fix(python): route blob query pandas through scanner

Bump version: 0.30.1-beta.0 → 0.30.1-beta.1
Bump version: 0.33.1-beta.0 → 0.33.1-beta.1
2026-06-04 04:40:40 +00:00 · 2026-06-03 22:59:25 +08:00 · 2026-06-03 11:17:37 +00:00 · 2026-06-03 11:16:51 +00:00 · 2026-06-03 19:15:44 +08:00 · 2026-06-02 14:02:22 -07:00
50 changed files with 2458 additions and 416 deletions
--- a/.agents/skills/README.md
+++ b/.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.
--- a/.agents/skills/lancedb-update-lance-dependency/SKILL.md
+++ b/.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.
--- a/.bumpversion.toml
+++ b/.bumpversion.toml
@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.30.0-beta.1"
+current_version = "0.30.1-beta.1"
 parse = """(?x)
    (?P<major>0|[1-9]\\d*)\\.
    (?P<minor>0|[1-9]\\d*)\\.
--- a/.github/workflows/codex-update-lance-dependency.yml
+++ b/.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
--- a/.github/workflows/lance-release-timer.yml
+++ b/.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
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1699,7 +1699,7 @@ version = "3.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "faf9468729b8cbcea668e36183cb69d317348c2e08e994829fb56ebfdfbaac34"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -2081,6 +2081,10 @@ name = "ctor"
 version = "1.0.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "378f0974ae2468eaf63aa036dbe9c926b0dc7ea64c156f2ea618bc2f75b934f0"
+dependencies = [
+ "link-section",
+ "linktime-proc-macro",
+]

 [[package]]
 name = "ctor-proc-macro"
@@ -2925,7 +2929,7 @@ dependencies = [
 "libc",
 "option-ext",
 "redox_users",
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -3122,7 +3126,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "39cab71617ae0d63f51a36d69f866391735b51691dbda63cf6f96d042b63efeb"
 dependencies = [
 "libc",
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -3302,8 +3306,8 @@ checksum = "42703706b716c37f96a77aea830392ad231f44c9e9a67872fa5548707e11b11c"

 [[package]]
 name = "fsst"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "rand 0.9.4",
@@ -4404,7 +4408,7 @@ dependencies = [
 "portable-atomic-util",
 "serde_core",
 "wasm-bindgen",
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -4570,8 +4574,8 @@ checksum = "e037a2e1d8d5fdbd49b16a4ea09d5d6401c1f29eca5ff29d03d3824dba16256a"

 [[package]]
 name = "lance"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arc-swap",
 "arrow",
@@ -4644,8 +4648,8 @@ dependencies = [

 [[package]]
 name = "lance-arrow"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4666,7 +4670,7 @@ dependencies = [
 [[package]]
 name = "lance-arrow-scalar"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4680,7 +4684,7 @@ dependencies = [
 [[package]]
 name = "lance-arrow-stats"
 version = "58.0.0"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-schema",
@@ -4689,8 +4693,8 @@ dependencies = [

 [[package]]
 name = "lance-bitpacking"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrayref",
 "paste",
@@ -4699,8 +4703,8 @@ dependencies = [

 [[package]]
 name = "lance-core"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4735,8 +4739,8 @@ dependencies = [

 [[package]]
 name = "lance-datafusion"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -4766,8 +4770,8 @@ dependencies = [

 [[package]]
 name = "lance-datagen"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -4785,8 +4789,8 @@ dependencies = [

 [[package]]
 name = "lance-encoding"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-arith",
 "arrow-array",
@@ -4821,8 +4825,8 @@ dependencies = [

 [[package]]
 name = "lance-file"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-arith",
 "arrow-array",
@@ -4853,8 +4857,8 @@ dependencies = [

 [[package]]
 name = "lance-index"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arc-swap",
 "arrow",
@@ -4920,8 +4924,8 @@ dependencies = [

 [[package]]
 name = "lance-io"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow",
 "arrow-arith",
@@ -4963,8 +4967,8 @@ dependencies = [

 [[package]]
 name = "lance-linalg"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -4980,8 +4984,8 @@ dependencies = [

 [[package]]
 name = "lance-namespace"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow",
 "async-trait",
@@ -4993,8 +4997,8 @@ dependencies = [

 [[package]]
 name = "lance-namespace-impls"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow",
 "arrow-ipc",
@@ -5043,8 +5047,8 @@ dependencies = [

 [[package]]
 name = "lance-select"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -5058,8 +5062,8 @@ dependencies = [

 [[package]]
 name = "lance-table"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -5099,8 +5103,8 @@ dependencies = [

 [[package]]
 name = "lance-testing"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "arrow-array",
 "arrow-schema",
@@ -5111,8 +5115,8 @@ dependencies = [

 [[package]]
 name = "lance-tokenizer"
-version = "7.2.0-beta.1"
-source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.1#b9995aba6115e8e4bc43179a45cbd0f9a170f305"
+version = "7.2.0-beta.3"
+source = "git+https://github.com/lance-format/lance.git?tag=v7.2.0-beta.3#7c070f760fa8e24c8015cb2afbd22c5e6b7898e8"
 dependencies = [
 "icu_segmenter",
 "jieba-rs",
@@ -5124,7 +5128,7 @@ dependencies = [

 [[package]]
 name = "lancedb"
-version = "0.30.0-beta.1"
+version = "0.30.1-beta.1"
 dependencies = [
 "ahash",
 "anyhow",
@@ -5207,7 +5211,7 @@ dependencies = [

 [[package]]
 name = "lancedb-nodejs"
-version = "0.30.0-beta.1"
+version = "0.30.1-beta.1"
 dependencies = [
 "arrow-array",
 "arrow-buffer",
@@ -5230,7 +5234,7 @@ dependencies = [

 [[package]]
 name = "lancedb-python"
-version = "0.33.0-beta.1"
+version = "0.33.1-beta.1"
 dependencies = [
 "arrow",
 "async-trait",
@@ -5412,6 +5416,18 @@ dependencies = [
 "thiserror 2.0.18",
 ]

+[[package]]
+name = "link-section"
+version = "0.16.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c8600ca3dbe044f07955b443ff606c50f45295b863289bbe7d0844d50cf11e4"
+
+[[package]]
+name = "linktime-proc-macro"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a44cd706ff0d503ee32b2071166510ca27e281228de10cd3aa8d35ff94560f81"
+
 [[package]]
 name = "linux-raw-sys"
 version = "0.12.1"
@@ -5871,7 +5887,7 @@ version = "0.50.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7957b9740744892f114936ab4a57b3f487491bbeafaf8083688b16841a4240e5"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -6041,9 +6057,9 @@ dependencies = [

 [[package]]
 name = "object_store_opendal"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "08298874eee5935c95bcaa393148834f9c53d904461ca15584a041d8a1c907c2"
+checksum = "0eb12a624a41fce745838d0ef3701ff6c47797c13cd18ad3612fd2a3134fdbd8"
 dependencies = [
 "async-trait",
 "bytes",
@@ -6098,11 +6114,11 @@ dependencies = [

 [[package]]
 name = "opendal"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "97b31d3d8e99a85d83b73ec26647f5607b80578ed9375810b6e44ffa3590a236"
+checksum = "96c9c85ce253ff87225e7669979d877a20c98a06604ec9d6dd5f4473e08f1ae1"
 dependencies = [
- "ctor 0.6.3",
+ "ctor 1.0.5",
 "opendal-core",
 "opendal-layer-concurrent-limit",
 "opendal-layer-logging",
@@ -6118,9 +6134,9 @@ dependencies = [

 [[package]]
 name = "opendal-core"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "1849dd2687e173e776d3af5fce1ba3ae47b9dd37a09d1c4deba850ef45fe00ca"
+checksum = "c4f8607c90e2c963a91467f50fb49fbc7fb3d573f88cea219ca59ccd3740b309"
 dependencies = [
 "anyhow",
 "base64 0.22.1",
@@ -6130,10 +6146,10 @@ dependencies = [
 "http-body 1.0.1",
 "jiff",
 "log",
- "md-5 0.10.6",
+ "md-5 0.11.0",
 "mea",
 "percent-encoding",
- "quick-xml 0.38.4",
+ "quick-xml 0.39.4",
 "reqsign-core",
 "reqwest 0.13.3",
 "serde",
@@ -6146,9 +6162,9 @@ dependencies = [

 [[package]]
 name = "opendal-layer-concurrent-limit"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "048b1b29c503263bdd80a9afe46a68cd02ea9bd361185b1feab4b151078998e9"
+checksum = "0d6f81ba6960e3fae1882f253b114b21d7e444e1534f209c7737a79f6243eb6f"
 dependencies = [
 "futures",
 "http 1.4.1",
@@ -6158,9 +6174,9 @@ dependencies = [

 [[package]]
 name = "opendal-layer-logging"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d2645adc988b12eda106e2679ae529facfbbaa868ceb706f6f8125c6af15c47b"
+checksum = "58ada45c6d81d1aa4c9305d0c7d4bc317c59c85866a0908a2d75a7a978aa5ee2"
 dependencies = [
 "log",
 "opendal-core",
@@ -6168,9 +6184,9 @@ dependencies = [

 [[package]]
 name = "opendal-layer-retry"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4eac134ffa4ddda6131a640a84a5315996424b9416c85052f8c64c1a33b70ad4"
+checksum = "7b2a25a718afb81fad81cb9a0580a1cb989221fa2317f888c6a37f8dad408eb7"
 dependencies = [
 "backon",
 "log",
@@ -6179,9 +6195,9 @@ dependencies = [

 [[package]]
 name = "opendal-layer-timeout"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "619586ab7480c2e3009f6d18eabab18957bc094778fd130bcc38924970a90f4c"
+checksum = "1e91f731724c213af81e9d03517859c8fc47b4578e64ad61ae4f099f10fe36e3"
 dependencies = [
 "opendal-core",
 "tokio",
@@ -6189,9 +6205,9 @@ dependencies = [

 [[package]]
 name = "opendal-service-azblob"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7452bf3ec61cfd81ac9ad9ada17825931e9e371d44a045c6bfab9596c0a2ac3b"
+checksum = "0030644366ef5d8cbe3a4a5822bf99a4aafddc1666e9d24b44d158d9062fc76a"
 dependencies = [
 "base64 0.22.1",
 "bytes",
@@ -6199,27 +6215,28 @@ dependencies = [
 "log",
 "opendal-core",
 "opendal-service-azure-common",
- "quick-xml 0.38.4",
+ "quick-xml 0.39.4",
 "reqsign-azure-storage",
 "reqsign-core",
 "reqsign-file-read-tokio",
 "serde",
- "sha2 0.10.9",
+ "sha2 0.11.0",
 "uuid",
 ]

 [[package]]
 name = "opendal-service-azdls"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8f9884c2d8cf8ba2bb077d79c877dac5863ba3bab9e2c9c1e41a2e0491404772"
+checksum = "6dea4908d490143a9b0b7f7a790e139ff829b06a023f670455ed3d44f664b361"
 dependencies = [
+ "base64 0.22.1",
 "bytes",
 "http 1.4.1",
 "log",
 "opendal-core",
 "opendal-service-azure-common",
- "quick-xml 0.38.4",
+ "quick-xml 0.39.4",
 "reqsign-azure-storage",
 "reqsign-core",
 "reqsign-file-read-tokio",
@@ -6229,9 +6246,9 @@ dependencies = [

