata.user_id", {
+      config: Index.btree(),
+      name: "nested_user_id_idx",
+    });
+    await nestedTable.createIndex("MetaData.userId", {
+      config: Index.btree(),
+      name: "mixed_case_metadata_user_id_idx",
+    });
+    await nestedTable.createIndex("`meta-data`.`user-id`", {
+      config: Index.btree(),
+      name: "escaped_names_idx",
+    });
+    await nestedTable.createIndex("literal.`a.b`", {
+      config: Index.btree(),
+      name: "literal_dot_idx",
+    });
     await nestedTable.createIndex("image.embedding", {
       name: "image_embedding_idx",
     });
-    const indices = await nestedTable.listIndices();
-    expect(indices).toContainEqual({
-      name: "image_embedding_idx",
-      indexType: "IvfPq",
-      columns: ["image.embedding"],
+    await nestedTable.createIndex("payload.text", {
+      config: Index.fts({ withPosition: false }),
+      name: "payload_text_idx",
     });
 
+    const indices = await nestedTable.listIndices();
+    expect(indices).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          name: "row_id_idx",
+          indexType: "BTree",
+          columns: ["rowId"],
+        }),
+        expect.objectContaining({
+          name: "row_dash_id_idx",
+          indexType: "BTree",
+          columns: ["`row-id`"],
+        }),
+        expect.objectContaining({
+          name: "top_user_id_idx",
+          indexType: "BTree",
+          columns: ["userId"],
+        }),
+        expect.objectContaining({
+          name: "nested_user_id_idx",
+          indexType: "BTree",
+          columns: ["metadata.user_id"],
+        }),
+        expect.objectContaining({
+          name: "mixed_case_metadata_user_id_idx",
+          indexType: "BTree",
+          columns: ["MetaData.userId"],
+        }),
+        expect.objectContaining({
+          name: "escaped_names_idx",
+          indexType: "BTree",
+          columns: ["`meta-data`.`user-id`"],
+        }),
+        expect.objectContaining({
+          name: "literal_dot_idx",
+          indexType: "BTree",
+          columns: ["literal.`a.b`"],
+        }),
+        expect.objectContaining({
+          name: "image_embedding_idx",
+          indexType: "IvfPq",
+          columns: ["image.embedding"],
+        }),
+        expect.objectContaining({
+          name: "payload_text_idx",
+          indexType: "FTS",
+          columns: ["payload.text"],
+        }),
+      ]),
+    );
+
+    const stats = await nestedTable.indexStats(
+      "mixed_case_metadata_user_id_idx",
+    );
+    expect(stats?.numIndexedRows).toEqual(300);
+    expect(stats?.indexType).toEqual("BTREE");
+
+    const filtered = await nestedTable
+      .query()
+      .where("MetaData.userId = 42")
+      .limit(1)
+      .toArray();
+    expect(filtered[0].MetaData.userId).toEqual(42);
+
+    const escapedFiltered = await nestedTable
+      .query()
+      .where("`row-id` = 43")
+      .limit(1)
+      .toArray();
+    expect(escapedFiltered[0]["row-id"]).toEqual(43);
+
     const explicit = await nestedTable
       .query()
       .nearestTo([0.0, 1.0])
@@ -829,7 +1096,37 @@ describe("When creating an index", () => {
       .nearestTo([0.0, 1.0])
       .limit(1)
       .toArray();
-    expect(inferred[0].id).toEqual(explicit[0].id);
+    expect(inferred[0].rowId).toEqual(explicit[0].rowId);
+
+    await nestedTable.add([
+      {
+        rowId: 300,
+        "row-id": 300,
+        userId: 300,
+        metadata: { ["user_id"]: 300 },
+        ["MetaData"]: { userId: 300 },
+        image: { embedding: [300.0, 301.0] },
+        payload: { text: "document 300" },
+        "meta-data": { "user-id": 300 },
+        literal: { "a.b": 300 },
+      },
+    ]);
+    await nestedTable.optimize();
+    const indicesAfterOptimize = await nestedTable.listIndices();
+    expect(indicesAfterOptimize).toEqual(
+      expect.arrayContaining([
+        expect.objectContaining({
+          name: "mixed_case_metadata_user_id_idx",
+          indexType: "BTree",
+          columns: ["MetaData.userId"],
+        }),
+        expect.objectContaining({
+          name: "image_embedding_idx",
+          indexType: "IvfPq",
+          columns: ["image.embedding"],
+        }),
+      ]),
+    );
   });
 
   it("should report multiple nested vector candidates", async () => {
@@ -963,11 +1260,13 @@ describe("When creating an index", () => {
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
     const indices = await tbl.listIndices();
     expect(indices.length).toBe(1);
-    expect(indices[0]).toEqual({
-      name: "vec_idx",
-      indexType: "IvfHnswSq",
-      columns: ["vec"],
-    });
+    expect(indices[0]).toEqual(
+      expect.objectContaining({
+        name: "vec_idx",
+        indexType: "IvfHnswSq",
+        columns: ["vec"],
+      }),
+    );
 
     // Search without specifying the column
     let rst = await tbl
@@ -1140,6 +1439,20 @@ describe("When creating an index", () => {
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
   });
 
+  test("create an FM index", async () => {
+    // FM-Index accelerates substring search on a string/binary column.
+    const db = await connect(tmpDir.name);
+    const fmTbl = await db.createTable("fm_table", [
+      { id: 0, text: "hello world" },
+      { id: 1, text: "foo bar" },
+    ]);
+    await fmTbl.createIndex("text", {
+      config: Index.fm(),
+    });
+    const indexDir = path.join(tmpDir.name, "fm_table.lance", "_indices");
+    expect(fs.readdirSync(indexDir)).toHaveLength(1);
+  });
+
   test("should be able to get index stats", async () => {
     await tbl.createIndex("id");
 
@@ -1150,7 +1463,6 @@ describe("When creating an index", () => {
     expect(stats?.distanceType).toBeUndefined();
     expect(stats?.indexType).toEqual("BTREE");
     expect(stats?.numIndices).toEqual(1);
-    expect(stats?.loss).toBeUndefined();
   });
 
   test("when getting stats on non-existent index", async () => {
@@ -1300,6 +1612,35 @@ describe("When creating an index", () => {
     expect(rst64Query.toString()).toEqual(rst64Search.toString());
     expect(rst64Query.numRows).toBe(2);
   });
+
+  it("should expose rich metadata fields on IndexConfig", async () => {
+    await tbl.createIndex("id", { config: Index.btree() });
+    await tbl.createIndex("vec");
+
+    const indicesByName = Object.fromEntries(
+      (await tbl.listIndices()).map((idx) => [idx.name, idx]),
+    );
+
+    const scalarIdx = indicesByName["id_idx"];
+    expect(scalarIdx).toBeDefined();
+    expect(typeof scalarIdx.indexUuid).toBe("string");
+    expect(scalarIdx.numIndexedRows).toBe(300);
+    expect(scalarIdx.numUnindexedRows).toBe(0);
+    expect(scalarIdx.numSegments).toBeGreaterThanOrEqual(1);
+    expect(scalarIdx.sizeBytes).toBeGreaterThan(0);
+    // Use toString check to avoid cross-realm instanceof failures with native Date objects
+    expect(Object.prototype.toString.call(scalarIdx.createdAt)).toBe(
+      "[object Date]",
+    );
+    expect((scalarIdx.createdAt as Date).getTime()).toBeGreaterThan(0);
+    expect(typeof scalarIdx.indexDetails).toBe("object");
+
+    const vectorIdx = indicesByName["vec_idx"];
+    expect(vectorIdx).toBeDefined();
+    expect(typeof vectorIdx.indexUuid).toBe("string");
+    expect(vectorIdx.numIndexedRows).toBe(300);
+    expect(typeof vectorIdx.indexDetails).toBe("object");
+  });
 });
 
 describe("When querying a table", () => {
@@ -1571,6 +1912,33 @@ describe("schema evolution", function () {
     expect(await table.schema()).toEqual(expectedSchema3);
   });
 
+  it("can update field metadata", async function () {
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("fm", [
+      { id: 1, category: "a" },
+      { id: 2, category: "b" },
+    ]);
+
+    const res = await table.updateFieldMetadata([
+      { path: "category", metadata: { unit: "label", pii: "false" } },
+    ]);
+    expect(res).toHaveProperty("version");
+    expect(res.version).toBe(2);
+
+    let cat = (await table.schema()).fields.find((f) => f.name === "category");
+    expect(cat?.metadata.get("unit")).toBe("label");
+    expect(cat?.metadata.get("pii")).toBe("false");
+
+    // merge: add a key, delete one via null, keep the rest
+    await table.updateFieldMetadata([
+      { path: "category", metadata: { source: "import", pii: null } },
+    ]);
+    cat = (await table.schema()).fields.find((f) => f.name === "category");
+    expect(cat?.metadata.get("unit")).toBe("label"); // preserved
+    expect(cat?.metadata.get("source")).toBe("import"); // added
+    expect(cat?.metadata.has("pii")).toBe(false); // deleted
+  });
+
   it("can cast to various types", async function () {
     const con = await connect(tmpDir.name);
 
@@ -2625,3 +2993,97 @@ describe("setLsmWriteSpec / unsetLsmWriteSpec", () => {
     ).rejects.toThrow();
   });
 });
+
+describe("LSM merge insert", () => {
+  let tmpDir: tmp.DirResult;
+
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  async function bucketTable(conn: Connection): Promise<Table> {
+    // The primary key column must be non-nullable.
+    const table = await conn.createEmptyTable(
+      "t",
+      new arrow.Schema([
+        new arrow.Field("id", new arrow.Utf8(), false),
+        new arrow.Field("value", new arrow.Float64(), true),
+      ]),
+    );
+    await table.add([
+      { id: "a", value: 1 },
+      { id: "b", value: 2 },
+    ]);
+    await table.setUnenforcedPrimaryKey("id");
+    // numBuckets = 1: every row routes to the single bucket.
+    await table.setLsmWriteSpec({
+      specType: "bucket",
+      column: "id",
+      numBuckets: 1,
+    });
+    return table;
+  }
+
+  it("routes merge_insert through the shard writer", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    const res = await table
+      .mergeInsert("id")
+      .whenMatchedUpdateAll()
+      .whenNotMatchedInsertAll()
+      .execute([
+        { id: "c", value: 3 },
+        { id: "d", value: 4 },
+      ]);
+    // LSM path: rows go to the MemWAL, so only numRows is populated.
+    expect(res.numRows).toBe(2);
+    expect(res.version).toBe(0);
+    expect(res.numInsertedRows).toBe(0);
+
+    await table.closeLsmWriters();
+  });
+
+  it("falls back to the standard path with useLsmWrite(false)", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    const res = await table
+      .mergeInsert("id")
+      .whenNotMatchedInsertAll()
+      .useLsmWrite(false)
+      .execute([
+        { id: "b", value: 9 },
+        { id: "e", value: 5 },
+      ]);
+    // Standard path commits: id="e" inserted ("b" already exists).
+    expect(res.numInsertedRows).toBe(1);
+    expect(await table.countRows()).toBe(3);
+  });
+
+  it("supports validateSingleShard(false)", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    const res = await table
+      .mergeInsert("id")
+      .whenMatchedUpdateAll()
+      .whenNotMatchedInsertAll()
+      .validateSingleShard(false)
+      .execute([{ id: "f", value: 6 }]);
+    expect(res.numRows).toBe(1);
+  });
+
+  it("rejects a non-upsert merge under an LSM spec", async () => {
+    const conn = await connect(tmpDir.name);
+    const table = await bucketTable(conn);
+
+    await expect(
+      table
+        .mergeInsert("id")
+        .whenNotMatchedInsertAll()
+        .execute([{ id: "g", value: 7 }]),
+    ).rejects.toThrow();
+  });
+});

nodejs/lancedb/connection.ts

@@ -84,6 +84,20 @@ export interface CreateTableOptions {
 }
 
 export interface OpenTableOptions {
+  /**
+   * Open the table scoped to this branch instead of the default branch.
+   *
+   * Reads and writes on the returned table operate in the branch's context.
+   */
+  branch?: string;
+  /**
+   * Open the table pinned to this version, producing a read-only view.
+   *
+   * Composes with {@link OpenTableOptions.branch}: when both are set, opens
+   * that branch at the version; otherwise opens `main` at the version. Call
+   * `checkoutLatest` to return to a writable state.
+   */
+  version?: number;
   /**
    * Configuration for object storage.
    *
@@ -483,7 +497,20 @@ export class LocalConnection extends Connection {
       options?.indexCacheSize,
     );
 
-    return new LocalTable(innerTable);
+    let table: Table = new LocalTable(innerTable);
+    // "main" is the default branch, so treat it as no branch. On a real branch,
+    // scope and pin in one step (yielding "version V of branch B"); otherwise
+    // pin the version, if any, against main.
+    const branch =
+      options?.branch != null && options.branch !== "main"
+        ? options.branch
+        : undefined;
+    if (branch != null) {
+      table = await (await table.branches()).checkout(branch, options?.version);
+    } else if (options?.version != null) {
+      await table.checkout(options.version);
+    }
+    return table;
   }
 
   async cloneTable(

nodejs/lancedb/index.ts

@@ -38,10 +38,12 @@ export {
   FragmentSummaryStats,
   Tags,
   TagContents,
+  BranchContents,
   MergeResult,
   AddResult,
   AddColumnsResult,
   AlterColumnsResult,
+  UpdateFieldMetadataResult,
   DeleteResult,
   DropColumnsResult,
   UpdateResult,
@@ -50,6 +52,7 @@ export {
   SplitHashOptions,
   SplitSequentialOptions,
   ShuffleOptions,
+  OAuthConfig as NativeOAuthConfig,
 } from "./native.js";
 
 export {
@@ -110,6 +113,7 @@ export {
 
 export {
   Table,
+  Branches,
   AddDataOptions,
   UpdateOptions,
   OptimizeOptions,
@@ -117,6 +121,7 @@ export {
   WriteProgress,
   LsmWriteSpec,
   ColumnAlteration,
+  FieldMetadataUpdate,
 } from "./table";
 
 export {
@@ -126,6 +131,8 @@ export {
   TokenResponse,
 } from "./header";
 
+export { OAuthConfig, OAuthFlowType } from "./oauth";
+
 export { MergeInsertBuilder, WriteExecutionOptions } from "./merge";
 
 export * as embedding from "./embedding";

nodejs/lancedb/indices.ts

@@ -702,6 +702,17 @@ export class Index {
     return new Index(LanceDbIndex.labelList());
   }
 
+  /**
+   * Create an FM-Index.
+   *
+   * An FM-Index is a scalar index on string or binary columns that accelerates
+   * substring search, i.e. `contains(col, 'needle')`. Unlike the tokenized
+   * full-text-search index, it matches arbitrary substrings of the raw bytes.
+   */
+  static fm() {
+    return new Index(LanceDbIndex.fm());
+  }
+
   /**
    * Create a full text search index
    *

nodejs/lancedb/merge.ts

@@ -87,6 +87,41 @@ export class MergeInsertBuilder {
       this.#schema,
     );
   }
+  /**
+   * Controls whether the merge uses the MemWAL LSM write path.
+   *
+   * By default (unset), a `mergeInsert` on a table with an LSM write spec is
+   * routed through Lance's MemWAL shard writer, and a table without one uses
+   * the standard path. Pass `false` to force the standard path even when a
+   * spec is set. Pass `true` to require a spec — `mergeInsert` rejects if none
+   * is installed.
+   *
+   * @param useLsmWrite - Whether to use the LSM write path.
+   */
+  useLsmWrite(useLsmWrite: boolean): MergeInsertBuilder {
+    return new MergeInsertBuilder(
+      this.#native.useLsmWrite(useLsmWrite),
+      this.#schema,
+    );
+  }
+  /**
+   * Controls how an LSM merge checks that its input targets a single shard.
+   *
+   * When a table has an LSM write spec, every row in a `mergeInsert` call must
+   * route to the same shard. When `true` (the default), every row is inspected
+   * to verify this. When `false`, only the first row is inspected and the
+   * shard it routes to is used for the whole input — a faster path for callers
+   * that have already pre-sharded their input. Has no effect on tables without
+   * an LSM write spec.
+   *
+   * @param validateSingleShard - Whether to check every row routes to one shard. Defaults to `true`.
+   */
+  validateSingleShard(validateSingleShard: boolean): MergeInsertBuilder {
+    return new MergeInsertBuilder(
+      this.#native.validateSingleShard(validateSingleShard),
+      this.#schema,
+    );
+  }
   /**
    * Executes the merge insert operation
    *

nodejs/lancedb/oauth.ts

@@ -0,0 +1,76 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+/**
+ * OAuth authentication flow types.
+ */
+export enum OAuthFlowType {
+  /** Client Credentials grant (service-to-service / M2M). */
+  ClientCredentials = "client_credentials",
+  /** Azure Managed Identity via IMDS. */
+  AzureManagedIdentity = "azure_managed_identity",
+}
+
+/**
+ * OAuth configuration for LanceDB authentication.
+ *
+ * This is the public TypeScript OAuth configuration type. The generated
+ * `NativeOAuthConfig` type has the same runtime shape but is an implementation
+ * detail of the napi-rs binding.
+ *
+ * All token acquisition and refresh is handled in the Rust layer.
+ * This config is passed through to Rust via napi-rs.
+ *
+ * @example Client Credentials (service-to-service):
+ * ```typescript
+ * const config: OAuthConfig = {
+ *   issuerUrl: "https://login.microsoftonline.com/{tenant}/v2.0",
+ *   clientId: "app-id",
+ *   clientSecret: "secret",
+ *   scopes: ["api://lancedb-api/.default"],
+ * };
+ * ```
+ *
+ * @example Azure Managed Identity:
+ * ```typescript
+ * const config: OAuthConfig = {
+ *   issuerUrl: "https://login.microsoftonline.com/{tenant}/v2.0",
+ *   clientId: "app-id",
+ *   scopes: ["api://lancedb-api/.default"],
+ *   flow: OAuthFlowType.AzureManagedIdentity,
+ * };
+ * ```
+ */
+export interface OAuthConfig {
+  /**
+   * OIDC issuer URL or OAuth authority URL.
+   * For Azure: `https://login.microsoftonline.com/{tenant_id}/v2.0`
+   */
+  issuerUrl: string;
+
+  /** Application / Client ID. */
+  clientId: string;
+
+  /**
+   * OAuth scopes to request.
+   * For Azure managed identity, exactly one scope or resource is required.
+   * For example: `["api://{app_id}/.default"]`
+   */
+  scopes: string[];
+
+  /** Authentication flow (default: ClientCredentials). */
+  flow?: OAuthFlowType;
+
+  /** Client secret (required for ClientCredentials). */
+  clientSecret?: string;
+
+  /** Client ID for user-assigned managed identity (AzureManagedIdentity). */
+  managedIdentityClientId?: string;
+
+  /**
+   * Seconds before expiry to trigger proactive refresh (default: 300).
+   * Keep this well below the token TTL; if it is greater than or equal to
+   * the TTL, each request refreshes the token.
+   */
+  refreshBufferSecs?: number;
+}

nodejs/lancedb/table.ts

@@ -25,13 +25,16 @@ import {
   AddColumnsSql,
   AddResult,
   AlterColumnsResult,
+  BranchContents,
   DeleteResult,
   DropColumnsResult,
   IndexConfig,
   IndexStatistics,
+  Branches as NativeBranches,
   OptimizeStats,
   TableStatistics,
   Tags,
+  UpdateFieldMetadataResult,
   UpdateResult,
   Table as _NativeTable,
 } from "./native";
@@ -161,7 +164,10 @@ export interface Version {
  *
  * `specType` is `"bucket"`, `"identity"`, or `"unsharded"`. For `"bucket"`,
  * `column` and `numBuckets` are required; for `"identity"`, `column` is
- * required.
+ * required and must be a deterministic function of the unenforced primary
+ * key (every row with a given primary key must always produce the same
+ * `column` value, or upserts of that key can land in different shards and a
+ * stale version can win).
  */
 export interface LsmWriteSpec {
   /** One of `"bucket"`, `"identity"`, or `"unsharded"`. */
@@ -505,6 +511,18 @@ export abstract class Table {
   abstract alterColumns(
     columnAlterations: ColumnAlteration[],
   ): Promise<AlterColumnsResult>;
+
+  /**
+   * Update per-field (column) metadata.
+   * @param {FieldMetadataUpdate[]} updates One or more per-field updates. Each
+   * update's metadata is merged into the field's existing metadata by default;
+   * a value of `null` deletes that key, and `replace: true` swaps the whole map.
+   * @returns {Promise<UpdateFieldMetadataResult>} resolves to the new table version.
+   */
+  abstract updateFieldMetadata(
+    updates: FieldMetadataUpdate[],
+  ): Promise<UpdateFieldMetadataResult>;
+
   /**
    * Drop one or more columns from the dataset
    *
@@ -567,6 +585,16 @@ export abstract class Table {
    * @returns {Promise<void>}
    */
   abstract unsetLsmWriteSpec(): Promise<void>;
+  /**
+   * Drain and close any cached MemWAL shard writers held for this table.
+   *
+   * When an {@link LsmWriteSpec} is installed, `mergeInsert` opens MemWAL
+   * shard writers and caches them for reuse across calls. This closes them,
+   * flushing pending data; writers reopen lazily on the next `mergeInsert`.
+   * It is a no-op when no writers are cached.
+   * @returns {Promise<void>}
+   */
+  abstract closeLsmWriters(): Promise<void>;
   /** Retrieve the version of the table */
 
   abstract version(): Promise<number>;
@@ -627,6 +655,22 @@ export abstract class Table {
    */
   abstract tags(): Promise<Tags>;
 
+  /**
+   * Get the branch manager for this table.
+   *
+   * Branches are isolated, writable lines of history forked from another
+   * branch (or version). Writes on a branch do not affect `main`.
+   */
+  abstract branches(): Promise<Branches>;
+
+  /**
+   * The branch this table handle is scoped to, or `null` for the main branch.
+   *
+   * A handle returned by {@link Branches.create} or {@link Branches.checkout}
+   * reports the branch it targets; a handle opened normally reports `null`.
+   */
+  abstract currentBranch(): string | null;
+
   /**
    * Restore the table to the currently checked out version
    *
@@ -1024,6 +1068,12 @@ export class LocalTable extends Table {
     return await this.inner.alterColumns(processedAlterations);
   }
 
+  async updateFieldMetadata(
+    updates: FieldMetadataUpdate[],
+  ): Promise<UpdateFieldMetadataResult> {
+    return await this.inner.updateFieldMetadata(updates);
+  }
+
   async dropColumns(columnNames: string[]): Promise<DropColumnsResult> {
     return await this.inner.dropColumns(columnNames);
   }
@@ -1041,6 +1091,10 @@ export class LocalTable extends Table {
     return await this.inner.unsetLsmWriteSpec();
   }
 
+  async closeLsmWriters(): Promise<void> {
+    return await this.inner.closeLsmWriters();
+  }
+
   async version(): Promise<number> {
     return await this.inner.version();
   }
@@ -1072,6 +1126,14 @@ export class LocalTable extends Table {
     return await this.inner.tags();
   }
 
+  async branches(): Promise<Branches> {
+    return new Branches(await this.inner.branches());
+  }
+
+  currentBranch(): string | null {
+    return this.inner.currentBranch() ?? null;
+  }
+
   async optimize(options?: Partial<OptimizeOptions>): Promise<OptimizeStats> {
     let cleanupOlderThanMs;
     if (
@@ -1186,3 +1248,73 @@ export interface ColumnAlteration {
   /** Set the new nullability. Note that a nullable column cannot be made non-nullable. */
   nullable?: boolean;
 }
+
+/** A per-field metadata update, addressed by dot-path. */
+export interface FieldMetadataUpdate {
+  /**
+   * Dot-separated path to the field. For a top-level column this is just its
+   * name; for a nested field it's the path, e.g. "a.b.c".
+   */
+  path: string;
+  /**
+   * Metadata key/value pairs. Merged into the field's existing metadata by
+   * default; a value of `null` deletes that key.
+   */
+  metadata: Record<string, string | null>;
+  /** If true, replace the field's entire metadata map instead of merging. */
+  replace?: boolean;
+}
+
+/**
+ * Branch manager for a {@link Table}.
+ *
+ * Unlike tags, `create` and `checkout` return a new {@link Table} handle scoped
+ * to the branch; writes on it do not affect `main`.
+ */
+export class Branches {
+  #inner: NativeBranches;
+
+  /**
+   * Construct a Branches manager. Internal use only.
+   * @hidden
+   */
+  constructor(inner: NativeBranches) {
+    this.#inner = inner;
+  }
+
+  /** List all branches, mapping name to branch metadata. */
+  async list(): Promise<Record<string, BranchContents>> {
+    return await this.#inner.list();
+  }
+
+  /**
+   * Create a branch and return a handle scoped to it.
+   *
+   * @param name Name of the new branch.
+   * @param fromRef Source branch to fork from. Defaults to `main`.
+   * @param fromVersion A specific version on `fromRef`. Defaults to latest.
+   */
+  async create(
+    name: string,
+    fromRef?: string,
+    fromVersion?: number,
+  ): Promise<Table> {
+    return new LocalTable(await this.#inner.create(name, fromRef, fromVersion));
+  }
+
+  /**
+   * Check out an existing branch and return a handle scoped to it.
+   *
+   * With `version` set, the returned handle is pinned to that version of the
+   * branch (a read-only, detached view); otherwise it tracks the branch's
+   * latest and stays writable.
+   */
+  async checkout(name: string, version?: number): Promise<Table> {
+    return new LocalTable(await this.#inner.checkout(name, version));
+  }
+
+  /** Delete a branch. */
+  async delete(name: string): Promise<void> {
+    return await this.#inner.delete(name);
+  }
+}

nodejs/npm/darwin-arm64/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.29.1-beta.0",
+	"version": "0.31.0-beta.4",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.29.1-beta.0",
+	"version": "0.31.0-beta.4",
 	"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": "0.29.1-beta.0",
+	"version": "0.31.0-beta.4",
 	"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": "0.29.1-beta.0",
+	"version": "0.31.0-beta.4",
 	"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": "0.29.1-beta.0",
+	"version": "0.31.0-beta.4",
 	"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": "0.29.1-beta.0",
+  "version": "0.31.0-beta.4",
   "os": [
     "win32"
   ],

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

@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.29.1-beta.0",
+	"version": "0.31.0-beta.4",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.29.1-beta.0",
+  "version": "0.31.0-beta.4",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.29.1-beta.0",
+      "version": "0.31.0-beta.4",
       "cpu": [
         "x64",
         "arm64"
@@ -26,7 +26,7 @@
         "@aws-sdk/client-s3": "3.1003.0",
         "@biomejs/biome": "^1.7.3",
         "@jest/globals": "^29.7.0",
-        "@napi-rs/cli": "3.5.1",
+        "@napi-rs/cli": "3.7.0",
         "@types/axios": "^0.14.0",
         "@types/jest": "^29.1.2",
         "@types/node": "22.7.4",
@@ -2942,9 +2942,9 @@
       }
     },
     "node_modules/@napi-rs/cli": {
-      "version": "3.5.1",
-      "resolved": "https://registry.npmjs.org/@napi-rs/cli/-/cli-3.5.1.tgz",
-      "integrity": "sha512-XBfLQRDcB3qhu6bazdMJsecWW55kR85l5/k0af9BIBELXQSsCFU0fzug7PX8eQp6vVdm7W/U3z6uP5WmITB2Gw==",
+      "version": "3.7.0",
+      "resolved": "https://registry.npmjs.org/@napi-rs/cli/-/cli-3.7.0.tgz",
+      "integrity": "sha512-3d3+rmxlOIV/G1zPWeX4PCxuYnhcCQM2BvY9rtimC8RO0dFR9gtYP+Grov+WoduZtfWRj5N1XvytWeRxxCk5zw==",
       "dev": true,
       "license": "MIT",
       "dependencies": {
@@ -2954,7 +2954,7 @@
         "@octokit/rest": "^22.0.1",
         "clipanion": "^4.0.0-rc.4",
         "colorette": "^2.0.20",
-        "emnapi": "^1.7.1",
+        "emnapi": "^1.10.0",
         "es-toolkit": "^1.41.0",
         "js-yaml": "^4.1.0",
         "obug": "^2.0.0",

nodejs/package.json

@@ -11,7 +11,7 @@
     "ann"
   ],
   "private": false,
-  "version": "0.29.1-beta.0",
+  "version": "0.31.0-beta.4",
   "main": "dist/index.js",
   "exports": {
     ".": "./dist/index.js",
@@ -43,7 +43,7 @@
     "@aws-sdk/client-s3": "3.1003.0",
     "@biomejs/biome": "^1.7.3",
     "@jest/globals": "^29.7.0",
-    "@napi-rs/cli": "3.5.1",
+    "@napi-rs/cli": "3.7.0",
     "@types/axios": "^0.14.0",
     "@types/jest": "^29.1.2",
     "@types/node": "22.7.4",

nodejs/pnpm-lock.yaml

@@ -31,8 +31,8 @@ importers:
         specifier: ^29.7.0
         version: 29.7.0
       '@napi-rs/cli':
-        specifier: 3.5.1
-        version: 3.5.1(@emnapi/core@1.10.0)(@emnapi/runtime@1.10.0)(@types/node@22.7.4)
+        specifier: 3.7.0
+        version: 3.7.0(@emnapi/core@1.10.0)(@emnapi/runtime@1.10.0)(@types/node@22.7.4)
       '@types/axios':
         specifier: ^0.14.0
         version: 0.14.4
@@ -887,8 +887,8 @@ packages:
   '@jridgewell/trace-mapping@0.3.31':
     resolution: {integrity: sha512-zzNR+SdQSDJzc8joaeP8QQoCQr8NuYx2dIIytl1QeBEZHJ9uW6hebsrYgbz8hJwUQao3TWCMtmfV8Nu1twOLAw==}
 
