fence remote function refresh submissions

## Problem

The canonical Function wire values and typed remote Job contract do not
yet provide a Python authoring surface or catalog client, so users
cannot package a scalar callable, register it, or reopen the exact
immutable Function version.

## Behavior

This adds scalar-only `@udf` authoring with deterministic annotation or
explicit Arrow schema validation, content-addressed Python artifacts,
and an internal scalar-to-Arrow-batch adapter descriptor. Registration
payloads model non-secret environment values and secret names only.

Remote connections can submit `create_function_async` and receive a
typed `Job<FunctionVersion>`, then reopen that exact version by name and
version ID. Synchronous connections can call `create_function` to submit
and wait for the immutable version in one operation. Local Function
catalog operations return a stable `NotSupported` error. Shared
Rust/Python golden payloads and mocked catalog responses freeze the
request, typed terminal result, and exact lookup contract.

## Validation

- Rust formatting, remote check, clippy, and focused LDB-1/LDB-2 tests
- Python formatting, lint, and focused LDB-1/LDB-2 tests
- Python API documentation build

.agents/skills/README.md

@@ -5,7 +5,3 @@ 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.
-
-The `lancedb` skill lives in the `plugins/lancedb` plugin (see `plugins/lancedb/skills/lancedb`)
-so it can be installed via the plugin marketplaces (`.claude-plugin/marketplace.json` and
-`.agents/plugins/marketplace.json`); the `lancedb` entry here is a symlink into that plugin.

.agents/skills/lancedb

@@ -1 +0,0 @@
-../../plugins/lancedb/skills/lancedb

.cargo/config.toml

@@ -9,6 +9,18 @@ debug = true
 codegen-units = 16
 lto = "thin"
 
+[profile.release-no-lto]
+inherits = "release"
+debug = true
+lto = false
+# Prioritize compile time when LTO is not relevant to the measurement.
+codegen-units = 16
+
+[profile.bench]
+inherits = "release"
+lto = "thin"
+codegen-units = 16
+
 [target.'cfg(all())']
 rustflags = [
     "-Wclippy::all",

.github/workflows/python.yml

@@ -229,7 +229,8 @@ jobs:
       # Make sure wheels are not included in the Rust cache
       - name: Delete wheels
         run: rm -rf target/wheels
-  pydantic1x:
+  min-deps:
+    name: "Minimum dependencies"
     timeout-minutes: 60
     runs-on: "ubuntu-24.04"
     defaults:
@@ -259,8 +260,7 @@ jobs:
           save-if: ${{ github.ref == 'refs/heads/main' }}
       - name: Install lancedb
         run: |
-          pip install "pydantic<2"
-          pip install pyarrow==16
+          pip install "pydantic==2.7.4" "pyarrow==16"
           pip install --extra-index-url https://pypi.fury.io/lance-format/ --extra-index-url https://pypi.fury.io/lancedb/ -e .[tests]
       - name: Run tests
         run: pytest -m "not slow and not s3_test" -x -v --durations=30 python/tests

AGENTS.md

@@ -18,6 +18,9 @@ Common commands:
 * Run specific test: `cargo test --quiet --features remote -p <package_name> --test <test_name>`
 * Lint: `cargo clippy --quiet --features remote --tests --examples`
 * Format Rust: `cargo fmt --all`
+* Use repository-defined Cargo profiles instead of ad hoc LTO overrides.
+* Use `release-with-debug` for benchmarks and profiling so optimized builds keep debug symbols without a rebuild.
+* Use `release-no-lto` only for local debugging, IO-bound benchmarks, or compile-time-sensitive performance investigation where LTO would not affect the measured bottleneck.
 * Format Python: `ruff format .`
 * Lint Python: `ruff check .`
 * Bootstrap Python dev env: `cd python && uv run --extra tests --extra dev maturin develop --extras tests,dev`

Cargo.lock

@@ -1740,9 +1740,9 @@ dependencies = [
 
 [[package]]
 name = "cmov"
-version = "0.5.3"
+version = "0.5.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3f88a43d011fc4a6876cb7344703e297c71dda42494fee094d5f7c76bf13f746"
+checksum = "0c9ea0ac24bc397ab3c98583a3c9ba74fa56b09a4449bbe172b9b1ddb016027a"
 
 [[package]]
 name = "colorchoice"

Cargo.toml

@@ -27,6 +27,7 @@ lance-testing = { "version" = "=11.0.0-beta.15", "tag" = "v11.0.0-beta.15", "git
 lance-datafusion = { "version" = "=11.0.0-beta.15", "tag" = "v11.0.0-beta.15", "git" = "https://github.com/lance-format/lance.git" }
 lance-encoding = { "version" = "=11.0.0-beta.15", "tag" = "v11.0.0-beta.15", "git" = "https://github.com/lance-format/lance.git" }
 lance-arrow = { "version" = "=11.0.0-beta.15", "tag" = "v11.0.0-beta.15", "git" = "https://github.com/lance-format/lance.git" }
+lancedb = { path = "rust/lancedb", default-features = false }
 ahash = "0.8"
 # Note that this one does not include pyarrow
 arrow = { version = "58.0.0", optional = false }
@@ -39,6 +40,7 @@ arrow-schema = "58.0.0"
 arrow-select = "58.0.0"
 arrow-cast = "58.0.0"
 async-trait = "0"
+bytes = "1"
 datafusion = { version = "54.0.0", default-features = false }
 datafusion-catalog = "54.0.0"
 datafusion-common = { version = "54.0.0", default-features = false }
@@ -65,7 +67,12 @@ url = "2"
 num-traits = "0.2"
 regex = "1.10"
 semver = "1.0.25"
-chrono = "0.4"
+serde = "1"
+serde_json = "1"
+tempfile = "3.5.0"
+tokio = { version = "1.23", features = ["rt-multi-thread", "sync"] }
+uuid = { version = "1.7.0", features = ["v4"] }
+chrono = { version = "0.4", default-features = false, features = ["clock"] }
 
 [profile.ci]
 debug = "line-tables-only"

ci/check_breaking_changes.py

@@ -2,6 +2,7 @@
 Check whether there are any breaking changes in the PRs between the base and head commits.
 If there are, assert that we have incremented the minor version.
 """
+
 import argparse
 import os
 from packaging.version import parse
@@ -27,7 +28,7 @@ if __name__ == "__main__":
     else:
         print("No breaking changes found.")
         exit(0)
-    
+
     last_stable_version = parse(args.last_stable_version)
     current_version = parse(args.current_version)
     if current_version.minor <= last_stable_version.minor:

ci/check_lance_release.py

@@ -1,5 +1,6 @@
 #!/usr/bin/env python3
 """Determine whether a newer Lance tag exists and expose results for CI."""
+
 from __future__ import annotations
 
 import argparse
@@ -36,8 +37,16 @@ class SemVer:
     prerelease: Tuple[Union[int, str], ...]
 
     def __lt__(self, other: "SemVer") -> bool:  # pragma: no cover - simple comparison
-        if (self.major, self.minor, self.patch) != (other.major, other.minor, other.patch):
-            return (self.major, self.minor, self.patch) < (other.major, other.minor, other.patch)
+        if (self.major, self.minor, self.patch) != (
+            other.major,
+            other.minor,
+            other.patch,
+        ):
+            return (self.major, self.minor, self.patch) < (
+                other.major,
+                other.minor,
+                other.patch,
+            )
         if self.prerelease == other.prerelease:
             return False
         if not self.prerelease:
@@ -142,7 +151,9 @@ def read_current_version(repo_root: Path) -> str:
         deps = data["workspace"]["dependencies"]
         entry = deps["lance"]
     except KeyError as exc:  # pragma: no cover - configuration guard
-        raise RuntimeError("Failed to locate workspace.dependencies.lance in Cargo.toml") from exc
+        raise RuntimeError(
+            "Failed to locate workspace.dependencies.lance in Cargo.toml"
+        ) from exc
 
     if isinstance(entry, str):
         raw_version = entry

ci/mock_openai.py

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 """A zero-dependency mock OpenAI embeddings API endpoint for testing purposes."""
+
 import argparse
 import json
 import http.server
@@ -22,11 +23,13 @@ class MockOpenAIRequestHandler(http.server.BaseHTTPRequestHandler):
 
         data = []
         for i in range(num_inputs):
-            data.append({
-                "object": "embedding",
-                "embedding": [0.1] * 1536,
-                "index": i,
-            })
+            data.append(
+                {
+                    "object": "embedding",
+                    "embedding": [0.1] * 1536,
+                    "index": i,
+                }
+            )
 
         response = {
             "object": "list",
@@ -35,7 +38,7 @@ class MockOpenAIRequestHandler(http.server.BaseHTTPRequestHandler):
             "usage": {
                 "prompt_tokens": 0,
                 "total_tokens": 0,
-            }
+            },
         }
 
         self.send_response(200)

ci/semver_sort.py

@@ -7,6 +7,7 @@ from packaging.version import parse, InvalidVersion
 
 if __name__ == "__main__":
     import argparse
+
     parser = argparse.ArgumentParser()
     parser.add_argument("prefix", default="v")
     args = parser.parse_args()

ci/set_lance_version.py

@@ -22,7 +22,7 @@ def run_command(command: str) -> str:
 def get_latest_stable_version() -> str:
     version_line = run_command("cargo info lance | grep '^version:'")
     # Example output: "version: 0.35.0 (latest 0.37.0)"
-    match = re.search(r'\(latest ([0-9.]+)\)', version_line)
+    match = re.search(r"\(latest ([0-9.]+)\)", version_line)
     if match:
         return match.group(1)
     # Fallback: use the first version after 'version:'
@@ -69,7 +69,7 @@ def extract_default_features(line: str) -> bool:
     """
     import re
 
-    match = re.search(r'default-features\s*=\s*false', line)
+    match = re.search(r"default-features\s*=\s*false", line)
     return match is not None
 
 
@@ -104,7 +104,7 @@ def dict_to_toml_line(package_name: str, config: dict) -> str:
             # This shouldn't happen with our current usage
             parts.append(f'"{key}" = {json.dumps(value)}')
 
-    return f'{package_name} = {{ {", ".join(parts)} }}\n'
+    return f"{package_name} = {{ {', '.join(parts)} }}\n"
 
 
 def update_cargo_toml(line_updater):

ci/validate_stable_lance.py

@@ -12,7 +12,7 @@ with open("Cargo.toml", "rb") as f:
             elif isinstance(dep, dict):
                 # Version doesn't have the beta tag in it, so we instead look
                 # at the git tag.
-                version = dep.get('tag', dep.get('version'))
+                version = dep.get("tag", dep.get("version"))
             else:
                 raise ValueError("Unexpected type for dependency: " + str(dep))
 

deny.toml

@@ -177,6 +177,11 @@ multiple-versions = "warn"
 # Wildcard version requirements (`foo = "*"`) are a footgun — they let any
 # future release in without review. Ban them outright.
 wildcards = "deny"
+# Lint every dependency declared by a workspace member against the shared
+# `[workspace.dependencies]` table: any crate used by more than one member must
+# go through `workspace = true`, and entries nothing uses are an error. This
+# keeps versions from drifting between the core crate and the bindings.
+workspace-dependencies = { duplicates = "deny", unused = "deny" }
 # Internal workspace crates reference each other via `path = "..."`, which
 # cargo-deny sees as a wildcard version. That's fine for private workspace
 # members (not published to crates.io), so allow it specifically for paths.

docs/requirements.txt

@@ -5,5 +5,5 @@ mkdocs-autorefs>=0.5,<=1.0
 mkdocstrings[python]>=0.24,<1.0
 griffe>=0.40,<1.0
 mkdocs-render-swagger-plugin>=0.1.0
-pydantic>=2.0,<3.0
-mkdocs-redirects>=1.2.0
+pydantic>=2.7.4,<3
+mkdocs-redirects>=1.2.0

docs/src/python/python.md

@@ -54,6 +54,54 @@ listing a storage directory.
 