 [[package]]
 name = "opendal-service-azure-common"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ffb0e45d6c8dcf66ce2da20e241bcb80e6e540e109a4ff20f318f6c9b4c54e0c"
+checksum = "9b489f13c42e69d69bdd72952b634356ec43a7881a20259b38b540fcecdf4051"
 dependencies = [
 "http 1.4.1",
 "opendal-core",
@@ -6239,9 +6256,9 @@ dependencies = [

 [[package]]
 name = "opendal-service-gcs"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "70a49477a10163431896d106136117f5670717f9c9e49cf6f710528800c6633a"
+checksum = "48de101aac565ed06af4b47903c24eafd249075553ec1fb18256751c45148d47"
 dependencies = [
 "async-trait",
 "bytes",
@@ -6249,7 +6266,7 @@ dependencies = [
 "log",
 "opendal-core",
 "percent-encoding",
- "quick-xml 0.38.4",
+ "quick-xml 0.39.4",
 "reqsign-core",
 "reqsign-file-read-tokio",
 "reqsign-google",
@@ -6260,9 +6277,9 @@ dependencies = [

 [[package]]
 name = "opendal-service-hf"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7b2ab7a2a8a11dfe257ef4db5c0de798acbcd0d6429c37382dad2154bc06a388"
+checksum = "c4922661976a1d40794a2adfbdb888cc3c23097690f825a92f773af38908a848"
 dependencies = [
 "bytes",
 "hf-xet",
@@ -6277,15 +6294,15 @@ dependencies = [

 [[package]]
 name = "opendal-service-oss"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "29c8a917829ad06d21b639558532cb0101fe49b040d946d673a73018683fac05"
+checksum = "328fa55e8888cbdfe00826bfea2a79042422b720e8369e9e021e46121dea5ace"
 dependencies = [
 "bytes",
 "http 1.4.1",
 "log",
 "opendal-core",
- "quick-xml 0.38.4",
+ "quick-xml 0.39.4",
 "reqsign-aliyun-oss",
 "reqsign-core",
 "reqsign-file-read-tokio",
@@ -6294,18 +6311,18 @@ dependencies = [

 [[package]]
 name = "opendal-service-s3"
-version = "0.56.0"
+version = "0.57.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9dadddeb9bb50b0d30927dd914c298c4ddca47e4c1cfa7674d311f0cf9b051c8"
+checksum = "313d46c9f5ae70bca26b7c3e3fbb9b639292625f28af73aa016f47e788af9deb"
 dependencies = [
 "base64 0.22.1",
 "bytes",
 "crc32c",
 "http 1.4.1",
 "log",
- "md-5 0.10.6",
+ "md-5 0.11.0",
 "opendal-core",
- "quick-xml 0.38.4",
+ "quick-xml 0.39.4",
 "reqsign-aws-v4",
 "reqsign-core",
 "reqsign-file-read-tokio",
@@ -7109,8 +7126,8 @@ version = "0.14.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "343d3bd7056eda839b03204e68deff7d1b13aba7af2b2fd16890697274262ee7"
 dependencies = [
- "heck 0.5.0",
- "itertools 0.14.0",
+ "heck 0.4.1",
+ "itertools 0.11.0",
 "log",
 "multimap",
 "petgraph",
@@ -7129,7 +7146,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "27c6023962132f4b30eb4c172c91ce92d933da334c59c23cddee82358ddafb0b"
 dependencies = [
 "anyhow",
- "itertools 0.14.0",
+ "itertools 0.11.0",
 "proc-macro2",
 "quote",
 "syn 2.0.117",
@@ -7288,7 +7305,6 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b66c2058c55a409d601666cffe35f04333cf1013010882cec174a7467cd4e21c"
 dependencies = [
 "memchr",
- "serde",
 ]

 [[package]]
@@ -7354,7 +7370,7 @@ dependencies = [
 "once_cell",
 "socket2 0.6.3",
 "tracing",
- "windows-sys 0.60.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -8092,7 +8108,7 @@ dependencies = [
 "errno",
 "libc",
 "linux-raw-sys",
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -8163,7 +8179,7 @@ dependencies = [
 "security-framework",
 "security-framework-sys",
 "webpki-root-certs",
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -8741,7 +8757,7 @@ version = "0.8.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c1c97747dbf44bb1ca44a561ece23508e99cb592e862f22222dcf42f51d1e451"
 dependencies = [
- "heck 0.5.0",
+ "heck 0.4.1",
 "proc-macro2",
 "quote",
 "syn 2.0.117",
@@ -8753,7 +8769,7 @@ version = "0.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "54254b8531cafa275c5e096f62d48c81435d1015405a91198ddb11e967301d40"
 dependencies = [
- "heck 0.5.0",
+ "heck 0.4.1",
 "proc-macro2",
 "quote",
 "syn 2.0.117",
@@ -9145,7 +9161,7 @@ dependencies = [
 "getrandom 0.4.2",
 "once_cell",
 "rustix",
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
@@ -10070,7 +10086,7 @@ version = "0.1.11"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c2a7b1c03c876122aa43f3020e6c3c3ee5c05081c9a00739faf7503aeba10d22"
 dependencies = [
- "windows-sys 0.61.2",
+ "windows-sys 0.59.0",
 ]

 [[package]]
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -13,20 +13,20 @@ categories = ["database-implementations"]
 rust-version = "1.91.0"