-  '@napi-rs/cli@3.5.1':
-    resolution: {integrity: sha512-XBfLQRDcB3qhu6bazdMJsecWW55kR85l5/k0af9BIBELXQSsCFU0fzug7PX8eQp6vVdm7W/U3z6uP5WmITB2Gw==}
+  '@napi-rs/cli@3.7.0':
+    resolution: {integrity: sha512-3d3+rmxlOIV/G1zPWeX4PCxuYnhcCQM2BvY9rtimC8RO0dFR9gtYP+Grov+WoduZtfWRj5N1XvytWeRxxCk5zw==}
     engines: {node: '>= 16'}
     hasBin: true
     peerDependencies:
@@ -4582,7 +4582,7 @@ snapshots:
       '@jridgewell/resolve-uri': 3.1.2
       '@jridgewell/sourcemap-codec': 1.5.5
 
-  '@napi-rs/cli@3.5.1(@emnapi/core@1.10.0)(@emnapi/runtime@1.10.0)(@types/node@22.7.4)':
+  '@napi-rs/cli@3.7.0(@emnapi/core@1.10.0)(@emnapi/runtime@1.10.0)(@types/node@22.7.4)':
     dependencies:
       '@inquirer/prompts': 8.4.3(@types/node@22.7.4)
       '@napi-rs/cross-toolchain': 1.0.3(@emnapi/core@1.10.0)(@emnapi/runtime@1.10.0)

nodejs/src/connection.rs

@@ -112,6 +112,12 @@ impl Connection {
 
         builder = builder.client_config(rust_config);
 
+        if let Some(oauth_config) = options.oauth_config {
+            let config: lancedb::remote::oauth::OAuthConfig =
+                oauth_config.try_into().default_error()?;
+            builder = builder.oauth_config(config);
+        }
+
         if let Some(api_key) = options.api_key {
             builder = builder.api_key(&api_key);
         }

nodejs/src/index.rs

@@ -4,7 +4,7 @@
 use std::sync::Mutex;
 
 use lancedb::index::Index as LanceDbIndex;
-use lancedb::index::scalar::{BTreeIndexBuilder, FtsIndexBuilder};
+use lancedb::index::scalar::{BTreeIndexBuilder, FmIndexBuilder, FtsIndexBuilder};
 use lancedb::index::vector::{
     IvfFlatIndexBuilder, IvfHnswPqIndexBuilder, IvfHnswSqIndexBuilder, IvfPqIndexBuilder,
     IvfRqIndexBuilder,
@@ -143,6 +143,13 @@ impl Index {
         }
     }
 
+    #[napi(factory)]
+    pub fn fm() -> Self {
+        Self {
+            inner: Mutex::new(Some(LanceDbIndex::Fm(FmIndexBuilder::default()))),
+        }
+    }
+
     #[napi(factory)]
     #[allow(clippy::too_many_arguments)]
     pub fn fts(

nodejs/src/lib.rs

@@ -24,15 +24,19 @@ mod util;
 #[napi(object)]
 #[derive(Debug)]
 pub struct ConnectionOptions {
-    /// (For LanceDB OSS only): The interval, in seconds, at which to check for
-    /// updates to the table from other processes. If None, then consistency is not
-    /// checked. For performance reasons, this is the default. For strong
-    /// consistency, set this to zero seconds. Then every read will check for
-    /// updates from other processes. As a compromise, you can set this to a
-    /// non-zero value for eventual consistency. If more than that interval
-    /// has passed since the last check, then the table will be checked for updates.
-    /// Note: this consistency only applies to read operations. Write operations are
+    /// The interval, in seconds, at which to check for updates to the table
+    /// from other processes. If None, then consistency is not checked. For
+    /// performance reasons, this is the default. For strong consistency, set
+    /// this to zero seconds. Then every read will check for updates from other
+    /// processes. As a compromise, you can set this to a non-zero value for
+    /// eventual consistency. If more than that interval has passed since the
+    /// last check, then the table will be checked for updates. Note: this
+    /// consistency only applies to read operations. Write operations are
     /// always consistent.
+    ///
+    /// Stronger consistency is not free. The smaller the interval, the more
+    /// often each read pays the cost of checking for updates against object
+    /// storage, raising per-read latency and cost.
     pub read_consistency_interval: Option<f64>,
     /// (For LanceDB OSS only): configuration for object storage.
     ///
@@ -61,6 +65,11 @@ pub struct ConnectionOptions {
     /// (For LanceDB cloud only): the host to use for LanceDB cloud. Used
     /// for testing purposes.
     pub host_override: Option<String>,
+    /// (For LanceDB cloud only): OAuth configuration for IdP-based
+    /// authentication (e.g., Azure Entra ID). When set, token acquisition
+    /// and refresh are handled entirely in Rust. TypeScript users should pass
+    /// the public `OAuthConfig` type exported from `@lancedb/lancedb`.
+    pub oauth_config: Option<remote::OAuthConfig>,
 }
 
 #[napi(object)]

nodejs/src/merge.rs

@@ -50,6 +50,20 @@ impl NativeMergeInsertBuilder {
         this
     }
 
+    #[napi]
+    pub fn use_lsm_write(&self, use_lsm_write: bool) -> Self {
+        let mut this = self.clone();
+        this.inner.use_lsm_write(use_lsm_write);
+        this
+    }
+
+    #[napi]
+    pub fn validate_single_shard(&self, validate_single_shard: bool) -> Self {
+        let mut this = self.clone();
+        this.inner.validate_single_shard(validate_single_shard);
+        this
+    }
+
     #[napi(catch_unwind)]
     pub async fn execute(&self, buf: Buffer) -> napi::Result<MergeResult> {
         let data = ipc_file_to_batches(buf.to_vec())

nodejs/src/remote.rs

@@ -3,6 +3,7 @@
 
 use std::collections::HashMap;
 
+use lancedb::error::Error;
 use napi_derive::*;
 
 /// Timeout configuration for remote HTTP client.
@@ -140,6 +141,84 @@ impl From<TlsConfig> for lancedb::remote::TlsConfig {
     }
 }
 
+/// OAuth configuration for LanceDB authentication.
+///
+/// This is the generated napi-rs binding shape. TypeScript users should prefer
+/// the public `OAuthConfig` type exported from `@lancedb/lancedb`.
+///
+/// All token acquisition and refresh is handled in the Rust layer.
+#[napi(object)]
+#[derive(Clone)]
+pub struct OAuthConfig {
+    /// OIDC issuer URL or OAuth authority URL.
+    /// For Azure: `https://login.microsoftonline.com/{tenant_id}/v2.0`
+    pub issuer_url: String,
+    /// Application / Client ID.
+    pub client_id: String,
+    /// OAuth scopes to request. For Azure managed identity, exactly one scope
+    /// or resource is required. For example: `["api://{app_id}/.default"]`
+    pub scopes: Vec<String>,
+    /// Authentication flow: "client_credentials" or "azure_managed_identity"
+    pub flow: Option<String>,
+    /// Client secret (required for client_credentials).
+    pub client_secret: Option<String>,
+    /// Client ID for user-assigned managed identity (azure_managed_identity).
+    pub managed_identity_client_id: Option<String>,
+    /// Seconds before expiry to trigger proactive refresh (default: 300).
+    /// Keep this well below the token TTL; if it is greater than or equal to
+    /// the TTL, each request refreshes the token.
+    pub refresh_buffer_secs: Option<u32>,
+}
+
+impl std::fmt::Debug for OAuthConfig {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("OAuthConfig")
+            .field("issuer_url", &self.issuer_url)
+            .field("client_id", &self.client_id)
+            .field("scopes", &self.scopes)
+            .field("flow", &self.flow)
+            .field(
+                "client_secret",
+                &self.client_secret.as_deref().map(|_| "<redacted>"),
+            )
+            .field(
+                "managed_identity_client_id",
+                &self.managed_identity_client_id,
+            )
+            .field("refresh_buffer_secs", &self.refresh_buffer_secs)
+            .finish()
+    }
+}
+
+impl TryFrom<OAuthConfig> for lancedb::remote::oauth::OAuthConfig {
+    type Error = Error;
+
+    fn try_from(config: OAuthConfig) -> Result<Self, Self::Error> {
+        use lancedb::remote::oauth::OAuthFlow;
+
+        let flow = match config.flow.as_deref().unwrap_or("client_credentials") {
+            "client_credentials" => OAuthFlow::ClientCredentials,
+            "azure_managed_identity" => OAuthFlow::AzureManagedIdentity {
+                client_id: config.managed_identity_client_id,
+            },
+            other => {
+                return Err(Error::InvalidInput {
+                    message: format!("Unknown OAuth flow type: {other}"),
+                });
+            }
+        };
+
+        Ok(Self {
+            issuer_url: config.issuer_url,
+            client_id: config.client_id,
+            client_secret: config.client_secret,
+            scopes: config.scopes,
+            flow,
+            refresh_buffer_secs: config.refresh_buffer_secs.map(|v| v as u64),
+        })
+    }
+}
+
 impl From<ClientConfig> for lancedb::remote::ClientConfig {
     fn from(config: ClientConfig) -> Self {
         Self {
@@ -156,3 +235,45 @@ impl From<ClientConfig> for lancedb::remote::ClientConfig {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_unknown_oauth_flow_returns_invalid_input() {
+        let config = OAuthConfig {
+            issuer_url: "https://issuer.example.com".to_string(),
+            client_id: "client-id".to_string(),
+            scopes: vec!["scope".to_string()],
+            flow: Some("typo".to_string()),
+            client_secret: None,
+            managed_identity_client_id: None,
+            refresh_buffer_secs: None,
+        };
+
+        let err = lancedb::remote::oauth::OAuthConfig::try_from(config).unwrap_err();
+        assert!(matches!(
+            err,
+            Error::InvalidInput { message }
+                if message == "Unknown OAuth flow type: typo"
+        ));
+    }
+
+    #[test]
+    fn test_oauth_config_debug_redacts_client_secret() {
+        let config = OAuthConfig {
+            issuer_url: "https://issuer.example.com".to_string(),
+            client_id: "client-id".to_string(),
+            scopes: vec!["scope".to_string()],
+            flow: Some("client_credentials".to_string()),
+            client_secret: Some("super-secret".to_string()),
+            managed_identity_client_id: None,
+            refresh_buffer_secs: None,
+        };
+
+        let debug = format!("{config:?}");
+        assert!(!debug.contains("super-secret"));
+        assert!(debug.contains("client_secret: Some(\"<redacted>\")"));
+    }
+}

nodejs/src/table.rs

@@ -3,10 +3,13 @@
 
 use std::collections::HashMap;
 
+use chrono::{DateTime, Utc};
+
 use lancedb::ipc::{ipc_file_to_batches, ipc_file_to_schema};
 use lancedb::table::{
-    AddDataMode, ColumnAlteration as LanceColumnAlteration, Duration, NewColumnTransform,
-    OptimizeAction, OptimizeOptions, Table as LanceDbTable,
+    AddDataMode, ColumnAlteration as LanceColumnAlteration, Duration,
+    FieldMetadataUpdate as LanceFieldMetadataUpdate, NewColumnTransform, OptimizeAction,
+    OptimizeOptions, Ref, Table as LanceDbTable,
 };
 use napi::bindgen_prelude::*;
 use napi::threadsafe_function::{ThreadsafeFunction, ThreadsafeFunctionCallMode};
@@ -355,6 +358,23 @@ impl Table {
         Ok(res.into())
     }
 
+    #[napi(catch_unwind)]
+    pub async fn update_field_metadata(
+        &self,
+        updates: Vec<FieldMetadataUpdate>,
+    ) -> napi::Result<UpdateFieldMetadataResult> {
+        let updates = updates
+            .into_iter()
+            .map(LanceFieldMetadataUpdate::from)
+            .collect::<Vec<_>>();
+        let res = self
+            .inner_ref()?
+            .update_field_metadata(&updates)
+            .await
+            .default_error()?;
+        Ok(res.into())
+    }
+
     #[napi(catch_unwind)]
     pub async fn drop_columns(&self, columns: Vec<String>) -> napi::Result<DropColumnsResult> {
         let col_refs = columns.iter().map(String::as_str).collect::<Vec<_>>();
@@ -391,6 +411,11 @@ impl Table {
             .default_error()
     }
 
+    #[napi(catch_unwind)]
+    pub async fn close_lsm_writers(&self) -> napi::Result<()> {
+        self.inner_ref()?.close_lsm_writers().await.default_error()
+    }
+
     #[napi(catch_unwind)]
     pub async fn version(&self) -> napi::Result<i64> {
         self.inner_ref()?
@@ -455,6 +480,19 @@ impl Table {
         })
     }
 
+    #[napi(catch_unwind)]
+    pub async fn branches(&self) -> napi::Result<Branches> {
+        Ok(Branches {
+            inner: self.inner_ref()?.clone(),
+        })
+    }
+
+    /// The branch this handle is scoped to, or `null` for the main branch.
+    #[napi]
+    pub fn current_branch(&self) -> napi::Result<Option<String>> {
+        Ok(self.inner_ref()?.current_branch())
+    }
+
     #[napi(catch_unwind)]
     pub async fn optimize(
         &self,
@@ -572,6 +610,43 @@ pub struct IndexConfig {
     /// Currently this is always an array of size 1. In the future there may
     /// be more columns to represent composite indices.
     pub columns: Vec<String>,
+    /// The UUID of the first segment of the index.
+    ///
+    /// `undefined` for remote tables, which do not yet surface this.
+    pub index_uuid: Option<String>,
+    /// The protobuf type URL, a precise type identifier for the index.
+    ///
+    /// `undefined` for remote tables.
+    pub type_url: Option<String>,
+    /// When the index was created.
+    ///
+    /// `undefined` for remote tables or indices created before timestamps were tracked.
+    pub created_at: Option<DateTime<Utc>>,
+    /// The number of rows indexed, across all segments.
+    ///
+    /// `undefined` for remote tables.
+    pub num_indexed_rows: Option<i64>,
+    /// The number of rows not yet covered by this index.
+    ///
+    /// `undefined` for remote tables.
+    pub num_unindexed_rows: Option<i64>,
+    /// The total size in bytes of all index files across all segments.
+    ///
+    /// `undefined` for remote tables or indices without size tracking.
+    pub size_bytes: Option<i64>,
+    /// The number of segments that make up the index.
+    ///
+    /// `undefined` for remote tables.
+    pub num_segments: Option<i32>,
+    /// The on-disk index format version.
+    ///
+    /// `undefined` for remote tables.
+    pub index_version: Option<i32>,
+    /// Index-type-specific details parsed as a JavaScript object.
+    ///
+    /// Falls back to a raw string if JSON parsing fails. `undefined` for
+    /// remote tables or when details are unavailable.
+    pub index_details: Option<serde_json::Value>,
 }
 
 impl From<lancedb::index::IndexConfig> for IndexConfig {
@@ -581,6 +656,17 @@ impl From<lancedb::index::IndexConfig> for IndexConfig {
             index_type,
             columns: value.columns,
             name: value.name,
+            index_uuid: value.index_uuid,
+            type_url: value.type_url,
+            created_at: value.created_at,
+            num_indexed_rows: value.num_indexed_rows.map(|n| n as i64),
+            num_unindexed_rows: value.num_unindexed_rows.map(|n| n as i64),
+            size_bytes: value.size_bytes.map(|n| n as i64),
+            num_segments: value.num_segments.map(|n| n as i32),
+            index_version: value.index_version,
+            index_details: value
+                .index_details
+                .and_then(|s| serde_json::from_str(&s).ok()),
         }
     }
 }
@@ -742,6 +828,29 @@ pub struct ColumnAlteration {
     pub nullable: Option<bool>,
 }
 
+/// A per-field metadata update, addressed by dot-path. Merges into the field's
+/// existing metadata by default; a `null` value deletes a key, and `replace`
+/// swaps the field's entire metadata map.
+#[napi(object)]
+pub struct FieldMetadataUpdate {
+    /// Dot-separated path to the field (e.g. "embedding" or "a.b.c").
+    pub path: String,
+    /// Metadata keys to set; a `null` value deletes that key.
+    pub metadata: HashMap<String, Option<String>>,
+    /// If true, replace the field's entire metadata map instead of merging.
+    pub replace: Option<bool>,
+}
+
+impl From<FieldMetadataUpdate> for LanceFieldMetadataUpdate {
+    fn from(js: FieldMetadataUpdate) -> Self {
+        Self {
+            path: js.path,
+            metadata: js.metadata,
+            replace: js.replace.unwrap_or(false),
+        }
+    }
+}
+
 impl TryFrom<ColumnAlteration> for LanceColumnAlteration {
     type Error = String;
     fn try_from(js: ColumnAlteration) -> std::result::Result<Self, Self::Error> {
@@ -792,9 +901,6 @@ pub struct IndexStatistics {
     pub distance_type: Option<String>,
     /// The number of parts this index is split into.
     pub num_indices: Option<u32>,
-    /// The KMeans loss value of the index,
-    /// it is only present for vector indices.
-    pub loss: Option<f64>,
 }
 impl From<lancedb::index::IndexStatistics> for IndexStatistics {
     fn from(value: lancedb::index::IndexStatistics) -> Self {
@@ -804,7 +910,6 @@ impl From<lancedb::index::IndexStatistics> for IndexStatistics {
             index_type: value.index_type.to_string(),
             distance_type: value.distance_type.map(|d| d.to_string()),
             num_indices: value.num_indices,
-            loss: value.loss,
         }
     }
 }
@@ -940,6 +1045,7 @@ pub struct MergeResult {
     pub num_updated_rows: i64,
     pub num_deleted_rows: i64,
     pub num_attempts: i64,
+    pub num_rows: i64,
 }
 
 impl From<lancedb::table::MergeResult> for MergeResult {
@@ -950,6 +1056,7 @@ impl From<lancedb::table::MergeResult> for MergeResult {
             num_updated_rows: value.num_updated_rows as i64,
             num_deleted_rows: value.num_deleted_rows as i64,
             num_attempts: value.num_attempts as i64,
+            num_rows: value.num_rows as i64,
         }
     }
 }
@@ -980,6 +1087,19 @@ impl From<lancedb::table::AlterColumnsResult> for AlterColumnsResult {
     }
 }
 
+#[napi(object)]
+pub struct UpdateFieldMetadataResult {
+    pub version: i64,
+}
+
+impl From<lancedb::table::UpdateFieldMetadataResult> for UpdateFieldMetadataResult {
+    fn from(value: lancedb::table::UpdateFieldMetadataResult) -> Self {
+        Self {
+            version: value.version as i64,
+        }
+    }
+}
+
 #[napi(object)]
 pub struct DropColumnsResult {
     pub version: i64,
@@ -999,6 +1119,13 @@ pub struct TagContents {
     pub manifest_size: i64,
 }
 
+#[napi]
+pub struct BranchContents {
+    pub parent_branch: Option<String>,
+    pub parent_version: i64,
+    pub manifest_size: i64,
+}
+
 #[napi]
 pub struct Tags {
     inner: LanceDbTable,
@@ -1067,3 +1194,75 @@ impl Tags {
             .default_error()
     }
 }
+
+#[napi]
+pub struct Branches {
+    inner: LanceDbTable,
+}
+
+#[napi]
+impl Branches {
+    #[napi]
+    pub async fn list(&self) -> napi::Result<HashMap<String, BranchContents>> {
+        let branches = self.inner.list_branches().await.default_error()?;
+        let result = branches
+            .into_iter()
+            .map(|(k, v)| {
+                (
+                    k,
+                    BranchContents {
+                        parent_branch: v.parent_branch,
+                        parent_version: v.parent_version as i64,
+                        manifest_size: v.manifest_size as i64,
+                    },
+                )
+            })
+            .collect();
+        Ok(result)
+    }
+
+    #[napi]
+    pub async fn create(
+        &self,
+        name: String,
+        from_ref: Option<String>,
+        from_version: Option<i64>,
+    ) -> napi::Result<Table> {
+        let from_ref = from_ref.filter(|b| b != "main");
+        let from_version = from_version
+            .map(|v| {
+                u64::try_from(v).map_err(|_| {
+                    napi::Error::from_reason("from_version must be a non-negative integer")
+                })
+            })
+            .transpose()?;
+        let from = Ref::Version(from_ref, from_version);
+        let table = self
+            .inner
+            .create_branch(&name, from)
+            .await
+            .default_error()?;
+        Ok(Table::new(table))
+    }
+
+    #[napi]
+    pub async fn checkout(&self, name: String, version: Option<i64>) -> napi::Result<Table> {
+        let version = version
+            .map(|v| {
+                u64::try_from(v)
+                    .map_err(|_| napi::Error::from_reason("version must be a non-negative integer"))
+            })
+            .transpose()?;
+        let table = self
+            .inner
+            .checkout_branch(&name, version)
+            .await
+            .default_error()?;
+        Ok(Table::new(table))
+    }
+
+    #[napi]
+    pub async fn delete(&self, name: String) -> napi::Result<()> {
+        self.inner.delete_branch(&name).await.default_error()
+    }
+}

python/.bumpversion.toml

@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.33.0-beta.0"
+current_version = "0.34.0-beta.4"
 parse = """(?x)
     (?P<major>0|[1-9]\\d*)\\.
     (?P<minor>0|[1-9]\\d*)\\.
@@ -23,6 +23,8 @@ allow_dirty = true
 commit = true
 message = "Bump version: {current_version} → {new_version}"
 commit_args = ""
+# bump-my-version >=1.4.0 rejects pre_commit_hooks containing shell syntax unless opted in.
+allow_shell_hooks = true
 
 # Update Cargo.lock after version bump
 pre_commit_hooks = [

python/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.33.0-beta.0"
+version = "0.34.0-beta.4"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"
@@ -26,7 +26,8 @@ lance-namespace-impls.workspace = true
 lance-io.workspace = true
 env_logger.workspace = true
 log.workspace = true
-pyo3 = { version = "0.28", features = ["extension-module", "abi3-py39"] }
+pyo3 = { version = "0.28", features = ["extension-module", "abi3-py39", "chrono"] }
+chrono = { version = "0.4", default-features = false, features = ["clock"] }
 pyo3-async-runtimes = { version = "0.28", features = [
     "attributes",
     "tokio-runtime",

python/python/lancedb/__init__.py

@@ -17,6 +17,17 @@ from .db import AsyncConnection, DBConnection, LanceDBConnection
 from .remote import ClientConfig
 from .remote.db import RemoteDBConnection
 from .expr import Expr, col, lit, func
+from .udf import (
+    udf,
+    table_udf,
+    Udf,
+    JobHandle,
+    JobFailedError,
+    MaterializedView,
+    AsyncJobHandle,
+    AsyncMaterializedView,
+)
+from .lineage import Lineage, Node, Edge, FunctionRef
 from .schema import vector
 from .table import AsyncTable, Table
 from ._lancedb import Session
@@ -89,12 +100,13 @@ def connect(
         If presented, connect to LanceDB cloud.
         Otherwise, connect to a database on file system or cloud storage.
         Can be set via environment variable `LANCEDB_API_KEY`.
+        OAuth configuration is currently supported only by ``connect_async``;
+        synchronous LanceDB Cloud connections require an API key.
     region: str, default "us-east-1"
         The region to use for LanceDB Cloud.
     host_override: str, optional
         The override url for LanceDB Cloud.
     read_consistency_interval: timedelta, default None
-        (For LanceDB OSS only)
         The interval at which to check for updates to the table from other
         processes. If None, then consistency is not checked. For performance
         reasons, this is the default. For strong consistency, set this to
@@ -104,6 +116,10 @@ def connect(
         the last check, then the table will be checked for updates. Note: this
         consistency only applies to read operations. Write operations are
         always consistent.
+
+        Stronger consistency is not free. The smaller the interval, the more
+        often each read pays the cost of checking for updates against object
+        storage, raising per-read latency and cost.
     client_config: ClientConfig or dict, optional
         Configuration options for the LanceDB Cloud HTTP client. If a dict, then
         the keys are the attributes of the ClientConfig class. If None, then the
@@ -147,6 +163,13 @@ def connect(
     >>> db = lancedb.connect("s3://my-bucket/lancedb",
     ...                      storage_options={"aws_access_key_id": "***"})
 
+    For tests and temporary data, use an in-memory database:
+
+    >>> db = lancedb.connect("memory://")
+
+    In-memory databases are not persisted. Tables are dropped when the last
+    connection or table handle referencing them is closed.
+
     Connect to LanceDB cloud:
 
     >>> db = lancedb.connect("db://my_database", api_key="ldb_...",
@@ -210,6 +233,7 @@ def connect(
             request_thread_pool=request_thread_pool,
             client_config=client_config,
             storage_options=storage_options,
+            read_consistency_interval=read_consistency_interval,
             **kwargs,
         )
     _check_s3_bucket_with_dots(str(uri), storage_options)
@@ -304,6 +328,15 @@ def deserialize_conn(
             manifest_enabled=parsed.get("manifest_enabled", False),
             namespace_client_properties=parsed.get("namespace_client_properties"),
         )
+    elif connection_type == "remote":
+        return RemoteDBConnection(
+            parsed["db_url"],
+            parsed["api_key"],
+            parsed.get("region", "us-east-1"),
+            host_override=parsed.get("host_override"),
+            client_config=parsed.get("client_config"),
+            storage_options=storage_options,
+        )
     else:
         raise ValueError(f"Unknown connection_type: {connection_type}")
 
@@ -320,6 +353,7 @@ async def connect_async(
     session: Optional[Session] = None,
     manifest_enabled: bool = False,
     namespace_client_properties: Optional[Dict[str, str]] = None,
+    oauth_config=None,
 ) -> AsyncConnection:
     """Connect to a LanceDB database.
 
@@ -336,7 +370,6 @@ async def connect_async(
     host_override: str, optional
         The override url for LanceDB Cloud.
     read_consistency_interval: timedelta, default None
-        (For LanceDB OSS only)
         The interval at which to check for updates to the table from other
         processes. If None, then consistency is not checked. For performance
         reasons, this is the default. For strong consistency, set this to
@@ -346,6 +379,10 @@ async def connect_async(
         the last check, then the table will be checked for updates. Note: this
         consistency only applies to read operations. Write operations are
         always consistent.
+
+        Stronger consistency is not free. The smaller the interval, the more
+        often each read pays the cost of checking for updates against object
+        storage, raising per-read latency and cost.
     client_config: ClientConfig or dict, optional
         Configuration options for the LanceDB Cloud HTTP client. If a dict, then
         the keys are the attributes of the ClientConfig class. If None, then the
@@ -366,6 +403,10 @@ async def connect_async(
     namespace_client_properties : dict, optional
         Additional directory namespace client properties to use with
         ``manifest_enabled=True``.
+    oauth_config : OAuthConfig, optional
+        OAuth configuration for LanceDB Cloud/Enterprise. This is supported by
+        ``connect_async`` only; synchronous ``connect`` uses API key
+        authentication for ``db://`` URIs.
 
     Examples
     --------
@@ -378,6 +419,8 @@ async def connect_async(
     ...     db = await lancedb.connect_async("s3://my-bucket/lancedb",
     ...                                      storage_options={
     ...                                          "aws_access_key_id": "***"})
+    ...     # For tests and temporary data, use an in-memory database
+    ...     db = await lancedb.connect_async("memory://")
     ...     # Connect to LanceDB cloud
     ...     db = await lancedb.connect_async("db://my_database", api_key="ldb_...",
     ...                                      client_config={
@@ -410,11 +453,24 @@ async def connect_async(
             session,
             manifest_enabled,
             namespace_client_properties,
+            oauth_config,
         )
     )
 
 
 __all__ = [
+    "udf",
+    "table_udf",
+    "Udf",
+    "JobHandle",
+    "JobFailedError",
+    "MaterializedView",
+    "AsyncJobHandle",
+    "AsyncMaterializedView",
+    "Lineage",
+    "Node",
+    "Edge",
+    "FunctionRef",
     "connect",
     "connect_async",
     "connect_namespace",

python/python/lancedb/_lancedb.pyi

@@ -1,4 +1,4 @@
-from datetime import timedelta
+from datetime import datetime, timedelta
 from typing import Dict, List, Optional, Tuple, Any, TypedDict, Union, Literal
 
 import pyarrow as pa
@@ -10,6 +10,7 @@ from .index import (
     IvfSq,
     Bitmap,
     LabelList,
+    Fm,
     HnswPq,
     HnswSq,
     HnswFlat,
@@ -47,6 +48,7 @@ class PyExpr:
     def lower(self) -> "PyExpr": ...
     def upper(self) -> "PyExpr": ...
     def contains(self, substr: "PyExpr") -> "PyExpr": ...
+    def isin(self, values: List["PyExpr"]) -> "PyExpr": ...
     def cast(self, data_type: pa.DataType) -> "PyExpr": ...
     def to_sql(self) -> str: ...
 
@@ -186,6 +188,7 @@ class Table:
             BTree,
             Bitmap,
             LabelList,
+            Fm,
             FTS,
         ],
         replace: Optional[bool],
@@ -202,12 +205,15 @@ class Table:
     async def prewarm_index(self, index_name: str) -> None: ...
     async def prewarm_data(self, columns: Optional[List[str]] = None) -> None: ...
     async def list_indices(self) -> list[IndexConfig]: ...
-    async def delete(self, filter: str) -> DeleteResult: ...
+    async def delete(self, filter: Union[str, PyExpr]) -> DeleteResult: ...
     async def add_columns(self, columns: list[tuple[str, str]]) -> AddColumnsResult: ...
     async def add_columns_with_schema(self, schema: pa.Schema) -> AddColumnsResult: ...
     async def alter_columns(
         self, columns: list[dict[str, Any]]
     ) -> AlterColumnsResult: ...
+    async def update_field_metadata(
+        self, updates: list[dict[str, Any]]
+    ) -> UpdateFieldMetadataResult: ...
     async def optimize(
         self,
         *,
@@ -220,8 +226,12 @@ class Table:
     async def set_unenforced_primary_key(self, columns: List[str]) -> None: ...
     async def set_lsm_write_spec(self, spec: LsmWriteSpec) -> None: ...
     async def unset_lsm_write_spec(self) -> None: ...
+    async def close_lsm_writers(self) -> None: ...
     @property
     def tags(self) -> Tags: ...
+    @property
+    def branches(self) -> Branches: ...
+    def current_branch(self) -> Optional[str]: ...
     def query(self) -> Query: ...
     def take_offsets(self, offsets: list[int]) -> TakeQuery: ...
     def take_row_ids(self, row_ids: list[int]) -> TakeQuery: ...
@@ -234,10 +244,30 @@ class Tags:
     async def delete(self, tag: str): ...
     async def update(self, tag: str, version: int): ...
 