 ::: lancedb.LsmWriteSpec
 
+## Functions and Jobs
+
+::: lancedb.functions.FunctionArtifact
+
+::: lancedb.functions.FunctionParameter
+
+::: lancedb.functions.FunctionResultField
+
+::: lancedb.functions.FunctionOutput
+
+::: lancedb.functions.FunctionSignature
+
+::: lancedb.functions.PythonEnvironmentSpec
+
+::: lancedb.functions.udf
+
+::: lancedb.functions.UdfDefinition
+
+::: lancedb.functions.FunctionRegistrationRequest
+
+::: lancedb.functions.FunctionArtifactRequest
+
+::: lancedb.functions.FunctionArtifactContent
+
+::: lancedb.functions.PythonAdapterSpec
+
+::: lancedb.functions.FunctionVersion
+
+::: lancedb.functions.PythonRuntimeSpec
+
+::: lancedb.functions.FunctionVersionRef
+
+::: lancedb.functions.ApplicationInput
+
+::: lancedb.functions.FunctionApplication
+
+::: lancedb.functions.InputBinding
+
+::: lancedb.functions.OutputMapping
+
+::: lancedb.functions.FunctionBinding
+
+::: lancedb.functions.RefreshColumnResult
+
+::: lancedb.job.Job
+
+::: lancedb.job.AsyncJob
+
 ## Expressions
 
 Type-safe expression builder for filters and projections. Use these instead
@@ -153,8 +201,9 @@ The same option is available on `lancedb.tokenize(...)` and the deprecated
 ```python
 import lancedb
 
-tokens = list(lancedb.tokenize("acme makes searchable data",
-                               custom_stop_words=["acme"]))
+tokens = list(
+    lancedb.tokenize("acme makes searchable data", custom_stop_words=["acme"])
+)
 ```
 
 ::: lancedb.tokenize

nodejs/Cargo.toml

@@ -16,12 +16,12 @@ crate-type = ["cdylib"]
 async-trait.workspace = true
 arrow-ipc.workspace = true
 arrow-array.workspace = true
-arrow-buffer = "58.0.0"
+arrow-buffer.workspace = true
 half.workspace = true
 arrow-schema.workspace = true
 env_logger.workspace = true
 futures.workspace = true
-lancedb = { path = "../rust/lancedb", default-features = false }
+lancedb.workspace = true
 lance-namespace.workspace = true
 napi = { version = "3.8.3", default-features = false, features = [
     "napi9",
@@ -29,8 +29,8 @@ napi = { version = "3.8.3", default-features = false, features = [
     "chrono_date",
     "serde-json",
 ] }
-chrono = { version = "0.4", default-features = false, features = ["clock"] }
-serde_json = "1"
+chrono.workspace = true
+serde_json.workspace = true
 napi-derive = "3.5.2"
 # Prevent dynamic linking of lzma, which comes from datafusion
 lzma-sys = { version = "0.1", features = ["static"] }

plugins/lancedb/skills/lancedb/SKILL.md

@@ -20,18 +20,16 @@ Do NOT assume local-only table helpers exist on remote tables. If the user asks
 
 1. Identify the SDK: Python, TypeScript, or both.
 2. Identify the table mode: local/embedded OSS, remote Enterprise/Cloud, or portable across both. If the user says "LanceDB Enterprise", choose the remote table path. If the task involves jobs in any way (listing, inspecting, creating, or canceling jobs), it is always the remote path and requires a remote server connection — see "Connecting to the LanceDB remote server" below before doing anything else.
-3. Read the matching language branch before writing or changing code:
-   - Python patterns: `references/python/patterns.md`
-   - Python API quick reference: `references/python/api_reference.md`
-   - Python performance guidance: `references/python/performance.md`
-   - TypeScript patterns: `references/typescript/patterns.md`
-   - TypeScript API quick reference: `references/typescript/api_reference.md`
-   - TypeScript performance guidance: `references/typescript/performance.md`
+3. Read the matching topic reference before writing or changing code:
    - Column metadata authoring (both SDKs): `references/column_metadata.md`
    - Branch operations (both SDKs): `references/branch_ops.md`
    - Remote server connection resolution (jobs, raw REST): `references/remote_connect.md`
    - Job operations REST API (list/describe/cancel/query_events): `references/remote_jobs.md`
-4. Start with `patterns.md` for the selected SDK. Read `api_reference.md` when choosing method names or return collectors. Read `performance.md` when the task involves ingestion, indexing, filtering, query tuning, diagnostics, or large datasets. Read `column_metadata.md` when the task is documenting, tagging, classifying, or grouping table columns (field descriptions, `lancedb:tag:*` tags, logical column families). Read `branch_ops.md` when the task involves branch lifecycle (list/create/delete), writing to a non-main branch, or verifying a change stayed off main. Read `remote_connect.md` when the task involves jobs or direct REST access to an Enterprise deployment, and `remote_jobs.md` for the job REST methods themselves (list, describe, cancel, query_events).
+
+   There is no bundled per-language guide. For exact method names, signatures, and options, look them up in the canonical sources instead of relying on memory:
+   - Python: `docs/src/python/python.md` (the hand-maintained API reference) and the source under `python/python/lancedb/` when working inside the LanceDB repo; otherwise <https://lancedb.github.io/lancedb/python/python/>.
+   - TypeScript: the generated typedoc under `docs/src/js/` and the source under `nodejs/lancedb/` when working inside the LanceDB repo; otherwise <https://lancedb.github.io/lancedb/js/globals/>.
+4. Apply the SDK invariants in "Per-SDK Invariants" below. Read `column_metadata.md` when the task is documenting, tagging, classifying, or grouping table columns (field descriptions, `lancedb:tag:*` tags, logical column families). Read `branch_ops.md` when the task involves branch lifecycle (list/create/delete), writing to a non-main branch, or verifying a change stayed off main. Read `remote_connect.md` when the task involves jobs or direct REST access to an Enterprise deployment, and `remote_jobs.md` for the job REST methods themselves (list, describe, cancel, query_events).
 5. For Python schemas, favor Pydantic models and validate records before writing. Use PyArrow schemas when Arrow-native, streaming, or highly dynamic data makes them materially better suited.
 6. Prefer `search()` or `query()` builders with explicit `select()` and `limit()` for reads.
 7. Avoid table-level full materialization in remote or portable code. This is the main local-vs-remote read pitfall.
@@ -54,6 +52,23 @@ The unsafe pattern is table-level or unbounded collection, plus local-only datas
 - Python: `table.to_pandas()`, `table.to_arrow()`, `table.to_polars()`; `table.to_lance()` is local/OSS-only dataset access, not materialization
 - TypeScript: `await table.toArrow()`, `await table.query().toArray()` without `limit()`
 
+## Per-SDK Invariants
+
+Python:
+
+- Result collectors: default to `.to_list()` (plain dicts, no extra dependency) or `.to_arrow()` (PyArrow ships with LanceDB). Use `.to_pandas()` / `.to_polars()` only when the project already declares that dependency — do not assume pandas or polars is installed.
+- Plain scans differ by client: the sync client has no `.query()` method — use `table.search()` with no argument; the async client uses `await async_table.query()`.
+
+TypeScript:
+
+- Collect bounded results with `.toArray()` (objects) or `.toArrow()` (Arrow) after `select()` and `limit()`.
+- For large reads, stream batches instead of collecting: `for await (const batch of table.query().where(...).select(...).limit(...)) { ... }`.
+
+Both SDKs:
+
+- Ingest in bulk or in batches of thousands of rows; never write per-row in a loop — each write creates a version and fragment, slowing ingestion and later queries.
+- Build a vector index once brute-force search is too slow (rule of thumb: beyond roughly 100K vectors locally), and scalar indexes for filtered columns and merge/upsert keys. Use index defaults unless the task states recall/latency requirements.
+
 ## Enterprise: never drop-then-reuse the same table name
 
 LanceDB Enterprise/Cloud splits a **control plane** (DDL: create/drop/rename) from a **data plane** (query nodes that serve reads). Query nodes cache the resolved dataset for a table name for up to `table_cache_ttl` — **default 300 seconds (5 minutes)**. After you drop or overwrite a table, the control plane updates immediately but the data plane keeps serving the *old* dataset until that cache entry expires. During the window the two planes disagree.