 [workspace.dependencies]
-lance = { "version" = "=7.2.0-beta.1", default-features = false, "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-core = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-datagen = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-file = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-io = { "version" = "=7.2.0-beta.1", default-features = false, "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-index = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-linalg = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-namespace-impls = { "version" = "=7.2.0-beta.1", default-features = false, "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-table = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-testing = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-datafusion = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-encoding = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
-lance-arrow = { "version" = "=7.2.0-beta.1", "tag" = "v7.2.0-beta.1", "git" = "https://github.com/lance-format/lance.git" }
+lance = { "version" = "=7.2.0-beta.3", default-features = false, "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-core = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-datagen = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-file = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-io = { "version" = "=7.2.0-beta.3", default-features = false, "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-index = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-linalg = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-namespace-impls = { "version" = "=7.2.0-beta.3", default-features = false, "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-table = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-testing = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-datafusion = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-encoding = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
+lance-arrow = { "version" = "=7.2.0-beta.3", "tag" = "v7.2.0-beta.3", "git" = "https://github.com/lance-format/lance.git" }
 ahash = "0.8"
 # Note that this one does not include pyarrow
 arrow = { version = "58.0.0", optional = false }
--- a/ci/update_lance_dependency.py
+++ b/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())
--- a/docs/src/java/java.md
+++ b/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>0.30.0-beta.1</version>
+    <version>0.30.1-beta.1</version>
 </dependency>
 ```

--- a/docs/src/js/classes/Table.md
+++ b/docs/src/js/classes/Table.md
@@ -994,6 +994,29 @@ 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
--- a/docs/src/js/globals.md
+++ b/docs/src/js/globals.md
@@ -65,6 +65,7 @@
 - [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)
@@ -101,6 +102,7 @@
 - [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)
--- a/docs/src/js/interfaces/FieldMetadataUpdate.md
+++ b/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.
--- a/docs/src/js/interfaces/UpdateFieldMetadataResult.md
+++ b/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;
+```
--- a/java/lancedb-core/pom.xml
+++ b/java/lancedb-core/pom.xml
@@ -8,7 +8,7 @@
    <parent>
      <groupId>com.lancedb</groupId>
      <artifactId>lancedb-parent</artifactId>
-      <version>0.30.0-beta.1</version>
+      <version>0.30.1-beta.1</version>
      <relativePath>../pom.xml</relativePath>
    </parent>

--- a/java/pom.xml
+++ b/java/pom.xml
@@ -6,7 +6,7 @@

    <groupId>com.lancedb</groupId>
    <artifactId>lancedb-parent</artifactId>
-    <version>0.30.0-beta.1</version>
+    <version>0.30.1-beta.1</version>
    <packaging>pom</packaging>
    <name>${project.artifactId}</name>
    <description>LanceDB Java SDK Parent POM</description>
--- a/nodejs/Cargo.toml
+++ b/nodejs/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "lancedb-nodejs"
 edition.workspace = true
-version = "0.30.0-beta.1"
+version = "0.30.1-beta.1"
 publish = false
 license.workspace = true
 description.workspace = true
--- a/nodejs/test/table.test.ts
+++ b/nodejs/test/table.test.ts
@@ -1571,6 +1571,33 @@ 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);

--- a/nodejs/lancedb/index.ts
+++ b/nodejs/lancedb/index.ts
@@ -42,6 +42,7 @@ export {
  AddResult,
  AddColumnsResult,
  AlterColumnsResult,
+  UpdateFieldMetadataResult,
  DeleteResult,
  DropColumnsResult,
  UpdateResult,
@@ -117,6 +118,7 @@ export {
  WriteProgress,
  LsmWriteSpec,
  ColumnAlteration,
+  FieldMetadataUpdate,
 } from "./table";

 export {
--- a/nodejs/lancedb/table.ts
+++ b/nodejs/lancedb/table.ts
@@ -32,6 +32,7 @@ import {
  OptimizeStats,
  TableStatistics,
  Tags,
+  UpdateFieldMetadataResult,
  UpdateResult,
  Table as _NativeTable,
 } from "./native";
@@ -508,6 +509,18 @@ 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
   *
@@ -1037,6 +1050,12 @@ 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);
  }
@@ -1203,3 +1222,19 @@ 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;
+}
--- a/nodejs/npm/darwin-arm64/package.json
+++ b/nodejs/npm/darwin-arm64/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-darwin-arm64",
-	"version": "0.30.0-beta.1",
+	"version": "0.30.1-beta.1",
 	"os": ["darwin"],
 	"cpu": ["arm64"],
 	"main": "lancedb.darwin-arm64.node",
--- a/nodejs/npm/linux-arm64-gnu/package.json
+++ b/nodejs/npm/linux-arm64-gnu/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-gnu",
-	"version": "0.30.0-beta.1",
+	"version": "0.30.1-beta.1",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-gnu.node",
--- a/nodejs/npm/linux-arm64-musl/package.json
+++ b/nodejs/npm/linux-arm64-musl/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-arm64-musl",
-	"version": "0.30.0-beta.1",
+	"version": "0.30.1-beta.1",
 	"os": ["linux"],
 	"cpu": ["arm64"],
 	"main": "lancedb.linux-arm64-musl.node",
--- a/nodejs/npm/linux-x64-gnu/package.json
+++ b/nodejs/npm/linux-x64-gnu/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-gnu",
-	"version": "0.30.0-beta.1",
+	"version": "0.30.1-beta.1",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-gnu.node",
--- a/nodejs/npm/linux-x64-musl/package.json
+++ b/nodejs/npm/linux-x64-musl/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-linux-x64-musl",
-	"version": "0.30.0-beta.1",
+	"version": "0.30.1-beta.1",
 	"os": ["linux"],
 	"cpu": ["x64"],
 	"main": "lancedb.linux-x64-musl.node",
--- a/nodejs/npm/win32-arm64-msvc/package.json
+++ b/nodejs/npm/win32-arm64-msvc/package.json
@@ -1,6 +1,6 @@
 {
  "name": "@lancedb/lancedb-win32-arm64-msvc",
-  "version": "0.30.0-beta.1",
+  "version": "0.30.1-beta.1",
  "os": [
    "win32"
  ],
--- a/nodejs/npm/win32-x64-msvc/package.json
+++ b/nodejs/npm/win32-x64-msvc/package.json
@@ -1,6 +1,6 @@
 {
 	"name": "@lancedb/lancedb-win32-x64-msvc",
-	"version": "0.30.0-beta.1",
+	"version": "0.30.1-beta.1",
 	"os": ["win32"],
 	"cpu": ["x64"],
 	"main": "lancedb.win32-x64-msvc.node",
--- a/nodejs/package-lock.json
+++ b/nodejs/package-lock.json
@@ -1,12 +1,12 @@
 {
  "name": "@lancedb/lancedb",
-  "version": "0.30.0-beta.1",
+  "version": "0.30.1-beta.1",
  "lockfileVersion": 3,
  "requires": true,
  "packages": {
    "": {
      "name": "@lancedb/lancedb",
-      "version": "0.30.0-beta.1",
+      "version": "0.30.1-beta.1",
      "cpu": [
        "x64",
        "arm64"
--- a/nodejs/package.json
+++ b/nodejs/package.json
@@ -11,7 +11,7 @@
    "ann"
  ],
  "private": false,
-  "version": "0.30.0-beta.1",
+  "version": "0.30.1-beta.1",
  "main": "dist/index.js",
  "exports": {
    ".": "./dist/index.js",
--- a/nodejs/src/table.rs
+++ b/nodejs/src/table.rs
@@ -5,8 +5,9 @@ use std::collections::HashMap;

 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, Table as LanceDbTable,
 };
 use napi::bindgen_prelude::*;
 use napi::threadsafe_function::{ThreadsafeFunction, ThreadsafeFunctionCallMode};
@@ -355,6 +356,23 @@ 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<_>>();
@@ -747,6 +765,29 @@ 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> {
@@ -987,6 +1028,19 @@ 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,
--- a/python/.bumpversion.toml
+++ b/python/.bumpversion.toml
@@ -1,5 +1,5 @@
 [tool.bumpversion]
-current_version = "0.33.0-beta.1"
+current_version = "0.33.1-beta.1"
 parse = """(?x)
    (?P<major>0|[1-9]\\d*)\\.
    (?P<minor>0|[1-9]\\d*)\\.
--- a/python/Cargo.toml
+++ b/python/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-python"
-version = "0.33.0-beta.1"
+version = "0.33.1-beta.1"
 publish = false
 edition.workspace = true
 description = "Python bindings for LanceDB"
--- a/python/python/lancedb/init.py
+++ b/python/python/lancedb/init.py
@@ -315,6 +315,15 @@ 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}")

--- a/python/python/lancedb/_lancedb.pyi
+++ b/python/python/lancedb/_lancedb.pyi
@@ -208,6 +208,9 @@ class Table:
    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,
        *,
@@ -460,6 +463,9 @@ class AddColumnsResult:
 class AlterColumnsResult:
    version: int

+class UpdateFieldMetadataResult:
+    version: int
+
 class DropColumnsResult:
    version: int

--- a/python/python/lancedb/permutation.py
+++ b/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
@@ -354,6 +355,49 @@ 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 +413,15 @@ 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 +441,7 @@ 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 +473,25 @@ 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 +549,7 @@ 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 +558,8 @@ 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 +635,10 @@ 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 +662,9 @@ 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 +673,8 @@ 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 +692,7 @@ 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 +708,28 @@ 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 +747,7 @@ class Permutation:
        """
        The number of rows in the permutation
        """
+        self._ensure_open()
        return self.reader.count_rows()

    @property
@@ -875,6 +906,7 @@ 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 +1008,7 @@ 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 +1044,11 @@ 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 +1067,11 @@ 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")
--- a/python/python/lancedb/query.py
+++ b/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,
--- a/python/python/lancedb/remote/db.py
+++ b/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."""

@@ -89,6 +148,11 @@ class RemoteDBConnection(DBConnection):
        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
@@ -111,6 +175,20 @@ class RemoteDBConnection(DBConnection):
    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,
@@ -331,7 +409,12 @@ class RemoteDBConnection(DBConnection):
            )

        table = LOOP.run(self._conn.open_table(name, namespace_path=namespace_path))
-        return RemoteTable(table, self.db_name)
+        return RemoteTable(
+            table,
+            self.db_name,
+            connection_state=self.serialize,
+            namespace_path=namespace_path,
+        )

    def clone_table(
        self,
@@ -380,7 +463,12 @@ 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(
@@ -525,7 +613,12 @@ 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):
--- a/python/python/lancedb/remote/table.py
+++ b/python/python/lancedb/remote/table.py
@@ -5,6 +5,7 @@ from datetime import timedelta
 import deprecation
 import logging
 from functools import cached_property
+import os
 from typing import (
    Any,
    Callable,
@@ -24,6 +25,7 @@ from lancedb._lancedb import (
    AddColumnsResult,
    AddResult,
    AlterColumnsResult,
+    UpdateFieldMetadataResult,
    DeleteResult,
    DropColumnsResult,
    IndexConfig,
@@ -63,14 +65,80 @@ 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
+        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)
+        table = db.open_table(self._name, namespace_path=self._namespace_path)
+        if self._checkout_version is not None:
+            table.checkout(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,
+        }
+
+    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._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})"
@@ -120,13 +188,19 @@ 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"""
@@ -777,6 +851,11 @@ class RemoteTable(Table):
    ) -> 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))

--- a/python/python/lancedb/table.py
+++ b/python/python/lancedb/table.py
@@ -89,6 +89,26 @@ from .index import lang_mapping

 BlobMode = Literal["lazy", "bytes", "descriptions"]

+_VALID_BLOB_MODES = ("lazy", "bytes", "descriptions")
+
+
+def _validate_blob_mode(blob_mode: BlobMode) -> None:
+    if blob_mode not in _VALID_BLOB_MODES:
+        modes = ", ".join(repr(mode) for mode in _VALID_BLOB_MODES)
+        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)
+
+
 _MODEL_BACKED_TOKENIZER_PREFIXES = ("jieba", "lindera")
 _MODEL_BACKED_TOKENIZER_ERRORS = (
    "unknown base tokenizer",
@@ -154,6 +174,7 @@ if TYPE_CHECKING:
        AddColumnsResult,
        AddResult,
        AlterColumnsResult,
+        UpdateFieldMetadataResult,
        DeleteResult,
        DropColumnsResult,
        LsmWriteSpec,
@@ -1799,6 +1820,29 @@ class Table(ABC):
            version: the new version number of the table after the alteration.
        """

+    @abstractmethod
+    def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        """
+        Update per-field (column) metadata.
+
+        Parameters
+        ----------
+        updates : dict
+            One or more dicts, each with:
+            - "path": str — dot-path to the field (e.g. "embedding" or "a.b.c").
+            - "metadata": dict[str, str | None] — keys to set; a value of ``None``
+              deletes that key.
+            - "replace": bool, optional — replace the field's whole metadata map
+              instead of merging (default False).
+
+        Returns
+        -------
+        UpdateFieldMetadataResult
+            version: the new table version after the update.
+        """
+
    @abstractmethod
    def drop_columns(self, columns: Iterable[str]) -> DropColumnsResult:
        """
@@ -2270,9 +2314,14 @@ class LanceTable(Table):
        -------
        pd.DataFrame
        """
-        if blob_mode == "lazy" and (
-            self._namespace_client is not None
-            or get_uri_scheme(self._dataset_path) == "memory"
+        _validate_blob_mode(blob_mode)
+        if blob_mode == "descriptions" or not _schema_has_blob_field(self.schema):
+            return self.to_arrow().to_pandas(**kwargs)
+
+        if (
+            blob_mode == "lazy"
+            and self._namespace_client is None
+            and get_uri_scheme(self._dataset_path) == "memory"
        ):
            return self.to_arrow().to_pandas(**kwargs)

@@ -3583,6 +3632,11 @@ class LanceTable(Table):
    ) -> 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))

@@ -3637,10 +3691,18 @@ class LanceTable(Table):
        """
        LOOP.run(self._table.migrate_v2_manifest_paths())

+    @deprecation.deprecated(
+        deprecated_in="0.33.1",
+        current_version=__version__,
+        details="Use update_field_metadata() instead.",
+    )
    def replace_field_metadata(self, field_name: str, new_metadata: Dict[str, str]):
        """
        Replace the metadata of a field in the schema

+        .. deprecated:: 0.33.1
+            Use :func:`update_field_metadata` instead.
+
        Parameters
        ----------
        field_name: str
@@ -4280,7 +4342,7 @@ class AsyncTable:
        can be executed with methods like [to_arrow][lancedb.query.AsyncQuery.to_arrow],
        [to_pandas][lancedb.query.AsyncQuery.to_pandas] and more.
        """
-        return AsyncQuery(self._inner.query())
+        return AsyncQuery(self._inner.query(), self)

    async def _to_lance(self, **kwargs) -> lance.LanceDataset:
        try:
@@ -4312,7 +4374,13 @@ class AsyncTable:
        -------
        pd.DataFrame
        """
-        if blob_mode == "lazy":
+        _validate_blob_mode(blob_mode)
+        if blob_mode == "descriptions" or not _schema_has_blob_field(
+            await self.schema()
+        ):
+            return (await self.to_arrow()).to_pandas(**kwargs)
+
+        if blob_mode == "lazy" and get_uri_scheme(await self.uri()) == "memory":
            return (await self.to_arrow()).to_pandas(**kwargs)
        return (await self._to_lance()).to_pandas(blob_mode=blob_mode, **kwargs)

@@ -5234,6 +5302,13 @@ class AsyncTable:
        """
        return await self._inner.alter_columns(alterations)

+    async def update_field_metadata(
+        self, *updates: dict[str, Any]
+    ) -> UpdateFieldMetadataResult:
+        """Update per-field metadata. See
+        [`Table.update_field_metadata`][lancedb.table.Table.update_field_metadata]."""
+        return await self._inner.update_field_metadata(updates)
+
    async def drop_columns(self, columns: Iterable[str]):
        """
        Drop columns from the table.
@@ -5349,7 +5424,7 @@ class AsyncTable:
        pa.RecordBatch
            A record batch containing the rows at the given offsets.
        """
-        return AsyncTakeQuery(self._inner.take_offsets(offsets))
+        return AsyncTakeQuery(self._inner.take_offsets(offsets), self)

    def take_row_ids(self, row_ids: list[int]) -> AsyncTakeQuery:
        """
@@ -5378,7 +5453,7 @@ class AsyncTable:
        AsyncTakeQuery
            A query object that can be executed to get the rows.
        """
-        return AsyncTakeQuery(self._inner.take_row_ids(row_ids))
+        return AsyncTakeQuery(self._inner.take_row_ids(row_ids), self)

    @property
    def tags(self) -> AsyncTags:
@@ -5518,12 +5593,20 @@ class AsyncTable:
        """
        await self._inner.migrate_manifest_paths_v2()

+    @deprecation.deprecated(
+        deprecated_in="0.33.1",
+        current_version=__version__,
+        details="Use update_field_metadata() instead.",
+    )
    async def replace_field_metadata(
        self, field_name: str, new_metadata: dict[str, str]
    ):
        """
        Replace the metadata of a field in the schema

+        .. deprecated:: 0.33.1
+            Use :func:`update_field_metadata` instead.
+
        Parameters
        ----------
        field_name: str
--- a/python/python/tests/test_namespace.py
+++ b/python/python/tests/test_namespace.py
@@ -76,6 +76,35 @@ class TestNamespaceConnection:
        assert len(result) == 0
        assert list(result.columns) == ["id", "vector", "text"]

+    def test_table_to_pandas_blob_lazy_through_namespace(self):
+        """Namespace-backed tables should use Lance blob-aware pandas conversion."""
+        pytest.importorskip("lance")
+        db = lancedb.connect_namespace("dir", {"root": self.temp_dir})
+        db.create_namespace(["test_ns"])
+        data = pa.table(
+            {
+                "id": pa.array([1, 2], pa.int64()),
+                "blob": pa.array([b"hello", b"world"], pa.large_binary()),
+            },
+            schema=pa.schema(
+                [
+                    pa.field("id", pa.int64()),
+                    pa.field(
+                        "blob",
+                        pa.large_binary(),
+                        metadata={"lance-encoding:blob": "true"},
+                    ),
+                ]
+            ),
+        )
+
+        table = db.create_table("blob_table", data, namespace_path=["test_ns"])
+        df = table.to_pandas(blob_mode="lazy").sort_values("id")
+
+        blob = df["blob"].iloc[0]
+        assert hasattr(blob, "readall")
+        assert blob.readall() == b"hello"
+
    def test_open_table_through_namespace(self):
        """Test opening an existing table through namespace."""
        db = lancedb.connect_namespace("dir", {"root": self.temp_dir})
--- a/python/python/tests/test_query.py
+++ b/python/python/tests/test_query.py
@@ -39,6 +39,35 @@ from utils import exception_output
 from importlib.util import find_spec


+def _blob_query_data():
+    return pa.table(
+        {
+            "id": pa.array([1, 2, 3, 4], pa.int64()),
+            "tag": pa.array(["drop", "keep", "keep", "keep"], pa.utf8()),
+            "vector": pa.array(
+                [[1.0, 0.0], [2.0, 0.0], [3.0, 0.0], [4.0, 0.0]],
+                type=pa.list_(pa.float32(), list_size=2),
+            ),
+            "blob": pa.array([b"one", b"two", b"three", b"four"], pa.large_binary()),
+        },
+        schema=pa.schema(
+            [
+                pa.field("id", pa.int64()),
+                pa.field("tag", pa.utf8()),
+                pa.field("vector", pa.list_(pa.float32(), list_size=2)),
+                pa.field(
+                    "blob", pa.large_binary(), metadata={"lance-encoding:blob": "true"}
+                ),
+            ]
+        ),
+    )
+
+
+def _assert_lazy_blob(value, expected: bytes):
+    assert hasattr(value, "readall")
+    assert value.readall() == expected
+
+
@pytest.fixture(scope="module")
 def table(tmpdir_factory) -> lancedb.table.Table:
    tmp_path = str(tmpdir_factory.mktemp("data"))
@@ -181,6 +210,216 @@ async def test_query_to_pandas_kwargs(table, table_async):
    assert async_df["id"].tolist() == [1, 2]


+@pytest.mark.parametrize("blob_mode", ["lazy", "bytes", "descriptions"])
+def test_plain_scan_query_to_pandas_blob_modes(tmp_db, blob_mode):
+    pytest.importorskip("lance")
+    table = tmp_db.create_table(
+        f"test_query_to_pandas_blob_{blob_mode}", _blob_query_data()
+    )
+
+    df = (
+        table.search()
+        .select(["id", "blob"])
+        .where("id = 1")
+        .to_pandas(blob_mode=blob_mode)
+    )
+
+    assert df["id"].tolist() == [1]
+    if blob_mode == "lazy":
+        _assert_lazy_blob(df["blob"].iloc[0], b"one")
+    elif blob_mode == "bytes":
+        assert df["blob"].tolist() == [b"one"]
+    else:
+        first = df["blob"].iloc[0]
+        assert first != b"one"
+        assert not hasattr(first, "readall")
+
+
+def test_plain_scan_query_to_pandas_blob_projection(tmp_db):
+    pytest.importorskip("lance")
+    table = tmp_db.create_table(
+        "test_query_to_pandas_blob_projection", _blob_query_data()
+    )
+
+    df = (
+        table.search()
+        .where("id >= 2")
+        .select({"id_alias": "id", "payload": "blob", "double_id": "id * 2"})
+        .limit(2)
+        .offset(1)
+        .to_pandas(blob_mode="bytes")
+    )
+
+    assert df["id_alias"].tolist() == [3, 4]
+    assert df["payload"].tolist() == [b"three", b"four"]
+    assert df["double_id"].tolist() == [6, 8]
+
+
+@pytest.mark.parametrize("blob_mode", ["bytes", "descriptions"])
+def test_plain_scan_query_to_pandas_blob_mode_does_not_collect_arrow(
+    tmp_db, monkeypatch, blob_mode
+):
+    pytest.importorskip("lance")
+    table = tmp_db.create_table(
+        "test_query_to_pandas_blob_no_arrow_collect", _blob_query_data()
+    )
+    query = table.search().where("id = 1").select(["id", "blob"])
+
+    def fail_to_arrow(*args, **kwargs):
+        raise AssertionError("to_arrow should not be called before native pandas")
+
+    monkeypatch.setattr(query, "to_arrow", fail_to_arrow)
+
+    df = query.to_pandas(blob_mode=blob_mode)
+
+    assert df["id"].tolist() == [1]
+    if blob_mode == "bytes":
+        assert df["blob"].tolist() == [b"one"]
+    else:
+        first = df["blob"].iloc[0]
+        assert first != b"one"
+        assert not hasattr(first, "readall")
+
+
+def test_plain_scan_query_to_pandas_blob_descriptions_flatten_uses_scanner(
+    tmp_db, monkeypatch
+):
+    pytest.importorskip("lance")
+    table = tmp_db.create_table(
+        "test_query_to_pandas_blob_desc_flatten", _blob_query_data()
+    )
+    query = table.search().where("id = 1").select(["id", "blob"])
+
+    def fail_to_arrow(*args, **kwargs):
+        raise AssertionError("to_arrow should not be called before scanner pandas")
+
+    monkeypatch.setattr(query, "to_arrow", fail_to_arrow)
+
+    df = query.to_pandas(blob_mode="descriptions", flatten=True)
+
+    assert df["id"].tolist() == [1]