+class Branches:
+    async def list(self) -> Dict[str, Any]: ...
+    async def create(
+        self,
+        name: str,
+        from_ref: Optional[str] = None,
+        from_version: Optional[int] = None,
+    ) -> Table: ...
+    async def checkout(self, name: str, version: Optional[int] = None) -> Table: ...
+    async def delete(self, name: str) -> None: ...
+
 class IndexConfig:
     name: str
     index_type: str
     columns: List[str]
+    index_uuid: Optional[str]
+    type_url: Optional[str]
+    created_at: Optional[datetime]
+    num_indexed_rows: Optional[int]
+    num_unindexed_rows: Optional[int]
+    size_bytes: Optional[int]
+    num_segments: Optional[int]
+    index_version: Optional[int]
+    index_details: Optional[Any]
 
 async def connect(
     uri: str,
@@ -250,6 +280,7 @@ async def connect(
     session: Optional[Session],
     manifest_enabled: bool = False,
     namespace_client_properties: Optional[Dict[str, str]] = None,
+    oauth_config: Optional[Any] = None,
 ) -> Connection: ...
 
 class RecordBatchStream:
@@ -420,6 +451,7 @@ class MergeResult:
     num_inserted_rows: int
     num_deleted_rows: int
     num_attempts: int
+    num_rows: int
 
 class LsmWriteSpec:
     """Specification selecting Lance's MemWAL LSM-style write path for
@@ -458,6 +490,9 @@ class AddColumnsResult:
 class AlterColumnsResult:
     version: int
 
+class UpdateFieldMetadataResult:
+    version: int
+
 class DropColumnsResult:
     version: int
 

python/python/lancedb/background_loop.py

@@ -2,6 +2,7 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 import asyncio
+import concurrent.futures
 import os
 import threading
 import warnings
@@ -37,6 +38,24 @@ class BackgroundEventLoop:
 
 LOOP = BackgroundEventLoop()
 
+
+def _new_embedding_executor() -> concurrent.futures.ThreadPoolExecutor:
+    return concurrent.futures.ThreadPoolExecutor(thread_name_prefix="lancedb-embedding")
+
+
+# Embedding functions can block for a long time -- a heavy local model or an
+# HTTP request to a remote embeddings API. Running them on asyncio's default
+# executor lets them starve the unrelated blocking I/O that shares that pool,
+# so they get a dedicated one. See
+# https://github.com/lancedb/lancedb/issues/3310.
+_EMBEDDING_EXECUTOR = _new_embedding_executor()
+
+
+def embedding_executor() -> concurrent.futures.ThreadPoolExecutor:
+    """Return the executor dedicated to running blocking embedding calls."""
+    return _EMBEDDING_EXECUTOR
+
+
 _FORK_WARNED = False
 
 
@@ -47,6 +66,12 @@ def _reset_after_fork():
     # the new state. The Rust-side tokio runtime is reset analogously by a
     # pthread_atfork hook installed in the _lancedb extension.
     LOOP._start()
+    # The embedding executor's worker threads are dead in the child as well.
+    # Replace it with a fresh pool (threads are spawned lazily, so this is
+    # cheap); we don't shut down the old one, since joining its dead workers
+    # could hang.
+    global _EMBEDDING_EXECUTOR
+    _EMBEDDING_EXECUTOR = _new_embedding_executor()
     global _FORK_WARNED
     if not _FORK_WARNED:
         _FORK_WARNED = True

python/python/lancedb/db.py

@@ -65,6 +65,7 @@ if TYPE_CHECKING:
     from .common import DATA, URI
     from .embeddings import EmbeddingFunctionConfig
     from ._lancedb import Session
+    from .udf import MaterializedView, AsyncMaterializedView
 
 from .namespace_utils import (
     _normalize_create_namespace_mode,
@@ -416,6 +417,8 @@ class DBConnection(EnforceOverrides):
         namespace_path: Optional[List[str]] = None,
         storage_options: Optional[Dict[str, str]] = None,
         index_cache_size: Optional[int] = None,
+        branch: Optional[str] = None,
+        version: Optional[int] = None,
     ) -> Table:
         """Open a Lance Table in the database.
 
@@ -444,6 +447,14 @@ class DBConnection(EnforceOverrides):
             connection will be inherited by the table, but can be overridden here.
             See available options at
             <https://docs.lancedb.com/storage/>
+        branch: str, optional
+            If provided, open a handle scoped to this branch instead of the
+            default branch. Reads and writes operate in the branch's context.
+        version: int, optional
+            If provided, open the table pinned to this version, producing a
+            read-only handle. Composes with ``branch``: when both are given,
+            opens that branch at the version; otherwise opens ``main`` at the
+            version. Call ``checkout_latest`` to return to a writable state.
 
         Returns
         -------
@@ -552,6 +563,259 @@ class DBConnection(EnforceOverrides):
         """
         raise NotImplementedError("serialize is not supported for this connection type")
 
+    # -- Derived compute: functions, materialized views, jobs -------------
+    # Server-backed features (LanceDB Enterprise / Cloud); local
+    # connections raise NotImplementedError for now.
+
+    def create_function(
+        self,
+        name,
+        language: str = "python",
+        return_type: Optional[str] = None,
+        body: Optional[str] = None,
+        options: Optional[Dict[str, str]] = None,
+        *,
+        replace: bool = False,
+    ):
+        """Register a UDF (CREATE FUNCTION).
+
+        Pass a ``@udf`` / ``@table_udf``-decorated function (preferred):
+
+            db.create_function(embed)
+
+        or the explicit fields:
+
+        Parameters
+        ----------
+        name: str or Udf
+            A decorated UDF object, or the function name.
+        language: str
+            Implementation language (currently "python").
+        return_type: str
+            SQL return type, e.g. "FLOAT", "FLOAT[1536]",
+            "STRUCT(a FLOAT, b VARCHAR)", "TABLE(chunk VARCHAR, idx INT)".
+        body: str
+            Function body: source text, or base64 cloudpickle bytes when
+            options["body_format"] == "cloudpickle".
+        options: dict, optional
+            input_columns, pip, num_gpus, batch_size, timeout,
+            error_policy, docker_image, body_format, ...
+        replace: bool
+            Drop an existing function of the same name first.
+        """
+        from .udf import Udf
+
+        if isinstance(name, Udf):
+            req = name.create_request()
+            name, language, return_type, body, options = (
+                req["name"],
+                req["language"],
+                req["return_type"],
+                req["body"],
+                req["options"],
+            )
+        if replace:
+            try:
+                self.drop_function(name)
+            except Exception:
+                pass
+        LOOP.run(self._conn.create_function(name, language, return_type, body, options))
+
+    def list_functions(self):
+        """List registered functions (SHOW FUNCTIONS)."""
+        return LOOP.run(self._conn.list_functions())
+
+    def drop_function(self, name: str):
+        """Drop a registered function (DROP FUNCTION)."""
+        LOOP.run(self._conn.drop_function(name))
+
+    def create_materialized_view(
+        self,
+        name: str,
+        source=None,
+        select=None,
+        *,
+        query: Optional[str] = None,
+        where: Optional[str] = None,
+        auto_refresh: bool = False,
+        with_no_data: bool = False,
+        replace: bool = False,
+        partition_by: Optional[str] = None,
+    ) -> "MaterializedView":
+        """Create a materialized view (CREATE MATERIALIZED VIEW); returns a
+        `MaterializedView` handle (``.wait()`` blocks until it is populated).
+
+        Two ways to specify the view body:
+
+        - ergonomic: pass ``source`` (a table name or table) and ``select``
+          items -- column names, expression strings ("embed(body)"),
+          (alias, expression) tuples, or ``@udf`` / ``@table_udf`` objects.
+          The SELECT is assembled and parsed server-side (one parser, shared
+          with SQL).
+        - raw: pass ``query=`` with a full SELECT, e.g.
+          "SELECT id, embed(body) AS vec FROM articles WHERE id > 1".
+
+        `partition_by` partitions the view's (single) table function on a source
+        column. If that column has an IVF vector index the server partitions by
+        its index clusters (image-dedup style); otherwise it groups by distinct
+        value. (Geneva's `partition_by` and `partition_by_indexed_column` unify
+        here -- the engine picks the strategy from the column.)
+        """
+        from .udf import build_view_query, MaterializedView
+
+        if query is None:
+            if source is None or select is None:
+                raise ValueError(
+                    "create_materialized_view needs either query= or both "
+                    "source and select"
+                )
+            query = build_view_query(source, select)
+        if where:
+            query += f" WHERE {where}"
+        if replace:
+            self._drop_view_if_exists(name)
+        job_id = LOOP.run(
+            self._conn.create_materialized_view(
+                name,
+                query=query,
+                auto_refresh=auto_refresh,
+                with_no_data=with_no_data,
+                partition_by=partition_by,
+            )
+        )
+        return MaterializedView(self, name, job_id=job_id)
+
+    def _drop_view_if_exists(self, name: str) -> None:
+        # `replace=True` is "drop if present"; only a not-found error is
+        # benign here. Anything else (perms, server fault) must surface rather
+        # than be masked by a later create failure.
+        try:
+            self.drop_materialized_view(name)
+        except Exception as e:
+            msg = str(e).lower()
+            if "not found" not in msg and "does not exist" not in msg:
+                raise
+
+    def job(self, job_id: str):
+        """A `JobHandle` for reconnecting to an inflight job by id -- e.g. an
+        id you stored, or one returned from the SQL / REST surface. Submit
+        methods (`refresh_column`, `MaterializedView.refresh`) already return a
+        handle directly, so you do not need this to wait on a fresh submission."""
+        from .udf import JobHandle
+
+        return JobHandle(self, job_id)
+
+    def lineage(
+        self,
+        table: str,
+        column: Optional[str] = None,
+        *,
+        direction: Optional[str] = None,
+        depth: Optional[int] = None,
+    ):
+        """Derived-compute lineage of a table/view, or one of its columns:
+        upstream sources, downstream dependents, and the function version +
+        location that produced each derived column (with a drift flag). Returns
+        a `Lineage`. `direction` is "upstream" | "downstream" | "both" (server
+        default both); `depth` limits column-hops (transitive when omitted)."""
+        # `self._conn` is the AsyncConnection; drive its async `lineage`
+        # (which parses the JSON) on the loop, mirroring create_materialized_view.
+        return LOOP.run(
+            self._conn.lineage(table, column, direction=direction, depth=depth)
+        )
+
+    def _refresh_materialized_view(
+        self,
+        name: str,
+        *,
+        full: bool = False,
+        src_version: Optional[int] = None,
+        num_workers: Optional[int] = None,
+        max_workers: Optional[int] = None,
+    ) -> str:
+        """Internal: submit a materialized-view refresh, return the job id.
+        The public surface is ``MaterializedView.refresh()`` (which returns a
+        `JobHandle`); this stays private so refresh is only reached through the
+        handle.
+
+        ``full=True`` forces a full rebuild (recompute and replace every row)
+        instead of the default incremental refresh.
+        """
+        return LOOP.run(
+            self._conn._refresh_materialized_view(
+                name,
+                full=full,
+                src_version=src_version,
+                num_workers=num_workers,
+                max_workers=max_workers,
+            )
+        )
+
+    def explain_refresh_materialized_view(
+        self,
+        name: str,
+        *,
+        full: bool = False,
+        src_version: Optional[int] = None,
+    ):
+        """Plan a refresh without running it (EXPLAIN REFRESH). Returns a
+        plan with .has_work / .source_version / .last_refreshed_version /
+        .full_refresh / .rebuild / .units_total. `full=True` plans a full
+        rebuild (incremental planning needs stable row IDs on the source)."""
+        return LOOP.run(
+            self._conn.explain_refresh_materialized_view(
+                name, full=full, src_version=src_version
+            )
+        )
+
+    def alter_materialized_view(self, name: str, *, auto_refresh: bool):
+        """Update a materialized view's options (ALTER MATERIALIZED VIEW)."""
+        LOOP.run(self._conn.alter_materialized_view(name, auto_refresh=auto_refresh))
+
+    def drop_materialized_view(self, name: str):
+        """Drop a materialized view definition (DROP MATERIALIZED VIEW)."""
+        LOOP.run(self._conn.drop_materialized_view(name))
+
+    def list_materialized_views(self):
+        """List registered materialized view definitions."""
+        return LOOP.run(self._conn.list_materialized_views())
+
+    def list_jobs(self):
+        """List inflight server-side jobs across the database's tables."""
+        return LOOP.run(self._conn.list_jobs())
+
+    def get_job(self, job_id: str, table: "str | None" = None):
+        """Look up one server-side job by id (the wait()/status poll path).
+
+        Passing ``table`` (the job's table) lets the server answer with an O(1)
+        single-node read instead of scanning the database's active jobs.
+        Returns the job's status, or None if it's unknown or no longer active.
+        """
+        return LOOP.run(self._conn.get_job(job_id, table))
+
+    def cancel_job(self, job_id: str) -> bool:
+        """Cancel an inflight server-side job by id (CANCEL JOB).
+
+        Returns True if a matching inflight job was found and flagged for
+        cancellation, False if none was inflight (already finished or
+        unknown id) -- cancellation is best-effort.
+        """
+        return LOOP.run(self._conn.cancel_job(job_id))
+
+    def job_history(self, job_id: "str | None" = None):
+        """Durable history of completed server-side jobs (SHOW JOB HISTORY).
+
+        Pass ``job_id`` to narrow to a single job. Unlike :meth:`list_jobs`
+        (live, inflight) these are the terminal records.
+        """
+        return LOOP.run(self._conn.job_history(job_id))
+
+    def errors(self, job_id: "str | None" = None, table: "str | None" = None):
+        """Per-row UDF errors recorded by ``error_policy=skip`` (SHOW ERRORS),
+        optionally filtered by ``job_id`` and/or ``table``.
+        """
+        return LOOP.run(self._conn.errors(job_id, table))
+
 
 class LanceDBConnection(DBConnection):
     """
@@ -958,6 +1222,8 @@ class LanceDBConnection(DBConnection):
         namespace_path: Optional[List[str]] = None,
         storage_options: Optional[Dict[str, str]] = None,
         index_cache_size: Optional[int] = None,
+        branch: Optional[str] = None,
+        version: Optional[int] = None,
     ) -> LanceTable:
         """Open a table in the database.
 
@@ -968,6 +1234,14 @@ class LanceDBConnection(DBConnection):
         namespace_path: List[str], optional
             The namespace to open the table from.  When non-empty, the
             table is resolved through the directory namespace client.
+        branch: str, optional
+            If provided, open a handle scoped to this branch instead of the
+            default branch. Reads and writes operate in the branch's context.
+        version: int, optional
+            If provided, open the table pinned to this version, producing a
+            read-only handle. Composes with ``branch``: when both are given,
+            opens that branch at the version; otherwise opens ``main`` at the
+            version. Call ``checkout_latest`` to return to a writable state.
 
         Returns
         -------
@@ -987,20 +1261,26 @@ class LanceDBConnection(DBConnection):
             )
 
         if namespace_path:
-            return self._namespace_conn().open_table(
+            tbl = self._namespace_conn().open_table(
+                name,
+                namespace_path=namespace_path,
+                storage_options=storage_options,
+                index_cache_size=index_cache_size,
+            )
+        else:
+            tbl = LanceTable.open(
+                self,
                 name,
                 namespace_path=namespace_path,
                 storage_options=storage_options,
                 index_cache_size=index_cache_size,
             )
 
-        return LanceTable.open(
-            self,
-            name,
-            namespace_path=namespace_path,
-            storage_options=storage_options,
-            index_cache_size=index_cache_size,
-        )
+        if branch is not None:
+            tbl = tbl.branches.checkout(branch, version)
+        elif version is not None:
+            tbl.checkout(version)
+        return tbl
 
     def clone_table(
         self,
@@ -1641,6 +1921,8 @@ class AsyncConnection(object):
         location: Optional[str] = None,
         namespace_client: Optional[Any] = None,
         managed_versioning: Optional[bool] = None,
+        branch: Optional[str] = None,
+        version: Optional[int] = None,
     ) -> AsyncTable:
         """Open a Lance Table in the database.
 
@@ -1676,6 +1958,14 @@ class AsyncConnection(object):
         managed_versioning: bool, optional
             Whether managed versioning is enabled for this table. If provided,
             avoids a redundant describe_table call when namespace_client is set.
+        branch: str, optional
+            If provided, open a handle scoped to this branch instead of the
+            default branch. Reads and writes operate in the branch's context.
+        version: int, optional
+            If provided, open the table pinned to this version, producing a
+            read-only handle. Composes with ``branch``: when both are given,
+            opens that branch at the version; otherwise opens ``main`` at the
+            version. Call ``checkout_latest`` to return to a writable state.
 
         Returns
         -------
@@ -1692,7 +1982,14 @@ class AsyncConnection(object):
             namespace_client=namespace_client,
             managed_versioning=managed_versioning,
         )
-        return AsyncTable(table)
+        tbl = AsyncTable(table)
+        # "main" is the default branch, so treat it as no branch: remote rejects
+        # every branch checkout (even "main"), and the version still applies.
+        if branch is not None and branch != "main":
+            tbl = await tbl.branches.checkout(branch, version)
+        elif version is not None:
+            await tbl.checkout(version)
+        return tbl
 
     async def clone_table(
         self,
@@ -1744,6 +2041,200 @@ class AsyncConnection(object):
         )
         return AsyncTable(table)
 
+    # -- Derived compute: functions, materialized views, jobs -------------
+    # Server-backed features (LanceDB Enterprise / Cloud); local
+    # connections raise NotImplementedError for now.
+
+    async def create_function(
+        self,
+        name,
+        language: str = "python",
+        return_type: Optional[str] = None,
+        body: Optional[str] = None,
+        options: Optional[Dict[str, str]] = None,
+        *,
+        replace: bool = False,
+    ):
+        """Register a UDF (CREATE FUNCTION). Accepts a ``@udf``/``@table_udf``
+        object (preferred) or the explicit (name, language, return_type, body,
+        options)."""
+        from .udf import Udf
+
+        if isinstance(name, Udf):
+            req = name.create_request()
+            name, language, return_type, body, options = (
+                req["name"],
+                req["language"],
+                req["return_type"],
+                req["body"],
+                req["options"],
+            )
+        if replace:
+            try:
+                await self.drop_function(name)
+            except Exception:
+                pass
+        await self._inner.create_function(name, language, return_type, body, options)
+
+    async def list_functions(self):
+        """List registered functions (SHOW FUNCTIONS)."""
+        return await self._inner.list_functions()
+
+    async def drop_function(self, name: str):
+        """Drop a registered function (DROP FUNCTION)."""
+        await self._inner.drop_function(name)
+
+    async def create_materialized_view(
+        self,
+        name: str,
+        source=None,
+        select=None,
+        *,
+        query: Optional[str] = None,
+        where: Optional[str] = None,
+        auto_refresh: bool = False,
+        with_no_data: bool = False,
+        replace: bool = False,
+        partition_by: Optional[str] = None,
+    ) -> "AsyncMaterializedView":
+        """Create a materialized view; returns an `AsyncMaterializedView`
+        handle (``.wait()`` blocks until populated). Pass either ``query=`` (a
+        full SELECT) or ``source`` + ``select`` items; `partition_by`
+        partitions the view's table function on a source column (index-cluster
+        if the column is IVF-indexed, else distinct-value). See the sync
+        method for the select grammar."""
+        from .udf import build_view_query, AsyncMaterializedView
+
+        if query is None:
+            if source is None or select is None:
+                raise ValueError(
+                    "create_materialized_view needs either query= or both "
+                    "source and select"
+                )
+            query = build_view_query(source, select)
+        if where:
+            query += f" WHERE {where}"
+        if replace:
+            try:
+                await self.drop_materialized_view(name)
+            except Exception as e:
+                msg = str(e).lower()
+                if "not found" not in msg and "does not exist" not in msg:
+                    raise
+        job_id = await self._inner.create_materialized_view(
+            name,
+            query,
+            auto_refresh=auto_refresh,
+            with_no_data=with_no_data,
+            partition_by=partition_by,
+        )
+        return AsyncMaterializedView(self, name, job_id=job_id)
+
+    def job(self, job_id: str):
+        """An `AsyncJobHandle` for reconnecting to an inflight job by id (a
+        stored id, or one from the SQL / REST surface). Submit methods already
+        return a handle, so this is only needed to re-attach to an existing
+        job."""
+        from .udf import AsyncJobHandle
+
+        return AsyncJobHandle(self, job_id)
+
+    async def lineage(
+        self,
+        table: str,
+        column: Optional[str] = None,
+        *,
+        direction: Optional[str] = None,
+        depth: Optional[int] = None,
+    ):
+        """Derived-compute lineage of a table/view (or column). See the sync
+        `Connection.lineage`. Returns a `Lineage`."""
+        from .lineage import Lineage
+
+        raw = await self._inner.table_lineage(table, column, direction, depth)
+        return Lineage.from_json(raw)
+
+    async def _refresh_materialized_view(
+        self,
+        name: str,
+        *,
+        full: bool = False,
+        src_version: Optional[int] = None,
+        num_workers: Optional[int] = None,
+        max_workers: Optional[int] = None,
+    ) -> str:
+        """Internal: submit a refresh, return the job id. The public surface is
+        ``AsyncMaterializedView.refresh()`` (returns an `AsyncJobHandle`).
+
+        ``full=True`` forces a full rebuild (recompute and replace every row)
+        instead of the default incremental refresh.
+        """
+        return await self._inner.refresh_materialized_view(
+            name,
+            full=full,
+            src_version=src_version,
+            num_workers=num_workers,
+            max_workers=max_workers,
+        )
+
+    async def explain_refresh_materialized_view(
+        self,
+        name: str,
+        *,
+        full: bool = False,
+        src_version: Optional[int] = None,
+    ):
+        """Plan a refresh without running it (EXPLAIN REFRESH)."""
+        return await self._inner.explain_refresh_materialized_view(
+            name, full=full, src_version=src_version
+        )
+
+    async def alter_materialized_view(self, name: str, *, auto_refresh: bool):
+        """Update a materialized view's options."""
+        await self._inner.alter_materialized_view(name, auto_refresh)
+
+    async def drop_materialized_view(self, name: str):
+        """Drop a materialized view definition."""
+        await self._inner.drop_materialized_view(name)
+
+    async def list_materialized_views(self):
+        """List registered materialized view definitions."""
+        return await self._inner.list_materialized_views()
+
+    async def list_jobs(self):
+        """List inflight server-side jobs across the database's tables."""
+        return await self._inner.list_jobs()
+
+    async def get_job(self, job_id: str, table: "str | None" = None):
+        """Look up one server-side job by id (the wait()/status poll path).
+        ``table`` (the job's table) enables an O(1) server-side lookup.
+        Returns the job's status, or None if unknown / no longer active."""
+        return await self._inner.get_job(job_id, table)
+
+    async def cancel_job(self, job_id: str) -> bool:
+        """Cancel an inflight server-side job by id (CANCEL JOB).
+
+        Returns True if a matching inflight job was found and flagged for
+        cancellation, False otherwise (best-effort).
+        """
+        return await self._inner.cancel_job(job_id)
+
+    async def job_history(self, job_id: "str | None" = None):
+        """Durable history of completed server-side jobs (SHOW JOB HISTORY).
+
+        Reads each table's durable job-history store. Pass ``job_id`` to narrow
+        to a single job. Unlike :meth:`list_jobs` (live, inflight) these are the
+        terminal records, with created/updated/completed timestamps.
+        """
+        return await self._inner.job_history(job_id)
+
+    async def errors(self, job_id: "str | None" = None, table: "str | None" = None):
+        """Per-row UDF errors recorded by ``error_policy=skip`` (SHOW ERRORS).
+
+        Optionally filtered by ``job_id`` and/or ``table``.
+        """
+        return await self._inner.errors(job_id, table)
+
     async def rename_table(
         self,
         cur_name: str,

python/python/lancedb/embeddings/colpali.py

@@ -81,6 +81,7 @@ class ColPaliEmbeddings(EmbeddingFunction):
             warnings.warn(
                 "use_token_pooling is deprecated, use pooling_strategy=None instead",
                 DeprecationWarning,
+                stacklevel=2,
             )
             self.pooling_strategy = None
 

python/python/lancedb/expr.py

@@ -19,7 +19,7 @@ operators::
 
 from __future__ import annotations
 
-from typing import Union
+from typing import Iterable, Union
 
 import pyarrow as pa
 
@@ -174,6 +174,11 @@ class Expr:
         """Return True where the string contains *substr*."""
         return Expr(self._inner.contains(_coerce(substr)._inner))
 
+    def isin(self, values: "Iterable[ExprLike]") -> "Expr":
+        """Return True where the value is one of *values* (SQL ``IN``)."""
+        inner = [_coerce(v)._inner for v in values]
+        return Expr(self._inner.isin(inner))
+
     # ── type cast ────────────────────────────────────────────────────────────
 
     def cast(self, data_type: Union[str, "pa.DataType"]) -> "Expr":

python/python/lancedb/index.py

@@ -93,6 +93,20 @@ class LabelList:
     pass
 
 
+@dataclass
+class Fm:
+    """Describe an FM-Index configuration.
+
+    `Fm` is a scalar index on string or binary columns that accelerates
+    substring search, i.e. `contains(col, 'needle')`. Unlike the tokenized
+    `FTS` index, it matches arbitrary substrings of the raw bytes.
+
+    For example, it works with `url`, `path`, `content`, etc.
+    """
+
+    pass
+
+
 @dataclass
 class FTS:
     """Describe a FTS index configuration.
@@ -281,6 +295,9 @@ class HnswPq:
     m: int = 20
     ef_construction: int = 300
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -386,6 +403,9 @@ class HnswSq:
     m: int = 20
     ef_construction: int = 300
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -579,6 +599,9 @@ class IvfFlat:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -609,6 +632,9 @@ class IvfSq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -739,6 +765,9 @@ class IvfPq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 @dataclass
@@ -792,6 +821,9 @@ class IvfRq:
     max_iterations: int = 50
     sample_rate: int = 256
     target_partition_size: Optional[int] = None
+    # Name of the accelerator (e.g. "cuda") to use for IVF training. When set,
+    # create_index() dispatches to pylance to build the index on the accelerator.
+    accelerator: Optional[str] = None
 