plugins/lancedb/skills/lancedb/references/python/api_reference.md

@@ -1,138 +0,0 @@
-# Python API Reference
-
-Quick method reference for Python LanceDB code. Cross-check source for non-trivial claims.
-
-## Connect
-
-If you're connecting to a remote database, use this:
-```python
-import lancedb
-
-db = lancedb.connect("db://my-db", api_key=api_key, host_override=host_override) # remote
-```
-(values may be found in LANCEDB_API_KEY and LANCEDB_HOST_OVERRIDE, either in env vars or a .env file)
-
-If you're connecting to a local table using OSS LanceDB, use this:
-```python
-db = lancedb.connect("./camelot-db") # local/OSS
-```
-If you're not sure which, or if you can't find the api_key or host_override params, ask the user.
-
-**Place the local database directory next to the script/entrypoint that opens it** (i.e. resolve the path relative to the script, `Path(__file__).parent / "camelot-db"`), not buried under a shared `data/` folder. The Lance dataset is the database, not a data file — keeping it beside its code makes ownership obvious and paths stable regardless of the working directory the script is launched from.
-
-**Do not name the directory `lancedb`** (e.g. `./lancedb`, `./data/lancedb`). It collides with the imported `lancedb` package name, which is confusing to read and easy to shadow in scripts. Give it a name derived from the repo or dataset with a clear prefix/suffix — for example `./<dataset>-db`, `./<repo>_lancedb`, or `./vectordb`.
-
-Async:
-
-```python
-db = await lancedb.connect_async("./camelot-db")
-```
-
-## Table Reads
-
-| Task | Preferred API |
-| --- | --- |
-| Vector search | `table.search(query_vector).limit(k)` |
-| Full scan with filters/projection (sync) | `table.search().where(...).select(...).limit(...)` |
-| Full scan with filters/projection (async) | `table.query().where(...).select(...).limit(...)` |
-| Filter | `.where("col > 10")` |
-| Projection | `.select(["id", "text"])` |
-| Bound result count | `.limit(20)` |
-| Collect bounded result as Python objects (default, no extra deps) | `.to_list()` on query/search result |
-| Collect bounded result as Arrow (default, `pyarrow` always available) | `.to_arrow()` on query/search result |
-| Collect bounded result as pandas (only if project uses pandas) | `.to_pandas()` on query/search result |
-| Collect bounded result as Polars (only if project uses polars) | `.to_polars()` on query/search result |
-
-## Sync vs Async Scan API
-
-The plain-scan entry point differs between the sync and async clients. **Verified against `lancedb` 0.34.0** — re-check if the pinned version changes:
-
-- **Sync** (`lancedb.connect(...)`): the table has **no `.query()` method**. Use `.search()` with no argument for a plain scan; it returns a query builder that supports `.where()`, `.select()`, `.limit()`, and the `.to_list()` / `.to_arrow()` / `.to_pandas()` / `.to_polars()` collectors.
-  ```python
-  rows = table.search().where("status = 'ready'").select(["id", "text"]).limit(20).to_list()
-  ```
-- **Async** (`lancedb.connect_async(...)`): the table has **both** `.query()` and `.search()`. Use `.query()` for a plain scan.
-  ```python
-  rows = await async_table.query().where("status = 'ready'").select(["id", "text"]).limit(20).to_list()
-  ```
-
-Do not call `table.query()` on a sync table — it raises `AttributeError`.
-
-## Local vs Remote Table Methods
-
-| API | Local table | Remote table | Agent guidance |
-| --- | --- | --- | --- |
-| `table.search(...)` | Yes | Yes | Preferred read path (sync + async) |
-| `table.query()` | Async only | Async only | Sync scan path is `table.search()`; `.query()` is the async scan builder |
-| `table.to_pandas()` | Yes | No / unsafe for portability | Avoid in portable code |
-| `table.to_arrow()` | Yes | No / unsafe for portability | Avoid in portable code |
-| `table.to_polars()` | Yes | No / unsafe for portability | Avoid in portable code |
-| `table.to_lance()` | Yes | No | Local/OSS escape hatch only |
-
-## Indexes
-
-Use `create_index(...)` for vector indexes and modern index configs. Use scalar indexes for filtered or merge keys.
-
-Common calls:
-
-```python
-table.create_index("vector")
-table.create_scalar_index("status")
-table.create_fts_index("text")
-```
-
-Check source docs before specifying advanced index config names or parameters.
-
-## Filtering And Recall Knobs
-
-```python
-table.search(query_vector).where("status = 'ready'")  # pre-filter by default
-table.search(query_vector).where("status = 'ready'", prefilter=False)
-table.search(query_vector).limit(10).refine_factor(20)
-table.search(query_vector).limit(10).nprobes(50)
-```
-
-Use post-filtering only when fewer than `limit` results are acceptable.
-
-## Diagnostics
-
-```python
-print(table.search(query_vector).where("year > 2000").limit(10).analyze_plan())
-print(table.index_stats("vector_idx"))
-```
-
-Use these before changing indexes or search tuning.
-
-## Column (Field) Metadata
-
-```python
-schema = table.schema                       # sync property; async: await table.schema()
-meta = schema.field("category").metadata    # dict[bytes, bytes] — Arrow metadata is bytes-keyed
-res = table.update_field_metadata(          # varargs: one dict per field; works local + remote
-    {"path": "category", "metadata": {"lancedb:description": "...", "lancedb:tag:field_type": "label"}}
-)
-res.version  # new table version
-```
-
-Merges by default; a `None` value deletes that key; `"replace": True` swaps the whole map. Nested fields use dot-paths (`"a.b.c"`). `replace_field_metadata` is deprecated. See `references/column_metadata.md` for key conventions (`lancedb:description`, `lancedb:tag:<name>`, `lancedb:logical-column`) and the authoring workflow.
-
-## Branches
-
-```python
-table.branches.list()                          # non-main branches; {} = only main
-exp = table.branches.create("exp")             # fork off main -> handle scoped to the branch
-wip = table.branches.checkout("wip")           # existing branch -> scoped handle (version= pins read-only)
-wip = db.open_table("t", branch="wip")         # or open scoped directly
-table.branches.delete("stale")                 # removes only the branch pointer
-table.current_branch()                         # None = main
-```
-
-There is no global switch — scoping is per table handle: any read/write on a branch handle lands on that branch; the original handle keeps targeting main. See `references/branch_ops.md` for the model and isolation checks.
-
-## Maintenance
-
-```python
-table.optimize()
-```
-
-Call this after every successful local/OSS ingestion. It handles compaction, cleanup of old versions according to retention, and index optimization. Do not add this for LanceDB Enterprise/Cloud remote tables; Enterprise handles compaction and cleanup automatically from cluster configuration.

plugins/lancedb/skills/lancedb/references/python/patterns.md

@@ -1,173 +0,0 @@
-# Python Patterns
-
-Use these patterns when writing Python code with `lancedb`.
-
-## Before Writing Code
-
-Choose the output type from what the project actually depends on. **Do not assume `pandas` or `polars` is installed** — they are heavy dependencies that many LanceDB projects do not use. `pyarrow`, by contrast, ships as a LanceDB dependency and is always available, so it is a safe default to lean on.
-
-Default output (after applying `select()` and `limit()`):
-
-- **Python objects**: `.to_list()` — a list of dicts, no extra dependencies. Prefer this for scripts, examples, and agent-generated code unless there is a reason to do otherwise.
-- **PyArrow**: `.to_arrow()` — a `pyarrow.Table`, when the surrounding code is Arrow-native or you need columnar/zero-copy handoff.
-
-Only reach for a DataFrame when the project *already* declares that dependency:
-
-- Pandas projects (pandas in `pyproject.toml`/requirements): `.to_pandas()`.
-- Polars projects (polars declared): `.to_polars()`.
-
-If unsure, check the dependency manifest or the imports in surrounding files. When in doubt, use `.to_list()` or `.to_arrow()`.
-
-## Schema Design and Validation
-
-Favor `LanceModel` and Pydantic validation for Python schemas. They keep field
-types readable, validate source records before a write, and map directly to a
-LanceDB schema. Use `Vector(dimension)` for fixed-size vectors:
-
-```python
-from lancedb.pydantic import LanceModel, Vector
-
-class Document(LanceModel):
-    id: int
-    text: str
-    vector: Vector(384, nullable=False)
-
-rows = [Document.model_validate(row) for row in source_rows]
-table = db.create_table("documents", schema=Document)
-table.add(rows)
-```
-
-Use PyArrow schemas instead when the pipeline is already Arrow-native, needs
-record-batch streaming, or has runtime schema requirements that would make a
-Pydantic model harder to understand. Declare Pydantic as a direct project
-dependency when application code imports it, even if LanceDB also depends on it.
-
-## Recommended Patterns
-
-### Bounded search or query
-
-Use this for application reads, examples, notebooks, and agent-generated scripts:
-
-```python
-results = (
-    table.search(query_vector)
-    .where("status = 'ready'")
-    .select(["id", "text"])
-    .limit(20)
-    .to_list()  # or .to_arrow(); .to_pandas()/.to_polars() only if the project uses them
-)
-```
-
-Why: `search()` works across local and remote tables and on both the sync and async clients. `select()` avoids fetching unused columns. `limit()` prevents accidental full-table reads. `.to_list()` and `.to_arrow()` avoid assuming pandas/polars is installed (see "Before Writing Code").
-
-For a **plain scan** (no query vector), the entry point differs by client:
-
-```python
-# Sync client: no .query() method — use .search() with no argument.
-rows = table.search().where("status = 'ready'").select(["id", "text"]).limit(20).to_list()
-
-# Async client: use .query().
-rows = await async_table.query().where("status = 'ready'").select(["id", "text"]).limit(20).to_list()
-```
-
-`table.query()` on a sync table raises `AttributeError` (verified on `lancedb` 0.34.0). See the "Sync vs Async Scan API" section in `api_reference.md`.
-
-### Bounded query result conversion
-
-It is fine to collect bounded query/search results:
-
-```python
-arrow_table = table.search().select(["id"]).limit(100).to_arrow()  # sync plain scan
-rows = table.search(query_vector).limit(10).to_list()
-df = table.search(query_vector).limit(10).to_pandas()  # only if pandas is a project dep
-```
-
-### Local-only Lance dataset API
-
-`table.to_lance()` does not itself materialize the full dataset. It returns the underlying `lance.LanceDataset`, making the table accessible through the PyLance dataset API. Use it when the task is explicitly local/OSS and needs Lance dataset methods not exposed by LanceDB:
-
-```python
-# Local/OSS only: RemoteTable does not expose table.to_lance().
-ds = table.to_lance()
-for batch in ds.to_batches(columns=["id", "text"], batch_size=10_000):
-    process(batch)
-```
-
-### Async Python
-
-Keep the same shape and bound the result before collecting:
-
-```python
-results = await (
-    async_table.query()
-    .where("status = 'ready'")
-    .select(["id", "text"])
-    .limit(20)
-    .to_list()  # or .to_arrow()
-)
-```
-
-## Anti-Patterns
-
-**Avoid the following anti-patterns in your code.**
-
-### Table-level full materialization
-
-Avoid whole-table collectors in portable or large-table code:
-
-```python
-df = table.to_pandas()
-arrow_table = table.to_arrow()
-polars_df = table.to_polars()
-```
-
-Why: local tables expose these whole-table collectors, but remote tables intentionally do not — a remote production table can be far larger than a local development table, so it is easy to accidentally pull the entire table into memory.
-
-`table.to_lance()` is different: it is not a full materialization call, but it is still local/OSS-only and should not appear in code meant to run against remote Enterprise tables.
-
-### Unbounded result collection
-
-Avoid query/search collection without a meaningful limit:
-
-```python
-rows = table.search().to_list()               # unbounded plain scan
-rows = table.search(query_vector).to_list()   # unbounded vector search
-```
-
-Prefer `select(...).limit(...)` before collecting; for large reads, stream in batches instead.
-
-### Per-row writes
-
-Avoid loops that write one row per call:
-
-```python
-for row in rows:
-    table.add([row])  # one commit + fragment per row
-```
-
-Each `add()` creates a new version and fragment. Pass the whole batch in a single call, or chunk very large inputs:
-
-```python
-table.add(rows)  # single commit
-# for very large inputs, add batches of several thousand rows
-```
-
-After the final successful write to an embedded OSS table, call
-`table.optimize()`. Skip this for Enterprise/Cloud tables because their
-maintenance is automatic.
-
-### Drop-then-reuse the same table name (Enterprise/Cloud)
-
-Avoid dropping or overwriting a remote table and then reusing that name right away:
-
-```python
-db.drop_table("my_table")
-table = db.create_table("my_table", data=rows)          # reads 500 for ~5 min
-table = db.create_table("my_table", data=rows, mode="overwrite")  # same problem
-```
-
-Why: Enterprise/Cloud splits DDL (control plane) from query serving (data plane). The data plane caches the dataset behind a table name for up to `table_cache_ttl` (default 300s / 5 min), so after a drop/overwrite the DDL succeeds but queries against the reused name return `500 Internal Server Error` until the cache expires — and a fresh `describe` may still show the old schema. Instead, write to a **fresh name**, use `list_tables()` and fail if it already exists, then `rename_table(fresh, final)` onto the final name only after the old table's drop has propagated (~5 min). See the "Enterprise: never drop-then-reuse the same table name" section in `SKILL.md`. Local/OSS tables have no separate data plane — overwrite freely there.
-
-### Guessing performance fixes
-
-Avoid changing `nprobes`, `refine_factor`, or index types before checking the query plan and index stats. Diagnose first, then tune one knob at a time.