+    assert any(column == "blob" or column.startswith("blob.") for column in df.columns)
+
+
+def test_plain_scan_query_to_pandas_scanner_state(tmp_db):
+    pytest.importorskip("lance")
+    data = _blob_query_data()
+    table = tmp_db.create_table("test_query_to_pandas_scanner_state", data.slice(0, 2))
+    table.add(data.slice(2, 2))
+
+    fragments = table.to_lance().get_fragments()
+    assert len(fragments) == 2
+
+    query = (
+        table.search()
+        .select(["id", "blob"])
+        .with_row_address()
+        .fragment_ids([fragments[1].fragment_id])
+    )
+    query_obj = query.to_query_object()
+    assert query_obj.with_row_address is True
+    assert query_obj.fragment_ids == [fragments[1].fragment_id]
+
+    df = query.to_pandas(blob_mode="descriptions")
+
+    assert df["id"].tolist() == [3, 4]
+    assert "_rowaddr" in df.columns
+    assert {rowaddr >> 32 for rowaddr in df["_rowaddr"]} == {fragments[1].fragment_id}
+
+    df_by_fragment = (
+        table.search()
+        .select(["id", "blob"])
+        .with_fragments([fragments[0]])
+        .to_pandas(blob_mode="descriptions")
+    )
+    assert df_by_fragment["id"].tolist() == [1, 2]
+
+
+@pytest.mark.asyncio
+async def test_async_plain_scan_query_to_pandas_blob_projection(tmp_db_async):
+    pytest.importorskip("lance")
+    table = await tmp_db_async.create_table(
+        "test_async_query_to_pandas_blob_projection", _blob_query_data()
+    )
+
+    lazy_df = await (
+        table.query().where("id = 1").select(["id", "blob"]).to_pandas(blob_mode="lazy")
+    )
+    assert lazy_df["id"].tolist() == [1]
+    _assert_lazy_blob(lazy_df["blob"].iloc[0], b"one")
+
+    bytes_df = await (
+        table.query()
+        .where("id >= 2")
+        .select({"id_alias": "id", "payload": "blob", "double_id": "id * 2"})
+        .limit(2)
+        .offset(1)
+        .to_pandas(blob_mode="bytes")
+    )
+    assert bytes_df["id_alias"].tolist() == [3, 4]
+    assert bytes_df["payload"].tolist() == [b"three", b"four"]
+    assert bytes_df["double_id"].tolist() == [6, 8]
+
+    desc_df = await (
+        table.query()
+        .where("id = 1")
+        .select(["blob"])
+        .to_pandas(blob_mode="descriptions")
+    )
+    first = desc_df["blob"].iloc[0]
+    assert first != b"one"
+    assert not hasattr(first, "readall")
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("blob_mode", ["bytes", "descriptions"])
+async def test_async_plain_scan_query_to_pandas_blob_mode_does_not_collect_arrow(
+    tmp_db_async, monkeypatch, blob_mode
+):
+    pytest.importorskip("lance")
+    table = await tmp_db_async.create_table(
+        "test_async_query_to_pandas_blob_no_arrow_collect", _blob_query_data()
+    )
+    query = table.query().where("id = 1").select(["id", "blob"])
+
+    async def fail_to_arrow(*args, **kwargs):
+        raise AssertionError("to_arrow should not be called before native pandas")
+
+    monkeypatch.setattr(query, "to_arrow", fail_to_arrow)
+
+    df = await query.to_pandas(blob_mode=blob_mode)
+
+    assert df["id"].tolist() == [1]
+    if blob_mode == "bytes":
+        assert df["blob"].tolist() == [b"one"]
+    else:
+        first = df["blob"].iloc[0]
+        assert first != b"one"
+        assert not hasattr(first, "readall")
+
+
+def test_vector_query_to_pandas_blob_mode_requires_native_path(tmp_db):
+    pytest.importorskip("lance")
+    table = tmp_db.create_table("test_vector_query_blob_mode", _blob_query_data())
+
+    with pytest.raises(RuntimeError, match="Lance native pandas conversion"):
+        table.search([1.0, 0.0]).select(["blob", "vector"]).limit(1).to_pandas(
+            blob_mode="lazy"
+        )
+
+
+def test_vector_query_to_pandas_blob_descriptions_requires_plain_scan(tmp_db):
+    pytest.importorskip("lance")
+    table = tmp_db.create_table(
+        "test_vector_query_blob_descriptions", _blob_query_data()
+    )
+
+    with pytest.raises(RuntimeError, match="plain scan query"):
+        table.search([1.0, 0.0]).select(["blob", "vector"]).limit(1).to_pandas(
+            blob_mode="descriptions"
+        )
+
+
 def test_order_by_plain_query(mem_db):
    table = mem_db.create_table(
        "test_order_by",
--- a/python/python/tests/test_remote_db.py
+++ b/python/python/tests/test_remote_db.py
@@ -1,12 +1,13 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
-import re
 from concurrent.futures import ThreadPoolExecutor
 import contextlib
 from datetime import timedelta
 import http.server
 import json
 import multiprocessing as mp
+import pickle
+import re
 import sys
 import threading
 import time
@@ -171,6 +172,155 @@ def test_table_len_sync():
        assert len(table) == 1


+def test_remote_connection_serializes():
+    def handler(request):
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(b'{"tables": []}')
+
+    with mock_lancedb_connection(handler) as db:
+        serialized = json.loads(db.serialize())
+        assert isinstance(serialized["client_config"], dict)
+        restored = lancedb.deserialize_conn(db.serialize())
+        assert restored.table_names() == []
+
+
+def test_remote_table_is_picklable():
+    def handler(request):
+        request.close_connection = True
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            payload = json.dumps(
+                {
+                    "version": 1,
+                    "schema": {
+                        "fields": [
+                            {"name": "id", "type": {"type": "int64"}, "nullable": False}
+                        ]
+                    },
+                }
+            )
+            request.wfile.write(payload.encode())
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"3")
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with mock_lancedb_connection(handler) as db:
+        table = db.open_table("test")
+        restored = pickle.loads(pickle.dumps(table))
+        assert restored.count_rows() == 3
+
+
+def test_remote_table_open_does_not_require_picklable_client_config():
+    from lancedb.remote import HeaderProvider
+
+    class LocalHeaderProvider(HeaderProvider):
+        def get_headers(self):
+            return {"X-Test-Header": "present"}
+
+    def handler(request):
+        request.close_connection = True
+        assert request.headers.get("X-Test-Header") == "present"
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b'{"version": 1, "schema": {"fields": []}}')
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"3")
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with http.server.HTTPServer(
+        ("localhost", 0), make_mock_http_handler(handler)
+    ) as server:
+        port = server.server_address[1]
+        handle = threading.Thread(target=server.serve_forever)
+        handle.start()
+        try:
+            db = lancedb.connect(
+                "db://dev",
+                api_key="fake",
+                host_override=f"http://localhost:{port}",
+                client_config={
+                    "retry_config": {"retries": 0},
+                    "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+                    "header_provider": LocalHeaderProvider(),
+                },
+            )
+            table = db.open_table("test")
+            assert table.count_rows() == 3
+            with pytest.raises(ValueError, match="header_provider"):
+                pickle.dumps(table)
+        finally:
+            server.shutdown()
+            handle.join()
+
+
+def test_remote_permutation_is_picklable():
+    from lancedb.permutation import Permutation
+
+    rows = list(range(10))
+
+    def handler(request):
+        request.close_connection = True
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            payload = json.dumps(
+                {
+                    "version": 1,
+                    "schema": {
+                        "fields": [
+                            {"name": "a", "type": {"type": "int64"}, "nullable": False}
+                        ]
+                    },
+                }
+            )
+            request.wfile.write(payload.encode())
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(str(len(rows)).encode())
+        elif request.path == "/v1/table/test/query/":
+            content_len = int(request.headers.get("Content-Length"))
+            body = json.loads(request.rfile.read(content_len))
+            if "filter" in body:
+                match = re.search(r"_rowoffset in \((.*?)\)", body["filter"])
+                offsets = [int(offset.strip()) for offset in match.group(1).split(",")]
+            else:
+                offsets = rows
+            table = pa.table({"a": [rows[offset] for offset in offsets]})
+
+            request.send_response(200)
+            request.send_header("Content-Type", "application/vnd.apache.arrow.file")
+            request.end_headers()
+            with pa.ipc.new_file(request.wfile, schema=table.schema) as writer:
+                writer.write_table(table)
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    with mock_lancedb_connection(handler) as db:
+        permutation = Permutation.identity(db.open_table("test"))
+        restored = pickle.loads(pickle.dumps(permutation))
+        assert restored.__getitems__([0, 2, 4]) == [{"a": 0}, {"a": 2}, {"a": 4}]
+
+
 def test_create_table_exist_ok():
    def handler(request):
        if request.path == "/v1/table/test/create/?mode=exist_ok":
@@ -1400,6 +1550,10 @@ def _remote_fork_child(port: int, queue) -> None:
    queue.put(db.table_names())


+def _remote_table_fork_child(table, queue) -> None:
+    queue.put(table.count_rows())
+
+
@pytest.mark.skipif(
    sys.platform != "linux",
    reason=(
@@ -1462,3 +1616,65 @@ def test_remote_connection_after_fork():
    finally:
        server.shutdown()
        server_thread.join()
+
+
+@pytest.mark.skipif(
+    sys.platform != "linux",
+    reason=(
+        "fork() is unavailable on Windows and unsafe on macOS "
+        "(Apple frameworks/TLS are not fork-safe)"
+    ),
+)
+def test_inherited_remote_table_reopens_after_fork():
+    def handler(request):
+        if request.path == "/v1/table/test/describe/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b'{"version": 1, "schema": {"fields": []}}')
+        elif request.path == "/v1/table/test/count_rows/":
+            request.send_response(200)
+            request.send_header("Content-Type", "application/json")
+            request.end_headers()
+            request.wfile.write(b"7")
+        else:
+            request.send_response(404)
+            request.end_headers()
+
+    server = http.server.HTTPServer(("localhost", 0), make_mock_http_handler(handler))
+    port = server.server_address[1]
+    server_thread = threading.Thread(target=server.serve_forever)
+    server_thread.start()
+    try:
+        db = lancedb.connect(
+            "db://dev",
+            api_key="fake",
+            host_override=f"http://localhost:{port}",
+            client_config={
+                "retry_config": {"retries": 0},
+                "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+            },
+        )
+        table = db.open_table("test")
+        assert table.count_rows() == 7
+
+        ctx = mp.get_context("fork")
+        queue = ctx.Queue()
+        proc = ctx.Process(target=_remote_table_fork_child, args=(table, queue))
+        proc.start()
+        proc.join(timeout=15)
+
+        if proc.is_alive():
+            proc.terminate()
+            proc.join(timeout=5)
+            if proc.is_alive():
+                proc.kill()
+                proc.join()
+            pytest.fail("Remote table hung after fork")
+
+        assert proc.exitcode == 0, f"child exited with code {proc.exitcode}"
+        assert not queue.empty(), "child produced no result"
+        assert queue.get() == 7
+    finally:
+        server.shutdown()
+        server_thread.join()
--- a/python/python/tests/test_table.py
+++ b/python/python/tests/test_table.py
@@ -26,6 +26,28 @@ from lancedb.table import LanceTable
 from pydantic import BaseModel


+def _blob_test_data():
+    return pa.table(
+        {