 
 __all__ = [
@@ -810,4 +842,5 @@ __all__ = [
     "FTS",
     "Bitmap",
     "LabelList",
+    "Fm",
 ]

python/python/lancedb/lineage.py

@@ -0,0 +1,177 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+"""Client-side model of derived-compute lineage.
+
+`Connection.lineage()` / `Table.lineage()` / `MaterializedView.lineage()` return
+a `Lineage`: the graph of what a column or materialized view derives from
+(upstream), what derives from it (downstream), and -- for each derived column --
+the function that produced it, the version it was produced with, and whether
+that is stale relative to the function the registry now holds.
+
+The server returns this as JSON (the wire contract); these classes deserialize
+it. Nothing here talks to the server.
+"""
+
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass, field
+from typing import List, Optional, Union
+
+
+@dataclass
+class FunctionRef:
+    """The function that produced a derived column, with version + location."""
+
+    name: str
+    #: Version that produced the data (stamped at compute time), if known.
+    as_computed_version: Optional[str] = None
+    #: Version the registry currently holds for this function name.
+    current_version: Optional[str] = None
+    #: True when the column was produced by an older function than the registry
+    #: now holds -- i.e. silently stale; re-refresh to catch up.
+    stale_vs_current: bool = False
+    language: Optional[str] = None
+    docker_image: Optional[str] = None
+    env_digest: Optional[str] = None
+    code_uri: Optional[str] = None
+
+    @classmethod
+    def _from(cls, d: dict) -> "FunctionRef":
+        return cls(
+            name=d["name"],
+            as_computed_version=d.get("as_computed_version"),
+            current_version=d.get("current_version"),
+            stale_vs_current=d.get("stale_vs_current", False),
+            language=d.get("language"),
+            docker_image=d.get("docker_image"),
+            env_digest=d.get("env_digest"),
+            code_uri=d.get("code_uri"),
+        )
+
+
+@dataclass
+class Node:
+    """A lineage node: a table, view, column, or function."""
+
+    kind: str  # "table" | "view" | "column" | "function"
+    id: str  # "table", "table.column", or "fn:name@version"
+    table: Optional[str] = None
+    function: Optional[FunctionRef] = None
+
+    @classmethod
+    def _from(cls, d: dict) -> "Node":
+        fn = d.get("function")
+        return cls(
+            kind=d["kind"],
+            id=d["id"],
+            table=d.get("table"),
+            function=FunctionRef._from(fn) if fn else None,
+        )
+
+
+@dataclass
+class Edge:
+    """`downstream` depends on `upstream`, produced by `via` (a function name,
+    or None for a passthrough)."""
+
+    downstream: str
+    upstream: str
+    via: Optional[str] = None
+
+    @classmethod
+    def _from(cls, d: dict) -> "Edge":
+        return cls(downstream=d["downstream"], upstream=d["upstream"], via=d.get("via"))
+
+
+@dataclass
+class Lineage:
+    """A derived-compute lineage graph (nodes + labeled edges)."""
+
+    target: str
+    nodes: List[Node] = field(default_factory=list)
+    edges: List[Edge] = field(default_factory=list)
+
+    @classmethod
+    def from_json(cls, raw: Union[str, bytes, dict]) -> "Lineage":
+        d = json.loads(raw) if isinstance(raw, (str, bytes)) else raw
+        return cls(
+            target=d.get("target", ""),
+            nodes=[Node._from(n) for n in d.get("nodes", [])],
+            edges=[Edge._from(e) for e in d.get("edges", [])],
+        )
+
+    def functions(self) -> List[FunctionRef]:
+        """The function nodes in the graph."""
+        return [n.function for n in self.nodes if n.function is not None]
+
+    def stale(self) -> List[FunctionRef]:
+        """Functions whose as-computed version is behind the current registry
+        version -- the columns they produced are silently out of date."""
+        return [f for f in self.functions() if f.stale_vs_current]
+
+    def to_dict(self) -> dict:
+        def prune(d: dict) -> dict:
+            return {k: v for k, v in d.items() if v is not None}
+
+        return {
+            "target": self.target,
+            "nodes": [
+                prune(
+                    {
+                        "kind": n.kind,
+                        "id": n.id,
+                        "table": n.table,
+                        "function": prune(vars(n.function)) if n.function else None,
+                    }
+                )
+                for n in self.nodes
+            ],
+            "edges": [prune(vars(e)) for e in self.edges],
+        }
+
+    def to_graphviz(self) -> str:
+        """Graphviz DOT for the lineage DAG: columns/tables as nodes, function
+        names on edges, drift edges dashed + red."""
+        stale_names = {f.name for f in self.stale()}
+        out = [
+            "digraph lineage {",
+            "  rankdir=LR;",
+            '  node [fontname="monospace"];',
+        ]
+        for n in self.nodes:
+            if n.kind == "function":
+                continue
+            shape = "ellipse" if n.kind in ("table", "view") else "box"
+            out.append(f'  "{n.id}" [shape={shape}];')
+        for e in self.edges:
+            attrs = ""
+            if e.via:
+                if e.via in stale_names:
+                    attrs = f' [label="{e.via}" color=red style=dashed]'
+                else:
+                    attrs = f' [label="{e.via}"]'
+            out.append(f'  "{e.upstream}" -> "{e.downstream}"{attrs};')
+        out.append("}")
+        return "\n".join(out)
+
+    def _repr_html_(self) -> str:
+        warn = ""
+        drift = self.stale()
+        if drift:
+            names = ", ".join(sorted({f.name for f in drift}))
+            warn = (
+                f'<p style="color:#b00000"><b>stale vs current:</b> {names} '
+                "(re-refresh to catch up)</p>"
+            )
+        rows = "".join(
+            f"<tr><td><code>{e.downstream}</code></td>"
+            f"<td>&larr; {e.via or ''}</td>"
+            f"<td><code>{e.upstream}</code></td></tr>"
+            for e in self.edges
+        )
+        return (
+            f"<b>lineage: <code>{self.target}</code></b>{warn}"
+            "<table><tr><th>derived</th><th>via</th><th>from</th></tr>"
+            f"{rows}</table>"
+        )

python/python/lancedb/merge.py

@@ -5,7 +5,9 @@
 from __future__ import annotations
 
 from datetime import timedelta
-from typing import TYPE_CHECKING, List, Optional
+from typing import TYPE_CHECKING, List, Optional, Union
+
+from .expr import Expr
 
 if TYPE_CHECKING:
     from .common import DATA
@@ -32,8 +34,11 @@ class LanceMergeInsertBuilder(object):
         self._when_not_matched_insert_all = False
         self._when_not_matched_by_source_delete = False
         self._when_not_matched_by_source_condition = None
+        self._when_not_matched_by_source_condition_expr = None
         self._timeout = None
         self._use_index = True
+        self._use_lsm_write = None
+        self._validate_single_shard = None
 
     def when_matched_update_all(
         self, *, where: Optional[str] = None
@@ -60,7 +65,7 @@ class LanceMergeInsertBuilder(object):
         return self
 
     def when_not_matched_by_source_delete(
-        self, condition: Optional[str] = None
+        self, condition: Union[str, Expr, None] = None
     ) -> LanceMergeInsertBuilder:
         """
         Rows that exist only in the target table (old data) will be
@@ -69,13 +74,16 @@ class LanceMergeInsertBuilder(object):
 
         Parameters
         ----------
-        condition: Optional[str], default None
+        condition: str or :class:`~lancedb.expr.Expr` or None, default None
             If None then all such rows will be deleted.  Otherwise the
-            condition will be used as an SQL filter to limit what rows
-            are deleted.
+            condition will be used as a filter to limit what rows are deleted.
+            Can be a SQL string or a type-safe :class:`~lancedb.expr.Expr`
+            built with :func:`~lancedb.expr.col` and :func:`~lancedb.expr.lit`.
         """
         self._when_not_matched_by_source_delete = True
-        if condition is not None:
+        if isinstance(condition, Expr):
+            self._when_not_matched_by_source_condition_expr = condition._inner
+        elif condition is not None:
             self._when_not_matched_by_source_condition = condition
         return self
 
@@ -96,6 +104,46 @@ class LanceMergeInsertBuilder(object):
         self._use_index = use_index
         return self
 
+    def use_lsm_write(self, use_lsm_write: bool) -> LanceMergeInsertBuilder:
+        """
+        Controls whether the merge uses the MemWAL LSM write path.
+
+        By default (unset), a `merge_insert` on a table with an LSM write spec
+        is routed through Lance's MemWAL shard writer, and a table without one
+        uses the standard path. Pass `False` to force the standard path even
+        when a spec is set. Pass `True` to require a spec — `merge_insert`
+        raises an error if none is installed.
+
+        Parameters
+        ----------
+        use_lsm_write: bool
+            Whether to use the LSM write path.
+        """
+        self._use_lsm_write = use_lsm_write
+        return self
+
+    def validate_single_shard(
+        self, validate_single_shard: bool
+    ) -> LanceMergeInsertBuilder:
+        """
+        Controls how an LSM merge checks that its input targets a single shard.
+
+        When a table has an LSM write spec, every row in a `merge_insert` call
+        must route to the same shard. When `True` (the default), every row is
+        inspected to verify this. When `False`, only the first row is inspected
+        and the shard it routes to is used for the whole input — a faster path
+        for callers that have already pre-sharded their input.
+
+        Has no effect on tables without an LSM write spec.
+
+        Parameters
+        ----------
+        validate_single_shard: bool
+            Whether to check every row routes to one shard. Defaults to `True`.
+        """
+        self._validate_single_shard = validate_single_shard
+        return self
+
     def execute(
         self,
         new_data: DATA,

python/python/lancedb/namespace.py

@@ -58,6 +58,7 @@ from lance_namespace import (
     ListTablesRequest,
     DescribeNamespaceRequest,
     DropTableRequest,
+    RenameTableRequest,
     ListNamespacesRequest,
     CreateNamespaceRequest,
     DropNamespaceRequest,
@@ -70,6 +71,9 @@ from lancedb.embeddings import EmbeddingFunctionConfig
 from ._lancedb import Session
 
 
+_MAX_QUERY_K = 2**31 - 1
+
+
 def _query_to_namespace_request(
     table_id: List[str],
     query: "Query",
@@ -144,7 +148,13 @@ def _query_to_namespace_request(
     if query.postfilter is not None:
         prefilter = not query.postfilter
 
-    k = query.limit if query.limit is not None else 10
+    if query.limit is not None:
+        k = query.limit
+    elif query.vector is None and query.full_text_query is None:
+        # limit k to max i32 value to avoid client overflows
+        k = _MAX_QUERY_K
+    else:
+        k = 10
 
     # Build request kwargs, only including non-None values for optional fields
     # that Pydantic doesn't accept as None
@@ -363,6 +373,19 @@ def _convert_pyarrow_schema_to_json(schema: pa.Schema) -> JsonArrowSchema:
     return JsonArrowSchema(fields=fields, metadata=meta)
 
 
+def _builds_namespace_natively(
+    namespace_client_impl: Optional[str],
+    namespace_client_properties: Optional[Dict[str, str]],
+) -> bool:
+    """Whether ``connect_namespace_client`` builds the namespace client natively
+    in Rust (installing the read-freshness context provider) rather than wrapping
+    the pre-built Python client.
+
+    Must mirror Rust ``build_namespace_natively`` in ``python/src/connection.rs``.
+    """
+    return namespace_client_impl == "rest" and bool(namespace_client_properties)
+
+
 class LanceNamespaceDBConnection(DBConnection):
     """
     A LanceDB connection that uses a namespace for table management.
@@ -422,6 +445,13 @@ class LanceNamespaceDBConnection(DBConnection):
         )
         self._namespace_client_impl = namespace_client_impl
         self._namespace_client_properties = namespace_client_properties
+        # When the namespace client is built natively (see Rust
+        # ``build_namespace_natively``), the underlying Rust table performs
+        # QueryTable pushdown through the read-freshness context provider, which
+        # the pure-Python ``query_table`` path bypasses.
+        self._route_pushdown_to_rust = _builds_namespace_natively(
+            namespace_client_impl, namespace_client_properties
+        )
         self._inner = AsyncConnection(
             _connect_namespace_client(
                 namespace_client,
@@ -533,6 +563,7 @@ class LanceNamespaceDBConnection(DBConnection):
             namespace_path=namespace_path,
             namespace_client=self._namespace_client,
             pushdown_operations=self._namespace_client_pushdown_operations,
+            route_pushdown_to_rust=self._route_pushdown_to_rust,
             _async=async_table,
         )
 
@@ -544,6 +575,8 @@ class LanceNamespaceDBConnection(DBConnection):
         namespace_path: Optional[List[str]] = None,
         storage_options: Optional[Dict[str, str]] = None,
         index_cache_size: Optional[int] = None,
+        branch: Optional[str] = None,
+        version: Optional[int] = None,
     ) -> Table:
         if namespace_path is None:
             namespace_path = []
@@ -562,14 +595,20 @@ class LanceNamespaceDBConnection(DBConnection):
                 raise TableNotFoundError(f"Table not found: {'$'.join(table_id)}")
             raise
 
-        return LanceTable(
+        tbl = LanceTable(
             self,
             name,
             namespace_path=namespace_path,
             namespace_client=self._namespace_client,
             pushdown_operations=self._namespace_client_pushdown_operations,
+            route_pushdown_to_rust=self._route_pushdown_to_rust,
             _async=async_table,
         )
+        if branch is not None:
+            tbl = tbl.branches.checkout(branch, version)
+        elif version is not None:
+            tbl.checkout(version)
+        return tbl
 
     @override
     def drop_table(self, name: str, namespace_path: Optional[List[str]] = None):
@@ -592,9 +631,14 @@ class LanceNamespaceDBConnection(DBConnection):
             cur_namespace_path = []
         if new_namespace_path is None:
             new_namespace_path = []
-        raise NotImplementedError(
-            "rename_table is not supported for namespace connections"
+        cur_table_id = cur_namespace_path + [cur_name]
+        new_namespace_id = new_namespace_path if new_namespace_path else None
+        request = RenameTableRequest(
+            id=cur_table_id,
+            new_table_name=new_name,
+            new_namespace_id=new_namespace_id,
         )
+        self._namespace_client.rename_table(request)
 
     @override
     def drop_database(self):
@@ -853,6 +897,8 @@ class AsyncLanceNamespaceDBConnection:
         storage_options: Optional[Dict[str, str]] = None,
         session: Optional[Session] = None,
         namespace_client_pushdown_operations: Optional[List[str]] = None,
+        namespace_client_impl: Optional[str] = None,
+        namespace_client_properties: Optional[Dict[str, str]] = None,
     ):
         """
         Initialize an async namespace-based LanceDB connection.
@@ -878,6 +924,12 @@ class AsyncLanceNamespaceDBConnection:
               namespace.create_table() instead of using declare_table + local write.
 
             Default is None (no pushdown, all operations run locally).
+        namespace_client_impl : Optional[str]
+            The namespace implementation name used to create this connection.
+            Required (with ``namespace_client_properties``) for the Rust client to
+            be built natively and install the read-freshness provider.
+        namespace_client_properties : Optional[Dict[str, str]]
+            The namespace properties used to create this connection.
         """
         self._namespace_client = namespace_client
         self.read_consistency_interval = read_consistency_interval
@@ -886,6 +938,14 @@ class AsyncLanceNamespaceDBConnection:
         self._namespace_client_pushdown_operations = set(
             namespace_client_pushdown_operations or []
         )
+        self._namespace_client_impl = namespace_client_impl
+        self._namespace_client_properties = namespace_client_properties
+        # See LanceNamespaceDBConnection: when built natively the Rust table runs
+        # QueryTable pushdown through the read-freshness provider, so defer to it
+        # rather than the urllib3 client (which omits x-lancedb-min-timestamp).
+        self._route_pushdown_to_rust = _builds_namespace_natively(
+            namespace_client_impl, namespace_client_properties
+        )
         self._inner = AsyncConnection(
             _connect_namespace_client(
                 namespace_client,
@@ -899,8 +959,8 @@ class AsyncLanceNamespaceDBConnection:
                 namespace_client_pushdown_operations=(
                     list(self._namespace_client_pushdown_operations)
                 ),
-                namespace_client_impl=None,
-                namespace_client_properties=None,
+                namespace_client_impl=namespace_client_impl,
+                namespace_client_properties=namespace_client_properties,
             )
         )
 
@@ -954,7 +1014,7 @@ class AsyncLanceNamespaceDBConnection:
         if mode.lower() not in ["create", "overwrite"]:
             raise ValueError("mode must be either 'create' or 'overwrite'")
         validate_table_name(name)
-        return await self._inner.create_table(
+        table = await self._inner.create_table(
             name,
             data,
             schema=schema,
@@ -966,6 +1026,12 @@ class AsyncLanceNamespaceDBConnection:
             embedding_functions=embedding_functions,
             storage_options=storage_options,
         )
+        return table._set_namespace_context(
+            namespace_path=namespace_path,
+            namespace_client=self._namespace_client,
+            pushdown_operations=self._namespace_client_pushdown_operations,
+            route_pushdown_to_rust=self._route_pushdown_to_rust,
+        )
 
     async def open_table(
         self,
@@ -974,12 +1040,14 @@ class AsyncLanceNamespaceDBConnection:
         namespace_path: Optional[List[str]] = None,
         storage_options: Optional[Dict[str, str]] = None,
         index_cache_size: Optional[int] = None,
+        branch: Optional[str] = None,
+        version: Optional[int] = None,
     ) -> AsyncTable:
         """Open an existing table from the namespace."""
         if namespace_path is None:
             namespace_path = []
         try:
-            return await self._inner.open_table(
+            table = await self._inner.open_table(
                 name,
                 namespace_path=namespace_path,
                 storage_options=storage_options,
@@ -990,6 +1058,18 @@ class AsyncLanceNamespaceDBConnection:
                 table_id = namespace_path + [name]
                 raise TableNotFoundError(f"Table not found: {'$'.join(table_id)}")
             raise
+        # "main" is the default branch, so treat it as no branch (mirrors the
+        # sync remote path); the version still applies.
+        if branch is not None and branch != "main":
+            table = await table.branches.checkout(branch, version)
+        elif version is not None:
+            await table.checkout(version)
+        return table._set_namespace_context(
+            namespace_path=namespace_path,
+            namespace_client=self._namespace_client,
+            pushdown_operations=self._namespace_client_pushdown_operations,
+            route_pushdown_to_rust=self._route_pushdown_to_rust,
+        )
 
     async def drop_table(self, name: str, namespace_path: Optional[List[str]] = None):
         """Drop a table from the namespace."""
@@ -1006,14 +1086,19 @@ class AsyncLanceNamespaceDBConnection:
         cur_namespace_path: Optional[List[str]] = None,
         new_namespace_path: Optional[List[str]] = None,
     ):
-        """Rename is not supported for namespace connections."""
+        """Rename a table in the namespace."""
         if cur_namespace_path is None:
             cur_namespace_path = []
         if new_namespace_path is None:
             new_namespace_path = []
-        raise NotImplementedError(
-            "rename_table is not supported for namespace connections"
+        cur_table_id = cur_namespace_path + [cur_name]
+        new_namespace_id = new_namespace_path if new_namespace_path else None
+        request = RenameTableRequest(
+            id=cur_table_id,
+            new_table_name=new_name,
+            new_namespace_id=new_namespace_id,
         )
+        self._namespace_client.rename_table(request)
 
     async def drop_database(self):
         """Deprecated method."""
@@ -1342,4 +1427,6 @@ def connect_namespace_async(
         storage_options=storage_options,
         session=session,
         namespace_client_pushdown_operations=namespace_client_pushdown_operations,
+        namespace_client_impl=namespace_client_impl,
+        namespace_client_properties=namespace_client_properties,
     )

python/python/lancedb/permutation.py

@@ -3,12 +3,13 @@
 
 import copy
 import json
+import os
 
 from deprecation import deprecated
 import pyarrow as pa
 
 from ._lancedb import async_permutation_builder, PermutationReader
-from .table import LanceTable
+from .table import LanceTable, Table
 from .background_loop import LOOP
 from .util import batch_to_tensor, batch_to_tensor_rows
 from typing import Any, Callable, Iterator, Literal, Optional, TYPE_CHECKING, Union
@@ -47,6 +48,14 @@ class PermutationBuilder:
         By default, the permutation builder will create a single split that contains all
         rows in the same order as the base table.
         """
+        if not hasattr(table, "_inner"):
+            raise TypeError(
+                f"PermutationBuilder requires a local LanceTable, "
+                f"got {type(table).__name__}. "
+                "The permutation API is not supported on remote tables. "
+                "Remote tables connect to LanceDB Cloud or Enterprise and do not have "
+                "direct access to the underlying Lance dataset needed for permutations."
+            )
         self._async = async_permutation_builder(table)
 
     def split_random(
@@ -354,6 +363,49 @@ class Transforms:
 DEFAULT_BATCH_SIZE = 100
 
 
+def _table_to_pickle_state(table: Table) -> dict[str, Any]:
+    from .remote.table import RemoteTable
+
+    if isinstance(table, RemoteTable):
+        return {
+            "kind": "remote",
+            "table": table,
+        }
+
+    if not isinstance(table, LanceTable):
+        raise ValueError(f"Cannot pickle table of type {type(table)!r}")
+
+    base_uri = table._conn.uri
+    if base_uri.startswith("memory://"):
+        return {
+            "kind": "memory",
+            "name": table.name,
+            "data": table.to_arrow(),
+        }
+
+    return {
+        "kind": "local",
+        "name": table.name,
+        "uri": base_uri,
+        "namespace": table._namespace_path,
+        "storage_options": table._conn.storage_options,
+    }
+
+
+def _table_from_pickle_state(state: dict[str, Any]) -> Table:
+    from . import connect
+
+    kind = state["kind"]
+    if kind == "remote":
+        return state["table"]
+    if kind == "memory":
+        return connect("memory://").create_table(state["name"], state["data"])
+    if kind == "local":
+        db = connect(state["uri"], storage_options=state["storage_options"])
+        return db.open_table(state["name"], namespace_path=state["namespace"] or None)
+    raise ValueError(f"Unknown table pickle state kind: {kind}")
+
+
 class Permutation:
     """
     A Permutation is a view of a dataset that can be used as input to model training
@@ -369,15 +421,15 @@ class Permutation:
 
     def __init__(
         self,
-        base_table: LanceTable,
-        permutation_table: Optional[LanceTable],
+        base_table: Table,
+        permutation_table: Optional[Table],
         split: int,
         selection: dict[str, str],
         batch_size: int,
         transform_fn: Callable[pa.RecordBatch, Any],
         offset: Optional[int] = None,
         limit: Optional[int] = None,
-        connection_factory: Optional[Callable[[str], LanceTable]] = None,
+        connection_factory: Optional[Callable[[str], Table]] = None,
         _reader: Optional[PermutationReader] = None,
     ):
         """
@@ -397,6 +449,7 @@ class Permutation:
         if _reader is None:
             _reader = LOOP.run(self._build_reader())
         self.reader: PermutationReader = _reader
+        self._pid = os.getpid()
 
     async def _build_reader(self) -> PermutationReader:
         reader = await PermutationReader.from_tables(
@@ -428,29 +481,25 @@ class Permutation:
         return new
 
     def with_connection_factory(
-        self, connection_factory: Callable[[str], LanceTable]
+        self, connection_factory: Callable[[str], Table]
     ) -> "Permutation":
         """
         Creates a new permutation that will use ``connection_factory`` to reopen
         the base table when this permutation is unpickled in a worker process.
 
-        The factory is a callable that takes a single argument — the base table
-        name — and returns a [LanceTable]. It must be picklable; the worker
+        The factory is a callable that takes a single argument, the base table
+        name, and returns a LanceDB table. It must be picklable; the worker
         will pickle it via standard ``pickle`` and call it to recover the base
         table. Picklable callables in practice means top-level (module-level)
         functions, ``functools.partial`` of such functions, or instances of
         picklable classes implementing ``__call__``. Lambdas and closures over
         local variables don't pickle with the default protocol.
 
-        Setting a factory is necessary when the URI alone is not enough to
-        re-open the connection — most importantly for LanceDB Cloud (``db://``)
-        connections, where ``api_key`` and ``region`` aren't recoverable from
-        the connection object after construction.
-
-        For local file or cloud-storage paths the factory is optional: if not
-        set, ``__getstate__`` falls back to capturing
-        ``(uri, storage_options, namespace_path)`` and re-opening via
-        ``lancedb.connect(uri, storage_options=...)``.
+        A factory is optional for normal local and remote LanceDB connections:
+        if not set, ``__getstate__`` captures the table's own picklable reopen
+        state. Use a factory when that default state is not enough, for example
+        when credentials should be loaded from the worker environment instead
+        of being embedded in the pickle.
 
         Examples
         --------
@@ -508,7 +557,7 @@ class Permutation:
         return new
 
     @classmethod
-    def identity(cls, table: LanceTable) -> "Permutation":
+    def identity(cls, table: Table) -> "Permutation":
         """
         Creates an identity permutation for the given table.
         """
@@ -517,8 +566,8 @@ class Permutation:
     @classmethod
     def from_tables(
         cls,
-        base_table: LanceTable,
-        permutation_table: Optional[LanceTable] = None,
+        base_table: Table,
+        permutation_table: Optional[Table] = None,
         split: Optional[Union[str, int]] = None,
     ) -> "Permutation":
         """
@@ -594,11 +643,10 @@ class Permutation:
 
         The base table is captured either via a user-supplied
         ``connection_factory`` (see [with_connection_factory]) or, as a
-        fallback, by introspecting ``(uri, storage_options, namespace_path)``
-        on the connection. The permutation table — always an in-memory
-        LanceDB table — is captured as a pyarrow Table (which pickles via
-        Arrow IPC natively). The reader is dropped from the wire format;
-        ``__setstate__`` rebuilds it from the restored tables.
+        fallback, by the table's own picklable reopen state. The permutation
+        table is captured as a pyarrow Table (which pickles via Arrow IPC
+        natively). The reader is dropped from the wire format and rebuilt
+        lazily on first use.
         """
         permutation_data: Optional[pa.Table] = None
         if self.permutation_table is not None:
@@ -622,39 +670,9 @@ class Permutation:
             # namespace from the existing connection.
             return common
 
-        # URI-introspection fallback: only viable for native (OSS) connections
-        # where (uri, storage_options) is enough to reopen. Remote / cloud
-        # connections don't expose recoverable api_key / region — those users
-        # must call with_connection_factory().
-        try:
-            base_uri = self.base_table._conn.uri
-            storage_options = self.base_table._conn.storage_options
-        except AttributeError as e:
-            raise ValueError(
-                "Cannot pickle this Permutation: the base table's connection "
-                "does not expose a uri/storage_options, which usually means it "
-                "is a remote (LanceDB Cloud) connection. Call "
-                "Permutation.with_connection_factory(...) first to provide a "
-                "picklable callable that re-opens the base table from a worker "
-                "process."
-            ) from e
-
-        if base_uri.startswith("memory://"):
-            # In-memory base tables don't exist in any worker process by
-            # default, so dump the entire base table into the pickle. This
-            # can be expensive for large datasets — users with large
-            # in-memory base tables should either persist them or set a
-            # connection_factory.
-            return {
-                **common,
-                "base_table_data": self.base_table.to_arrow(),
-            }
-
         return {
             **common,
-            "base_table_uri": base_uri,
-            "base_table_namespace": self.base_table._namespace_path,
-            "base_table_storage_options": storage_options,
+            "base_table_state": _table_to_pickle_state(self.base_table),
         }
 
     def __setstate__(self, state: dict[str, Any]) -> None:
@@ -663,6 +681,8 @@ class Permutation:
         connection_factory = state["connection_factory"]
         if connection_factory is not None:
             base_table = connection_factory(state["base_table_name"])
+        elif "base_table_state" in state:
+            base_table = _table_from_pickle_state(state["base_table_state"])
         elif "base_table_data" in state:
             # In-memory base table inlined into the pickle; rebuild the same
             # way we rebuild the in-memory permutation table.
@@ -680,7 +700,7 @@ class Permutation:
                 namespace_path=state["base_table_namespace"] or None,
             )
 
-        permutation_table: Optional[LanceTable] = None
+        permutation_table: Optional[Table] = None
         if state["permutation_data"] is not None:
             mem_db = connect("memory://")
             permutation_table = mem_db.create_table(
@@ -696,10 +716,28 @@ class Permutation:
         self.offset = state["offset"]
         self.limit = state["limit"]
         self.connection_factory = connection_factory
+        self.reader = None
+        self._pid = None
+
+    def _ensure_open(self) -> None:
+        pid = os.getpid()
+        if self.reader is not None and getattr(self, "_pid", None) == pid:
+            return
+        # The reader owns Rust-side table handles. Rebuild it after unpickle or
+        # fork even though the Python table wrappers reopen themselves.
+        if hasattr(self.base_table, "_ensure_open"):
+            self.base_table._ensure_open()
+        if self.permutation_table is not None and hasattr(
+            self.permutation_table, "_ensure_open"
+        ):
+            self.permutation_table._ensure_open()
         self.reader = LOOP.run(self._build_reader())
+        self._pid = pid
 
     @property
     def schema(self) -> pa.Schema:
+        self._ensure_open()
+
         async def do_output_schema():
             return await self.reader.output_schema(self.selection)
 
@@ -717,6 +755,7 @@ class Permutation:
         """
         The number of rows in the permutation
         """
+        self._ensure_open()
         return self.reader.count_rows()
 
     @property
@@ -875,6 +914,7 @@ class Permutation:
         If skip_last_batch is True, the last batch will be skipped if it is not a
         multiple of batch_size.
         """
+        self._ensure_open()
 
         async def get_iter():
             return await self.reader.read(self.selection, batch_size=batch_size)
@@ -976,6 +1016,7 @@ class Permutation:
         so `with_format` and `with_transform` affect this method in the same way
         they affect iteration.
         """
+        self._ensure_open()
 
         async def do_take_offsets():
             return await self.reader.take_offsets(offsets, selection=self.selection)
@@ -1011,9 +1052,11 @@ class Permutation:
         """
         Skip the first `skip` rows of the permutation
         """
+        self._ensure_open()
         new = copy.copy(self)
         new.offset = skip
         new.reader = LOOP.run(new._build_reader())
+        new._pid = os.getpid()
         return new
 
     @deprecated(details="Use with_take instead")
@@ -1032,9 +1075,11 @@ class Permutation:
         """
         Limit the permutation to `limit` rows (following any `skip`)
         """
+        self._ensure_open()
         new = copy.copy(self)
         new.limit = limit
         new.reader = LOOP.run(new._build_reader())
+        new._pid = os.getpid()
         return new
 
     @deprecated(details="Use with_repeat instead")

python/python/lancedb/pydantic.py

@@ -275,7 +275,18 @@ def _py_type_to_arrow_type(py_type: Type[Any], field: FieldInfo) -> pa.DataType:
         tz = get_extras(field, "tz")
         return pa.timestamp("us", tz=tz)
     elif getattr(py_type, "__origin__", None) in (list, tuple):
-        child = py_type.__args__[0]
+        # A bare, unparameterised ``typing.List`` / ``typing.Tuple`` matches this
+        # branch (its ``__origin__`` is ``list`` / ``tuple``) but has no
+        # ``__args__``, so we cannot infer the element type. Raise a clear
+        # ``TypeError`` instead of crashing with an opaque ``AttributeError``.
+        args = getattr(py_type, "__args__", None)
+        if not args:
+            raise TypeError(
+                "Converting Pydantic type to Arrow Type: unsupported type "
+                f"{py_type}. Specify the element type, e.g. List[int] instead "
+                "of a bare List."
+            )
+        child = args[0]
         return _pydantic_list_child_to_arrow(child, field)
     raise TypeError(
         f"Converting Pydantic type to Arrow Type: unsupported type {py_type}."

python/python/lancedb/query.py

@@ -41,6 +41,14 @@ from .rerankers.rrf import RRFReranker
 from .rerankers.util import check_reranker_result
 from .util import flatten_columns
 
+BlobMode = Literal["lazy", "bytes", "descriptions"]
+
+_BLOB_MODE_TO_HANDLING = {
+    "lazy": "blobs_descriptions",
+    "bytes": "all_binary",
+    "descriptions": "blobs_descriptions",
+}
+
 if TYPE_CHECKING:
     import sys
 
@@ -55,7 +63,7 @@ if TYPE_CHECKING:
     from ._lancedb import VectorQuery as LanceVectorQuery
     from .common import VEC
     from .pydantic import LanceModel
-    from .table import Table
+    from .table import AsyncTable, Table
 
     if sys.version_info >= (3, 11):
         from typing import Self
@@ -65,6 +73,179 @@ if TYPE_CHECKING:
 T = TypeVar("T", bound="LanceModel")
 