plugins/lancedb/skills/lancedb/references/python/performance.md

@@ -1,131 +0,0 @@
-# Python Performance Guidance
-
-Use this when writing Python code that ingests data, queries large tables, builds indexes, or investigates latency.
-
-## Ingestion
-
-### Recommended: validate schemas and records with Pydantic
-
-Favor `LanceModel` for readable Python schema definitions and validate source
-records before writing. Use PyArrow directly for Arrow-native or streaming
-pipelines where it is the clearer representation.
-
-```python
-from lancedb.pydantic import LanceModel, Vector
-
-class Document(LanceModel):
-    id: int
-    text: str
-    vector: Vector(384, nullable=False)
-
-rows = [Document.model_validate(row) for row in source_rows]
-table = db.create_table("documents", schema=Document)
-table.add(rows)
-```
-
-### Recommended: bulk ingestion for materialized data
-
-```python
-table.add(arrow_table)
-table.add(df)
-table.add(pa.dataset("data/", format="parquet"))
-```
-
-For very large initial loads, create the table empty first, then call `add(...)`. Passing data directly to `create_table(name, data)` can skip the auto-parallel write path.
-
-### Recommended: iterator ingestion for generated or streamed data
-
-```python
-def batches():
-    for raw in source:
-        vectors = model.encode(raw["text"])
-        yield pa.RecordBatch.from_pydict({**raw, "vector": vectors})
-
-table.add(batches())
-```
-
-Use chunks of several thousand rows or more when practical. Tiny batches and per-row writes create many small fragments.
-
-### Anti-pattern: per-row `add()`
-
-```python
-for row in rows:
-    table.add([row])
-```
-
-Each call creates a version and fragment. This slows ingestion and later queries.
-
-## Indexing
-
-- Build a vector index once brute-force vector search becomes too slow. As a rule of thumb, local brute force is fine below roughly 100K vectors; beyond that, build an index.
-- Use `IVF_PQ` as the general-purpose default. Enterprise builds this automatically.
-- Use scalar indexes for filtered columns and merge/upsert keys.
-- Use `BTREE` for mostly distinct numeric/string/temporal columns, `BITMAP` for booleans and low-cardinality columns, and `LABEL_LIST` for list membership queries.
-- Keep full-text defaults unless phrase queries require position data.
-
-## Querying
-
-Always be explicit:
-
-```python
-table.search(query_vector).select(["id", "title"]).limit(20)
-```
-
-- `select()` reduces bytes read and transferred.
-- `limit()` prevents accidental full-table materialization.
-- Pre-filtering is the default and guarantees returned rows satisfy the predicate.
-- Use post-filtering only when fewer than `limit` results are acceptable.
-
-## Recall Tuning
-
-Tune one knob at a time:
-
-- Quantized indexes: raise `refine_factor` to rescore more candidates on full vectors.
-- HNSW-backed indexes: raise `ef`; start around `1.5 * k`, increase toward `10 * k` if recall is short.
-- IVF candidate breadth: `nprobes` is auto-tuned; override only when a selective pre-filter leaves too few neighbors.
-
-## Maintenance
-
-After every successful embedded OSS/local ingestion, call `table.optimize()`.
-Do not add this to LanceDB Enterprise/Cloud remote table code; remote compaction
-and cleanup are handled automatically based on the Enterprise cluster
-configuration.
-
-Why local maintenance is needed:
-
-- Frequent writes can create many small fragments. Queries then need to scan across more files, which can increase latency.
-- Updates, deletes, and appends create new table versions. Old versions are retained for time travel and rollback, which can grow disk usage.
-- Indexes may have newly added rows that are not yet fully optimized into the index structure.
-
-For local/OSS tables, run `optimize()` after the final successful ingestion
-write. Also run it after later batches of update/delete operations or on a
-regular maintenance schedule:
-
-```python
-table.optimize()
-```
-
-If the user wants more aggressive local disk cleanup, pass a shorter cleanup retention window:
-
-```python
-from datetime import timedelta
-
-table.optimize(cleanup_older_than=timedelta(days=1))
-```
-
-Do not use very short cleanup windows when the application depends on time travel, rollback, or old versions.
-
-## Diagnostics
-
-Before changing code or indexes, inspect:
-
-```python
-print(table.search(query_vector).where("year > 2000").limit(10).analyze_plan())
-print(table.index_stats("vector_idx"))
-```
-
-Look for high scan bytes, missing indexes, fragmented data, and unindexed rows.
-
-## Python Multiprocessing
-
-When using multiprocessing, use `spawn` rather than `fork`. LanceDB is multi-threaded internally, and `fork` plus a multi-threaded process is unsafe.

plugins/lancedb/skills/lancedb/references/typescript/api_reference.md

@@ -1,105 +0,0 @@
-# TypeScript API Reference
-
-Quick method reference for TypeScript LanceDB code. Cross-check source for non-trivial claims.
-
-## Connect
-
-```typescript
-import * as lancedb from "@lancedb/lancedb";
-
-const db = await lancedb.connect("./camelot-db");
-```
-
-**Place the local database directory next to the script/entrypoint that opens it** (resolve the path relative to the module, e.g. via `import.meta.dirname` / `__dirname`), not buried under a shared `data/` folder. The Lance dataset is the database, not a data file — keeping it beside its code makes ownership obvious and paths stable regardless of the working directory the script is launched from.
-
-**Do not name the directory `lancedb`** (e.g. `./lancedb`, `./data/lancedb`). It collides with the imported `lancedb` package/namespace, which is confusing to read. Give it a name derived from the repo or dataset with a clear prefix/suffix — for example `./<dataset>-db`, `./<repo>_lancedb`, or `./vectordb`.
-
-Remote connections use `db://...` plus Enterprise/Cloud credentials and deployment settings. Check current source/docs for exact connection options.
-
-## Table Reads
-
-| Task | Preferred API |
-| --- | --- |
-| Vector search | `table.search(queryVector).limit(k)` |
-| Full scan with filters/projection | `table.query().where(...).select(...).limit(...)` |
-| Filter | `.where("col > 10")` |
-| Projection | `.select(["id", "text"])` |
-| Bound result count | `.limit(20)` |
-| Collect bounded result as objects | `.toArray()` on query/search result |
-| Collect bounded result as Arrow | `.toArrow()` on query/search result |
-| Stream result batches | `for await (const batch of table.query()...)` |
-
-## Local vs Remote Safety
-
-| API | Agent guidance |
-| --- | --- |
-| `table.search(...)` | Preferred read path |
-| `table.query()` | Preferred scan/filter path |
-| `await table.toArrow()` | Avoid in portable or large-table code |
-| `await table.query().toArray()` with no `limit()` | Avoid; unbounded collection |
-| `await table.query().toArrow()` with no `limit()` | Avoid; unbounded collection |
-
-## Indexes
-
-```typescript
-await table.createIndex("vector");
-await table.createIndex("status");
-```
-
-Use vector indexes for large vector search workloads and scalar indexes for filtered columns or merge/upsert keys. Check source/docs before specifying advanced index options.
-
-## Filtering And Recall Knobs
-
-```typescript
-await table.search(queryVector).where("status = 'ready'").limit(10).toArray();
-await table.search(queryVector).limit(10).refineFactor(20).toArray();
-await table.search(queryVector).limit(10).nprobes(50).toArray();
-await table.search(queryVector).limit(10).ef(100).toArray();
-await table.search(queryVector).where("status = 'ready'").postfilter().limit(10).toArray();
-```
-
-Use `postfilter()` only when fewer than `limit` results are acceptable.
-
-## Diagnostics
-
-```typescript
-console.log(await table.search(queryVector).where("year > 2000").limit(10).analyzePlan());
-console.log(await table.indexStats("vector_idx"));
-```
-
-Use these before changing indexes or search tuning.
-
-## Column (Field) Metadata
-
-```typescript
-const schema = await table.schema();
-const meta = schema.fields.find((f) => f.name === "category")?.metadata; // Map<string, string>
-const res = await table.updateFieldMetadata([
-  { path: "category", metadata: { "lancedb:description": "...", "lancedb:tag:field_type": "label" } },
-]);
-res.version; // new table version
-```
-
-Merges by default; a `null` value deletes that key; `replace: true` swaps the whole map. Nested fields use dot-paths (`"a.b.c"`). See `references/column_metadata.md` for key conventions (`lancedb:description`, `lancedb:tag:<name>`, `lancedb:logical-column`) and the authoring workflow.
-
-## Branches
-
-```typescript
-const branches = await table.branches();       // async manager
-await branches.list();                         // non-main branches; {} = only main
-const exp = await branches.create("exp");      // fork off main -> Table scoped to the branch
-const wip = await branches.checkout("wip");    // existing branch -> scoped Table (version arg pins read-only)
-const wip2 = await db.openTable("t", { branch: "wip" }); // or open scoped directly
-await branches.delete("stale");                // removes only the branch pointer
-table.currentBranch();                         // null = main
-```
-
-There is no global switch — scoping is per table handle: any read/write on a branch handle lands on that branch; the original handle keeps targeting main. See `references/branch_ops.md` for the model and isolation checks.
-
-## Maintenance
-
-```typescript
-await table.optimize();
-```
-
-Call this after every successful local/OSS ingestion. It handles compaction, cleanup of old versions according to retention, and index optimization. Do not add this for LanceDB Enterprise/Cloud remote tables; Enterprise handles compaction and cleanup automatically from cluster configuration.

plugins/lancedb/skills/lancedb/references/typescript/patterns.md

@@ -1,100 +0,0 @@
-# TypeScript Patterns
-
-Use these patterns when writing TypeScript code with `@lancedb/lancedb`.
-
-## Recommended Patterns
-
-### Bounded query
-
-Use this for application reads, scripts, and examples:
-
-```typescript
-const rows = await table
-  .query()
-  .where("status = 'ready'")
-  .select(["id", "text"])
-  .limit(20)
-  .toArray();
-```
-
-### Bounded vector search
-
-```typescript
-const rows = await table
-  .search(queryVector)
-  .select(["id", "text"])
-  .limit(20)
-  .toArray();
-```
-
-### Batch streaming for larger reads
-
-When the task needs many rows, avoid collecting everything at once:
-
-```typescript
-for await (const batch of table
-  .query()
-  .where("status = 'ready'")
-  .select(["id", "text"])
-  .limit(10_000)) {
-  process(batch);
-}
-```
-
-## Anti-Patterns
-
-**Avoid the following anti-patterns in your code.**
-
-### Table-level full materialization
-
-Avoid whole-table collectors in portable or large-table code:
-
-```typescript
-const tableArrow = await table.toArrow();
-```
-
-Why: local tables expose these whole-table collectors, but remote tables intentionally do not — a remote production table can be far larger than a local development table, so it is easy to accidentally pull the entire table into memory.
-
-### Unbounded result collection
-
-Avoid query/search collection without a meaningful limit:
-
-```typescript
-const rows = await table.query().toArray();              // unbounded plain scan
-const rows = await table.search(queryVector).toArray();  // unbounded vector search
-```
-
-Prefer `select(...).limit(...)` before collecting; for large reads, stream in batches instead.
-
-### Per-row writes
-
-Avoid loops that write one row per call:
-
-```typescript
-for (const row of rows) {
-  await table.add([row]); // one commit + fragment per row
-}
-```
-
-Each `add()` creates a new version and fragment. Pass the whole batch in a single call, or chunk very large inputs:
-
-```typescript
-await table.add(rows); // single commit
-// for very large inputs, add in chunks of several thousand rows
-```
-
-### Drop-then-reuse the same table name (Enterprise/Cloud)
-
-Avoid dropping or overwriting a remote table and then reusing that name right away:
-
-```typescript
-await db.dropTable("my_table");
-const table = await db.createTable("my_table", rows);                    // reads 500 for ~5 min
-const table = await db.createTable("my_table", rows, { mode: "overwrite" }); // same problem
-```
-
-Why: Enterprise/Cloud splits DDL (control plane) from query serving (data plane). The data plane caches the dataset behind a table name for up to `table_cache_ttl` (default 300s / 5 min), so after a drop/overwrite the DDL succeeds but queries against the reused name return `500 Internal Server Error` until the cache expires — and a fresh `describe` may still show the old schema. Instead, write to a **fresh name**, use `tableNames()` and fail if it already exists, then `renameTable(fresh, final)` onto the final name only after the old table's drop has propagated (~5 min). See the "Enterprise: never drop-then-reuse the same table name" section in `SKILL.md`. Local/OSS tables have no separate data plane — overwrite freely there.
-
-### Guessing performance fixes
-
-Avoid changing `nprobes`, `refineFactor`, `ef`, or index settings before checking `analyzePlan()` and `indexStats(...)`. Diagnose first, then tune one knob at a time.