+            "id": pa.array([1, 2], pa.int64()),
+            "blob": pa.array([b"hello", b"world"], pa.large_binary()),
+        },
+        schema=pa.schema(
+            [
+                pa.field("id", pa.int64()),
+                pa.field(
+                    "blob", pa.large_binary(), metadata={"lance-encoding:blob": "true"}
+                ),
+            ]
+        ),
+    )
+
+
+def _assert_lazy_blob(value, expected: bytes):
+    assert hasattr(value, "readall")
+    assert value.readall() == expected
+
+
 def test_basic(mem_db: DBConnection):
    data = [
        {"vector": [3.1, 4.1], "item": "foo", "price": 10.0},
@@ -57,27 +79,30 @@ def test_table_to_pandas_default_matches_arrow(tmp_db: DBConnection):
    pd.testing.assert_frame_equal(table.to_pandas(), expected)


-def test_table_to_pandas_blob_bytes(tmp_db: DBConnection):
+def test_table_to_pandas_invalid_blob_mode_non_blob_table(tmp_db: DBConnection):
+    data = pa.table({"id": [1, 2], "text": ["one", "two"]})
+    table = tmp_db.create_table("test_to_pandas_invalid_blob_mode", data=data)
+
+    with pytest.raises(ValueError, match="blob_mode must be one of"):
+        table.to_pandas(blob_mode="invalid")
+
+
+@pytest.mark.parametrize("blob_mode", ["lazy", "bytes", "descriptions"])
+def test_table_to_pandas_blob_modes(tmp_db: DBConnection, blob_mode):
    pytest.importorskip("lance")
-    data = pa.table(
-        {
-            "id": pa.array([1, 2], pa.int64()),
-            "blob": pa.array([b"hello", b"world"], pa.large_binary()),
-        },
-        schema=pa.schema(
-            [
-                pa.field("id", pa.int64()),
-                pa.field(
-                    "blob", pa.large_binary(), metadata={"lance-encoding:blob": "true"}
-                ),
-            ]
-        ),
-    )
-    table = tmp_db.create_table("test_to_pandas_blob_bytes", data=data)
+    table = tmp_db.create_table(f"test_to_pandas_blob_{blob_mode}", _blob_test_data())

-    df = table.to_pandas(blob_mode="bytes")
+    df = table.to_pandas(blob_mode=blob_mode)

-    assert df["blob"].tolist() == [b"hello", b"world"]
+    if blob_mode == "lazy":
+        _assert_lazy_blob(df["blob"].iloc[0], b"hello")
+        _assert_lazy_blob(df["blob"].iloc[1], b"world")
+    elif blob_mode == "bytes":
+        assert df["blob"].tolist() == [b"hello", b"world"]
+    else:
+        first = df["blob"].iloc[0]
+        assert first != b"hello"
+        assert not hasattr(first, "readall")


 def test_table_to_pandas_kwargs(tmp_db: DBConnection):
@@ -93,22 +118,8 @@ def test_table_to_pandas_kwargs(tmp_db: DBConnection):
@pytest.mark.asyncio
 async def test_async_table_to_pandas_blob_bytes(tmp_db_async: AsyncConnection):
    pytest.importorskip("lance")
-    data = pa.table(
-        {
-            "id": pa.array([1, 2], pa.int64()),
-            "blob": pa.array([b"hello", b"world"], pa.large_binary()),
-        },
-        schema=pa.schema(
-            [
-                pa.field("id", pa.int64()),
-                pa.field(
-                    "blob", pa.large_binary(), metadata={"lance-encoding:blob": "true"}
-                ),
-            ]
-        ),
-    )
    table = await tmp_db_async.create_table(
-        "test_async_to_pandas_blob_bytes", data=data
+        "test_async_to_pandas_blob_bytes", data=_blob_test_data()
    )

    df = await table.to_pandas(blob_mode="bytes")
@@ -116,6 +127,19 @@ async def test_async_table_to_pandas_blob_bytes(tmp_db_async: AsyncConnection):
    assert df["blob"].tolist() == [b"hello", b"world"]


+@pytest.mark.asyncio
+async def test_async_table_to_pandas_invalid_blob_mode_non_blob_table(
+    tmp_db_async: AsyncConnection,
+):
+    table = await tmp_db_async.create_table(
+        "test_async_to_pandas_invalid_blob_mode",
+        data=pa.table({"id": [1, 2], "text": ["one", "two"]}),
+    )
+
+    with pytest.raises(ValueError, match="blob_mode must be one of"):
+        await table.to_pandas(blob_mode="invalid")
+
+
@pytest.mark.asyncio
 async def test_async_table_to_pandas_kwargs(tmp_db_async: AsyncConnection):
    pd = pytest.importorskip("pandas")
@@ -2472,6 +2496,30 @@ def test_alter_columns(mem_db: DBConnection):
    assert table.to_arrow().column_names == ["new_id"]


+def test_update_field_metadata(mem_db: DBConnection):
+    data = pa.table({"id": [0, 1], "category": ["a", "b"]})
+    table = mem_db.create_table("my_table", data=data)
+
+    res = table.update_field_metadata(
+        {"path": "category", "metadata": {"unit": "label", "pii": "false"}}
+    )
+    assert res.version == 2
+    # Arrow field metadata is bytes-keyed
+    assert table.schema.field("category").metadata == {
+        b"unit": b"label",
+        b"pii": b"false",
+    }
+
+    # merge: add a key, delete one via None, keep the rest
+    table.update_field_metadata(
+        {"path": "category", "metadata": {"source": "import", "pii": None}}
+    )
+    assert table.schema.field("category").metadata == {
+        b"unit": b"label",
+        b"source": b"import",
+    }
+
+
@pytest.mark.asyncio
 async def test_alter_columns_async(mem_db_async: AsyncConnection):
    data = pa.table({"id": [0, 1]})
--- a/python/python/tests/test_torch.py
+++ b/python/python/tests/test_torch.py
@@ -1,10 +1,15 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors

+import contextlib
 import functools
+import http.server
+import json
 import multiprocessing as mp
 import pickle
+import re
 import sys
+import threading

 import lancedb
 import pyarrow as pa
@@ -15,6 +20,107 @@ from lancedb.util import tbl_to_tensor
 torch = pytest.importorskip("torch")


+REMOTE_ROWS = list(range(100))
+
+
+def _make_mock_http_handler(handler):
+    class MockLanceDBHandler(http.server.BaseHTTPRequestHandler):
+        def do_GET(self):
+            handler(self)
+
+        def do_POST(self):
+            handler(self)
+
+    return MockLanceDBHandler
+
+
+def _remote_schema_payload():
+    return {
+        "version": 1,
+        "schema": {
+            "fields": [
+                {"name": "a", "type": {"type": "int64"}, "nullable": False},
+            ]
+        },
+    }
+
+
+def _offsets_from_filter(filter_sql: str | None) -> list[int]:
+    if filter_sql is None:
+        return REMOTE_ROWS
+    match = re.search(r"_rowoffset in \((.*?)\)", filter_sql)
+    if match is None:
+        return REMOTE_ROWS
+    raw_offsets = match.group(1).strip()
+    if raw_offsets == "":
+        return []
+    return [int(offset.strip()) for offset in raw_offsets.split(",")]
+
+
+def _remote_dataset_handler(request):
+    request.close_connection = True
+    if request.path == "/v1/table/test/describe/":
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(json.dumps(_remote_schema_payload()).encode())
+    elif request.path == "/v1/table/test/count_rows/":
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.end_headers()
+        request.wfile.write(str(len(REMOTE_ROWS)).encode())
+    elif request.path == "/v1/table/test/query/":
+        content_len = int(request.headers.get("Content-Length"))
+        body = json.loads(request.rfile.read(content_len))
+        offsets = _offsets_from_filter(body.get("filter"))
+        requested_columns = body.get("columns") or ["a"]
+        if isinstance(requested_columns, dict):
+            requested_columns = list(requested_columns)
+
+        data = {}
+        for column in requested_columns:
+            if column == "a":
+                data[column] = [REMOTE_ROWS[offset] for offset in offsets]
+            elif column == "_rowoffset":
+                data[column] = offsets
+            elif column == "_rowid":
+                data[column] = offsets
+
+        table = pa.table(data)
+        request.send_response(200)
+        request.send_header("Content-Type", "application/vnd.apache.arrow.file")
+        request.end_headers()
+        with pa.ipc.new_file(request.wfile, schema=table.schema) as writer:
+            writer.write_table(table)
+    else:
+        request.send_response(404)
+        request.end_headers()
+
+
+@contextlib.contextmanager
+def _remote_dataset_table():
+    with http.server.ThreadingHTTPServer(
+        ("localhost", 0), _make_mock_http_handler(_remote_dataset_handler)
+    ) as server:
+        port = server.server_address[1]
+        handle = threading.Thread(target=server.serve_forever)
+        handle.start()
+        try:
+            db = lancedb.connect(
+                "db://dev",
+                api_key="fake",
+                host_override=f"http://localhost:{port}",
+                client_config={
+                    "retry_config": {"retries": 0},
+                    "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+                },
+            )
+            yield db.open_table("test")
+        finally:
+            server.shutdown()
+            handle.join()
+
+
 def _open_native_table(uri: str, table_name: str):
    """Top-level connection factory used by the explicit-factory pickle test.

@@ -107,6 +213,39 @@ def test_permutation_dataloader_multiprocessing(tmp_db):
    assert seen == 1000


+def test_remote_table_dataloader_multiprocessing():
+    with _remote_dataset_table() as table:
+        dataloader = torch.utils.data.DataLoader(
+            table,
+            collate_fn=tbl_to_tensor,
+            batch_size=10,
+            num_workers=2,
+            multiprocessing_context="spawn",
+        )
+        seen = 0
+        for batch in dataloader:
+            assert batch.size(0) == 1
+            assert batch.size(1) == 10
+            seen += batch.size(1)
+        assert seen == len(REMOTE_ROWS)
+
+
+def test_remote_permutation_dataloader_multiprocessing():
+    with _remote_dataset_table() as table:
+        permutation = Permutation.identity(table)
+        dataloader = torch.utils.data.DataLoader(
+            permutation,
+            batch_size=10,
+            num_workers=2,
+            multiprocessing_context="spawn",
+        )
+        seen = 0
+        for batch in dataloader:
+            assert batch["a"].size(0) == 10
+            seen += batch["a"].size(0)
+        assert seen == len(REMOTE_ROWS)
+
+
 def test_permutation_pickle_with_connection_factory(tmp_path):
    """When the user provides a connection_factory, pickling should round-trip
    through that factory rather than introspecting the connection URI. Useful
@@ -171,6 +310,35 @@ def _multiworker_dataloader_target(db_uri: str, result_queue):
    result_queue.put(count)


+def _remote_multiworker_dataloader_target(port: int, result_queue):
+    import lancedb
+    from lancedb.permutation import Permutation
+
+    db = lancedb.connect(
+        "db://dev",
+        api_key="fake",
+        host_override=f"http://localhost:{port}",
+        client_config={
+            "retry_config": {"retries": 0},
+            "timeout_config": {"connect_timeout": 2, "read_timeout": 2},
+        },
+    )
+    table = db.open_table("test")
+    permutation = Permutation.identity(table)
+
+    dataloader = torch.utils.data.DataLoader(
+        permutation,
+        batch_size=10,
+        num_workers=2,
+        multiprocessing_context="fork",
+    )
+    count = 0
+    for batch in dataloader:
+        assert batch["a"].size(0) == 10
+        count += 1
+    result_queue.put(count)
+
+
@pytest.mark.skipif(
    sys.platform != "linux",
    reason=(
@@ -208,3 +376,46 @@ def test_permutation_dataloader_fork_workers(tmp_path):
    assert proc.exitcode == 0, f"child exited with code {proc.exitcode}"
    assert not queue.empty(), "child produced no batches"
    assert queue.get() == 100
+
+
+@pytest.mark.skipif(
+    sys.platform != "linux",
+    reason=(
+        "fork() is unavailable on Windows and unsafe on macOS "
+        "(Apple frameworks/TLS are not fork-safe)"
+    ),
+)
+def test_remote_permutation_dataloader_fork_workers():
+    with http.server.ThreadingHTTPServer(
+        ("localhost", 0), _make_mock_http_handler(_remote_dataset_handler)
+    ) as server:
+        port = server.server_address[1]
+        handle = threading.Thread(target=server.serve_forever)
+        handle.start()
+        try:
+            ctx = mp.get_context("spawn")
+            queue = ctx.Queue()
+            proc = ctx.Process(
+                target=_remote_multiworker_dataloader_target,
+                args=(port, queue),
+            )
+            proc.start()
+            proc.join(timeout=30)
+
+            if proc.is_alive():
+                proc.terminate()
+                proc.join(timeout=5)
+                if proc.is_alive():
+                    proc.kill()
+                    proc.join()
+                pytest.fail(
+                    "Remote permutation hung when iterated in a fork-based "
+                    "DataLoader worker"
+                )
+
+            assert proc.exitcode == 0, f"child exited with code {proc.exitcode}"
+            assert not queue.empty(), "child produced no batches"
+            assert queue.get() == 10
+        finally:
+            server.shutdown()
+            handle.join()
--- a/python/src/lib.rs
+++ b/python/src/lib.rs
@@ -16,7 +16,7 @@ use query::{FTSQuery, HybridQuery, Query, VectorQuery};
 use session::Session;
 use table::{
    AddColumnsResult, AddResult, AlterColumnsResult, DeleteResult, DropColumnsResult, LsmWriteSpec,
-    MergeResult, Table, UpdateResult,
+    MergeResult, Table, UpdateFieldMetadataResult, UpdateResult,
 };

 pub mod arrow;
@@ -50,6 +50,7 @@ pub fn _lancedb(_py: Python, m: &Bound<'_, PyModule>) -> PyResult<()> {
    m.add_class::<RecordBatchStream>()?;
    m.add_class::<AddColumnsResult>()?;
    m.add_class::<AlterColumnsResult>()?;
+    m.add_class::<UpdateFieldMetadataResult>()?;
    m.add_class::<AddResult>()?;
    m.add_class::<MergeResult>()?;
    m.add_class::<LsmWriteSpec>()?;
--- a/python/src/table.rs
+++ b/python/src/table.rs
@@ -16,12 +16,12 @@ use arrow::{
    pyarrow::{FromPyArrow, PyArrowType, ToPyArrow},
 };
 use lancedb::table::{
-    AddDataMode, ColumnAlteration, Duration, NewColumnTransform, OptimizeAction, OptimizeOptions,
-    Table as LanceDbTable,
+    AddDataMode, ColumnAlteration, Duration, FieldMetadataUpdate, NewColumnTransform,
+    OptimizeAction, OptimizeOptions, Table as LanceDbTable,
 };
 use pyo3::{
    Bound, FromPyObject, Py, PyAny, PyRef, PyResult, Python,
-    exceptions::{PyKeyError, PyRuntimeError, PyValueError},
+    exceptions::{PyRuntimeError, PyValueError},
    pyclass, pymethods,
    types::{IntoPyDict, PyAnyMethods, PyDict, PyDictMethods},
 };
@@ -357,6 +357,27 @@ impl From<lancedb::table::AlterColumnsResult> for AlterColumnsResult {
    }
 }

+#[pyclass(get_all, from_py_object)]
+#[derive(Clone, Debug)]
+pub struct UpdateFieldMetadataResult {
+    pub version: u64,
+}
+
+#[pymethods]
+impl UpdateFieldMetadataResult {
+    pub fn __repr__(&self) -> String {
+        format!("UpdateFieldMetadataResult(version={})", self.version)
+    }
+}
+
+impl From<lancedb::table::UpdateFieldMetadataResult> for UpdateFieldMetadataResult {
+    fn from(result: lancedb::table::UpdateFieldMetadataResult) -> Self {
+        Self {
+            version: result.version,
+        }
+    }
+}
+
 #[pyclass(get_all, from_py_object)]
 #[derive(Clone, Debug)]
 pub struct DropColumnsResult {
@@ -1102,31 +1123,57 @@ impl Table {
        field_name: String,
        metadata: &Bound<'_, PyDict>,
    ) -> PyResult<Bound<'a, PyAny>> {
-        let mut new_metadata = HashMap::<String, String>::new();
-        for (column_name, value) in metadata.into_iter() {
-            let key: String = column_name.extract()?;
-            let value: String = value.extract()?;
-            new_metadata.insert(key, value);
+        // Deprecated: forwards to the update_field_metadata path (replace mode).
+        let mut update = FieldMetadataUpdate::new(field_name).replace();
+        for (key, value) in metadata.into_iter() {
+            update = update.set(key.extract::<String>()?, value.extract::<String>()?);
        }

        let inner = self_.inner_ref()?.clone();
        future_into_py(self_.py(), async move {
-            let native_tbl = inner
-                .as_native()
-                .ok_or_else(|| PyValueError::new_err("This cannot be run on a remote table"))?;
-            let schema = native_tbl.manifest().await.infer_error()?.schema;
-            let field = schema
-                .field(&field_name)
-                .ok_or_else(|| PyKeyError::new_err(format!("Field {} not found", field_name)))?;
-
-            native_tbl
-                .replace_field_metadata(vec![(field.id as u32, new_metadata)])
-                .await
-                .infer_error()?;
-
+            inner.update_field_metadata(&[update]).await.infer_error()?;
            Ok(())
        })
    }
+
+    pub fn update_field_metadata<'a>(
+        self_: PyRef<'a, Self>,
+        updates: Vec<Bound<PyDict>>,
+    ) -> PyResult<Bound<'a, PyAny>> {
+        let updates = updates
+            .iter()
+            .map(|update| {
+                let path: String = update
+                    .get_item("path")?
+                    .ok_or_else(|| PyValueError::new_err("Missing path"))?
+                    .extract()?;
+                let mut field_update = FieldMetadataUpdate::new(path);
+                if let Some(metadata) = update.get_item("metadata")? {
+                    let metadata_dict = metadata.cast::<PyDict>()?;
+                    for (key, value) in metadata_dict.iter() {
+                        let key: String = key.extract()?;
+                        if value.is_none() {
+                            field_update = field_update.remove(key);
+                        } else {
+                            field_update = field_update.set(key, value.extract::<String>()?);
+                        }
+                    }
+                }
+                if let Some(replace) = update.get_item("replace")?
+                    && replace.extract::<bool>()?
+                {
+                    field_update = field_update.replace();
+                }
+                Ok(field_update)
+            })
+            .collect::<PyResult<Vec<_>>>()?;
+
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            let result = inner.update_field_metadata(&updates).await.infer_error()?;
+            Ok(UpdateFieldMetadataResult::from(result))
+        })
+    }
 }

 #[derive(FromPyObject)]
--- a/rust/lancedb/Cargo.toml
+++ b/rust/lancedb/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.30.0-beta.1"
+version = "0.30.1-beta.1"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
--- a/rust/lancedb/src/remote/table.rs
+++ b/rust/lancedb/src/remote/table.rs
@@ -18,13 +18,13 @@ use crate::index::waiter::wait_for_index;
 use crate::query::{QueryFilter, QueryRequest, Select, VectorQueryRequest};
 use crate::table::AddColumnsResult;
 use crate::table::AddResult;
-use crate::table::AlterColumnsResult;
 use crate::table::DeleteResult;
 use crate::table::DropColumnsResult;
 use crate::table::MergeResult;
 use crate::table::Tags;
 use crate::table::UpdateResult;
 use crate::table::query::create_multi_vector_plan;
+use crate::table::{AlterColumnsResult, FieldMetadataUpdate, UpdateFieldMetadataResult};
 use crate::table::{AnyQuery, Filter, Predicate, PreprocessingOutput, TableStatistics};
 use crate::utils::background_cache::BackgroundCache;
 use crate::utils::{
@@ -1968,6 +1968,35 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
        Ok(result)
    }

+    async fn update_field_metadata(
+        &self,
+        updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        self.check_mutable().await?;
+        let body = serde_json::json!({ "updates": updates });
+        let request = self
+            .client
+            .post(&format!(
+                "/v1/table/{}/update_field_metadata/",
+                self.identifier
+            ))
+            .json(&body);
+        let (request_id, response) = self.send(request, true).await?;
+        let response = self.check_table_response(&request_id, response).await?;
+        let body = response.text().await.err_to_http(request_id.clone())?;
+
+        let result: UpdateFieldMetadataResult =
+            serde_json::from_str(&body).map_err(|e| Error::Http {
+                source: format!("Failed to parse update_field_metadata response: {}", e).into(),
+                request_id,
+                status_code: None,
+            })?;
+
+        self.invalidate_schema_cache();
+        self.track_write_version(result.version);
+        Ok(result)
+    }
+
    async fn drop_columns(&self, columns: &[&str]) -> Result<DropColumnsResult> {
        self.check_mutable().await?;
        let body = serde_json::json!({ "columns": columns });
@@ -2261,6 +2290,7 @@ mod tests {

    use crate::remote::client::{ClientConfig, RetryConfig};
    use crate::table::AddDataMode;
+    use crate::table::FieldMetadataUpdate;

    use arrow::{array::AsArray, compute::concat_batches, datatypes::Int32Type};
    use arrow_array::{Int32Array, RecordBatch, RecordBatchIterator, record_batch};
@@ -6460,4 +6490,25 @@ mod tests {
        assert!(!headers.contains_key("x-lancedb-min-version"));
        assert!(!headers.contains_key("x-lancedb-min-timestamp"));
    }
+
+    #[tokio::test]
+    async fn test_update_field_metadata() {
+        let table = Table::new_with_handler("my_table", |request| {
+            assert_eq!(request.method(), "POST");
+            assert_eq!(
+                request.url().path(),
+                "/v1/table/my_table/update_field_metadata/"
+            );
+            http::Response::builder()
+                .status(200)
+                .body(r#"{"version": 7, "fields": {"category": {"unit": "label"}}}"#)
+                .unwrap()
+        });
+
+        let result = table
+            .update_field_metadata(&[FieldMetadataUpdate::new("category").set("unit", "label")])
+            .await
+            .unwrap();
+        assert_eq!(result.version, 7);
+    }
 }
--- a/rust/lancedb/src/table.rs
+++ b/rust/lancedb/src/table.rs
@@ -91,7 +91,10 @@ pub use lance::dataset::scanner::DatasetRecordBatchStream;
 use lance::dataset::statistics::DatasetStatisticsExt;
 pub use lance_index::optimize::OptimizeOptions;
 pub use optimize::{CompactionOptions, OptimizeAction, OptimizeStats};
-pub use schema_evolution::{AddColumnsResult, AlterColumnsResult, DropColumnsResult};
+pub use schema_evolution::{
+    AddColumnsResult, AlterColumnsResult, DropColumnsResult, FieldMetadataUpdate,
+    UpdateFieldMetadataResult,
+};
 use serde_with::skip_serializing_none;
 pub use update::{UpdateBuilder, UpdateResult};

@@ -660,6 +663,19 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
            message: "create_insert_exec not implemented".to_string(),
        })
    }
+    /// Update per-field metadata. Merges into existing metadata by default;
+    /// [`FieldMetadataUpdate::remove`] deletes a key and
+    /// [`FieldMetadataUpdate::replace`] swaps the field's whole map.
+    ///
+    /// The default returns `NotSupported`; Lance-backed and remote tables override it.
+    async fn update_field_metadata(
+        &self,
+        _updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        Err(Error::NotSupported {
+            message: "update_field_metadata is not supported on this table type".into(),
+        })
+    }
 }

 /// A Table is a collection of strong typed Rows.
@@ -1340,6 +1356,14 @@ impl Table {
        self.inner.alter_columns(alterations).await
    }

+    /// Update per-field metadata (merges by default).
+    pub async fn update_field_metadata(
+        &self,
+        updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        self.inner.update_field_metadata(updates).await
+    }
+
    /// Remove columns from the table.
    pub async fn drop_columns(&self, columns: &[&str]) -> Result<DropColumnsResult> {
        self.inner.drop_columns(columns).await
@@ -2580,6 +2604,7 @@ impl NativeTable {
    ///   field id and the second element is a hashmap of metadata key-value
    ///   pairs.
    ///
+    #[deprecated(since = "0.33.1", note = "Use `update_field_metadata` instead")]
    pub async fn replace_field_metadata(
        &self,
        new_values: impl IntoIterator<Item = (u32, HashMap<String, String>)>,