 
+def _validate_blob_mode(blob_mode: BlobMode) -> None:
+    if blob_mode not in _BLOB_MODE_TO_HANDLING:
+        modes = ", ".join(repr(mode) for mode in _BLOB_MODE_TO_HANDLING)
+        raise ValueError(f"blob_mode must be one of {modes}, got {blob_mode!r}")
+
+
+def _field_is_blob(field: pa.Field) -> bool:
+    metadata = field.metadata or {}
+    return metadata.get(b"lance-encoding:blob") == b"true" or (
+        metadata.get("lance-encoding:blob") == "true"
+    )
+
+
+def _schema_has_blob_field(schema: pa.Schema) -> bool:
+    return any(_field_is_blob(field) for field in schema)
+
+
+def _blob_mode_requires_native_pandas(blob_mode: BlobMode, schema: pa.Schema) -> bool:
+    return blob_mode in _BLOB_MODE_TO_HANDLING and _schema_has_blob_field(schema)
+
+
+def _unsupported_blob_pandas_error(reason: str) -> RuntimeError:
+    return RuntimeError(
+        "blob columns require Lance native scanner conversion for query "
+        f"to_pandas(), but {reason}. Use a plain scan query or remove blob "
+        "columns from the projection."
+    )
+
+
+def _query_is_plain_scan(query: Query) -> bool:
+    return (
+        query.vector is None
+        and query.full_text_query is None
+        and not query.postfilter
+        and not query.order_by
+    )
+
+
+def _filter_to_sql(filter: Optional[Union[str, Expr]]) -> Optional[str]:
+    if filter is None:
+        return None
+    if isinstance(filter, Expr):
+        return filter.to_sql()
+    return filter
+
+
+def _projection_to_scanner_kwargs(
+    columns: Optional[
+        Union[
+            List[str], List[Tuple[str, Union[str, Expr]]], Dict[str, Union[str, Expr]]
+        ]
+    ],
+) -> Dict[str, Any]:
+    if columns is None:
+        return {}
+    if isinstance(columns, list):
+        if all(isinstance(column, str) for column in columns):
+            return {"columns": columns}
+        if all(isinstance(column, tuple) and len(column) == 2 for column in columns):
+            return {
+                "columns": {
+                    name: expr.to_sql() if isinstance(expr, Expr) else expr
+                    for name, expr in columns
+                }
+            }
+        # Let Lance raise the detailed projection validation error.
+        return {"columns": columns}
+
+    projection = {}
+    for name, expr in columns.items():
+        if isinstance(expr, Expr):
+            expr = expr.to_sql()
+        projection[name] = expr
+    return {"columns": projection}
+
+
+def _scanner_kwargs_for_query(
+    query: Query, blob_mode: BlobMode, dataset: Optional[Any] = None
+) -> Dict[str, Any]:
+    fragments = _scanner_fragments_for_query(query, dataset)
+    kwargs = {
+        **_projection_to_scanner_kwargs(query.columns),
+        "filter": _filter_to_sql(query.filter),
+        "limit": query.limit,
+        "offset": query.offset,
+        "with_row_id": query.with_row_id,
+        "with_row_address": query.with_row_address,
+        "fast_search": query.fast_search,
+        "blob_handling": _BLOB_MODE_TO_HANDLING[blob_mode],
+        "fragments": fragments,
+    }
+    return {key: value for key, value in kwargs.items() if value is not None}
+
+
+def _scanner_fragments_for_query(query: Query, dataset: Optional[Any]) -> Optional[Any]:
+    if query.fragments is not None and query.fragment_ids is not None:
+        raise ValueError("fragments and fragment_ids cannot both be set")
+    if query.fragments is not None:
+        return query.fragments
+    if query.fragment_ids is None:
+        return None
+    if dataset is None:
+        raise ValueError("fragment_ids require a Lance dataset")
+
+    requested = set(query.fragment_ids)
+    fragments = [
+        fragment
+        for fragment in dataset.get_fragments()
+        if fragment.fragment_id in requested
+    ]
+    found = {fragment.fragment_id for fragment in fragments}
+    missing = requested - found
+    if missing:
+        missing_ids = ", ".join(str(fragment_id) for fragment_id in sorted(missing))
+        raise ValueError(f"fragment_ids not found in dataset: {missing_ids}")
+    return fragments
+
+
+def _ensure_lazy_blob_frame(
+    df: "pd.DataFrame", schema: pa.Schema, blob_mode: BlobMode
+) -> "pd.DataFrame":
+    if blob_mode != "lazy" or not _schema_has_blob_field(schema) or len(df) == 0:
+        return df
+
+    for field in schema:
+        if not _field_is_blob(field) or field.name not in df.columns:
+            continue
+        value = df[field.name].iloc[0]
+        if value is not None and not hasattr(value, "readall"):
+            raise _unsupported_blob_pandas_error(
+                "the Lance scanner did not return lazy blob files"
+            )
+    return df
+
+
+def _scanner_to_table(scanner: Any) -> pa.Table:
+    if hasattr(scanner, "to_pyarrow"):
+        reader = scanner.to_pyarrow()
+        return reader.read_all()
+    if hasattr(scanner, "to_table"):
+        return scanner.to_table()
+    reader = scanner.to_reader()
+    return reader.read_all()
+
+
+def _scanner_to_pandas(scanner: Any, blob_mode: BlobMode, **kwargs) -> "pd.DataFrame":
+    schema = getattr(scanner, "projected_schema", None)
+    if schema is None:
+        schema = getattr(scanner, "schema", None)
+    if schema is None:
+        schema = getattr(scanner, "dataset_schema", None)
+    if callable(schema):
+        schema = schema()
+    if hasattr(scanner, "to_pandas"):
+        try:
+            df = scanner.to_pandas(blob_mode=blob_mode, **kwargs)
+        except TypeError as err:
+            message = str(err)
+            if "blob_mode" not in message and "unexpected keyword" not in message:
+                raise
+            df = scanner.to_pandas(**kwargs)
+        if schema is not None:
+            return _ensure_lazy_blob_frame(df, schema, blob_mode)
+        return df
+
+    tbl = _scanner_to_table(scanner)
+    if blob_mode == "lazy" and _schema_has_blob_field(tbl.schema):
+        raise _unsupported_blob_pandas_error(
+            "the Lance scanner does not expose to_pandas"
+        )
+    return tbl.to_pandas(**kwargs)
+
+
 # Pydantic validation function for vector queries
 def ensure_vector_query(
     val: Any,
@@ -499,6 +680,13 @@ class Query(pydantic.BaseModel):
     # if true, include the row id in the results
     with_row_id: Optional[bool] = None
 
+    # if true, include the row address in the results
+    with_row_address: Optional[bool] = None
+
+    # Lance fragments or fragment ids to scan on scanner-backed plain queries
+    fragments: Optional[Any] = None
+    fragment_ids: Optional[List[int]] = None
+
     # offset to start fetching results from
     offset: Optional[int] = None
 
@@ -691,6 +879,9 @@ class LanceQueryBuilder(ABC):
         self._where = None
         self._postfilter = None
         self._with_row_id = None
+        self._with_row_address = None
+        self._fragments = None
+        self._fragment_ids = None
         self._vector = None
         self._text = None
         self._ef = None
@@ -718,6 +909,7 @@ class LanceQueryBuilder(ABC):
         self,
         flatten: Optional[Union[int, bool]] = None,
         *,
+        blob_mode: BlobMode = "lazy",
         timeout: Optional[timedelta] = None,
         **kwargs,
     ) -> "pd.DataFrame":
@@ -737,11 +929,41 @@ class LanceQueryBuilder(ABC):
         timeout: Optional[timedelta]
             The maximum time to wait for the query to complete.
             If None, wait indefinitely.
+        blob_mode: str, default "lazy"
+            Controls how blob columns are returned for plain scan queries.
+            Vector, FTS, hybrid, and other non-native query shapes keep the
+            existing Arrow conversion path and only support blob descriptions.
         **kwargs
             Forwarded to pyarrow.Table.to_pandas after query execution and
             optional flattening.
         """
+        _validate_blob_mode(blob_mode)
+        output_schema = getattr(self, "output_schema", None)
+        if output_schema is not None:
+            schema = output_schema()
+            if _blob_mode_requires_native_pandas(blob_mode, schema):
+                native_error = None
+                if (flatten is None or blob_mode == "descriptions") and timeout is None:
+                    try:
+                        df = self._plain_scan_to_pandas(
+                            blob_mode, flatten=flatten, **kwargs
+                        )
+                        if df is not None:
+                            return df
+                    except Exception as err:
+                        native_error = err
+                reason = (
+                    "this query shape cannot use Lance native pandas conversion"
+                    if native_error is None
+                    else str(native_error)
+                )
+                raise _unsupported_blob_pandas_error(reason) from native_error
+
         tbl = flatten_columns(self.to_arrow(timeout=timeout), flatten)
+        if _blob_mode_requires_native_pandas(blob_mode, tbl.schema):
+            raise _unsupported_blob_pandas_error(
+                "this query shape cannot use Lance native pandas conversion"
+            )
         return tbl.to_pandas(**kwargs)
 
     @abstractmethod
@@ -947,6 +1169,32 @@ class LanceQueryBuilder(ABC):
         self._with_row_id = with_row_id
         return self
 
+    def with_row_address(self, with_row_address: bool = True) -> Self:
+        """Set whether to return row addresses.
+
+        Parameters
+        ----------
+        with_row_address: bool, default True
+            If True, return the _rowaddr column in the results.
+
+        Returns
+        -------
+        LanceQueryBuilder
+            The LanceQueryBuilder object.
+        """
+        self._with_row_address = with_row_address
+        return self
+
+    def with_fragments(self, fragments: Any) -> Self:
+        """Set the Lance fragments to scan for plain scanner-backed queries."""
+        self._fragments = fragments
+        return self
+
+    def fragment_ids(self, fragment_ids: List[int]) -> Self:
+        """Set the Lance fragment ids to scan for plain scanner-backed queries."""
+        self._fragment_ids = fragment_ids
+        return self
+
     def explain_plan(self, verbose: Optional[bool] = False) -> str:
         """Return the execution plan for this query.
 
@@ -1086,6 +1334,25 @@ class LanceQueryBuilder(ABC):
         """
         raise NotImplementedError
 
+    def _plain_scan_to_pandas(
+        self,
+        blob_mode: BlobMode,
+        flatten: Optional[Union[int, bool]] = None,
+        **kwargs,
+    ) -> Optional["pd.DataFrame"]:
+        query = self.to_query_object()
+        if not _query_is_plain_scan(query):
+            return None
+
+        dataset = self._table.to_lance()
+        scanner = dataset.scanner(
+            **_scanner_kwargs_for_query(query, blob_mode, dataset)
+        )
+        if flatten is not None:
+            tbl = flatten_columns(_scanner_to_table(scanner), flatten)
+            return tbl.to_pandas(**kwargs)
+        return _scanner_to_pandas(scanner, blob_mode, **kwargs)
+
     @abstractmethod
     def to_query_object(self) -> Query:
         """Return a serializable representation of the query
@@ -1357,6 +1624,9 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
             refine_factor=self._refine_factor,
             vector_column=self._vector_column,
             with_row_id=self._with_row_id,
+            with_row_address=self._with_row_address,
+            fragments=self._fragments,
+            fragment_ids=self._fragment_ids,
             offset=self._offset,
             fast_search=self._fast_search,
             ef=self._ef,
@@ -1559,6 +1829,9 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
             limit=self._limit,
             postfilter=self._postfilter,
             with_row_id=self._with_row_id,
+            with_row_address=self._with_row_address,
+            fragments=self._fragments,
+            fragment_ids=self._fragment_ids,
             full_text_query=FullTextSearchQuery(
                 query=self._query, columns=self._fts_columns
             ),
@@ -1629,6 +1902,9 @@ class LanceEmptyQueryBuilder(LanceQueryBuilder):
             filter=self._where,
             limit=self._limit,
             with_row_id=self._with_row_id,
+            with_row_address=self._with_row_address,
+            fragments=self._fragments,
+            fragment_ids=self._fragment_ids,
             offset=self._offset,
             order_by=self._order_by,
         )
@@ -2207,7 +2483,11 @@ class AsyncQueryBase(object):
     Base class for all async queries (take, scan, vector, fts, hybrid)
     """
 
-    def __init__(self, inner: Union[LanceQuery, LanceVectorQuery, LanceTakeQuery]):
+    def __init__(
+        self,
+        inner: Union[LanceQuery, LanceVectorQuery, LanceTakeQuery],
+        table: Optional["AsyncTable"] = None,
+    ):
         """
         Construct an AsyncQueryBase
 
@@ -2215,6 +2495,10 @@ class AsyncQueryBase(object):
         [AsyncTable.query][lancedb.table.AsyncTable.query] method to create a query.
         """
         self._inner = inner
+        self._table = table
+        self._with_row_address = None
+        self._fragments = None
+        self._fragment_ids = None
 
     def to_query_object(self) -> Query:
         """
@@ -2223,7 +2507,11 @@ class AsyncQueryBase(object):
         This is currently experimental but can be useful as the query object is pure
         python and more easily serializable.
         """
-        return Query.from_inner(self._inner.to_query_request())
+        query = Query.from_inner(self._inner.to_query_request())
+        query.with_row_address = self._with_row_address
+        query.fragments = self._fragments
+        query.fragment_ids = self._fragment_ids
+        return query
 
     def select(self, columns: Union[List[str], dict[str, str]]) -> Self:
         """
@@ -2280,6 +2568,27 @@ class AsyncQueryBase(object):
         self._inner.with_row_id()
         return self
 
+    def with_row_address(self, with_row_address: bool = True) -> Self:
+        """
+        Include the _rowaddr column in scanner-backed plain query results.
+        """
+        self._with_row_address = with_row_address
+        return self
+
+    def with_fragments(self, fragments: Any) -> Self:
+        """
+        Restrict scanner-backed plain query results to the given Lance fragments.
+        """
+        self._fragments = fragments
+        return self
+
+    def fragment_ids(self, fragment_ids: List[int]) -> Self:
+        """
+        Restrict scanner-backed plain query results to the given Lance fragment ids.
+        """
+        self._fragment_ids = fragment_ids
+        return self
+
     async def to_batches(
         self,
         *,
@@ -2357,6 +2666,8 @@ class AsyncQueryBase(object):
         self,
         flatten: Optional[Union[int, bool]] = None,
         timeout: Optional[timedelta] = None,
+        *,
+        blob_mode: BlobMode = "lazy",
         **kwargs,
     ) -> "pd.DataFrame":
         """
@@ -2390,13 +2701,63 @@ class AsyncQueryBase(object):
             The maximum time to wait for the query to complete.
             If not specified, no timeout is applied. If the query does not
             complete within the specified time, an error will be raised.
+        blob_mode: str, default "lazy"
+            Controls how blob columns are returned for plain scan queries.
+            Vector, FTS, hybrid, and other non-native query shapes keep the
+            existing Arrow conversion path and only support blob descriptions.
         **kwargs
             Forwarded to pyarrow.Table.to_pandas after query execution and
             optional flattening.
         """
-        return (
-            flatten_columns(await self.to_arrow(timeout=timeout), flatten)
-        ).to_pandas(**kwargs)
+        _validate_blob_mode(blob_mode)
+        if hasattr(self._inner, "output_schema"):
+            schema = await self.output_schema()
+            if _blob_mode_requires_native_pandas(blob_mode, schema):
+                native_error = None
+                if (flatten is None or blob_mode == "descriptions") and timeout is None:
+                    try:
+                        df = await self._plain_scan_to_pandas(
+                            blob_mode, flatten=flatten, **kwargs
+                        )
+                        if df is not None:
+                            return df
+                    except Exception as err:
+                        native_error = err
+                reason = (
+                    "this query shape cannot use Lance native pandas conversion"
+                    if native_error is None
+                    else str(native_error)
+                )
+                raise _unsupported_blob_pandas_error(reason) from native_error
+
+        tbl = flatten_columns(await self.to_arrow(timeout=timeout), flatten)
+        if _blob_mode_requires_native_pandas(blob_mode, tbl.schema):
+            raise _unsupported_blob_pandas_error(
+                "this query shape cannot use Lance native pandas conversion"
+            )
+        return tbl.to_pandas(**kwargs)
+
+    async def _plain_scan_to_pandas(
+        self,
+        blob_mode: BlobMode,
+        flatten: Optional[Union[int, bool]] = None,
+        **kwargs,
+    ) -> Optional["pd.DataFrame"]:
+        if self._table is None:
+            return None
+
+        query = self.to_query_object()
+        if not _query_is_plain_scan(query):
+            return None
+
+        dataset = await self._table._to_lance()
+        scanner = dataset.scanner(
+            **_scanner_kwargs_for_query(query, blob_mode, dataset)
+        )
+        if flatten is not None:
+            tbl = flatten_columns(_scanner_to_table(scanner), flatten)
+            return tbl.to_pandas(**kwargs)
+        return _scanner_to_pandas(scanner, blob_mode, **kwargs)
 
     async def to_polars(
         self,
@@ -2503,14 +2864,18 @@ class AsyncStandardQuery(AsyncQueryBase):
     Base class for "standard" async queries (all but take currently)
     """
 
-    def __init__(self, inner: Union[LanceQuery, LanceVectorQuery]):
+    def __init__(
+        self,
+        inner: Union[LanceQuery, LanceVectorQuery],
+        table: Optional["AsyncTable"] = None,
+    ):
         """
         Construct an AsyncStandardQuery
 
         This method is not intended to be called directly.  Instead, use the
         [AsyncTable.query][lancedb.table.AsyncTable.query] method to create a query.
         """
-        super().__init__(inner)
+        super().__init__(inner, table)
 
     def where(self, predicate: Union[str, Expr]) -> Self:
         """
@@ -2616,14 +2981,14 @@ class AsyncStandardQuery(AsyncQueryBase):
 
 
 class AsyncQuery(AsyncStandardQuery):
-    def __init__(self, inner: LanceQuery):
+    def __init__(self, inner: LanceQuery, table: Optional["AsyncTable"] = None):
         """
         Construct an AsyncQuery
 
         This method is not intended to be called directly.  Instead, use the
         [AsyncTable.query][lancedb.table.AsyncTable.query] method to create a query.
         """
-        super().__init__(inner)
+        super().__init__(inner, table)
         self._inner = inner
 
     @classmethod
@@ -2707,10 +3072,11 @@ class AsyncQuery(AsyncStandardQuery):
             new_self = self._inner.nearest_to(query_vectors[0])
             for v in query_vectors[1:]:
                 new_self.add_query_vector(v)
-            return AsyncVectorQuery(new_self)
+            return AsyncVectorQuery(new_self, self._table)
         else:
             return AsyncVectorQuery(
-                self._inner.nearest_to(AsyncQuery._query_vec_to_array(query_vector))
+                self._inner.nearest_to(AsyncQuery._query_vec_to_array(query_vector)),
+                self._table,
             )
 
     def nearest_to_text(
@@ -2743,17 +3109,18 @@ class AsyncQuery(AsyncStandardQuery):
 
         if isinstance(query, str):
             return AsyncFTSQuery(
-                self._inner.nearest_to_text({"query": query, "columns": columns})
+                self._inner.nearest_to_text({"query": query, "columns": columns}),
+                self._table,
             )
         # FullTextQuery object
-        return AsyncFTSQuery(self._inner.nearest_to_text({"query": query}))
+        return AsyncFTSQuery(self._inner.nearest_to_text({"query": query}), self._table)
 
 
 class AsyncFTSQuery(AsyncStandardQuery):
     """A query for full text search for LanceDB."""
 
-    def __init__(self, inner: LanceFTSQuery):
-        super().__init__(inner)
+    def __init__(self, inner: LanceFTSQuery, table: Optional["AsyncTable"] = None):
+        super().__init__(inner, table)
         self._inner = inner
         self._reranker = None
 
@@ -2835,10 +3202,11 @@ class AsyncFTSQuery(AsyncStandardQuery):
             new_self = self._inner.nearest_to(query_vectors[0])
             for v in query_vectors[1:]:
                 new_self.add_query_vector(v)
-            return AsyncHybridQuery(new_self)
+            return AsyncHybridQuery(new_self, self._table)
         else:
             return AsyncHybridQuery(
-                self._inner.nearest_to(AsyncQuery._query_vec_to_array(query_vector))
+                self._inner.nearest_to(AsyncQuery._query_vec_to_array(query_vector)),
+                self._table,
             )
 
     async def to_batches(
@@ -3029,7 +3397,7 @@ class AsyncVectorQueryBase:
 
 
 class AsyncVectorQuery(AsyncStandardQuery, AsyncVectorQueryBase):
-    def __init__(self, inner: LanceVectorQuery):
+    def __init__(self, inner: LanceVectorQuery, table: Optional["AsyncTable"] = None):
         """
         Construct an AsyncVectorQuery
 
@@ -3039,7 +3407,7 @@ class AsyncVectorQuery(AsyncStandardQuery, AsyncVectorQueryBase):
         a vector query.  Or you can use
         [AsyncTable.vector_search][lancedb.table.AsyncTable.vector_search]
         """
-        super().__init__(inner)
+        super().__init__(inner, table)
         self._inner = inner
         self._reranker = None
         self._query_string = None
@@ -3093,10 +3461,13 @@ class AsyncVectorQuery(AsyncStandardQuery, AsyncVectorQueryBase):
 
         if isinstance(query, str):
             return AsyncHybridQuery(
-                self._inner.nearest_to_text({"query": query, "columns": columns})
+                self._inner.nearest_to_text({"query": query, "columns": columns}),
+                self._table,
             )
         # FullTextQuery object
-        return AsyncHybridQuery(self._inner.nearest_to_text({"query": query}))
+        return AsyncHybridQuery(
+            self._inner.nearest_to_text({"query": query}), self._table
+        )
 
     async def to_batches(
         self,
@@ -3123,8 +3494,8 @@ class AsyncHybridQuery(AsyncStandardQuery, AsyncVectorQueryBase):
     in the `rerank` method to convert the scores to ranks and then normalize them.
     """
 
-    def __init__(self, inner: LanceHybridQuery):
-        super().__init__(inner)
+    def __init__(self, inner: LanceHybridQuery, table: Optional["AsyncTable"] = None):
+        super().__init__(inner, table)
         self._inner = inner
         self._norm = "score"
         self._reranker = RRFReranker()
@@ -3165,8 +3536,8 @@ class AsyncHybridQuery(AsyncStandardQuery, AsyncVectorQueryBase):
         max_batch_length: Optional[int] = None,
         timeout: Optional[timedelta] = None,
     ) -> AsyncRecordBatchReader:
-        fts_query = AsyncFTSQuery(self._inner.to_fts_query())
-        vec_query = AsyncVectorQuery(self._inner.to_vector_query())
+        fts_query = AsyncFTSQuery(self._inner.to_fts_query(), self._table)
+        vec_query = AsyncVectorQuery(self._inner.to_vector_query(), self._table)
 
         # save the row ID choice that was made on the query builder and force it
         # to actually fetch the row ids because we need this for reranking
@@ -3266,8 +3637,16 @@ class AsyncTakeQuery(AsyncQueryBase):
     Builder for parameterizing and executing take queries.
     """
 
-    def __init__(self, inner: LanceTakeQuery):
-        super().__init__(inner)
+    def __init__(self, inner: LanceTakeQuery, table: Optional["AsyncTable"] = None):
+        super().__init__(inner, table)
+
+    async def _plain_scan_to_pandas(
+        self,
+        blob_mode: BlobMode,
+        flatten: Optional[Union[int, bool]] = None,
+        **kwargs,
+    ) -> Optional["pd.DataFrame"]:
+        return None
 
 
 class BaseQueryBuilder(object):
@@ -3319,6 +3698,27 @@ class BaseQueryBuilder(object):
         self._inner.with_row_id()
         return self
 
+    def with_row_address(self, with_row_address: bool = True) -> Self:
+        """
+        Include the _rowaddr column in scanner-backed plain query results.
+        """
+        self._inner.with_row_address(with_row_address)
+        return self
+
+    def with_fragments(self, fragments: Any) -> Self:
+        """
+        Restrict scanner-backed plain query results to the given Lance fragments.
+        """
+        self._inner.with_fragments(fragments)
+        return self
+
+    def fragment_ids(self, fragment_ids: List[int]) -> Self:
+        """
+        Restrict scanner-backed plain query results to the given Lance fragment ids.
+        """
+        self._inner.fragment_ids(fragment_ids)
+        return self
+
     def output_schema(self) -> pa.Schema:
         """
         Return the output schema for the query
@@ -3400,6 +3800,8 @@ class BaseQueryBuilder(object):
         self,
         flatten: Optional[Union[int, bool]] = None,
         timeout: Optional[timedelta] = None,
+        *,
+        blob_mode: BlobMode = "lazy",
         **kwargs,
     ) -> "pd.DataFrame":
         """
@@ -3433,11 +3835,15 @@ class BaseQueryBuilder(object):
             The maximum time to wait for the query to complete.
             If not specified, no timeout is applied. If the query does not
             complete within the specified time, an error will be raised.
+        blob_mode: str, default "lazy"
+            Controls how blob columns are returned for plain scan queries.
         **kwargs
             Forwarded to pyarrow.Table.to_pandas after query execution and
             optional flattening.
         """
-        return LOOP.run(self._inner.to_pandas(flatten, timeout, **kwargs))
+        return LOOP.run(
+            self._inner.to_pandas(flatten, timeout, blob_mode=blob_mode, **kwargs)
+        )
 
     def to_polars(
         self,

python/python/lancedb/remote/__init__.py

@@ -9,6 +9,7 @@ from typing import List, Optional
 from lancedb import __version__
 
 from .header import HeaderProvider
+from .oauth import OAuthConfig, OAuthFlowType
 
 __all__ = [
     "TimeoutConfig",
@@ -16,6 +17,8 @@ __all__ = [
     "TlsConfig",
     "ClientConfig",
     "HeaderProvider",
+    "OAuthConfig",
+    "OAuthFlowType",
 ]
 
 

python/python/lancedb/remote/db.py

@@ -3,6 +3,7 @@
 
 
 from datetime import timedelta
+import json
 import logging
 from concurrent.futures import ThreadPoolExecutor
 import sys
@@ -17,7 +18,7 @@ else:
 
 # Remove this import to fix circular dependency
 # from lancedb import connect_async
-from lancedb.remote import ClientConfig
+from lancedb.remote import ClientConfig, RetryConfig, TimeoutConfig, TlsConfig
 import pyarrow as pa
 
 from ..common import DATA
@@ -36,6 +37,64 @@ from ..table import Table
 from ..util import validate_table_name
 
 
+def _duration_seconds(value: Optional[timedelta]) -> Optional[float]:
+    return value.total_seconds() if value is not None else None
+
+
+def _timeout_config_to_dict(
+    config: Optional[TimeoutConfig],
+) -> Optional[dict[str, Any]]:
+    if config is None:
+        return None
+    return {
+        "timeout": _duration_seconds(config.timeout),
+        "connect_timeout": _duration_seconds(config.connect_timeout),
+        "read_timeout": _duration_seconds(config.read_timeout),
+        "pool_idle_timeout": _duration_seconds(config.pool_idle_timeout),
+    }
+
+
+def _retry_config_to_dict(config: RetryConfig) -> dict[str, Any]:
+    return {
+        "retries": config.retries,
+        "connect_retries": config.connect_retries,
+        "read_retries": config.read_retries,
+        "backoff_factor": config.backoff_factor,
+        "backoff_jitter": config.backoff_jitter,
+        "statuses": config.statuses,
+    }
+
+
+def _tls_config_to_dict(config: Optional[TlsConfig]) -> Optional[dict[str, Any]]:
+    if config is None:
+        return None
+    return {
+        "cert_file": config.cert_file,
+        "key_file": config.key_file,
+        "ssl_ca_cert": config.ssl_ca_cert,
+        "assert_hostname": config.assert_hostname,
+    }
+
+
+def _client_config_to_dict(config: ClientConfig) -> dict[str, Any]:
+    if config.header_provider is not None:
+        raise ValueError(
+            "Cannot serialize a remote connection with a header_provider. "
+            "Use static api_key/extra_headers or provide a worker-side "
+            "connection factory instead."
+        )
+    return {
+        "user_agent": config.user_agent,
+        "retry_config": _retry_config_to_dict(config.retry_config),
+        "timeout_config": _timeout_config_to_dict(config.timeout_config),
+        "extra_headers": config.extra_headers,
+        "id_delimiter": config.id_delimiter,
+        "tls_config": _tls_config_to_dict(config.tls_config),
+        "header_provider": None,
+        "user_id": config.user_id,
+    }
+
+
 class RemoteDBConnection(DBConnection):
     """A connection to a remote LanceDB database."""
 