plugins/lancedb/skills/lancedb/references/typescript/performance.md

@@ -1,78 +0,0 @@
-# TypeScript Performance Guidance
-
-Use this when writing TypeScript code that ingests data, queries large tables, builds indexes, or investigates latency.
-
-## Ingestion
-
-- Prefer bulk or batched writes.
-- Avoid per-row write loops; they create many small commits/fragments.
-- For generated data, accumulate reasonable batches before adding.
-- For file-backed data, prefer APIs that stream from Arrow/Parquet-style inputs when available.
-
-## Indexing
-
-- Build a vector index once brute-force vector search becomes too slow. As a rule of thumb, local brute force is fine below roughly 100K vectors; beyond that, build an index.
-- Use the general-purpose vector index defaults unless the task has explicit recall/latency requirements.
-- Build scalar indexes for filtered columns and merge/upsert keys.
-- Use full-text index phrase options only when phrase queries require them.
-
-## Querying
-
-Always be explicit:
-
-```typescript
-await table.search(queryVector).select(["id", "title"]).limit(20).toArray();
-```
-
-- `select()` reduces bytes read and transferred.
-- `limit()` prevents accidental full-table collection.
-- Pre-filtering is the default behavior. Use `postfilter()` only when fewer than `limit` results are acceptable.
-
-## Recall Tuning
-
-Tune one knob at a time:
-
-- Quantized indexes: raise `refineFactor(...)` to rescore more candidates on full vectors.
-- HNSW-backed indexes: raise `ef(...)`; start around `1.5 * k`, increase toward `10 * k` if recall is short.
-- IVF candidate breadth: `nprobes(...)` is usually auto-tuned; override only when a selective pre-filter leaves too few neighbors.
-
-## Maintenance
-
-After every successful embedded OSS/local ingestion, call `table.optimize()`.
-Do not add this to LanceDB Enterprise/Cloud remote table code; remote compaction
-and cleanup are handled automatically based on the Enterprise cluster
-configuration.
-
-Why local maintenance is needed:
-
-- Frequent writes can create many small fragments. Queries then need to scan across more files, which can increase latency.
-- Updates, deletes, and appends create new table versions. Old versions are retained for time travel and rollback, which can grow disk usage.
-- Indexes may have newly added rows that are not yet fully optimized into the index structure.
-
-For local/OSS tables, run `optimize()` after the final successful ingestion
-write. Also run it after later batches of update/delete operations or on a
-regular maintenance schedule:
-
-```typescript
-await table.optimize();
-```
-
-If the user wants more aggressive local disk cleanup, pass a shorter cleanup retention window:
-
-```typescript
-const olderThan = new Date(Date.now() - 24 * 60 * 60 * 1000);
-await table.optimize({ cleanupOlderThan: olderThan });
-```
-
-Do not use very short cleanup windows when the application depends on time travel, rollback, or old versions.
-
-## Diagnostics
-
-Before changing code or indexes, inspect:
-
-```typescript
-console.log(await table.search(queryVector).where("year > 2000").limit(10).analyzePlan());
-console.log(await table.indexStats("vector_idx"));
-```
-
-Look for high scan cost, missing indexes, fragmented data, and unindexed rows.

plugins/lancedb/skills/lancedb/scripts/check_materialization.py

@@ -64,7 +64,9 @@ def scan_python(path: Path, text: str) -> list[Finding]:
 
 def statement_around(text: str, start: int, end: int) -> str:
     before = max(text.rfind(";", 0, start), text.rfind("\n\n", 0, start))
-    after_candidates = [pos for pos in (text.find(";", end), text.find("\n\n", end)) if pos != -1]
+    after_candidates = [
+        pos for pos in (text.find(";", end), text.find("\n\n", end)) if pos != -1
+    ]
     after = min(after_candidates) if after_candidates else len(text)
     return text[before + 1 : after].strip()
 

python/Cargo.toml

@@ -15,10 +15,10 @@ name = "_lancedb"
 crate-type = ["cdylib"]
 
 [dependencies]
-arrow = { version = "58.0.0", features = ["pyarrow"] }
-async-trait = "0.1"
-bytes = "1"
-lancedb = { path = "../rust/lancedb", default-features = false }
+arrow = { workspace = true, features = ["pyarrow"] }
+async-trait.workspace = true
+bytes.workspace = true
+lancedb.workspace = true
 datafusion-common.workspace = true
 lance-core.workspace = true
 lance-namespace.workspace = true
@@ -27,17 +27,17 @@ lance-io.workspace = true
 env_logger.workspace = true
 log.workspace = true
 pyo3 = { version = "0.28", features = ["extension-module", "abi3-py310", "chrono"] }
-chrono = { version = "0.4", default-features = false, features = ["clock"] }
+chrono.workspace = true
 pyo3-async-runtimes = { version = "0.28", features = [
     "attributes",
     "tokio-runtime",
 ] }
-pin-project = "1.1.5"
+pin-project.workspace = true
 futures.workspace = true
-serde = "1"
-serde_json = "1"
+serde.workspace = true
+serde_json.workspace = true
 snafu.workspace = true
-tokio = { version = "1.40", features = ["sync", "rt-multi-thread"] }
+tokio.workspace = true
 libc = "0.2"
 
 [build-dependencies]

python/pyproject.toml

@@ -8,7 +8,7 @@ dependencies = [
     "overrides>=0.7; python_version<'3.12'",
     "packaging>=23.0",
     "pyarrow>=16",
-    "pydantic>=1.10",
+    "pydantic>=2.7.4,<3",
     "tqdm>=4.27.0",
     "lance-namespace>=0.3.2"
 ]

python/python/lancedb/__init__.py

@@ -22,6 +22,16 @@ from .remote.db import RemoteDBConnection
 from .expr import Expr, col, lit, func
 from .schema import blob, vector, BlobType
 from .job import AsyncJob, Job
+from .functions import (
+    FunctionArtifactRequest as FunctionArtifactRequest,
+    FunctionApplication as FunctionApplication,
+    FunctionBinding as FunctionBinding,
+    FunctionRegistrationRequest as FunctionRegistrationRequest,
+    FunctionVersion as FunctionVersion,
+    PythonRuntimeSpec as PythonRuntimeSpec,
+    UdfDefinition as UdfDefinition,
+    udf as udf,
+)
 from .table import AsyncTable, Table
 from .types import BaseTokenizerType
 from ._lancedb import Session

python/python/lancedb/_lancedb.pyi

@@ -147,6 +147,8 @@ class Connection(object):
         limit: Optional[int],
     ) -> list[str]: ...  # Deprecated: Use list_tables instead
     def job(self, job_id: str) -> Job: ...
+    async def create_function_async(self, request_json: str) -> FunctionJob: ...
+    async def get_function(self, name: str, version: str) -> str: ...
     async def list_jobs(self) -> List[JobInfo]: ...
     async def get_job(self, job_id: str) -> Optional[JobDescription]: ...
     async def cancel_job(self, job_id: str) -> bool: ...
@@ -226,6 +228,19 @@ class Job:
     async def wait(self) -> None: ...
     async def cancel(self) -> None: ...
 
+class FunctionJob:
+    @property
+    def id(self) -> Optional[str]: ...
+    async def status(self) -> str: ...
+    async def wait(self) -> str: ...
+    async def cancel(self) -> None: ...
+
+class RefreshJob:
+    id: Optional[str]
+    async def status(self) -> str: ...
+    async def wait(self) -> str: ...
+    async def cancel(self) -> None: ...
+
 class JobInfo:
     @property
     def job_id(self) -> str: ...
@@ -341,8 +356,11 @@ class Table:
     async def add_computed_columns(
         self, columns: list[tuple[str, str]]
     ) -> AddColumnsResult: ...
+    async def add_function_columns(
+        self, application_json: str, output_name: Optional[str]
+    ) -> AddColumnsResult: ...
     async def refresh_column(self, column: str) -> RefreshColumnResult: ...
-    async def refresh_column_async(self, column: str) -> Job: ...
+    async def refresh_column_async(self, column: str) -> RefreshJob: ...
     async def add_columns_with_schema(self, schema: pa.Schema) -> AddColumnsResult: ...
     async def alter_columns(
         self, columns: list[dict[str, Any]]

python/python/lancedb/db.py

@@ -45,7 +45,8 @@ from lance_namespace.errors import NamespaceNotEmptyError, TableNotFoundError
 
 from . import __version__
 from ._lancedb import connect as lancedb_connect  # type: ignore
-from .job import AsyncJob, Job
+from .functions import FunctionVersion, UdfDefinition
+from .job import AsyncJob, Job, _function_job
 from .table import (
     AsyncTable,
     LanceTable,
@@ -616,6 +617,31 @@ class DBConnection(EnforceOverrides):
         """
         raise NotImplementedError("serialize is not supported for this connection type")
 
+    def create_function(self, definition: UdfDefinition) -> FunctionVersion:
+        """Register a scalar Python UDF and wait for its immutable version.
+
+        This is the blocking counterpart of :meth:`create_function_async`.
+        Local connections raise ``NotImplementedError``.
+        """
+        return self.create_function_async(definition).wait()
+
+    def create_function_async(self, definition: UdfDefinition) -> Job[FunctionVersion]:
+        """Register a scalar Python UDF through the remote Function catalog.
+
+        Submission returns a typed job. The immutable Function version becomes
+        available only when :meth:`Job.wait` succeeds. Local connections raise
+        ``NotImplementedError``.
+        """
+        raise NotImplementedError(
+            "Function catalog operations are not supported for this connection type"
+        )
+
+    def get_function(self, name: str, *, version: str) -> FunctionVersion:
+        """Open one exact immutable Function version from the remote catalog."""
+        raise NotImplementedError(
+            "Function catalog operations are not supported for this connection type"
+        )
+
     def job(self, job_id: str) -> Job:
         """A [Job][lancedb.job.Job] handle for a server-side job by id.
 