@@ -2886,6 +2911,13 @@ impl BaseTable for NativeTable {
        schema_evolution::execute_alter_columns(self, alterations).await
    }

+    async fn update_field_metadata(
+        &self,
+        updates: &[FieldMetadataUpdate],
+    ) -> Result<UpdateFieldMetadataResult> {
+        schema_evolution::execute_update_field_metadata(self, updates).await
+    }
+
    async fn drop_columns(&self, columns: &[&str]) -> Result<DropColumnsResult> {
        schema_evolution::execute_drop_columns(self, columns).await
    }
@@ -3136,7 +3168,6 @@ pub struct FragmentSummaryStats {
 #[cfg(test)]
 #[allow(deprecated)]
 mod tests {
-    use std::collections::HashMap;
    use std::sync::Arc;
    use std::sync::atomic::{AtomicBool, Ordering};
    use std::time::Duration;
@@ -4449,10 +4480,10 @@ mod tests {
            Some(&"test_val2_update".to_string())
        );

-        let mut new_field_metadata = HashMap::<String, String>::new();
-        new_field_metadata.insert("test_field_key1".into(), "test_field_val1".into());
        native_tbl
-            .replace_field_metadata(vec![(field.id as u32, new_field_metadata)])
+            .update_field_metadata(&[
+                FieldMetadataUpdate::new("i").set("test_field_key1", "test_field_val1")
+            ])
            .await
            .unwrap();

--- a/rust/lancedb/src/table/datafusion/udtf/fts.rs
+++ b/rust/lancedb/src/table/datafusion/udtf/fts.rs
@@ -870,10 +870,8 @@ mod tests {
            .await
            .unwrap();

-        assert_eq!(
-            result.iter().map(|batch| batch.num_rows()).sum::<usize>(),
-            0
-        );
+        // Should return empty or nearly empty result
+        assert!(result[0].num_rows() <= 1);
    }

    #[tokio::test]
--- a/rust/lancedb/src/table/schema_evolution.rs
+++ b/rust/lancedb/src/table/schema_evolution.rs
@@ -10,6 +10,7 @@

 use lance::dataset::{ColumnAlteration, NewColumnTransform};
 use serde::{Deserialize, Serialize};
+use std::collections::HashMap;

 use super::NativeTable;
 use crate::Result;
@@ -44,6 +45,52 @@ pub struct DropColumnsResult {
    pub version: u64,
 }

+/// A single field's metadata update, addressed by dot-path.
+///
+/// Merges into the field's existing metadata by default. Use [`Self::remove`] to
+/// delete a key, or [`Self::replace`] to swap the field's entire metadata map.
+#[derive(Debug, Clone, PartialEq, Eq, Default, Serialize)]
+pub struct FieldMetadataUpdate {
+    /// Dot-separated path to the field (e.g. `"embedding"` or `"address.zip"`).
+    pub path: String,
+    /// Keys to set (`Some`) or delete (`None`).
+    pub metadata: HashMap<String, Option<String>>,
+    /// If `true`, replace the field's entire metadata map instead of merging.
+    pub replace: bool,
+}
+
+impl FieldMetadataUpdate {
+    pub fn new(path: impl Into<String>) -> Self {
+        Self {
+            path: path.into(),
+            metadata: HashMap::new(),
+            replace: false,
+        }
+    }
+
+    pub fn set(mut self, key: impl Into<String>, value: impl Into<String>) -> Self {
+        self.metadata.insert(key.into(), Some(value.into()));
+        self
+    }
+
+    pub fn remove(mut self, key: impl Into<String>) -> Self {
+        self.metadata.insert(key.into(), None);
+        self
+    }
+
+    pub fn replace(mut self) -> Self {
+        self.replace = true;
+        self
+    }
+}
+
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default)]
+pub struct UpdateFieldMetadataResult {
+    /// The commit version associated with the operation.
+    #[serde(default)]
+    pub version: u64,
+}
+
 /// Internal implementation of the add columns logic.
 ///
 /// Adds new columns to the table using the provided transforms.
@@ -90,6 +137,32 @@ pub(crate) async fn execute_drop_columns(
    Ok(DropColumnsResult { version })
 }

+/// Internal implementation of the update field metadata logic.
+///
+/// Merges or replaces per-field metadata, addressing fields by dot-path.
+pub(crate) async fn execute_update_field_metadata(
+    table: &NativeTable,
+    updates: &[FieldMetadataUpdate],
+) -> Result<UpdateFieldMetadataResult> {
+    table.dataset.ensure_mutable()?;
+    let mut dataset = (*table.dataset.get().await?).clone();
+
+    let mut builder = dataset.update_field_metadata();
+    for update in updates {
+        let entries = update.metadata.iter().map(|(k, v)| (k.clone(), v.clone()));
+        builder = if update.replace {
+            builder.replace(&update.path, entries)?
+        } else {
+            builder.update(&update.path, entries)?
+        };
+    }
+    builder.await?;
+
+    let version = dataset.version().version;
+    table.dataset.update(dataset);
+    Ok(UpdateFieldMetadataResult { version })
+}
+
 #[cfg(test)]
 mod tests {
    use arrow_array::{Int32Array, StringArray, record_batch};
@@ -97,6 +170,7 @@ mod tests {
    use futures::TryStreamExt;
    use lance::dataset::ColumnAlteration;

+    use super::FieldMetadataUpdate;
    use crate::connect;
    use crate::query::{ExecutableQuery, QueryBase, Select};
    use crate::table::NewColumnTransform;
@@ -610,4 +684,46 @@ mod tests {
        let v4 = table.version().await.unwrap();
        assert_eq!(drop_result.version, v4);
    }
+
+    #[tokio::test]
+    async fn test_update_field_metadata() {
+        let conn = connect("memory://").execute().await.unwrap();
+        let batch = record_batch!(
+            ("id", Int32, [1, 2, 3]),
+            ("category", Utf8, ["A", "B", "C"])
+        )
+        .unwrap();
+        let table = conn
+            .create_table("test_update_field_metadata", batch)
+            .execute()
+            .await
+            .unwrap();
+
+        // Set metadata on a field.
+        table
+            .update_field_metadata(&[FieldMetadataUpdate::new("category")
+                .set("unit", "label")
+                .set("pii", "false")])
+            .await
+            .unwrap();
+        let schema = table.schema().await.unwrap();
+        let field = schema.field_with_name("category").unwrap();
+        assert_eq!(
+            field.metadata().get("unit").map(String::as_str),
+            Some("label")
+        );
+
+        // Merge: add a key, delete one, keep the rest.
+        table
+            .update_field_metadata(&[FieldMetadataUpdate::new("category")
+                .set("source", "import")
+                .remove("pii")])
+            .await
+            .unwrap();
+        let schema = table.schema().await.unwrap();
+        let md = schema.field_with_name("category").unwrap().metadata();
+        assert_eq!(md.get("unit").map(String::as_str), Some("label")); // preserved
+        assert_eq!(md.get("source").map(String::as_str), Some("import")); // added
+        assert!(!md.contains_key("pii")); // deleted
+    }
 }
Author	SHA1	Message	Date
BubbleCal	01c338e51c	fix(python): route blob query pandas through scanner	2026-06-03 22:59:25 +08:00
Lance Release	9483b534af	Bump version: 0.30.1-beta.0 → 0.30.1-beta.1	2026-06-03 11:17:37 +00:00
Lance Release	ac3411e81e	Bump version: 0.33.1-beta.0 → 0.33.1-beta.1	2026-06-03 11:16:51 +00:00
Yang Cen	6f18eb4cce	feat(python): support blob modes in query to_pandas (#3487 ) ## Feature - What is the new feature? - Adds `blob_mode` support to sync and async Python query `to_pandas()` APIs. - Enables plain scan queries to return blob columns as lazy `BlobFile` objects, raw bytes, or blob descriptions. - Lets namespace-backed local tables use Lance native blob-aware pandas conversion for lazy blobs. - Why do we need this feature? - Table and Lance dataset/scanner APIs already support blob-aware pandas conversion, but LanceDB query builders did not expose that capability. - Geneva and other callers should be able to use query-level `to_pandas(blob_mode=...)` without manually constructing Lance scanners. - How does it work? - Plain scan queries route through Lance scanner native `to_pandas(blob_mode=...)`, preserving filter, projection, limit, offset, row id, and alias/expression projection behavior. - Non-native query shapes keep existing Arrow fallback semantics and raise a clear error when they return blob columns with `blob_mode="lazy"` or `blob_mode="bytes"`. - Focused tests cover table/query blob modes, filter/select/limit/offset/alias query cases, async query behavior, vector-query error boundaries, and namespace-backed lazy blobs. ## Validation - `cd python && .venv/bin/maturin develop --uv --extras tests,dev --profile dev` - `cd python && uv run --frozen --no-sync pytest python/tests/test_table.py::test_table_to_pandas_blob_modes python/tests/test_table.py::test_async_table_to_pandas_blob_bytes python/tests/test_query.py::test_plain_scan_query_to_pandas_blob_modes python/tests/test_query.py::test_plain_scan_query_to_pandas_blob_projection python/tests/test_query.py::test_async_plain_scan_query_to_pandas_blob_projection python/tests/test_query.py::test_vector_query_to_pandas_blob_mode_requires_native_path python/tests/test_namespace.py::TestNamespaceConnection::test_table_to_pandas_blob_lazy_through_namespace -q` - `cd python && uv run --frozen --no-sync ruff format --check .` - `cd python && uv run --frozen --no-sync ruff check .` - `git diff --check`	2026-06-03 19:15:44 +08:00
Brendan Clement	379684391e	feat: deprecate replace_field_metadata for update_field_metadata (#3484 ) ### Summary Deprecates the Python replace_field_metadata (on Table and AsyncTable) in favor of update_field_metadata. Mirrors Lance, which already deprecated Dataset.replace_field_metadata for update_field_metadata. Stacked on top of #3482 as this was a follow-up task after adding update_field_metadata	2026-06-02 14:02:22 -07:00
Brendan Clement	d065be0474	feat: add update_field_metadata to edit per-field metadata (#3482 ) ### Summary Adds update_field_metadata to the client SDK (Rust core, Python, and TypeScript) so clients can edit per-field (column) Arrow metadata (schema.fields[].metadata) ### Testing - added unit tests - ran E2E against a local server on both local and remote tables (set → merge → delete), across Python sync/async and TypeScript ### Next steps - deprecate replace_field_metadata in the python lancedb favor of this (typescript didn't have replace_field_metadata method). This matches Lance's API direction (Lance already deprecated replace_field_metadata for update_field_metadata)	2026-06-02 07:00:00 -07:00
Xuanwo	7b874905fd	ci: move Lance dependency bump flow into skill (#3475 ) Moves the Lance dependency bump process into an in-repository skill so local agents and GitHub Actions share the same workflow definition. The update workflow is now an explicit, optional-tag entrypoint; latest-release resolution, duplicate PR handling, Java/Rust dependency updates, and Sophon follow-up are documented in the skill and backed by a small deterministic helper.	2026-06-02 16:05:37 +08:00
Xuanwo	a327044e2f	feat(python): support remote tables in PyTorch dataloaders (#3432 ) This PR makes remote LanceDB tables usable from PyTorch multiprocessing workers. Remote tables now carry enough safe JSON connection state to reopen themselves after pickle/spawn or fork, and permutations lazily rebuild their reader from restored tables instead of trying to reuse process-local handles. This addresses the remote-table gap in the PyTorch dataset path while preserving the explicit connection factory escape hatch for custom worker-side credential loading or non-serializable header providers. Validated with targeted remote table, permutation, and PyTorch DataLoader tests.	2026-06-02 15:38:28 +08:00
Lance Release	f20ec99dec	Bump version: 0.30.0-beta.1 → 0.30.1-beta.0	2026-06-01 12:41:45 +00:00
Lance Release	60f961584c	Bump version: 0.33.0-beta.1 → 0.33.1-beta.0	2026-06-01 12:41:02 +00:00
Xuanwo	ac699d7ecf	chore: bump lance to 7.2.0-beta.3 (#3471 ) This updates the workspace Lance dependencies from `v7.1.0-beta.4` to `v7.2.0-beta.3` and refreshes `Cargo.lock`. The lockfile now points at Lance commit `7c070f760fa8e24c8015cb2afbd22c5e6b7898e8` and includes the transitive dependency updates required by the new beta.	2026-06-01 20:40:07 +08:00