@@ -50,6 +109,7 @@ class RemoteDBConnection(DBConnection):
         connection_timeout: Optional[float] = None,
         read_timeout: Optional[float] = None,
         storage_options: Optional[Dict[str, str]] = None,
+        read_consistency_interval: Optional[timedelta] = None,
     ):
         """Connect to a remote LanceDB database."""
         if isinstance(client_config, dict):
@@ -64,6 +124,7 @@ class RemoteDBConnection(DBConnection):
                 "request_thread_pool is no longer used and will be removed in "
                 "a future release.",
                 DeprecationWarning,
+                stacklevel=2,
             )
 
         if connection_timeout is not None:
@@ -72,6 +133,7 @@ class RemoteDBConnection(DBConnection):
                 "release. Please use client_config.timeout_config.connect_timeout "
                 "instead.",
                 DeprecationWarning,
+                stacklevel=2,
             )
             client_config.timeout_config.connect_timeout = timedelta(
                 seconds=connection_timeout
@@ -82,12 +144,18 @@ class RemoteDBConnection(DBConnection):
                 "read_timeout is deprecated and will be removed in a future release. "
                 "Please use client_config.timeout_config.read_timeout instead.",
                 DeprecationWarning,
+                stacklevel=2,
             )
             client_config.timeout_config.read_timeout = timedelta(seconds=read_timeout)
 
         parsed = urlparse(db_url)
         if parsed.scheme != "db":
             raise ValueError(f"Invalid scheme: {parsed.scheme}, only accepts db://")
+        self.db_url = db_url
+        self.api_key = api_key
+        self.region = region
+        self.host_override = host_override
+        self.storage_options = storage_options
         self.db_name = parsed.netloc
 
         self.client_config = client_config
@@ -103,12 +171,27 @@ class RemoteDBConnection(DBConnection):
                 host_override=host_override,
                 client_config=client_config,
                 storage_options=storage_options,
+                read_consistency_interval=read_consistency_interval,
             )
         )
 
     def __repr__(self) -> str:
         return f"RemoteConnect(name={self.db_name})"
 
+    @override
+    def serialize(self) -> str:
+        return json.dumps(
+            {
+                "connection_type": "remote",
+                "db_url": self.db_url,
+                "api_key": self.api_key,
+                "region": self.region,
+                "host_override": self.host_override,
+                "client_config": _client_config_to_dict(self.client_config),
+                "storage_options": self.storage_options,
+            }
+        )
+
     @override
     def list_namespaces(
         self,
@@ -303,6 +386,8 @@ class RemoteDBConnection(DBConnection):
         namespace_path: Optional[List[str]] = None,
         storage_options: Optional[Dict[str, str]] = None,
         index_cache_size: Optional[int] = None,
+        branch: Optional[str] = None,
+        version: Optional[int] = None,
     ) -> Table:
         """Open a Lance Table in the database.
 
@@ -313,6 +398,14 @@ class RemoteDBConnection(DBConnection):
         namespace_path: List[str], optional
             The namespace to open the table from.
             None or empty list represents root namespace.
+        branch: str, optional
+            If provided, open a handle scoped to this branch instead of the
+            default branch. Reads and writes operate in the branch's context.
+        version: int, optional
+            If provided, open the table pinned to this version, producing a
+            read-only handle. Composes with ``branch``: when both are given,
+            opens that branch at the version; otherwise opens ``main`` at the
+            version. Call ``checkout_latest`` to return to a writable state.
 
         Returns
         -------
@@ -329,7 +422,17 @@ class RemoteDBConnection(DBConnection):
             )
 
         table = LOOP.run(self._conn.open_table(name, namespace_path=namespace_path))
-        return RemoteTable(table, self.db_name)
+        tbl = RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=namespace_path,
+        )
+        if branch is not None:
+            tbl = tbl.branches.checkout(branch, version)
+        elif version is not None:
+            tbl.checkout(version)
+        return tbl
 
     def clone_table(
         self,
@@ -378,7 +481,12 @@ class RemoteDBConnection(DBConnection):
                 is_shallow=is_shallow,
             )
         )
-        return RemoteTable(table, self.db_name)
+        return RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=target_namespace_path,
+        )
 
     @override
     def create_table(
@@ -523,7 +631,12 @@ class RemoteDBConnection(DBConnection):
                 fill_value=fill_value,
             )
         )
-        return RemoteTable(table, self.db_name)
+        return RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=namespace_path,
+        )
 
     @override
     def drop_table(self, name: str, namespace_path: Optional[List[str]] = None):

python/python/lancedb/remote/errors.py

@@ -27,6 +27,9 @@ class LanceDBClientError(RuntimeError):
         self.request_id = request_id
         self.status_code = status_code
 
+    def __reduce__(self) -> tuple[type, tuple]:
+        return (self.__class__, (str(self), self.request_id, self.status_code))
+
 
 class HttpError(LanceDBClientError):
     """An error that occurred during an HTTP request.
@@ -101,3 +104,19 @@ class RetryError(LanceDBClientError):
         self.max_request_failures = max_request_failures
         self.max_connect_failures = max_connect_failures
         self.max_read_failures = max_read_failures
+
+    def __reduce__(self) -> tuple[type, tuple]:
+        return (
+            self.__class__,
+            (
+                str(self),
+                self.request_id,
+                self.request_failures,
+                self.connect_failures,
+                self.read_failures,
+                self.max_request_failures,
+                self.max_connect_failures,
+                self.max_read_failures,
+                self.status_code,
+            ),
+        )

python/python/lancedb/remote/oauth.py

@@ -0,0 +1,75 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import List, Optional
+
+
+class OAuthFlowType(str, Enum):
+    """OAuth authentication flow types."""
+
+    CLIENT_CREDENTIALS = "client_credentials"
+    """Client Credentials grant (service-to-service / M2M)."""
+
+    AZURE_MANAGED_IDENTITY = "azure_managed_identity"
+    """Azure Managed Identity via IMDS."""
+
+
+@dataclass
+class OAuthConfig:
+    """OAuth configuration for LanceDB authentication.
+
+    All token acquisition and refresh is handled in the Rust layer.
+    This config is passed through to Rust via PyO3.
+
+    Parameters
+    ----------
+    issuer_url : str
+        OIDC issuer URL or OAuth authority URL.
+        For Azure: ``https://login.microsoftonline.com/{tenant_id}/v2.0``
+    client_id : str
+        Application / Client ID.
+    scopes : List[str]
+        OAuth scopes to request.
+        For Azure managed identity, exactly one scope or resource is required.
+        For example: ``["api://{app_id}/.default"]``
+    flow : OAuthFlowType
+        Authentication flow to use. Default: CLIENT_CREDENTIALS.
+    client_secret : Optional[str]
+        Client secret (required for CLIENT_CREDENTIALS).
+    managed_identity_client_id : Optional[str]
+        Client ID for user-assigned managed identity (AZURE_MANAGED_IDENTITY).
+    refresh_buffer_secs : Optional[int]
+        Seconds before expiry to trigger proactive refresh (default: 300).
+        Keep this well below the token TTL; if it is greater than or equal to
+        the TTL, each request refreshes the token.
+
+    Examples
+    --------
+    Client Credentials (service-to-service):
+
+    >>> config = OAuthConfig(
+    ...     issuer_url="https://login.microsoftonline.com/{tenant}/v2.0",
+    ...     client_id="app-id",
+    ...     client_secret="secret",
+    ...     scopes=["api://lancedb-api/.default"],
+    ... )
+
+    Azure Managed Identity:
+
+    >>> config = OAuthConfig(
+    ...     issuer_url="https://login.microsoftonline.com/{tenant}/v2.0",
+    ...     client_id="app-id",
+    ...     scopes=["api://lancedb-api/.default"],
+    ...     flow=OAuthFlowType.AZURE_MANAGED_IDENTITY,
+    ... )
+    """
+
+    issuer_url: str
+    client_id: str
+    scopes: List[str]
+    flow: OAuthFlowType = OAuthFlowType.CLIENT_CREDENTIALS
+    client_secret: Optional[str] = field(default=None, repr=False)
+    managed_identity_client_id: Optional[str] = None
+    refresh_buffer_secs: Optional[int] = None

python/python/lancedb/remote/table.py

@@ -2,15 +2,34 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 from datetime import timedelta
+import deprecation
 import logging
 from functools import cached_property
-from typing import Any, Callable, Dict, Iterable, List, Optional, Union, Literal
+import os
+from typing import (
+    Any,
+    Callable,
+    Dict,
+    Iterable,
+    List,
+    Optional,
+    TYPE_CHECKING,
+    Union,
+    Literal,
+    overload,
+)
+
+if TYPE_CHECKING:
+    from ..udf import JobHandle
 import warnings
 
+from lancedb import __version__
+
 from lancedb._lancedb import (
     AddColumnsResult,
     AddResult,
     AlterColumnsResult,
+    UpdateFieldMetadataResult,
     DeleteResult,
     DropColumnsResult,
     IndexConfig,
@@ -32,6 +51,7 @@ from lancedb.index import (
     LabelList,
 )
 from lancedb.remote.db import LOOP
+from lancedb.table import IndexConfigType, KNOWN_METRICS
 import pyarrow as pa
 
 from lancedb.common import DATA, VEC, VECTOR_COLUMN_NAME
@@ -40,7 +60,7 @@ from lancedb.embeddings import EmbeddingFunctionRegistry
 from lancedb.table import _normalize_progress
 
 from ..query import LanceVectorQueryBuilder, LanceQueryBuilder, LanceTakeQueryBuilder
-from ..table import AsyncTable, BlobMode, IndexStatistics, Query, Table, Tags
+from ..table import AsyncTable, BlobMode, Branches, IndexStatistics, Query, Table, Tags
 from ..types import BaseTokenizerType
 
 
@@ -49,14 +69,90 @@ class RemoteTable(Table):
         self,
         table: AsyncTable,
         db_name: str,
+        *,
+        connection_state: Optional[Union[str, Callable[[], str]]] = None,
+        namespace_path: Optional[List[str]] = None,
     ):
-        self._table = table
+        self._table_handle = table
+        self._name = table.name
         self.db_name = db_name
+        self._connection_state = connection_state
+        self._namespace_path = list(namespace_path or [])
+        self._checkout_version: Optional[int] = None
+        # The branch this handle is scoped to (None == main). Persisted so a
+        # fork/pickle reopen restores the branch instead of reverting to main.
+        self._branch: Optional[str] = None
+        self._pid = os.getpid()
+
+    def _serialized_connection_state(self) -> str:
+        if self._connection_state is None:
+            raise RuntimeError(
+                "Cannot reopen this remote table because it does not carry "
+                "serialized connection state"
+            )
+        if callable(self._connection_state):
+            self._connection_state = self._connection_state()
+        return self._connection_state
+
+    @property
+    def _table(self) -> AsyncTable:
+        self._ensure_open()
+        assert self._table_handle is not None
+        return self._table_handle
+
+    @_table.setter
+    def _table(self, table: AsyncTable) -> None:
+        self._table_handle = table
+        self._name = table.name
+        self._pid = os.getpid()
+
+    def _ensure_open(self) -> None:
+        pid = os.getpid()
+        if self._table_handle is not None and self._pid == pid:
+            return
+
+        # Pickle clears the handle; fork inherits a handle created in the
+        # parent process. In both cases reopen before touching the Rust client.
+        from lancedb import deserialize_conn
+
+        db = deserialize_conn(self._serialized_connection_state(), for_worker=True)
+        # Reopen on the same branch and pinned version (branch=None / version=None
+        # reproduce the plain main-latest open).
+        table = db.open_table(
+            self._name,
+            namespace_path=self._namespace_path,
+            branch=self._branch,
+            version=self._checkout_version,
+        )
+
+        self._table_handle = table._table
+        self.db_name = table.db_name
+        self._pid = pid
+
+    def __getstate__(self) -> dict:
+        return {
+            "connection_state": self._serialized_connection_state(),
+            "db_name": self.db_name,
+            "name": self.name,
+            "namespace_path": self._namespace_path,
+            "checkout_version": self._checkout_version,
+            "branch": self._branch,
+        }
+
+    def __setstate__(self, state: dict) -> None:
+        self._table_handle = None
+        self._name = state["name"]
+        self.db_name = state["db_name"]
+        self._connection_state = state["connection_state"]
+        self._namespace_path = state["namespace_path"]
+        self._checkout_version = state["checkout_version"]
+        self._branch = state.get("branch")
+        self._pid = None
 
     @property
     def name(self) -> str:
         """The name of the table"""
-        return self._table.name
+        return self._name
 
     def __repr__(self) -> str:
         return f"RemoteTable({self.db_name}.{self.name})"
@@ -78,6 +174,34 @@ class RemoteTable(Table):
     def tags(self) -> Tags:
         return Tags(self._table)
 
+    @property
+    def branches(self) -> Branches:
+        """Branch management for the table.
+
+        ``create``/``checkout`` return a new table handle scoped to the branch;
+        writes on it do not affect ``main``.
+        """
+        return Branches(self)
+
+    def current_branch(self) -> Optional[str]:
+        """The branch this table handle is scoped to, or ``None`` for ``main``."""
+        return self._table.current_branch()
+
+    def _wrap_branch_handle(
+        self, async_table: AsyncTable, version: Optional[int] = None
+    ) -> "RemoteTable":
+        # A branch handle stays a RemoteTable with the same connection context.
+        # Record the branch and version pin so a fork/pickle reopen restores both.
+        handle = RemoteTable(
+            async_table,
+            self.db_name,
+            connection_state=self._connection_state,
+            namespace_path=self._namespace_path,
+        )
+        handle._branch = async_table.current_branch()
+        handle._checkout_version = version
+        return handle
+
     @cached_property
     def embedding_functions(self) -> Dict[str, EmbeddingFunctionConfig]:
         """
@@ -106,13 +230,19 @@ class RemoteTable(Table):
         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))
+        result = LOOP.run(self._table.checkout(version))
+        self._checkout_version = self.version
+        return result
 
     def checkout_latest(self):
-        return LOOP.run(self._table.checkout_latest())
+        result = LOOP.run(self._table.checkout_latest())
+        self._checkout_version = None
+        return result
 
     def restore(self, version: Optional[Union[int, str]] = None):
-        return LOOP.run(self._table.restore(version))
+        result = LOOP.run(self._table.restore(version))
+        self._checkout_version = None
+        return result
 
     def list_indices(self) -> Iterable[IndexConfig]:
         """List all the indices on the table"""
@@ -122,6 +252,11 @@ class RemoteTable(Table):
         """List all the stats of a specified index"""
         return LOOP.run(self._table.index_stats(index_uuid))
 
+    @deprecation.deprecated(
+        deprecated_in="0.25.0",
+        current_version=__version__,
+        details="Use create_index() with config=BTree()/Bitmap()/LabelList() instead.",
+    )
     def create_scalar_index(
         self,
         column: str,
@@ -131,7 +266,12 @@ class RemoteTable(Table):
         wait_timeout: Optional[timedelta] = None,
         name: Optional[str] = None,
     ):
-        """Creates a scalar index
+        """Creates a scalar index.
+
+        .. deprecated:: 0.25.0
+            Use :meth:`create_index` with a BTree, Bitmap, or LabelList config instead.
+            Example: ``table.create_index("column", config=BTree())``
+
         Parameters
         ----------
         column : str
@@ -162,6 +302,11 @@ class RemoteTable(Table):
             )
         )
 
+    @deprecation.deprecated(
+        deprecated_in="0.25.0",
+        current_version=__version__,
+        details="Use create_index() with config=FTS() instead.",
+    )
     def create_fts_index(
         self,
         column: str,
@@ -182,6 +327,12 @@ class RemoteTable(Table):
         prefix_only: bool = False,
         name: Optional[str] = None,
     ):
+        """Create a full-text search index on a column.
+
+        .. deprecated:: 0.25.0
+            Use :meth:`create_index` with an FTS config instead.
+            Example: ``table.create_index("text_column", config=FTS())``
+        """
         config = FTS(
             with_position=with_position,
             base_tokenizer=base_tokenizer,
@@ -205,9 +356,43 @@ class RemoteTable(Table):
             )
         )
 
+    # New unified API overload
+    @overload
     def create_index(
         self,
-        metric="l2",
+        column: str,
+        /,
+        *,
+        config: IndexConfigType,
+        wait_timeout: Optional[timedelta] = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+    ) -> None: ...
+
+    # Legacy API overload (deprecated)
+    @overload
+    def create_index(
+        self,
+        metric: Literal["l2", "cosine", "dot", "hamming"] = ...,
+        vector_column_name: str = ...,
+        index_cache_size: Optional[int] = ...,
+        num_partitions: Optional[int] = ...,
+        num_sub_vectors: Optional[int] = ...,
+        replace: Optional[bool] = ...,
+        accelerator: Optional[str] = ...,
+        index_type: Literal[
+            "VECTOR", "IVF_FLAT", "IVF_SQ", "IVF_PQ", "IVF_HNSW_SQ", "IVF_HNSW_PQ"
+        ] = ...,
+        wait_timeout: Optional[timedelta] = ...,
+        *,
+        num_bits: int = ...,
+        name: Optional[str] = ...,
+        train: bool = ...,
+    ) -> None: ...
+
+    def create_index(
+        self,
+        metric: str = "l2",
         vector_column_name: str = VECTOR_COLUMN_NAME,
         index_cache_size: Optional[int] = None,
         num_partitions: Optional[int] = None,
@@ -218,89 +403,113 @@ class RemoteTable(Table):
         wait_timeout: Optional[timedelta] = None,
         *,
         num_bits: int = 8,
+        config: Optional[IndexConfigType] = None,
         name: Optional[str] = None,
         train: bool = True,
     ):
-        """Create an index on the table.
+        """Create an index on a column.
 
-        Parameters
-        ----------
-        metric : str
-            The metric to use for the index. Default is "l2".
-        vector_column_name : str
-            The name of the vector column. Default is "vector".
+        This method supports both the new unified API and the legacy API
+        for backwards compatibility. The new API takes the column name as the
+        first positional argument and an index configuration object via
+        ``config``; the legacy API takes the distance metric as the first
+        argument plus separate ``vector_column_name`` / ``num_partitions`` /
+        etc. parameters, and emits a ``DeprecationWarning``.
 
         Examples
         --------
-        >>> import lancedb
-        >>> import uuid
-        >>> from lancedb.schema import vector
-        >>> db = lancedb.connect("db://...", api_key="...", # doctest: +SKIP
-        ...                      region="...") # doctest: +SKIP
-        >>> table_name = uuid.uuid4().hex
-        >>> schema = pa.schema(
-        ...     [
-        ...             pa.field("id", pa.uint32(), False),
-        ...            pa.field("vector", vector(128), False),
-        ...             pa.field("s", pa.string(), False),
-        ...     ]
+        New API (recommended):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "vector", config=IvfPq(distance_type="l2")
         ... )
-        >>> table = db.create_table( # doctest: +SKIP
-        ...     table_name, # doctest: +SKIP
-        ...     schema=schema, # doctest: +SKIP
+        >>> table.create_index("category", config=BTree())  # doctest: +SKIP
+        >>> table.create_index("content", config=FTS())  # doctest: +SKIP
+
+        Legacy API (deprecated):
+
+        >>> table.create_index(  # doctest: +SKIP
+        ...     "l2", vector_column_name="vector"
         ... )
-        >>> table.create_index("l2", "vector") # doctest: +SKIP
         """
+        # Detect whether this is a legacy API call
+        is_legacy = self._is_legacy_create_index_call(
+            metric,
+            config,
+            num_partitions,
+            num_sub_vectors,
+            vector_column_name,
+            accelerator,
+            index_cache_size,
+            replace,
+        )
 
-        if accelerator is not None:
-            logging.warning(
-                "GPU accelerator is not yet supported on LanceDB cloud."
-                "If you have 100M+ vectors to index,"
-                "please contact us at contact@lancedb.com"
-            )
-        if replace is not None:
-            logging.warning(
-                "replace is not supported on LanceDB cloud."
-                "Existing indexes will always be replaced."
+        if is_legacy:
+            warnings.warn(
+                "The create_index() API with metric/num_partitions parameters is "
+                "deprecated and will be removed in a future version. "
+                "Please migrate to the new unified API:\n"
+                "  # Old (deprecated):\n"
+                "  table.create_index('l2', vector_column_name='my_vector')\n"
+                "  # New (recommended):\n"
+                "  table.create_index('my_vector', config=IvfPq(distance_type='l2'))",
+                DeprecationWarning,
+                stacklevel=2,
             )
 
-        index_type = index_type.upper()
-        if index_type == "VECTOR" or index_type == "IVF_PQ":
-            config = IvfPq(
-                distance_type=metric,
-                num_partitions=num_partitions,
-                num_sub_vectors=num_sub_vectors,
-                num_bits=num_bits,
-            )
-        elif index_type == "IVF_RQ":
-            config = IvfRq(
-                distance_type=metric,
-                num_partitions=num_partitions,
-                num_bits=num_bits,
-            )
-        elif index_type == "IVF_SQ":
-            config = IvfSq(distance_type=metric, num_partitions=num_partitions)
-        elif index_type == "IVF_HNSW_PQ":
-            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, num_partitions=num_partitions)
-        elif index_type == "IVF_HNSW_FLAT":
-            config = HnswFlat(distance_type=metric, num_partitions=num_partitions)
-        elif index_type == "IVF_FLAT":
-            config = IvfFlat(distance_type=metric, num_partitions=num_partitions)
+            column = vector_column_name
+
+            if accelerator is not None:
+                logging.warning(
+                    "GPU accelerator is not yet supported on LanceDB cloud."
+                    "If you have 100M+ vectors to index,"
+                    "please contact us at contact@lancedb.com"
+                )
+            if replace is not None:
+                logging.warning(
+                    "replace is not supported on LanceDB cloud."
+                    "Existing indexes will always be replaced."
+                )
+
+            idx_type = index_type.upper()
+            if idx_type == "VECTOR" or idx_type == "IVF_PQ":
+                config = IvfPq(
+                    distance_type=metric,
+                    num_partitions=num_partitions,
+                    num_sub_vectors=num_sub_vectors,
+                    num_bits=num_bits,
+                )
+            elif idx_type == "IVF_RQ":
+                config = IvfRq(
+                    distance_type=metric,
+                    num_partitions=num_partitions,
+                    num_bits=num_bits,
+                )
+            elif idx_type == "IVF_SQ":
+                config = IvfSq(distance_type=metric, num_partitions=num_partitions)
+            elif idx_type == "IVF_HNSW_PQ":
+                raise ValueError(
+                    "IVF_HNSW_PQ is not supported on LanceDB cloud."
+                    "Please use IVF_HNSW_SQ instead."
+                )
+            elif idx_type == "IVF_HNSW_SQ":
+                config = HnswSq(distance_type=metric, num_partitions=num_partitions)
+            elif idx_type == "IVF_HNSW_FLAT":
+                config = HnswFlat(distance_type=metric, num_partitions=num_partitions)
+            elif idx_type == "IVF_FLAT":
+                config = IvfFlat(distance_type=metric, num_partitions=num_partitions)
+            else:
+                raise ValueError(
+                    f"Unknown vector index type: {idx_type}. Valid options are"
+                    " 'IVF_FLAT', 'IVF_PQ', 'IVF_RQ', 'IVF_SQ',"
+                    " 'IVF_HNSW_PQ', 'IVF_HNSW_SQ', 'IVF_HNSW_FLAT'"
+                )
         else:
-            raise ValueError(
-                f"Unknown vector index type: {index_type}. Valid options are"
-                " 'IVF_FLAT', 'IVF_PQ', 'IVF_RQ', 'IVF_SQ',"
-                " 'IVF_HNSW_PQ', 'IVF_HNSW_SQ', 'IVF_HNSW_FLAT'"
-            )
+            column = metric
 
         LOOP.run(
             self._table.create_index(
-                vector_column_name,
+                column,
                 config=config,
                 wait_timeout=wait_timeout,
                 name=name,
@@ -308,6 +517,37 @@ class RemoteTable(Table):
             )
         )
 
+    def _is_legacy_create_index_call(
+        self,
+        first_arg: str,
+        config: Optional[IndexConfigType],
+        num_partitions: Optional[int],
+        num_sub_vectors: Optional[int],
+        vector_column_name: str,
+        accelerator: Optional[str],
+        index_cache_size: Optional[int],
+        replace: Optional[bool],
+    ) -> bool:
+        """Detect if this is a legacy create_index call."""
+        if config is not None:
+            return False
+        if any(
+            x is not None
+            for x in (
+                num_partitions,
+                num_sub_vectors,
+                accelerator,
+                index_cache_size,
+                replace,
+            )
+        ):
+            return True
+        if vector_column_name != VECTOR_COLUMN_NAME:
+            return True
+        if first_arg.lower() in KNOWN_METRICS:
+            return True
+        return False
+
     def add(
         self,
         data: DATA,
@@ -609,7 +849,8 @@ class RemoteTable(Table):
         """
         warnings.warn(
             "cleanup_old_versions() is a no-op on LanceDB Cloud. "
-            "Tables are automatically cleaned up and optimized."
+            "Tables are automatically cleaned up and optimized.",
+            stacklevel=2,
         )
         pass
 
@@ -621,7 +862,8 @@ class RemoteTable(Table):
         """
         warnings.warn(
             "compact_files() is a no-op on LanceDB Cloud. "
-            "Tables are automatically compacted and optimized."
+            "Tables are automatically compacted and optimized.",
+            stacklevel=2,
         )
         pass
 
@@ -638,21 +880,161 @@ class RemoteTable(Table):
         """
         warnings.warn(
             "optimize() is a no-op on LanceDB Cloud. "
-            "Indices are optimized automatically."
+            "Indices are optimized automatically.",
+            stacklevel=2,
         )
         pass
 
     def count_rows(self, filter: Optional[str] = None) -> int:
         return LOOP.run(self._table.count_rows(filter))
 
-    def add_columns(self, transforms: Dict[str, str]) -> AddColumnsResult:
-        return LOOP.run(self._table.add_columns(transforms))
+    def add_columns(
+        self,
+        transforms: Optional[Dict[str, str]] = None,
+        *,
+        computed: Optional[Dict[str, tuple]] = None,
+    ) -> Optional[AddColumnsResult]:
+        result = None
+        if transforms is not None:
+            result = LOOP.run(self._table.add_columns(transforms))
+        if computed:
+            LOOP.run(self._table.add_columns(computed=computed))
+        return result
+
+    def refresh_column(
+        self,
+        columns,
+        *,
+        where: Optional[str] = None,
+        num_workers: Optional[int] = None,
+        max_workers: Optional[int] = None,
+        batch_size: Optional[int] = None,
+        priority: Optional[str] = None,
+    ) -> "JobHandle":
+        """Trigger recompute of computed columns (REFRESH COLUMN).
+
+        The expression is resolved server-side from each column's stored
+        binding; columns bound to the same struct-returning function
+        refresh together. Returns a `JobHandle` to wait on, poll, or cancel
+        (``tbl.refresh_column("c").wait()``). Server-backed feature
+        (LanceDB Enterprise / Cloud).
+
+        num_workers / max_workers / batch_size / priority are per-refresh
+        scheduling knobs (how to run THIS refresh) and override any default
+        the function carries. `priority` is a Kueue tier
+        (training | interactive | backfill).
+        """
+        from ..udf import JobHandle
+
+        if isinstance(columns, str):
+            columns = [columns]
+        job_id = LOOP.run(
+            self._table.refresh_column(
+                list(columns),
+                where=where,
+                num_workers=num_workers,
+                max_workers=max_workers,
+                batch_size=batch_size,
+                priority=priority,
+            )
+        )
+        return JobHandle(self._job_conn(), job_id)
+
+    def lineage(self, column=None, *, direction=None, depth=None):
+        """Derived-compute lineage of this table, or one of its columns:
+        upstream sources, downstream dependents, and the function version +
+        location that produced each derived column (with a drift flag). Returns
+        a `Lineage`. See `Connection.lineage`."""
+        return self._job_conn().lineage(
+            self._name, column, direction=direction, depth=depth
+        )
+
+    def _job_conn(self):
+        """A client connection for polling jobs this table spawns. Built lazily
+        from the table's serialized connection state and cached (not pickled --
+        a forked/unpickled table rebuilds it on next use)."""
+        from lancedb import deserialize_conn
+
+        conn = getattr(self, "_job_conn_cache", None)
+        if conn is None:
+            conn = deserialize_conn(self._serialized_connection_state())
+            self._job_conn_cache = conn
+        return conn
+
+    def load_columns(
+        self,
+        source: Union[str, Iterable[str]],
+        pk: str,
+        columns: Union[Iterable[str], Dict[str, str]],
+        *,
+        source_format: str = "parquet",
+        source_pk: Optional[str] = None,
+        on_missing: str = "carry",
+        source_storage_options: Optional[Dict[str, str]] = None,
+        num_workers: Optional[int] = None,
+        max_workers: Optional[int] = None,
+        batch_size: Optional[int] = None,
+        commit_granularity: Optional[int] = None,
+        priority: Optional[str] = None,
+    ) -> str:
+        """Fill existing columns from an external source by primary-key join.
+
+        The distributed-job equivalent of Geneva's ``Table.load_columns()``:
+        imports precomputed values (e.g. embeddings) from Parquet/Lance/IPC into
+        this table, matching on a primary key. Returns the load job id.
+        Server-backed feature (LanceDB Enterprise / Cloud).
+
+        Parameters
+        ----------
+        source: str | list[str]
+            One source URI or a list of URIs.
+        pk: str
+            Destination primary-key column. Also the source key unless
+            ``source_pk`` is given.
+        columns: list[str] | dict[str, str]
+            Value columns to load. A list loads same-named columns; a dict maps
+            ``{target: source}``.
+        source_format: str
+            ``"parquet"`` (default), ``"lance"``, or ``"ipc"``.
+        source_pk: str, optional
+            Source primary-key column when it differs from ``pk``.
+        on_missing: str
+            Behavior for destination rows with no source match:
+            ``"carry"`` (default, keep existing), ``"null"``, or ``"error"``.
+        """
+        if isinstance(source, str):
+            source = [source]
+        if isinstance(columns, dict):
+            mappings = [(target, src) for target, src in columns.items()]
+        else:
+            mappings = [(c, None) for c in columns]
+        return LOOP.run(
+            self._table.load_columns(
+                list(source),
+                source_format,
+                pk,
+                mappings,
+                source_key=source_pk,
+                source_storage_options=source_storage_options,
+                on_missing=on_missing,
+                num_workers=num_workers,
+                max_workers=max_workers,
+                batch_size=batch_size,
+                commit_granularity=commit_granularity,
+                priority=priority,
+            )
+        )
 
     def alter_columns(
         self, *alterations: Iterable[Dict[str, str]]
     ) -> AlterColumnsResult:
         return LOOP.run(self._table.alter_columns(*alterations))
 