@@ -1256,6 +1282,15 @@ class LanceDBConnection(DBConnection):
         """
         return Job(self._conn.job(job_id))
 
+    @override
+    def create_function_async(self, definition: UdfDefinition) -> Job[FunctionVersion]:
+        job = LOOP.run(self._conn.create_function_async(definition))
+        return Job(job)
+
+    @override
+    def get_function(self, name: str, *, version: str) -> FunctionVersion:
+        return LOOP.run(self._conn.get_function(name, version=version))
+
     @override
     def list_jobs(self) -> List[JobInfo]:
         """List server-side jobs across the database's tables."""
@@ -2023,6 +2058,25 @@ class AsyncConnection(object):
         """
         return AsyncJob(self._inner.job(job_id))
 
+    async def create_function_async(
+        self, definition: UdfDefinition
+    ) -> AsyncJob[FunctionVersion]:
+        """Register a scalar Python UDF through the remote Function catalog.
+
+        The returned typed job resolves to the immutable Function version.
+        Local connections raise ``NotImplementedError``.
+        """
+        if not isinstance(definition, UdfDefinition):
+            raise TypeError("create_function_async requires a @udf definition")
+        inner = await self._inner.create_function_async(
+            definition.registration_request.to_canonical_json()
+        )
+        return _function_job(inner)
+
+    async def get_function(self, name: str, *, version: str) -> FunctionVersion:
+        """Open one exact immutable Function version from the remote catalog."""
+        return FunctionVersion.from_json(await self._inner.get_function(name, version))
+
     async def list_jobs(self) -> List[JobInfo]:
         """List server-side jobs across the database's tables."""
         return await self._inner.list_jobs()

python/python/lancedb/embeddings/base.py

@@ -26,7 +26,6 @@ class EmbeddingFunction(BaseModel, ABC):
     3. ndims() which returns the number of dimensions of the vector column
     """
 
-    __slots__ = ("__weakref__",)  # pydantic 1.x compatibility
     max_retries: int = (
         7  # Setting 0 disables retires. Maybe this should not be enabled by default,
     )

python/python/lancedb/embeddings/bedrock.py

@@ -7,8 +7,7 @@ from functools import cached_property
 from typing import List, Union
 
 import numpy as np
-
-from lancedb.pydantic import PYDANTIC_VERSION
+from pydantic import ConfigDict
 
 from ..util import attempt_import_or_raise
 from .base import TextEmbeddingFunction
@@ -67,13 +66,7 @@ class BedRockText(TextEmbeddingFunction):
     source_input_type: str = "search_document"
     query_input_type: str = "search_query"
 
-    if PYDANTIC_VERSION.major < 2:  # Pydantic 1.x compat
-
-        class Config:
-            keep_untouched = (cached_property,)
-    else:
-        model_config = dict()
-        model_config["ignored_types"] = (cached_property,)
+    model_config = ConfigDict(ignored_types=(cached_property,))
 
     def ndims(self):
         # return len(self._generate_embedding("test"))

python/python/lancedb/embeddings/gemini_text.py

@@ -7,8 +7,7 @@ from functools import cached_property
 from typing import List, Optional, Union
 
 import numpy as np
-
-from lancedb.pydantic import PYDANTIC_VERSION
+from pydantic import ConfigDict
 
 from ..util import attempt_import_or_raise
 from .base import TextEmbeddingFunction
@@ -87,13 +86,7 @@ class GeminiText(TextEmbeddingFunction):
     query_task_type: str = "retrieval_query"
     source_task_type: str = "retrieval_document"
 
-    if PYDANTIC_VERSION.major < 2:  # Pydantic 1.x compat
-
-        class Config:
-            keep_untouched = (cached_property,)
-    else:
-        model_config = dict()
-        model_config["ignored_types"] = (cached_property,)
+    model_config = ConfigDict(ignored_types=(cached_property,))
 
     def ndims(self):
         if self.dim:

python/python/lancedb/embeddings/imagebind.py

@@ -7,14 +7,13 @@ from typing import List, Union
 
 import numpy as np
 import pyarrow as pa
+from pydantic import ConfigDict
 
 from ..util import attempt_import_or_raise
 from .base import EmbeddingFunction
 from .registry import register
 from .utils import AUDIO, IMAGES, TEXT
 
-from lancedb.pydantic import PYDANTIC_VERSION
-
 
 @register("imagebind")
 class ImageBindEmbeddings(EmbeddingFunction):
@@ -31,13 +30,7 @@ class ImageBindEmbeddings(EmbeddingFunction):
     device: str = "cpu"
     normalize: bool = False
 
-    if PYDANTIC_VERSION.major < 2:  # Pydantic 1.x compat
-
-        class Config:
-            keep_untouched = (cached_property,)
-    else:
-        model_config = dict()
-        model_config["ignored_types"] = (cached_property,)
+    model_config = ConfigDict(ignored_types=(cached_property,))
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)

python/python/lancedb/embeddings/transformers.py

@@ -7,8 +7,7 @@ from typing import List, Any
 
 import numpy as np
 
-from pydantic import PrivateAttr
-from lancedb.pydantic import PYDANTIC_VERSION
+from pydantic import ConfigDict, PrivateAttr
 
 from ..util import attempt_import_or_raise
 from .base import EmbeddingFunction
@@ -59,13 +58,7 @@ class TransformersEmbeddingFunction(EmbeddingFunction):
         )
         self._model.to(self.device)
 
-    if PYDANTIC_VERSION.major < 2:  # Pydantic 1.x compat
-
-        class Config:
-            keep_untouched = (cached_property,)
-    else:
-        model_config = dict()
-        model_config["ignored_types"] = (cached_property,)
+    model_config = ConfigDict(ignored_types=(cached_property,))
 
     def ndims(self):
         self._ndims = self._model.config.hidden_size

python/python/lancedb/expr.py

@@ -85,8 +85,9 @@ class Expr:
     # for dict keys / set membership.
     __hash__ = None  # type: ignore[assignment]
 
-    def __init__(self, inner: PyExpr) -> None:
+    def __init__(self, inner: PyExpr, *, column_path: str | None = None) -> None:
         self._inner = inner
+        self._column_path = column_path
 
     # ── comparisons ──────────────────────────────────────────────────────────
 
@@ -273,7 +274,7 @@ def col(name: str) -> Expr:
     >>> col("age") > lit(18)
     Expr((age > 18))
     """
-    return Expr(expr_col(name))
+    return Expr(expr_col(name), column_path=name)
 
 
 def lit(value: Union[bool, int, float, str, bytes, date, datetime, Decimal]) -> Expr:

python/python/lancedb/functions.py

@@ -0,0 +1,952 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+"""Canonical values exchanged with LanceDB Enterprise Function services.
+
+These immutable models contain client/wire state only. Catalog persistence,
+environment bake, secret resolution, and execution are owned by Sophon.
+"""
+
+from __future__ import annotations
+
+import ast
+import base64
+import functools
+import hashlib
+import inspect
+import json
+import math
+import re
+import sys
+import textwrap
+import types
+import uuid
+from collections.abc import Mapping
+from datetime import date, datetime
+from typing import (
+    Annotated,
+    Any,
+    Callable,
+    Optional,
+    Union,
+    get_args,
+    get_origin,
+    get_type_hints,
+    overload,
+)
+
+import pyarrow as pa
+from pydantic import (
+    BaseModel,
+    ConfigDict,
+    Field,
+    conint,
+    field_validator,
+    model_validator,
+)
+
+_Int32 = conint(strict=True, ge=-(2**31), le=2**31 - 1)
+_UInt32 = conint(strict=True, ge=0, le=2**32 - 1)
+_UInt64 = conint(strict=True, ge=0, le=2**64 - 1)
+
+
+class _FrozenDict(dict):
+    def _immutable(self, *args, **kwargs):
+        raise TypeError("remote canonical values are immutable")
+
+    __setitem__ = _immutable
+    __delitem__ = _immutable
+    clear = _immutable
+    pop = _immutable
+    popitem = _immutable
+    setdefault = _immutable
+    update = _immutable
+
+    def __ior__(self, other):
+        self._immutable()
+
+
+def _freeze_value(value):
+    if isinstance(value, Mapping):
+        return _FrozenDict({key: _freeze_value(child) for key, child in value.items()})
+    if isinstance(value, (list, tuple)):
+        return tuple(_freeze_value(child) for child in value)
+    return value
+
+
+def _validate_literal(value):
+    if isinstance(value, float):
+        raise ValueError(
+            "floating-point Function literals are not part of the Slice 1 "
+            "canonical wire contract"
+        )
+    if isinstance(value, int) and not isinstance(value, bool):
+        if not -(2**63) <= value <= 2**64 - 1:
+            raise ValueError(
+                "Function integer literal is outside the canonical JSON range"
+            )
+    elif isinstance(value, Mapping):
+        for child in value.values():
+            _validate_literal(child)
+    elif isinstance(value, (list, tuple)):
+        for child in value:
+            _validate_literal(child)
+    return value
+
+
+def _known_wire_value(value):
+    if isinstance(value, _RemoteValue):
+        return value._known_dict()
+    if isinstance(value, Mapping):
+        return {key: _known_wire_value(child) for key, child in value.items()}
+    if isinstance(value, (list, tuple)):
+        return [_known_wire_value(child) for child in value]
+    return value
+
+
+class _RemoteValue(BaseModel):
+    model_config = ConfigDict(extra="ignore", frozen=True)
+
+    @model_validator(mode="after")
+    def _freeze_mappings(self):
+        for name, value in self.__dict__.items():
+            object.__setattr__(self, name, _freeze_value(value))
+        return self
+
+    @classmethod
+    def from_json(cls, payload: str):
+        return cls.model_validate_json(payload)
+
+    def _known_dict(self) -> dict[str, Any]:
+        known = {}
+        for name, field in self.__class__.model_fields.items():
+            value = getattr(self, name)
+            if value is None:
+                continue
+            if not field.is_required():
+                default_factory = field.default_factory
+                if default_factory is not None and value == default_factory():
+                    continue
+                if default_factory is None and value == field.default:
+                    continue
+            known[name] = _known_wire_value(value)
+        return known
+
+    def _copy(self, *, update: Mapping[str, Any]):
+        update = {name: _freeze_value(value) for name, value in update.items()}
+        return self.model_copy(update=update)
+
+    def to_canonical_json(self) -> str:
+        return json.dumps(
+            self._known_dict(),
+            ensure_ascii=False,
+            allow_nan=False,
+            sort_keys=True,
+            separators=(",", ":"),
+        )
+
+
+class _OpenRemoteValue(_RemoteValue):
+    """Forward-readable value whose extras stay out of canonical encoding."""
+
+    model_config = ConfigDict(extra="allow", frozen=True)
+
+    def _unknown_field_names(self) -> set[str]:
+        return set((self.__pydantic_extra__ or {}).keys())
+
+
+class FunctionArtifact(_RemoteValue):
+    """Content-addressed Python artifact identity."""
+
+    kind: str
+    digest: str
+    entrypoint: str
+
+
+class FunctionArtifactContent(_RemoteValue):
+    """Encoded artifact bytes uploaded during remote registration."""
+
+    encoding: str
+    data: str
+
+
+class PythonAdapterSpec(_RemoteValue):
+    """Internal scalar-callable to Arrow-batch adapter selection."""
+
+    kind: str
+    version: _UInt32
+
+
+class FunctionArtifactRequest(_RemoteValue):
+    """Source artifact uploaded while registering a Function."""
+
+    kind: str
+    digest: str
+    entrypoint: str
+    content: FunctionArtifactContent
+    adapter: PythonAdapterSpec
+
+
+class FunctionParameter(_RemoteValue):
+    name: str
+    arrow_type: str
+    nullable: bool
+
+
+class FunctionResultField(_OpenRemoteValue):
+    name: str
+    arrow_type: str
+    nullable: bool
+
+
+class FunctionOutput(_OpenRemoteValue):
+    """Scalar or ordered named-struct output; unknown kinds remain decodable."""
+
+    kind: str
+    arrow_type: Optional[str] = None
+    nullable: Optional[bool] = None
+    fields: tuple[FunctionResultField, ...] = ()
+
+
+class FunctionSignature(_RemoteValue):
+    inputs: tuple[FunctionParameter, ...]
+    output: FunctionOutput
+
+
+class PythonEnvironmentSpec(_RemoteValue):
+    """One Sophon-managed Python environment source."""
+
+    kind: str
+    packages: tuple[str, ...] = ()
+    path: Optional[str] = None
+    modules: tuple[str, ...] = ()
+    image: Optional[str] = None
+
+
+class PythonRuntimeSpec(_RemoteValue):
+    """Remote runtime definition with non-secret environment values.
+
+    V1 supports ``kind="python"``. Newer runtime kinds remain readable, while
+    their unknown payload fields are intentionally not retained by the client.
+    """
+
+    kind: str
+    python_version: Optional[str] = None
+    environment: Optional[PythonEnvironmentSpec] = None
+    env: Optional[Mapping[str, str]] = None
+
+    @model_validator(mode="after")
+    def _validate_runtime_kind(self):
+        if self.kind == "python":
+            if self.python_version is None:
+                raise ValueError("python runtime requires python_version")
+            if self.environment is None:
+                raise ValueError("python runtime requires environment")
+        else:
+            object.__setattr__(self, "python_version", None)
+            object.__setattr__(self, "environment", None)
+            object.__setattr__(self, "env", None)
+        return self
+
+
+class FunctionVersion(_RemoteValue):
+    """An exact immutable Function version returned by Enterprise.
+
+    Scheduling resources, priority, concurrency, and retry policy belong to
+    the submitting Job and are not part of this identity.
+    """
+
+    name: str
+    version: str
+    artifact: FunctionArtifact
+    signature: FunctionSignature
+    runtime: PythonRuntimeSpec
+    runtime_digest: str
+    environment_digest: str
+    required_secrets: tuple[str, ...] = ()
+    created_at: str
+
+    def __call__(self, **inputs: Any) -> FunctionApplication:
+        """Bind this exact version to table columns as one grouped application."""
+        from lancedb.expr import Expr
+
+        parameters = tuple(parameter.name for parameter in self.signature.inputs)
+        missing = [parameter for parameter in parameters if parameter not in inputs]
+        unknown = sorted(set(inputs) - set(parameters))
+        if missing or unknown:
+            details = []
+            if missing:
+                details.append(f"missing inputs: {missing!r}")
+            if unknown:
+                details.append(f"unknown inputs: {unknown!r}")
+            raise TypeError("invalid Function inputs (" + "; ".join(details) + ")")
+
+        bindings = []
+        for parameter in parameters:
+            value = inputs[parameter]
+            if not isinstance(value, Expr) or value._column_path is None:
+                raise TypeError(
+                    f"Function input {parameter!r} must be a direct col(...) reference"
+                )
+            bindings.append(
+                ApplicationInput(
+                    parameter=parameter,
+                    kind="column",
+                    value={"path": value._column_path},
+                )
+            )
+        return FunctionApplication(
+            function=FunctionVersionRef(name=self.name, version=self.version),
+            inputs=tuple(bindings),
+            output=self.signature.output,
+            group_id=f"fg_{uuid.uuid4().hex}",
+        )
+
+
+class FunctionRegistrationRequest(_RemoteValue):
+    """Stable remote registration envelope produced by :func:`udf`.
+
+    Only secret names are represented. Secret values are resolved inside the
+    remote service and have no client request field.
+    """
+
+    name: str
+    artifact: FunctionArtifactRequest
+    signature: FunctionSignature
+    runtime: PythonRuntimeSpec
+    required_secrets: tuple[str, ...] = ()
+
+
+class FunctionVersionRef(_OpenRemoteValue):
+    name: str
+    version: str
+
+
+class ApplicationInput(_OpenRemoteValue):
+    """One parameter value.
+
+    Slice 1 freezes integers, strings, booleans, nulls, arrays, and objects.
+    Floating-point literal encoding is deferred until Python authoring is
+    introduced with a language-neutral numeric representation.
+    """
+
+    parameter: str
+    kind: str
+    value: Any
+
+    @field_validator("value")
+    @classmethod
+    def _validate_value(cls, value):
+        return _validate_literal(value)
+
+
+class FunctionApplication(_OpenRemoteValue):
+    """Immutable pre-declaration application of an exact Function version."""
+
+    function: FunctionVersionRef
+    inputs: tuple[ApplicationInput, ...]
+    output: FunctionOutput
+    group_id: str
+    columns: Mapping[str, str] = Field(default_factory=dict)
+
+    def _known_dict(self) -> dict[str, Any]:
+        value = super()._known_dict()
+        for name in self._unknown_field_names():
+            value.pop(name, None)
+        return value
+
+    def _ensure_declarable(self) -> None:
+        unknown = {f"application.{name}" for name in self._unknown_field_names()}
+        unknown.update(
+            f"function.{name}" for name in self.function._unknown_field_names()
+        )
+        for index, input_value in enumerate(self.inputs):
+            unknown.update(
+                f"inputs[{index}].{name}" for name in input_value._unknown_field_names()
+            )
+        unknown.update(f"output.{name}" for name in self.output._unknown_field_names())
+        for index, field in enumerate(self.output.fields):
+            unknown.update(
+                f"output.fields[{index}].{name}"
+                for name in field._unknown_field_names()
+            )
+        if unknown:
+            raise ValueError(
+                "Function application contains fields from a newer contract: "
+                f"{sorted(unknown)!r}"
+            )
+
+    def rename(self, *, columns: Mapping[str, str]) -> FunctionApplication:
+        """Return a copy with result-field to table-column aliases."""
+        if self.output.kind != "named_struct":
+            raise ValueError("rename(columns=...) requires a named-struct application")
+        result_fields = {field.name for field in self.output.fields}
+        unknown = set(columns) - result_fields
+        if unknown:
+            raise ValueError(f"unknown Function result fields: {sorted(unknown)!r}")
+        merged = dict(self.columns)
+        merged.update(columns)
+        destinations = tuple(
+            merged.get(field.name, field.name) for field in self.output.fields
+        )
+        if len(set(destinations)) != len(destinations):
+            raise ValueError("FunctionApplication rename destinations must be unique")
+        return self._copy(update={"columns": merged})
+
+
+class InputBinding(_RemoteValue):
+    parameter: str
+    field_id: _Int32
+    field_path: str
+    arrow_type: str
+    nullable: bool
+
+
+class OutputMapping(_RemoteValue):
+    """One stable result-field mapping.
+
+    Assignment state is outside the Slice 1 client contract. During the NULL
+    transition Lance exposes no public cell-flag identifier to persist here.
+    """
+
+    result_field: str
+    output_name: str
+    output_field_id: _Int32
+    output_ordinal: _UInt32
+    arrow_type: str
+    nullable: bool
+
+
+class FunctionBinding(_RemoteValue):
+    """Immutable grouped binding persisted by the Enterprise table service."""
+
+    binding_id: str
+    revision: _UInt64
+    function: FunctionVersionRef
+    group_id: str
+    inputs: tuple[InputBinding, ...]
+    outputs: tuple[OutputMapping, ...]
+    input_schema: Optional[Mapping[str, Any]] = None
+    output_schema: Optional[Mapping[str, Any]] = None
+
+
+class RefreshColumnResult(_RemoteValue):
+    """Terminal result of a remote Function-column refresh Job."""
+
+    rows_assigned: _UInt64
+    rows_failed: _UInt64
+    rows_remaining: _UInt64
+    source_version: _UInt64
+    published_version: Optional[_UInt64] = None
+
+    @property
+    def rows_filled(self) -> int:
+        """Deprecated compatibility alias for :attr:`rows_assigned`."""
+        return self.rows_assigned
+
+    @property
+    def version(self) -> Optional[int]:
+        """Deprecated compatibility alias for :attr:`published_version`."""
+        return self.published_version
+
+
+_FUNCTION_NAME = re.compile(r"^[A-Za-z_][A-Za-z0-9_.-]*$")
+_SECRET_NAME = re.compile(r"^[A-Za-z_][A-Za-z0-9_]*$")
+
+
+def _canonical_arrow_type(data_type: pa.DataType) -> str:
+    primitive_types = (
+        (pa.bool_(), "bool"),
+        (pa.int8(), "int8"),
+        (pa.int16(), "int16"),
+        (pa.int32(), "int32"),
+        (pa.int64(), "int64"),
+        (pa.uint8(), "uint8"),
+        (pa.uint16(), "uint16"),
+        (pa.uint32(), "uint32"),
+        (pa.uint64(), "uint64"),
+        (pa.float16(), "float16"),
+        (pa.float32(), "float32"),
+        (pa.float64(), "float64"),
+        (pa.string(), "utf8"),
+        (pa.large_utf8(), "large_utf8"),
+        (pa.binary(), "binary"),
+        (pa.large_binary(), "large_binary"),
+        (pa.date32(), "date32"),
+        (pa.date64(), "date64"),
+    )
+    for candidate, name in primitive_types:
+        if data_type == candidate:
+            return name
+    if pa.types.is_fixed_size_binary(data_type):
+        return f"fixed_size_binary[{data_type.byte_width}]"
+    if pa.types.is_list(data_type):
+        return f"list<{_canonical_arrow_type(data_type.value_type)}>"
+    if pa.types.is_large_list(data_type):
+        return f"large_list<{_canonical_arrow_type(data_type.value_type)}>"
+    if pa.types.is_fixed_size_list(data_type):
+        return (
+            f"fixed_size_list<{_canonical_arrow_type(data_type.value_type)}>"
+            f"[{data_type.list_size}]"