+    def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        return LOOP.run(self._table.update_field_metadata(*updates))
+
     def drop_columns(self, columns: Iterable[str]) -> DropColumnsResult:
         return LOOP.run(self._table.drop_columns(columns))
 
@@ -668,6 +1050,10 @@ class RemoteTable(Table):
         """Not supported on LanceDB Cloud."""
         return LOOP.run(self._table.unset_lsm_write_spec())
 
+    def close_lsm_writers(self) -> None:
+        """No-op on LanceDB Cloud (no local shard writers)."""
+        return LOOP.run(self._table.close_lsm_writers())
+
     def drop_index(self, index_name: str):
         return LOOP.run(self._table.drop_index(index_name))
 

python/python/lancedb/rerankers/linear_combination.py

@@ -102,8 +102,15 @@ class LinearCombinationReranker(Reranker):
 
         combined_list = []
         for row_id, result in results.items():
+            # Convert vector distance to a relevance score in [0, 1] where
+            # higher is better.  Missing vector entries are penalised with
+            # `_invert_score(fill)` = 1 - fill (= 0.0 for the default fill=1).
             vector_score = self._invert_score(result.get("_distance", fill))
-            fts_score = result.get("_score", fill)
+            # FTS scores (BM25) are already in a "higher = more relevant" space.
+            # Missing FTS entries are penalised symmetrically: we use
+            # `1 - fill` so that the same `fill` value drives both missing-vector
+            # and missing-FTS penalties in the same direction.
+            fts_score = result.get("_score", 1 - fill)
             result["_relevance_score"] = self._combine_score(vector_score, fts_score)
             combined_list.append(result)
 
@@ -123,8 +130,12 @@ class LinearCombinationReranker(Reranker):
         return tbl
 
     def _combine_score(self, vector_score, fts_score):
-        # these scores represent distance
-        return 1 - (self.weight * vector_score + (1 - self.weight) * fts_score)
+        # Both vector_score (inverted distance) and fts_score are in a
+        # "higher = more relevant" space.  A straight weighted average gives
+        # higher _relevance_score to better matches, as expected.
+        # Previously this returned `1 - (...)` which inverted the final
+        # ranking so that the *least* relevant document ranked first.
+        return self.weight * vector_score + (1 - self.weight) * fts_score
 
     def _invert_score(self, dist: float):
         # Invert the score between relevance and distance

python/python/lancedb/rerankers/mrr.py

@@ -125,6 +125,9 @@ class MRRReranker(Reranker):
         This cannot reuse rerank_hybrid because MRR semantics require treating
         each vector result as a separate ranking system.
         """
+        if not vector_results:
+            raise ValueError("vector_results must not be empty")
+
         if not all(isinstance(v, type(vector_results[0])) for v in vector_results):
             raise ValueError(
                 "All elements in vector_results should be of the same type"

python/python/lancedb/rerankers/rrf.py

@@ -82,6 +82,9 @@ class RRFReranker(Reranker):
         results from multiple vector searches as it doesn't support reranking
         vector results individually.
         """
+        if not vector_results:
+            raise ValueError("vector_results must not be empty")
+
         # Make sure all elements are of the same type
         if not all(isinstance(v, type(vector_results[0])) for v in vector_results):
             raise ValueError(

python/python/lancedb/scannable.py

@@ -86,7 +86,10 @@ def _from_list(data: list) -> Scannable:
 
 @to_scannable.register(dict)
 def _from_dict(data: dict) -> Scannable:
-    raise ValueError("Cannot add a single dictionary to a table. Use a list.")
+    raise ValueError(
+        "Cannot create or add rows from a single dictionary. "
+        "Use a list of dictionaries instead."
+    )
 
 
 @to_scannable.register(LanceModel)

python/python/lancedb/udf.py

@@ -0,0 +1,753 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+"""UDF authoring for LanceDB derived compute (server-backed).
+
+`@udf` / `@table_udf` turn a plain Python function into a registrable
+server-side UDF: a cloudpickled (or source) body, a SQL signature inferred
+from type hints, and the runtime options (pip deps, GPUs, batching, ...).
+Register and use them through the existing connection/table API:
+
+    import lancedb
+    from lancedb import udf, table_udf
+
+    db = lancedb.connect("db://my_db", api_key="...", host_override="...")
+
+    @udf(pip=["torch>=2.0"], num_gpus=1)
+    def embed(text: str) -> list[float]:
+        return model.encode(text).tolist()
+
+    db.create_function(embed)                 # CREATE FUNCTION (once)
+    tbl = db.open_table("docs")
+    tbl.add_columns(computed={"vec": embed("text")})  # bind embed(text) -> vec
+    tbl.refresh_column("vec").wait()          # materialize (returns a JobHandle)
+    view = db.create_materialized_view("chunks", tbl, ["id", chunk_fn])
+
+`embed("text")` applies the registered function to the `text` column and yields
+the expression `embed(text)`; the function itself stays decoupled from any
+column, so the same `embed` works on any column or table.
+
+These operations are server-backed (LanceDB Enterprise / Cloud); the
+decorator itself works locally (define + call), only registration needs a
+remote connection.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import base64
+import dataclasses
+import functools
+import inspect
+import re
+import sys
+import textwrap
+import time
+import typing
+
+# -- type hints -> SQL type strings -------------------------------------
+
+_SCALARS = {
+    int: "BIGINT",
+    # Pragmatic default for ML workloads: python float maps to FLOAT
+    # (Float32). Use an explicit `returns=` for DOUBLE.
+    float: "FLOAT",
+    str: "VARCHAR",
+    bool: "BOOLEAN",
+    bytes: "BLOB",
+}
+
+
+class TypeInferenceError(TypeError):
+    pass
+
+
+def sql_type(hint) -> str:
+    """SQL type string for a python type hint."""
+    if hint in _SCALARS:
+        return _SCALARS[hint]
+    origin = typing.get_origin(hint)
+    if origin in (list, typing.List):
+        (item,) = typing.get_args(hint) or (None,)
+        if item in _SCALARS:
+            return f"{_SCALARS[item]}[]"
+        raise TypeInferenceError(
+            f"unsupported list item type {item!r}; use an explicit returns="
+        )
+    fields = _struct_fields(hint)
+    if fields is not None:
+        inner = ", ".join(f"{name} {sql_type(h)}" for name, h in fields)
+        return f"STRUCT({inner})"
+    raise TypeInferenceError(
+        f"cannot infer a SQL type for {hint!r}; pass an explicit type string"
+    )
+
+
+def _struct_fields(hint):
+    """(name, hint) pairs for a TypedDict or dataclass, else None."""
+    if dataclasses.is_dataclass(hint):
+        return [(f.name, f.type) for f in dataclasses.fields(hint)]
+    # TypedDict detection: a dict subclass with __annotations__.
+    if (
+        isinstance(hint, type)
+        and issubclass(hint, dict)
+        and typing.get_type_hints(hint)
+    ):
+        return list(typing.get_type_hints(hint).items())
+    return None
+
+
+def return_type(fn, override: "str | None", table: bool) -> str:
+    """SQL return type for a function: explicit override wins, else the
+    return annotation. Table functions render as TABLE(...) and accept
+    struct-shaped hints (TypedDict/dataclass, optionally list-wrapped)."""
+    if override is not None:
+        s = override.strip()
+        if table and not s.upper().startswith("TABLE"):
+            if s.upper().startswith("STRUCT"):
+                return "TABLE" + s[len("STRUCT") :]
+            raise TypeInferenceError(
+                "a table function's returns= must be TABLE(...) or STRUCT(...)"
+            )
+        return s
+
+    hints = typing.get_type_hints(fn)
+    ret = hints.get("return")
+    if ret is None:
+        raise TypeInferenceError(
+            f"function {fn.__name__!r} needs a return annotation or returns="
+        )
+    if table:
+        # Accept list[Row] / Row where Row is a TypedDict or dataclass.
+        if typing.get_origin(ret) in (list, typing.List):
+            (ret,) = typing.get_args(ret)
+        fields = _struct_fields(ret)
+        if fields is None:
+            raise TypeInferenceError(
+                "a table function must return rows shaped as a TypedDict or "
+                "dataclass (optionally list-wrapped); or pass returns=..."
+            )
+        inner = ", ".join(f"{name} {sql_type(h)}" for name, h in fields)
+        return f"TABLE({inner})"
+    return sql_type(ret)
+
+
+def param_types(fn) -> "list[tuple[str, str]]":
+    """(name, sql type) per parameter, from annotations. Each UDF
+    parameter binds to a source column of the same name by default."""
+    hints = typing.get_type_hints(fn)
+    out = []
+    for name, p in inspect.signature(fn).parameters.items():
+        if p.kind in (p.VAR_POSITIONAL, p.VAR_KEYWORD):
+            raise TypeInferenceError("*args/**kwargs are not supported in UDFs")
+        hint = hints.get(name)
+        if hint is None:
+            raise TypeInferenceError(
+                f"parameter {name!r} of {fn.__name__!r} needs a type annotation"
+            )
+        out.append((name, sql_type(hint)))
+    return out
+
+
+# -- column expressions -------------------------------------------------
+
+
+class ColumnExpr(str):
+    """A computed-column expression produced by applying a registered
+    function to column names, e.g. ``embed("data") -> "embed(data)"``.
+
+    It IS the expression string everywhere a string is expected (views, SQL,
+    logging), and additionally carries the function's declared return type so
+    ``add_columns(computed=...)`` can declare the column without a hand-written
+    type. ``field_types`` holds the per-field SQL types of a STRUCT return, for
+    fanning one expression out to several columns.
+    """
+
+    data_type: "str | None"
+    field_types: "list[str] | None"
+
+    def __new__(cls, expr: str, data_type=None, field_types=None):
+        obj = super().__new__(cls, expr)
+        obj.data_type = data_type
+        obj.field_types = field_types
+        return obj
+
+
+def _normalize_computed(computed: dict) -> dict:
+    """Normalize the user-facing ``computed=`` mapping to the canonical
+    ``{name: (sql_type, expression)}`` form.
+
+    Accepts, per entry:
+      - value is a `ColumnExpr` (from ``fn("col")``): the column's SQL type
+        comes from the function's return type -- no hand-written type needed. A
+        tuple key (``("chunk", "idx")``) fans a STRUCT return out to one
+        (type, expression) entry per field, in declared order.
+      - value is a legacy ``(sql_type, expression)`` tuple: passed through (the
+        escape hatch, e.g. bare-name function strings).
+    """
+    out: dict = {}
+    for key, val in computed.items():
+        if isinstance(val, ColumnExpr):
+            expr = str(val)
+            if isinstance(key, (tuple, list)):
+                if not val.field_types:
+                    raise ValueError(
+                        f"columns {tuple(key)} need a STRUCT-returning function; "
+                        f"{expr} returns a single value"
+                    )
+                if len(val.field_types) != len(key):
+                    raise ValueError(
+                        f"{len(key)} columns but {len(val.field_types)} struct fields "
+                        f"in {expr}"
+                    )
+                for name, t in zip(key, val.field_types):
+                    out[name] = (t, expr)
+            else:
+                if val.data_type is None:
+                    raise ValueError(f"cannot infer a type for {expr}; pass types=")
+                out[key] = (val.data_type, expr)
+        else:
+            out[key] = val
+    return out
+
+
+# -- the @udf / @table_udf decorators -----------------------------------
+
+
+class Udf:
+    def __init__(
+        self,
+        fn,
+        *,
+        returns: "str | None" = None,
+        table: bool = False,
+        name: "str | None" = None,
+        pip: "list[str] | None" = None,
+        pip_index_url: "str | None" = None,
+        pip_extra_index_urls: "list[str] | None" = None,
+        find_links: "list[str] | None" = None,
+        requirements: "str | list[str] | None" = None,
+        conda: "list[str] | None" = None,
+        conda_channels: "list[str] | None" = None,
+        env: "dict[str, str] | list[str] | None" = None,
+        num_cpus: "int | None" = None,
+        num_gpus: "int | None" = None,
+        batch_size: "int | None" = None,
+        timeout: "float | None" = None,
+        error_policy: "str | None" = None,
+        max_skip_ratio: "float | None" = None,
+        retries: "int | None" = None,
+        docker_image: "str | None" = None,
+        description: "str | None" = None,
+        prefer_source: bool = False,
+    ):
+        functools.update_wrapper(self, fn)
+        self.fn = fn
+        self.name = name or fn.__name__
+        self.table = table
+        self.params = param_types(fn)
+        self.returns = return_type(fn, returns, table)
+        self.prefer_source = prefer_source
+        self.options: "dict[str, str]" = {}
+        if conda and (pip or requirements):
+            raise ValueError("pass conda or pip/requirements, not both")
+        if conda_channels and not conda:
+            raise ValueError("conda_channels requires conda")
+        if pip:
+            self.options["pip"] = ",".join(pip)
+        if pip_extra_index_urls:
+            self.options["pip_extra_index_urls"] = ",".join(pip_extra_index_urls)
+        if find_links:
+            self.options["find_links"] = ",".join(find_links)
+        if requirements:
+            self.options["requirements"] = _format_requirements(requirements)
+        if conda:
+            self.options["conda"] = ",".join(conda)
+        if conda_channels:
+            self.options["conda_channels"] = ",".join(conda_channels)
+        if env:
+            self.options["env"] = _format_env(env)
+        for key, val in [
+            ("pip_index_url", pip_index_url),
+            ("num_cpus", num_cpus),
+            ("num_gpus", num_gpus),
+            ("batch_size", batch_size),
+            ("timeout", timeout),
+            ("error_policy", error_policy),
+            ("max_skip_ratio", max_skip_ratio),
+            ("retries", retries),
+            ("docker_image", docker_image),
+        ]:
+            if val is not None:
+                self.options[key] = str(val)
+        # Keep the source in the description (when available) so the
+        # catalog stays inspectable even for pickled bodies.
+        if description is not None:
+            self.options["description"] = description
+        else:
+            try:
+                self.options["description"] = textwrap.dedent(inspect.getsource(fn))
+            except (OSError, TypeError):
+                pass
+
+    def __call__(self, *args, **kwargs):
+        """Call with real values to run locally; call with column-name
+        strings to build an expression for backfills and views, e.g.
+        ``embed("data")`` -> the expression ``embed(data)`` (a `ColumnExpr`
+        carrying the function's return type for `add_columns(computed=...)`)."""
+        if args and all(isinstance(a, str) for a in args) and not kwargs:
+            return self.expression(*args)
+        return self.fn(*args, **kwargs)
+
+    def expression(self, *columns: str) -> ColumnExpr:
+        """The expression applying this function to `columns` (default: the
+        function's own parameter names). Returns a `ColumnExpr` -- a string
+        that also carries the declared return type (and struct field types)."""
+        cols = columns or [p for p, _ in self.params]
+        expr = f"{self.name}({', '.join(cols)})"
+        field_types = None
+        if self.returns.upper().startswith("STRUCT"):
+            field_types = struct_field_types(self.returns)
+        return ColumnExpr(expr, data_type=self.returns, field_types=field_types)
+
+    def _body(self) -> "tuple[str, str]":
+        """(body literal, body_format). Source when requested and
+        retrievable; cloudpickle otherwise (handles closures)."""
+        if self.prefer_source:
+            try:
+                src = textwrap.dedent(inspect.getsource(self.fn))
+                # Strip the decorator line(s) so the stored body is a
+                # plain function definition.
+                lines = src.splitlines(keepends=True)
+                while lines and lines[0].lstrip().startswith("@"):
+                    lines.pop(0)
+                return "".join(lines), "source"
+            except (OSError, TypeError):
+                pass
+        import cloudpickle
+
+        raw = cloudpickle.dumps(self.fn)
+        return base64.b64encode(raw).decode("ascii"), "cloudpickle"
+
+    def _body_and_options(self) -> "tuple[str, dict[str, str]]":
+        """The body literal plus the finalized options (body_format /
+        python_version / cloudpickle-pip bookkeeping for a non-source
+        body)."""
+        body, body_format = self._body()
+        options = dict(self.options)
+        if body_format != "source":
+            options["body_format"] = body_format
+            # Pickled code objects only load under the same interpreter
+            # minor version; record ours so the worker can fail with a
+            # clear message instead of a bytecode error.
+            options["python_version"] = self.pickle_environment()
+            # The worker deserializes the body with cloudpickle; make sure
+            # the job's pip environment provides it. Conda bakes inject
+            # cloudpickle server-side, so do not create an invalid pip+conda
+            # declaration here.
+            if "conda" not in options:
+                pip = [d for d in options.get("pip", "").split(",") if d]
+                if not any(d.startswith("cloudpickle") for d in pip):
+                    pip.append("cloudpickle")
+                options["pip"] = ",".join(pip)
+        return body, options
+
+    def create_request(self) -> dict:
+        """Keyword arguments for `connection.create_function`."""
+        body, options = self._body_and_options()
+        return {
+            "name": self.name,
+            "language": "python",
+            "return_type": self.returns,
+            "body": body,
+            "options": options,
+        }
+
+    def create_statement(self) -> str:
+        """The equivalent `CREATE FUNCTION` SQL (for SQL-surface callers)."""
+        params = ", ".join(f"{n} {t}" for n, t in self.params)
+        body, options = self._body_and_options()
+        with_clause = ""
+        if options:
+            rendered = ", ".join(
+                f"{k} = '{_escape(v)}'" for k, v in sorted(options.items())
+            )
+            with_clause = f" WITH ({rendered})"
+        return (
+            f"CREATE FUNCTION {self.name}({params}) RETURNS {self.returns} "
+            f"LANGUAGE python AS '{_escape_body(body)}'{with_clause}"
+        )
+
+    def pickle_environment(self) -> str:
+        """Python version the body pickles under -- workers should match
+        the minor version for cloudpickle compatibility."""
+        return f"{sys.version_info.major}.{sys.version_info.minor}"
+
+
+def _escape(s: str) -> str:
+    return str(s).replace("'", "''")
+
+
+def _format_requirements(requirements: "str | list[str]") -> str:
+    if isinstance(requirements, str):
+        return requirements
+    return "\n".join(str(req) for req in requirements)
+
+
+def _format_env(env: "dict[str, str] | list[str]") -> str:
+    if isinstance(env, dict):
+        return "; ".join(f"{key}={value}" for key, value in env.items())
+    return "; ".join(str(entry) for entry in env)
+
+
+def _escape_body(body: str) -> str:
+    # The server unescapes \n / \t in single-quoted bodies; encode real
+    # newlines accordingly and escape quotes.
+    return (
+        body.replace("\\", "\\\\")
+        .replace("'", "''")
+        .replace("\n", "\\n")
+        .replace("\t", "\\t")
+    )
+
+
+def udf(fn=None, **kwargs):
+    """Decorate a function as a scalar (or struct-returning) UDF.
+
+    @udf
+    def doubled(val: int) -> float: ...
+
+    @udf(pip=["torch>=2"], num_gpus=1)
+    def embed(body: str) -> list[float]: ...
+    """
+    if fn is not None:
+        return Udf(fn, **kwargs)
+    return lambda f: Udf(f, **kwargs)
+
+
+def table_udf(fn=None, **kwargs):
+    """Decorate a table function (UDTF): each input row may emit zero or
+    more output rows. Only usable in materialized views.
+
+        class Chunk(TypedDict):
+            chunk: str
+            chunk_idx: int
+
+        @table_udf
+        def chunker(body: str) -> list[Chunk]: ...
+    """
+    kwargs["table"] = True
+    if fn is not None:
+        return Udf(fn, **kwargs)
+    return lambda f: Udf(f, **kwargs)
+
+
+# -- view / job handles (thin references over a connection) -------------
+
+
+def struct_field_types(returns: str) -> "list[str]":
+    """Field type strings of a STRUCT(...) SQL type, in declared order."""
+    inner = returns.strip()[len("STRUCT(") : -1]
+    fields, depth, start = [], 0, 0
+    for i, c in enumerate(inner):
+        if c in "([":
+            depth += 1
+        elif c in ")]":
+            depth -= 1
+        elif c == "," and depth == 0:
+            fields.append(inner[start:i].strip())
+            start = i + 1
+    fields.append(inner[start:].strip())
+    # Each field is "name TYPE"; drop the name.
+    return [f.split(None, 1)[1] for f in fields]
+
+
+def build_view_query(source, select) -> str:
+    """Assemble a view SELECT from a source (name or table) and select
+    items: a column name, an expression string, a (alias, expression)
+    tuple, or a @udf/@table_udf object."""
+    src = source.name if hasattr(source, "name") else source
+    items = []
+    for item in select:
+        if isinstance(item, Udf):
+            items.append(item.expression())
+        elif isinstance(item, tuple):
+            alias, expr = item
+            expr = expr.expression() if isinstance(expr, Udf) else expr
+            items.append(f"{expr} AS {alias}")
+        else:
+            items.append(item)
+    return f"SELECT {', '.join(items)} FROM {src}"
+
+
+def _job_id_matches(handle_id: str, listed_id: str) -> bool:
+    # The refresh/backfill endpoints return the submission id (a uuid), but
+    # the agent names the manifest job "<table>-<type>-<first 8 of the
+    # submission id>" -- which is what list_jobs and cancel report. Match the
+    # canonical id directly, or by that submission prefix.
+    if listed_id == handle_id:
+        return True
+    prefix = handle_id[:8]
+    return len(prefix) >= 4 and prefix in listed_id
+
+
+class MaterializedView:
+    """A reference to a materialized view (name + connection). Operations are
+    server-backed connection calls bound to the name.
+
+    ``create_materialized_view`` returns one of these; ``job_id`` is the
+    initial-population job (None when the view was created with no data), so
+    ``db.create_materialized_view(...).wait()`` blocks until it is populated.
+    """
+
+    def __init__(self, conn, name: str, job_id: "str | None" = None):
+        self.conn = conn
+        self.name = name
+        #: initial-population job id from create, or None (with_no_data).
+        self.job_id = job_id
+
+    def wait(self, timeout: float = 3600.0, poll: float = 2.0) -> str:
+        """Block until the initial-population job (from create) finishes.
+        A no-op when the view was created with no data."""
+        if self.job_id is None:
+            return "finished"
+        return JobHandle(self.conn, self.job_id, table=self.name).wait(
+            timeout=timeout, poll=poll
+        )
+
+    def refresh(self, full: bool = False) -> "JobHandle":
+        """Refresh the materialized view; returns a `JobHandle` to wait on,
+        poll, or cancel (``view.refresh().wait()``).
+
+        ``full=True`` forces a full rebuild (recompute and replace every row)
+        instead of the default incremental refresh. A full rebuild preserves
+        the view's indexes -- they are reindexed by the distributed indexer.
+        """
+        job_id = self.conn._refresh_materialized_view(self.name, full=full)
+        return JobHandle(self.conn, job_id, table=self.name)
+
+    def explain_refresh(self, full: bool = False):
+        """Plan a refresh without running it (EXPLAIN REFRESH)."""
+        return self.conn.explain_refresh_materialized_view(self.name, full=full)
+
+    def alter(self, auto_refresh: bool) -> None:
+        self.conn.alter_materialized_view(self.name, auto_refresh=auto_refresh)
+
+    def drop(self) -> None:
+        self.conn.drop_materialized_view(self.name)
+
+    # A materialized view is a first-class table: it can be indexed and
+    # searched like any other. These open the materialized dataset by name and
+    # delegate. Indexes declared this way are recorded against the view, so the
+    # engine re-applies them after a full refresh rebuilds the dataset (a full
+    # refresh overwrites the dataset, which would otherwise drop its indices).
+    def _table(self):
+        return self.conn.open_table(self.name)
+
+    def create_index(self, *args, **kwargs):
+        """Build an index on the materialized view (see Table.create_index)."""
+        return self._table().create_index(*args, **kwargs)
+
+    def create_scalar_index(self, *args, **kwargs):
+        """Build a scalar index on the materialized view."""
+        return self._table().create_scalar_index(*args, **kwargs)
+
+    def create_fts_index(self, *args, **kwargs):
+        """Build a full-text-search index on the materialized view."""
+        return self._table().create_fts_index(*args, **kwargs)
+
+    def search(self, *args, **kwargs):
+        """Search the materialized view (vector / FTS / hybrid)."""
+        return self._table().search(*args, **kwargs)
+
+    def lineage(self, column=None, *, direction=None, depth=None):
+        """Lineage of the materialized view (or one of its columns). Delegates
+        to the backing table; the server already includes the view's sources
+        and downstream dependents. Returns a `Lineage`."""
+        return self._table().lineage(column, direction=direction, depth=depth)
+
+
+_PROGRESS = re.compile(r"(\d+)/(\d+)")
+
+
+class JobFailedError(RuntimeError):
+    """Raised by ``JobHandle.wait()`` when the server reports the job ``failed``.
+
+    Carries the server-side error so a doomed backfill (e.g. a multi-column
+    ``REFRESH COLUMN`` of a scalar UDF) surfaces its real cause promptly,
+    instead of the caller blocking until ``wait()``'s timeout.
+    """
+
+    def __init__(self, job_id: str, error: "str | None"):
+        self.job_id = job_id
+        self.error = error
+        super().__init__(f"job {job_id} failed: {error or 'unknown error'}")
+
+
+class JobHandle:
+    """A reference to an inflight server-side job, with polling helpers."""
+
+    #: How long an unseen job is treated as still materializing (submission
+    #: -> agent cycle -> manifest write is async).
+    GRACE_SECONDS = 20.0
+
+    def __init__(self, conn, job_id: str, table: "str | None" = None):
+        self.conn = conn
+        self.id = job_id
+        #: The job's table, when known (refresh_column / MV refresh). Lets the
+        #: server resolve this job with an O(1) single-node read; without it the
+        #: lookup scans the database's active jobs (still correct).
+        self.table = table
+        self._created = time.monotonic()
+        self._seen = False
+
+    def _job(self):
+        # Poll by id (one job), not list_jobs (every active job): the server
+        # matches the submission/manifest id and reads just this table's node.
+        return self.conn.get_job(self.id, self.table)
+
+    def status(self) -> str:
+        """pending / running / cancelling / stale, or 'finished' once the
+        job has left the inflight listing."""
+        job = self._job()
+        if job is not None:
+            self._seen = True
+            return job.state
+        if not self._seen and time.monotonic() - self._created < self.GRACE_SECONDS:
+            return "pending"
+        return "finished"
+
+    def progress(self) -> "tuple[int, int] | None":
+        """(units_done, units_total) while running, else None."""
+        job = self._job()
+        if job is not None and job.units_total is not None:
+            return job.units_done or 0, job.units_total
+        return None
+
+    def wait(self, timeout: float = 3600.0, poll: float = 2.0) -> str:
+        deadline = time.monotonic() + timeout
+        while time.monotonic() < deadline:
+            state = self.status()
+            if state in ("finished", "stale"):
+                return state
+            if state == "failed":
+                # Terminal failure -- surface the server error now, don't block
+                # until `timeout`. `finalize` wrote it to the job's status node.
+                job = self._job()
+                raise JobFailedError(self.id, job.error if job is not None else None)
+            if state == "pending":
+                time.sleep(min(poll, 0.5))
+                continue
+            job = self._job()
+            if job is not None and job.committed:
+                return "finished"
+            time.sleep(poll)
+        raise TimeoutError(f"job {self.id} still {self.status()} after {timeout}s")
+
+    def cancel(self) -> None:
+        # Cancel by the canonical manifest id (what cancel matches), found
+        # via the submission prefix; fall back to the raw id.
+        job = self._job()
+        self.conn.cancel_job(job.job_id if job is not None else self.id)
+
+
+class AsyncMaterializedView:
+    """Async reference to a materialized view (name + async connection)."""
+
+    def __init__(self, conn, name: str, job_id: "str | None" = None):
+        self.conn = conn
+        self.name = name
+        #: initial-population job id from create, or None (with_no_data).
+        self.job_id = job_id
+
+    async def wait(self, timeout: float = 3600.0, poll: float = 2.0) -> str:
+        """Block until the initial-population job (from create) finishes.
+        A no-op when the view was created with no data."""
+        if self.job_id is None:
+            return "finished"
+        return await AsyncJobHandle(self.conn, self.job_id, table=self.name).wait(
+            timeout=timeout, poll=poll
+        )
+
+    async def refresh(self, full: bool = False) -> "AsyncJobHandle":
+        """Refresh the materialized view; returns an `AsyncJobHandle` to wait
+        on, poll, or cancel.
+
+        ``full=True`` forces a full rebuild instead of an incremental refresh
+        (indexes are preserved and reindexed by the distributed indexer).
+        """
+        job_id = await self.conn._refresh_materialized_view(self.name, full=full)
+        return AsyncJobHandle(self.conn, job_id, table=self.name)
+
+    async def explain_refresh(self, full: bool = False):
+        return await self.conn.explain_refresh_materialized_view(self.name, full=full)
+
+    async def alter(self, auto_refresh: bool) -> None:
+        await self.conn.alter_materialized_view(self.name, auto_refresh=auto_refresh)
+
+    async def drop(self) -> None:
+        await self.conn.drop_materialized_view(self.name)
+
+    async def lineage(self, column=None, *, direction=None, depth=None):
+        """Lineage of the materialized view (or column). Returns a `Lineage`."""
+        return await self.conn.lineage(
+            self.name, column, direction=direction, depth=depth
+        )
+
+
+class AsyncJobHandle:
+    """Async reference to an inflight server-side job, with polling helpers."""
+
+    GRACE_SECONDS = 20.0
+
+    def __init__(self, conn, job_id: str, table: "str | None" = None):
+        self.conn = conn
+        self.id = job_id
+        #: See JobHandle.table -- enables an O(1) by-id lookup when known.
+        self.table = table
+        self._created = time.monotonic()
+        self._seen = False
+
+    async def _job(self):
+        # Poll by id, not list_jobs (see JobHandle._job).
+        return await self.conn.get_job(self.id, self.table)
+
+    async def status(self) -> str:
+        job = await self._job()
+        if job is not None:
+            self._seen = True
+            return job.state
+        if not self._seen and time.monotonic() - self._created < self.GRACE_SECONDS:
+            return "pending"
+        return "finished"
+
+    async def progress(self) -> "tuple[int, int] | None":
+        job = await self._job()
+        if job is not None and job.units_total is not None:
+            return job.units_done or 0, job.units_total
+        return None
+
+    async def wait(self, timeout: float = 3600.0, poll: float = 2.0) -> str:
+        deadline = time.monotonic() + timeout
+        while time.monotonic() < deadline:
+            state = await self.status()
+            if state in ("finished", "stale"):
+                return state
+            if state == "failed":
+                # Terminal failure -- surface the server error now, don't block
+                # until `timeout`. `finalize` wrote it to the job's status node.
+                job = await self._job()
+                raise JobFailedError(self.id, job.error if job is not None else None)
+            if state == "pending":
+                await asyncio.sleep(min(poll, 0.5))
+                continue
+            job = await self._job()
+            if job is not None and job.committed:
+                return "finished"
+            await asyncio.sleep(poll)
+        raise TimeoutError(
+            f"job {self.id} still {await self.status()} after {timeout}s"
+        )
+
+    async def cancel(self) -> None:
+        job = await self._job()
+        await self.conn.cancel_job(job.job_id if job is not None else self.id)

python/python/lancedb/util.py

@@ -373,9 +373,15 @@ def _(value: list):
 @value_to_sql.register(dict)
 def _(value: dict):
     # https://datafusion.apache.org/user-guide/sql/scalar_functions.html#named-struct
+    # Render the field name through value_to_sql(str(...)) as well so that keys
+    # containing characters meaningful in SQL (e.g. a single quote) are escaped
+    # the same way string values are. A bare f"'{k}'" would emit invalid SQL for
+    # a key like "it's".
     return (
         "named_struct("
-        + ", ".join(f"'{k}', {value_to_sql(v)}" for k, v in value.items())
+        + ", ".join(
+            f"{value_to_sql(str(k))}, {value_to_sql(v)}" for k, v in value.items()
+        )
         + ")"
     )
 
@@ -385,6 +391,21 @@ def _(value: np.ndarray):
     return value_to_sql(value.tolist())
 
 
+@value_to_sql.register(np.bool_)
+def _(value: np.bool_):
+    return value_to_sql(bool(value))
+
+
+@value_to_sql.register(np.integer)
+def _(value: np.integer):
+    return value_to_sql(int(value))
+
+
+@value_to_sql.register(np.floating)
+def _(value: np.floating):
+    return value_to_sql(float(value))
+
+
 def deprecated(func):
     """This is a decorator which can be used to mark functions
     as deprecated. It will result in a warning being emitted

python/python/tests/docs/test_merge_insert.py

@@ -57,7 +57,7 @@ async def test_upsert_async(mem_db_async):
     await table.count_rows()  # 3
     res
     # MergeResult(version=2, num_updated_rows=1,
-    # num_inserted_rows=1, num_deleted_rows=0)
+    # num_inserted_rows=1, num_deleted_rows=0, num_rows=2)
     # --8<-- [end:upsert_basic_async]
     assert await table.count_rows() == 3
     assert res.version == 2
@@ -86,7 +86,7 @@ def test_insert_if_not_exists(mem_db):
     table.count_rows()  # 3
     res
     # MergeResult(version=2, num_updated_rows=0,
-    # num_inserted_rows=1, num_deleted_rows=0)
+    # num_inserted_rows=1, num_deleted_rows=0, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert table.count_rows() == 3
     assert res.version == 2
@@ -116,7 +116,7 @@ async def test_insert_if_not_exists_async(mem_db_async):
     await table.count_rows()  # 3
     res
     # MergeResult(version=2, num_updated_rows=0,
-    # num_inserted_rows=1, num_deleted_rows=0)
+    # num_inserted_rows=1, num_deleted_rows=0, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert await table.count_rows() == 3
     assert res.version == 2
@@ -150,7 +150,7 @@ def test_replace_range(mem_db):
     table.count_rows("doc_id = 1")  # 1
     res
     # MergeResult(version=2, num_updated_rows=1,
-    # num_inserted_rows=0, num_deleted_rows=1)
+    # num_inserted_rows=0, num_deleted_rows=1, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert table.count_rows("doc_id = 1") == 1
     assert res.version == 2
@@ -185,7 +185,7 @@ async def test_replace_range_async(mem_db_async):
     await table.count_rows("doc_id = 1")  # 1
     res
     # MergeResult(version=2, num_updated_rows=1,
-    # num_inserted_rows=0, num_deleted_rows=1)
+    # num_inserted_rows=0, num_deleted_rows=1, num_rows=1)
     # --8<-- [end:insert_if_not_exists]
     assert await table.count_rows("doc_id = 1") == 1
     assert res.version == 2

python/python/tests/test_db.py

@@ -466,7 +466,8 @@ async def test_create_table_v2_manifest_paths_async(tmp_path):
     assert await tbl.uses_v2_manifest_paths()
     manifests_dir = tmp_path / "test_v2_manifest_paths.lance" / "_versions"
     for manifest in os.listdir(manifests_dir):
-        assert re.match(r"\d{20}\.manifest", manifest)
+        if manifest.endswith(".manifest"):
+            assert re.match(r"\d{20}\.manifest", manifest)
 
     # Start a table in V1 mode then migrate
     tbl = await db_no_v2_paths.create_table(
@@ -476,13 +477,15 @@ async def test_create_table_v2_manifest_paths_async(tmp_path):
     assert not await tbl.uses_v2_manifest_paths()
     manifests_dir = tmp_path / "test_v2_migration.lance" / "_versions"
     for manifest in os.listdir(manifests_dir):
-        assert re.match(r"\d\.manifest", manifest)
+        if manifest.endswith(".manifest"):
+            assert re.match(r"\d\.manifest", manifest)
 
     await tbl.migrate_manifest_paths_v2()
     assert await tbl.uses_v2_manifest_paths()
 
     for manifest in os.listdir(manifests_dir):
-        assert re.match(r"\d{20}\.manifest", manifest)
+        if manifest.endswith(".manifest"):
+            assert re.match(r"\d{20}\.manifest", manifest)
 
 
 @pytest.mark.asyncio

python/python/tests/test_errors.py

@@ -0,0 +1,56 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+import pickle
+
+from lancedb.remote.errors import HttpError, LanceDBClientError, RetryError
+
+
+def test_pickle_lancedb_client_error():
+    err = LanceDBClientError("something went wrong", "req-123", 400)
+    restored = pickle.loads(pickle.dumps(err))
+    assert str(restored) == "something went wrong"
+    assert restored.request_id == "req-123"
+    assert restored.status_code == 400
+
+
+def test_pickle_lancedb_client_error_no_status_code():
+    err = LanceDBClientError("fail", "req-456")
+    restored = pickle.loads(pickle.dumps(err))
+    assert str(restored) == "fail"
+    assert restored.request_id == "req-456"
+    assert restored.status_code is None
+
+
+def test_pickle_http_error():
+    err = HttpError("not found", "req-789", 404)
+    restored = pickle.loads(pickle.dumps(err))
+    assert isinstance(restored, HttpError)
+    assert str(restored) == "not found"
+    assert restored.request_id == "req-789"
+    assert restored.status_code == 404
+
+
+def test_pickle_retry_error():
+    err = RetryError(
+        "max retries exceeded",
+        "req-abc",
+        request_failures=3,
+        connect_failures=1,
+        read_failures=2,
+        max_request_failures=5,
+        max_connect_failures=3,
+        max_read_failures=3,
+        status_code=503,
+    )
+    restored = pickle.loads(pickle.dumps(err))
+    assert isinstance(restored, RetryError)
+    assert str(restored) == "max retries exceeded"
+    assert restored.request_id == "req-abc"
+    assert restored.request_failures == 3
+    assert restored.connect_failures == 1
+    assert restored.read_failures == 2
+    assert restored.max_request_failures == 5
+    assert restored.max_connect_failures == 3
+    assert restored.max_read_failures == 3
+    assert restored.status_code == 503

python/python/tests/test_fts.py

@@ -215,11 +215,12 @@ def test_reject_legacy_tantivy_index(table):
 
 @pytest.mark.parametrize("with_position", [True, False])
 def test_create_inverted_index(table, with_position):
-    table.create_fts_index(
-        "text",
-        with_position=with_position,
-        name="custom_fts_index",
-    )
+    with pytest.warns(DeprecationWarning, match="create_fts_index"):
+        table.create_fts_index(
+            "text",
+            with_position=with_position,
+            name="custom_fts_index",
+        )
     indices = table.list_indices()
     fts_indices = [i for i in indices if i.index_type == "FTS"]
     assert any(i.name == "custom_fts_index" for i in fts_indices)

python/python/tests/test_index.py

@@ -20,6 +20,7 @@ from lancedb.index import (
     IvfRq,
     Bitmap,
     LabelList,
+    Fm,
     HnswPq,
     HnswSq,
     HnswFlat,
@@ -90,7 +91,9 @@ async def test_create_scalar_index(some_table: AsyncTable):
     # Can recreate if replace=True
     await some_table.create_index("id", replace=True)
     indices = await some_table.list_indices()
-    assert str(indices) == '[Index(BTree, columns=["id"], name="id_idx")]'
+    assert str(indices).startswith(
+        '[IndexConfig(name="id_idx", index_type="BTree", columns=["id"]'
+    )
     assert len(indices) == 1
     assert indices[0].index_type == "BTree"
     assert indices[0].columns == ["id"]
@@ -105,6 +108,27 @@ async def test_create_scalar_index(some_table: AsyncTable):
     assert len(indices) == 0
 
 
+@pytest.mark.asyncio
+async def test_index_config_repr(db_async):
+    # Use >= 1000 rows so the thousands separator in the repr is exercised.
+    nrows = 1500
+    table = await db_async.create_table(
+        "repr_table", pa.Table.from_pydict({"id": list(range(nrows))})
+    )
+    await table.create_index("id", config=BTree())
+    indices = await table.list_indices()
+    assert len(indices) == 1
+
+    r = repr(indices[0])
+    assert r.startswith('IndexConfig(name="id_idx", index_type="BTree", columns=["id"]')
+    # Integer counts use `_` thousands separators (valid Python int syntax).
+    assert "num_indexed_rows=1_500" in r
+    assert "num_unindexed_rows=0" in r
+    # created_at renders as a datetime so the value round-trips.
+    assert "created_at=datetime.datetime(" in r
+    assert r.endswith(")")
+
+
 @pytest.mark.asyncio
 async def test_create_nested_scalar_index_lists_canonical_paths(db_async):
     metadata_type = pa.struct(
@@ -113,8 +137,14 @@ async def test_create_nested_scalar_index_lists_canonical_paths(db_async):
             pa.field("user.id", pa.int32()),
         ]
     )
+    mixed_case_metadata_type = pa.struct([pa.field("userId", pa.int32())])
+    escaped_metadata_type = pa.struct([pa.field("user-id", pa.int32())])
+    literal_type = pa.struct([pa.field("a.b", pa.int32())])
     data = pa.Table.from_arrays(
         [
+            pa.array([1, 2, 3], type=pa.int32()),
+            pa.array([1, 2, 3], type=pa.int32()),
+            pa.array([1, 2, 3], type=pa.int32()),
             pa.array([1, 2, 3], type=pa.int32()),
             pa.array(
                 [
@@ -124,37 +154,91 @@ async def test_create_nested_scalar_index_lists_canonical_paths(db_async):
                 ],
                 type=metadata_type,
             ),
+            pa.array(
+                [{"userId": 10}, {"userId": 20}, {"userId": 30}],
+                type=mixed_case_metadata_type,
+            ),
+            pa.array(
+                [{"user-id": 10}, {"user-id": 20}, {"user-id": 30}],
+                type=escaped_metadata_type,
+            ),
+            pa.array(
+                [{"a.b": 10}, {"a.b": 20}, {"a.b": 30}],
+                type=literal_type,
+            ),
+        ],
+        names=[
+            "rowId",
+            "row-id",
+            "userId",
+            "user_id",
+            "metadata",
+            "MetaData",
+            "meta-data",
+            "literal",
         ],
-        names=["user_id", "metadata"],
     )
     table = await db_async.create_table("nested_scalar_index", data)
 
-    await table.create_index("user_id", config=BTree(), name="top_user_id_idx")
+    await table.create_index("rowId", config=BTree(), name="row_id_idx")
+    await table.create_index("`row-id`", config=BTree(), name="row_dash_id_idx")
+    await table.create_index("userId", config=BTree(), name="top_user_id_idx")
+    await table.create_index("user_id", config=BTree(), name="top_snake_user_id_idx")
     await table.create_index(
         "metadata.user_id", config=BTree(), name="nested_user_id_idx"
     )
     await table.create_index(
         "metadata.`user.id`", config=BTree(), name="escaped_user_id_idx"
     )
+    await table.create_index(
+        "MetaData.userId", config=BTree(), name="mixed_case_metadata_user_id_idx"
+    )
+    await table.create_index(
+        "`meta-data`.`user-id`", config=BTree(), name="escaped_names_idx"
+    )
+    await table.create_index("literal.`a.b`", config=BTree(), name="literal_dot_idx")
 
     columns_by_name = {
         index.name: index.columns for index in await table.list_indices()
     }
-    assert columns_by_name["top_user_id_idx"] == ["user_id"]
+    assert columns_by_name["row_id_idx"] == ["rowId"]
+    assert columns_by_name["row_dash_id_idx"] == ["`row-id`"]
+    assert columns_by_name["top_user_id_idx"] == ["userId"]
+    assert columns_by_name["top_snake_user_id_idx"] == ["user_id"]
     assert columns_by_name["nested_user_id_idx"] == ["metadata.user_id"]
     assert columns_by_name["escaped_user_id_idx"] == ["metadata.`user.id`"]
+    assert columns_by_name["mixed_case_metadata_user_id_idx"] == ["MetaData.userId"]
+    assert columns_by_name["escaped_names_idx"] == ["`meta-data`.`user-id`"]
+    assert columns_by_name["literal_dot_idx"] == ["literal.`a.b`"]
+
+    for index_name in columns_by_name:
+        stats = await table.index_stats(index_name)
+        assert stats is not None
+        assert stats.num_indexed_rows == 3
 
 
 @pytest.mark.asyncio
 async def test_create_fixed_size_binary_index(some_table: AsyncTable):
     await some_table.create_index("fsb", config=BTree())
     indices = await some_table.list_indices()
-    assert str(indices) == '[Index(BTree, columns=["fsb"], name="fsb_idx")]'
+    assert str(indices).startswith(
+        '[IndexConfig(name="fsb_idx", index_type="BTree", columns=["fsb"]'
+    )
     assert len(indices) == 1
     assert indices[0].index_type == "BTree"
     assert indices[0].columns == ["fsb"]
 
 
+@pytest.mark.asyncio
+async def test_create_fm_index(some_table: AsyncTable):
+    # FM-Index accelerates substring search on string/binary columns.
+    await some_table.create_index("data", config=Fm())
+    indices = await some_table.list_indices()
+    assert len(indices) == 1
+    assert indices[0].index_type == "Fm"
+    assert indices[0].columns == ["data"]
+
+
 @pytest.mark.asyncio
 async def test_create_bitmap_index(some_table: AsyncTable):
     await some_table.create_index("id", config=Bitmap())
@@ -162,12 +246,13 @@ async def test_create_bitmap_index(some_table: AsyncTable):
     await some_table.create_index("data", config=Bitmap())
     indices = await some_table.list_indices()
     assert len(indices) == 3
+    # list_indices returns indices in alphabetical order by name
     assert indices[0].index_type == "Bitmap"
-    assert indices[0].columns == ["id"]
+    assert indices[0].columns == ["data"]
     assert indices[1].index_type == "Bitmap"
-    assert indices[1].columns == ["is_active"]
+    assert indices[1].columns == ["id"]
     assert indices[2].index_type == "Bitmap"
-    assert indices[2].columns == ["data"]
+    assert indices[2].columns == ["is_active"]
 
     index_name = indices[0].name
     stats = await some_table.index_stats(index_name)
@@ -187,14 +272,65 @@ async def test_create_bitmap_index(some_table: AsyncTable):
 async def test_create_label_list_index(some_table: AsyncTable):
     await some_table.create_index("tags", config=LabelList())
     indices = await some_table.list_indices()
-    assert str(indices) == '[Index(LabelList, columns=["tags"], name="tags_idx")]'
+    assert str(indices).startswith(
+        '[IndexConfig(name="tags_idx", index_type="LabelList", columns=["tags"]'
+    )
+    plan = await some_table.query().where("array_has(tags, 'tag0')").explain_plan()
+    assert "ScalarIndexQuery" in plan
+
+
+@pytest.mark.asyncio
+async def test_create_large_list_label_list_index(db_async):
+    data = pa.Table.from_pydict(
+        {"tags": [[f"tag{i % 2}", "shared"] for i in range(16)]},
+        schema=pa.schema([pa.field("tags", pa.large_list(pa.string()))]),
+    )
+    table = await db_async.create_table("large_list_label_list_index", data)
+
+    await table.create_index("tags", config=LabelList())
+    indices = await table.list_indices()
+    assert str(indices).startswith(
+        '[IndexConfig(name="tags_idx", index_type="LabelList", columns=["tags"]'
+    )
+    plan = await table.query().where("array_has(tags, 'shared')").explain_plan()
+    assert "ScalarIndexQuery" in plan
+
+
+@pytest.mark.asyncio
+async def test_create_label_list_index_rejects_list_struct(db_async):
+    item_type = pa.struct(
+        [
+            pa.field("tag", pa.string()),
+            pa.field(
+                "metadata",
+                pa.struct([pa.field("userId", pa.string())]),
+            ),
+        ]
+    )
+    data = pa.Table.from_pylist(
+        [
+            {
+                "items": [
+                    {"tag": "tag0", "metadata": {"userId": "user0"}},
+                    {"tag": "shared", "metadata": {"userId": "user1"}},
+                ]
+            }
+        ],
+        schema=pa.schema([pa.field("items", pa.list_(item_type))]),
+    )
+    table = await db_async.create_table("list_struct_label_list_index", data)
+
+    with pytest.raises(Exception, match="LabelList index cannot be created"):
+        await table.create_index("items", config=LabelList())
 
 
 @pytest.mark.asyncio
 async def test_full_text_search_index(some_table: AsyncTable):
     await some_table.create_index("tags", config=FTS(with_position=False))
     indices = await some_table.list_indices()
-    assert str(indices) == '[Index(FTS, columns=["tags"], name="tags_idx")]'
+    assert str(indices).startswith(
+        '[IndexConfig(name="tags_idx", index_type="FTS", columns=["tags"]'
+    )
 
     await some_table.prewarm_index("tags_idx")
 
@@ -225,7 +361,6 @@ async def test_create_vector_index(some_table: AsyncTable):
     assert stats.num_indexed_rows == await some_table.count_rows()
     assert stats.num_unindexed_rows == 0
     assert stats.num_indices == 1
-    assert stats.loss >= 0.0
 
 
 @pytest.mark.asyncio
@@ -249,7 +384,6 @@ async def test_create_4bit_ivfpq_index(some_table: AsyncTable):
     assert stats.num_indexed_rows == await some_table.count_rows()
     assert stats.num_unindexed_rows == 0
     assert stats.num_indices == 1
-    assert stats.loss >= 0.0
 
 
 @pytest.mark.asyncio

python/python/tests/test_job_handle.py

@@ -0,0 +1,92 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+"""JobHandle.wait() terminal-state handling.
+
+Regression coverage for the cluster backfill-failure hang: the server reports a
+doomed job as ``state="failed"`` within seconds, but ``wait()`` used to ignore
+``failed`` and block until its (default 3600s) timeout. These tests pin that a
+``failed`` job raises ``JobFailedError`` promptly, carrying the server error.
+"""
+
+import asyncio
+import time
+
+import pytest
+
+from lancedb.udf import JobHandle, AsyncJobHandle, JobFailedError
+
+
+class FakeJobInfo:
+    """Mirror of the pyo3 builtins.JobInfo fields wait()/status() read."""
+
+    def __init__(self, state, error=None, committed=False, units_total=None):
+        self.state = state
+        self.error = error
+        self.committed = committed
+        self.units_total = units_total
+        self.units_done = None
+        self.job_id = "job-1"
+
+
+class FakeConn:
+    """get_job() walks a scripted list of JobInfo (or None) snapshots, holding
+    the last one once exhausted, so wait() polls a deterministic timeline."""
+
+    def __init__(self, snapshots):
+        self._snaps = list(snapshots)
+        self.calls = 0
+
+    def get_job(self, job_id, table=None):
+        snap = self._snaps[min(self.calls, len(self._snaps) - 1)]
+        self.calls += 1
+        return snap
+
+
+class AsyncFakeConn(FakeConn):
+    async def get_job(self, job_id, table=None):
+        return FakeConn.get_job(self, job_id, table)
+
+
+def test_wait_raises_on_failed_promptly():
+    # pending -> failed: wait() must raise the server error, not TimeoutError.
+    conn = FakeConn(
+        [None, FakeJobInfo("failed", error="multi-column backfill needs a STRUCT")]
+    )
+    jh = JobHandle(conn, "job-1", table="t")
+    t0 = time.monotonic()
+    with pytest.raises(JobFailedError) as exc:
+        jh.wait(timeout=30, poll=0.01)
+    assert time.monotonic() - t0 < 5  # prompt, nowhere near the 30s timeout
+    assert "STRUCT" in str(exc.value)
+    assert exc.value.error == "multi-column backfill needs a STRUCT"
+    assert exc.value.job_id == "job-1"
+
+
+def test_wait_returns_finished_on_success():
+    # running -> finished (job left the inflight listing) returns normally.
+    conn = FakeConn([FakeJobInfo("running", units_total=2), None])
+    jh = JobHandle(conn, "job-1", table="t")
+    jh._seen = True  # already observed, so a None now means "finished" not grace
+    assert jh.wait(timeout=30, poll=0.01) == "finished"
+
+
+def test_wait_returns_finished_on_committed():
+    # A committed job that is still listed resolves to finished.
+    conn = FakeConn([FakeJobInfo("running", committed=True, units_total=2)])
+    jh = JobHandle(conn, "job-1", table="t")
+    jh._seen = True
+    assert jh.wait(timeout=30, poll=0.01) == "finished"
+
+
+def test_async_wait_raises_on_failed_promptly():
+    conn = AsyncFakeConn([None, FakeJobInfo("failed", error="boom")])
+    jh = AsyncJobHandle(conn, "job-1", table="t")
+
+    async def run():
+        t0 = time.monotonic()
+        with pytest.raises(JobFailedError) as exc:
+            await jh.wait(timeout=30, poll=0.01)
+        assert time.monotonic() - t0 < 5
+        assert exc.value.error == "boom"
+
+    asyncio.run(run())

python/python/tests/test_lsm_write_spec.py

@@ -40,16 +40,6 @@ def _make_table(tmp_path):
 def test_set_lsm_write_spec_validates(tmp_path):
     _db, table = _make_table(tmp_path)
 
-    # No PK set yet.
-    with pytest.raises(Exception, match="primary key"):
-        table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 4))
-
-    table.set_unenforced_primary_key("id")
-
-    # Column mismatch.
-    with pytest.raises(Exception, match="match"):
-        table.set_lsm_write_spec(LsmWriteSpec.bucket("v", 4))
-
     # Out-of-range num_buckets.
     with pytest.raises(Exception, match="num_buckets"):
         table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 0))
@@ -70,7 +60,6 @@ def test_unset_lsm_write_spec(tmp_path):
         table.unset_lsm_write_spec()
 
     # Install a spec, then remove it; afterwards a fresh spec can be set.
-    table.set_unenforced_primary_key("id")
     table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 4))
     table.unset_lsm_write_spec()
     # A second unset errors — there is no spec left to remove.

python/python/tests/test_merge_insert_lsm.py

@@ -0,0 +1,196 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+"""Tests for the MemWAL LSM ``merge_insert`` dispatch."""
+
+from datetime import timedelta
+
+import lancedb
+import pyarrow as pa
+import pytest
+from lancedb._lancedb import LsmWriteSpec
+
+SCHEMA = pa.schema(
+    [
+        pa.field("id", pa.int64(), nullable=False),
+        pa.field("value", pa.int64(), nullable=False),
+    ]
+)
+
+REGION_SCHEMA = pa.schema(
+    [
+        pa.field("id", pa.int64(), nullable=False),
+        pa.field("region", pa.utf8(), nullable=False),
+    ]
+)
+
+
+def _reader(ids):
+    batch = pa.RecordBatch.from_arrays(
+        [
+            pa.array(ids, type=pa.int64()),
+            pa.array(list(range(len(ids))), type=pa.int64()),
+        ],
+        schema=SCHEMA,
+    )
+    return pa.RecordBatchReader.from_batches(SCHEMA, [batch])
+
+
+def _region_reader(rows):
+    batch = pa.RecordBatch.from_arrays(
+        [
+            pa.array([row[0] for row in rows], type=pa.int64()),
+            pa.array([row[1] for row in rows], type=pa.utf8()),
+        ],
+        schema=REGION_SCHEMA,
+    )
+    return pa.RecordBatchReader.from_batches(REGION_SCHEMA, [batch])
+
+
+def _bucket_table(tmp_path):
+    """A table with ``id`` as the primary key and a single-bucket LSM spec."""
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _reader([1, 2, 3]))
+    table.set_unenforced_primary_key("id")
+    # num_buckets = 1: every row routes to the single bucket.
+    table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 1))
+    return table
+
+
+def test_lsm_merge_insert_bucket(tmp_path):
+    table = _bucket_table(tmp_path)
+    # Empty `on` defaults to the primary key.
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([3, 4, 5]))
+    )
+    # LSM path: rows go to the MemWAL, so only num_rows is populated.
+    assert result.num_rows == 3
+    assert result.version == 0
+    assert result.num_inserted_rows == 0
+    assert result.num_updated_rows == 0
+
+
+def test_lsm_merge_insert_unsharded(tmp_path):
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _reader([1, 2, 3]))
+    table.set_unenforced_primary_key("id")
+    table.set_lsm_write_spec(LsmWriteSpec.unsharded())
+    result = (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([10, 11, 12, 13]))
+    )
+    assert result.num_rows == 4
+
+
+def test_lsm_merge_insert_identity(tmp_path):
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _region_reader([(1, "us"), (2, "us")]))
+    table.set_unenforced_primary_key("id")
+    table.set_lsm_write_spec(LsmWriteSpec.identity("region"))
+    # All rows share one identity value, so they route to one shard.
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_region_reader([(3, "us"), (4, "us")]))
+    )
+    assert result.num_rows == 2
+
+
+def test_lsm_merge_insert_use_lsm_write_false(tmp_path):
+    table = _bucket_table(tmp_path)  # rows id = 1, 2, 3
+    # use_lsm_write(False) opts out: the standard path runs and commits.
+    result = (
+        table.merge_insert("id")
+        .when_not_matched_insert_all()
+        .use_lsm_write(False)
+        .execute(_reader([3, 4, 5]))
+    )
+    assert result.num_inserted_rows == 2
+    assert table.count_rows() == 5
+
+
+def test_lsm_merge_insert_validate_single_shard_off(tmp_path):
+    table = _bucket_table(tmp_path)
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .validate_single_shard(False)
+        .execute(_reader([6, 7, 8]))
+    )
+    assert result.num_rows == 3
+
+
+def test_lsm_merge_insert_use_lsm_write_true_requires_spec(tmp_path):
+    # A table with a primary key but no LSM write spec installed.
+    db = lancedb.connect(tmp_path, read_consistency_interval=timedelta(seconds=0))
+    table = db.create_table("t", _reader([1, 2, 3]))
+    table.set_unenforced_primary_key("id")
+    with pytest.raises(Exception, match="use_lsm_write"):
+        (
+            table.merge_insert("id")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .use_lsm_write(True)
+            .execute(_reader([4]))
+        )
+
+
+def test_lsm_merge_insert_rejects_on_not_primary_key(tmp_path):
+    table = _bucket_table(tmp_path)
+    with pytest.raises(Exception, match="primary key"):
+        (
+            table.merge_insert("value")
+            .when_matched_update_all()
+            .when_not_matched_insert_all()
+            .execute(_reader([1]))
+        )
+
+
+def test_lsm_merge_insert_rejects_non_upsert(tmp_path):
+    table = _bucket_table(tmp_path)
+    # Insert-only (no when_matched_update_all) is not the upsert shape.
+    with pytest.raises(Exception, match="upsert"):
+        table.merge_insert([]).when_not_matched_insert_all().execute(_reader([4]))
+
+
+def test_lsm_close_writers(tmp_path):
+    table = _bucket_table(tmp_path)
+    (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([7, 8]))
+    )
+    table.close_lsm_writers()
+    # The writer reopens lazily on the next merge_insert.
+    result = (
+        table.merge_insert([])
+        .when_matched_update_all()
+        .when_not_matched_insert_all()
+        .execute(_reader([9]))
+    )
+    assert result.num_rows == 1
+
+
+@pytest.mark.asyncio
+async def test_async_lsm_merge_insert(tmp_path):
+    db = await lancedb.connect_async(
+        tmp_path, read_consistency_interval=timedelta(seconds=0)
+    )
+    table = await db.create_table("t", _reader([1, 2, 3]))
+    await table.set_unenforced_primary_key("id")
+    await table.set_lsm_write_spec(LsmWriteSpec.bucket("id", 1))
+
+    builder = (
+        table.merge_insert([]).when_matched_update_all().when_not_matched_insert_all()
+    )
+    result = await builder.execute(_reader([3, 4, 5]))
+    assert result.num_rows == 3
+    await table.close_lsm_writers()