+        )
+    if pa.types.is_struct(data_type):
+        fields = ",".join(
+            f"{field.name}:{_canonical_arrow_type(field.type)}" for field in data_type
+        )
+        return f"struct<{fields}>"
+    if pa.types.is_timestamp(data_type):
+        timezone = f",tz={data_type.tz}" if data_type.tz is not None else ""
+        return f"timestamp[{data_type.unit}{timezone}]"
+    if pa.types.is_time32(data_type) or pa.types.is_time64(data_type):
+        return f"time[{data_type.unit}]"
+    if pa.types.is_duration(data_type):
+        return f"duration[{data_type.unit}]"
+    if pa.types.is_decimal(data_type):
+        bit_width = data_type.bit_width
+        return f"decimal{bit_width}({data_type.precision},{data_type.scale})"
+    raise TypeError(f"unsupported Arrow type for Function signature: {data_type}")
+
+
+def _annotation_type(annotation: Any) -> tuple[pa.DataType, bool]:
+    nullable = False
+    origin = get_origin(annotation)
+    if origin in (Union, types.UnionType):
+        arguments = get_args(annotation)
+        non_none = tuple(
+            argument for argument in arguments if argument is not type(None)
+        )
+        if len(non_none) != 1 or len(non_none) == len(arguments):
+            raise TypeError(f"unsupported union annotation: {annotation!r}")
+        annotation = non_none[0]
+        nullable = True
+
+    origin = get_origin(annotation)
+    if origin is Annotated:
+        base, *metadata = get_args(annotation)
+        arrow_types = [value for value in metadata if isinstance(value, pa.DataType)]
+        if len(arrow_types) != 1:
+            raise TypeError(
+                "Annotated Function types require exactly one PyArrow DataType"
+            )
+        _, base_nullable = _annotation_type(base)
+        return arrow_types[0], nullable or base_nullable
+
+    if isinstance(annotation, pa.DataType):
+        return annotation, nullable
+    if annotation is bool:
+        return pa.bool_(), nullable
+    if annotation is int:
+        return pa.int64(), nullable
+    if annotation is float:
+        return pa.float64(), nullable
+    if annotation is str:
+        return pa.string(), nullable
+    if annotation is bytes:
+        return pa.binary(), nullable
+    if annotation is date:
+        return pa.date32(), nullable
+    if annotation is datetime:
+        return pa.timestamp("us"), nullable
+    if get_origin(annotation) is list:
+        arguments = get_args(annotation)
+        if len(arguments) != 1:
+            raise TypeError(f"unsupported list annotation: {annotation!r}")
+        value_type, value_nullable = _annotation_type(arguments[0])
+        if value_nullable:
+            raise TypeError("nullable Function list elements are not supported")
+        return pa.list_(value_type), nullable
+    raise TypeError(f"unsupported Function annotation: {annotation!r}")
+
+
+def _callable_parameters(function: Callable[..., Any]) -> tuple[inspect.Parameter, ...]:
+    parameters = tuple(inspect.signature(function).parameters.values())
+    for parameter in parameters:
+        if parameter.kind in (
+            inspect.Parameter.POSITIONAL_ONLY,
+            inspect.Parameter.VAR_POSITIONAL,
+            inspect.Parameter.VAR_KEYWORD,
+        ):
+            raise TypeError("Function callables require named, non-variadic parameters")
+        if parameter.default is not inspect.Parameter.empty:
+            raise TypeError("Function callable defaults are not supported")
+    return parameters
+
+
+def _function_output(output: pa.DataType | pa.Field | pa.Schema) -> FunctionOutput:
+    if isinstance(output, pa.Schema):
+        fields = tuple(output)
+    elif isinstance(output, pa.Field) and pa.types.is_struct(output.type):
+        if output.nullable:
+            raise ValueError("Function output must be non-nullable")
+        fields = tuple(output.type)
+    elif isinstance(output, pa.DataType) and pa.types.is_struct(output):
+        fields = tuple(output)
+    else:
+        field = (
+            output
+            if isinstance(output, pa.Field)
+            else pa.field("result", output, nullable=False)
+        )
+        if not isinstance(field, pa.Field):
+            raise TypeError(
+                "output_schema must be a PyArrow DataType, Field, or Schema"
+            )
+        if field.nullable:
+            raise ValueError("Function output must be non-nullable")
+        return FunctionOutput(
+            kind="scalar",
+            arrow_type=_canonical_arrow_type(field.type),
+            nullable=False,
+        )
+
+    if not fields:
+        raise ValueError("named-struct Function output must contain at least one field")
+    if any(field.nullable for field in fields):
+        raise ValueError("Function output fields must be non-nullable")
+    names = [field.name for field in fields]
+    if len(set(names)) != len(names):
+        raise ValueError("Function output field names must be unique")
+    return FunctionOutput(
+        kind="named_struct",
+        fields=tuple(
+            FunctionResultField(
+                name=field.name,
+                arrow_type=_canonical_arrow_type(field.type),
+                nullable=False,
+            )
+            for field in fields
+        ),
+    )
+
+
+def _infer_signature(
+    function: Callable[..., Any],
+    input_schema: Optional[pa.Schema],
+    output_schema: Optional[pa.DataType | pa.Field | pa.Schema],
+) -> FunctionSignature:
+    parameters = _callable_parameters(function)
+    if (input_schema is None) != (output_schema is None):
+        raise ValueError("input_schema and output_schema must be provided together")
+
+    if input_schema is not None:
+        if not isinstance(input_schema, pa.Schema):
+            raise TypeError("input_schema must be a PyArrow Schema")
+        expected = tuple(parameter.name for parameter in parameters)
+        actual = tuple(input_schema.names)
+        if actual != expected:
+            raise ValueError(
+                "input_schema fields must exactly match callable parameters in order: "
+                f"expected {expected!r}, got {actual!r}"
+            )
+        inputs = tuple(
+            FunctionParameter(
+                name=field.name,
+                arrow_type=_canonical_arrow_type(field.type),
+                nullable=field.nullable,
+            )
+            for field in input_schema
+        )
+        return FunctionSignature(inputs=inputs, output=_function_output(output_schema))
+
+    try:
+        annotations = get_type_hints(function, include_extras=True)
+    except Exception as error:
+        raise TypeError(f"failed to resolve Function annotations: {error}") from error
+    missing = [
+        parameter.name for parameter in parameters if parameter.name not in annotations
+    ]
+    if missing or "return" not in annotations:
+        names = missing + ([] if "return" in annotations else ["return"])
+        raise TypeError(f"missing Function annotations: {names!r}")
+    inputs = []
+    for parameter in parameters:
+        data_type, nullable = _annotation_type(annotations[parameter.name])
+        inputs.append(
+            FunctionParameter(
+                name=parameter.name,
+                arrow_type=_canonical_arrow_type(data_type),
+                nullable=nullable,
+            )
+        )
+    output_type, output_nullable = _annotation_type(annotations["return"])
+    if output_nullable:
+        raise ValueError("Function output must be non-nullable")
+    return FunctionSignature(
+        inputs=tuple(inputs),
+        output=_function_output(pa.field("result", output_type, nullable=False)),
+    )
+
+
+def _is_udf_decorator(node: ast.expr) -> bool:
+    if isinstance(node, ast.Call):
+        node = node.func
+    return (isinstance(node, ast.Name) and node.id == "udf") or (
+        isinstance(node, ast.Attribute) and node.attr == "udf"
+    )
+
+
+def _literal_source(value: Any) -> str:
+    if value is None or type(value) in (bool, int, str, bytes):
+        return repr(value)
+    if type(value) is float and math.isfinite(value):
+        return repr(value)
+    if type(value) is tuple:
+        children = ", ".join(_literal_source(child) for child in value)
+        if len(value) == 1:
+            children += ","
+        return f"({children})"
+    raise TypeError(
+        "Function source references an unsupported global value of type "
+        f"{type(value).__name__}"
+    )
+
+
+def _package_source(function: Callable[..., Any]) -> bytes:
+    if not inspect.isfunction(function) or inspect.iscoroutinefunction(function):
+        raise TypeError("@udf requires a synchronous Python function")
+    try:
+        source = textwrap.dedent(inspect.getsource(function))
+    except (OSError, TypeError) as error:
+        raise ValueError("@udf requires inspectable Python source") from error
+    module = ast.parse(source)
+    definitions = [
+        node
+        for node in module.body
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef))
+        and node.name == function.__name__
+    ]
+    if len(definitions) != 1 or not isinstance(definitions[0], ast.FunctionDef):
+        raise ValueError("@udf source must contain exactly one synchronous function")
+    definition = definitions[0]
+    if any(not _is_udf_decorator(decorator) for decorator in definition.decorator_list):
+        raise ValueError("@udf cannot package additional Python decorators")
+    definition.decorator_list = []
+
+    closure = inspect.getclosurevars(function)
+    if closure.nonlocals:
+        raise ValueError("@udf cannot package functions that capture closure values")
+    if closure.unbound:
+        raise ValueError(
+            f"@udf source contains unresolved global names: {sorted(closure.unbound)!r}"
+        )
+    globals_source = []
+    for name, value in sorted(closure.globals.items()):
+        if isinstance(value, types.ModuleType):
+            globals_source.append(f"import {value.__name__} as {name}")
+        else:
+            globals_source.append(f"{name} = {_literal_source(value)}")
+
+    function_source = ast.unparse(definition)
+    parts = ["from __future__ import annotations"]
+    if globals_source:
+        parts.extend(["", *globals_source])
+    parts.extend(["", function_source, ""])
+    try:
+        import cloudpickle
+    except ImportError as error:
+        raise RuntimeError("@udf remote registration requires cloudpickle") from error
+    return cloudpickle.dumps(function)
+
+
+class UdfDefinition:
+    """A scalar Python callable prepared for remote Function registration.
+
+    Instances are created with :func:`udf`. Calling an instance executes the
+    original scalar Python function, which keeps local unit testing ordinary.
+    Remote execution adapts that scalar callable to the internal Arrow batch
+    ABI described by the registration artifact.
+    """
+
+    def __init__(
+        self,
+        function: Callable[..., Any],
+        *,
+        name: Optional[str],
+        input_schema: Optional[pa.Schema],
+        output_schema: Optional[pa.DataType | pa.Field | pa.Schema],
+        pip: tuple[str, ...],
+        env: Mapping[str, str],
+        secrets: tuple[str, ...],
+        python_version: Optional[str],
+    ):
+        function_name = name or function.__name__
+        if not _FUNCTION_NAME.fullmatch(function_name):
+            raise ValueError(f"invalid Function name: {function_name!r}")
+        packages = tuple(sorted(set(pip)))
+        if any(not package or package != package.strip() for package in packages):
+            raise ValueError("pip requirements must be non-empty and trimmed")
+        environment = dict(env)
+        if any(
+            not isinstance(key, str) or not isinstance(value, str)
+            for key, value in environment.items()
+        ):
+            raise TypeError("Function env keys and values must be strings")
+        required_secrets = tuple(sorted(set(secrets)))
+        invalid_secrets = [
+            secret for secret in required_secrets if not _SECRET_NAME.fullmatch(secret)
+        ]
+        if invalid_secrets:
+            raise ValueError(f"invalid Function secret names: {invalid_secrets!r}")
+        overlap = set(environment) & set(required_secrets)
+        if overlap:
+            raise ValueError(
+                f"Function env and secret names must be disjoint: {sorted(overlap)!r}"
+            )
+
+        signature = _infer_signature(function, input_schema, output_schema)
+        source = _package_source(function)
+        digest = f"sha256:{hashlib.sha256(source).hexdigest()}"
+        runtime = PythonRuntimeSpec(
+            kind="python",
+            python_version=python_version
+            or f"{sys.version_info.major}.{sys.version_info.minor}",
+            environment=PythonEnvironmentSpec(kind="pip", packages=packages),
+            env=environment,
+        )
+        self._function = function
+        self._request = FunctionRegistrationRequest(
+            name=function_name,
+            artifact=FunctionArtifactRequest(
+                kind="python_callable",
+                digest=digest,
+                entrypoint=function.__name__,
+                content=FunctionArtifactContent(
+                    encoding="base64",
+                    data=base64.b64encode(source).decode("ascii"),
+                ),
+                adapter=PythonAdapterSpec(
+                    kind="scalar_to_arrow_batch",
+                    version=1,
+                ),
+            ),
+            signature=signature,
+            runtime=runtime,
+            required_secrets=required_secrets,
+        )
+        functools.update_wrapper(self, function)
+
+    @property
+    def registration_request(self) -> FunctionRegistrationRequest:
+        """The immutable request sent by ``create_function_async``."""
+        return self._request
+
+    def __call__(self, *args, **kwargs):
+        return self._function(*args, **kwargs)
+
+
+@overload
+def udf(function: Callable[..., Any]) -> UdfDefinition: ...
+
+
+@overload
+def udf(
+    function: None = None,
+    *,
+    name: Optional[str] = None,
+    input_schema: Optional[pa.Schema] = None,
+    output_schema: Optional[pa.DataType | pa.Field | pa.Schema] = None,
+    pip: tuple[str, ...] | list[str] = (),
+    env: Optional[Mapping[str, str]] = None,
+    secrets: tuple[str, ...] | list[str] = (),
+    python_version: Optional[str] = None,
+) -> Callable[[Callable[..., Any]], UdfDefinition]: ...
+
+
+def udf(
+    function: Optional[Callable[..., Any]] = None,
+    *,
+    name: Optional[str] = None,
+    input_schema: Optional[pa.Schema] = None,
+    output_schema: Optional[pa.DataType | pa.Field | pa.Schema] = None,
+    pip: tuple[str, ...] | list[str] = (),
+    env: Optional[Mapping[str, str]] = None,
+    secrets: tuple[str, ...] | list[str] = (),
+    python_version: Optional[str] = None,
+):
+    """Prepare a scalar Python callable for remote Function registration.
+
+    Input and output signatures are inferred from supported annotations. For
+    Arrow types annotations cannot express precisely, pass ``input_schema``
+    and ``output_schema`` together. Nullable outputs are rejected because V1
+    uses physical NULL to represent unassigned computed-column rows.
+
+    Parameters
+    ----------
+    function : Callable, optional
+        The synchronous scalar callable to package.
+    name : str, optional
+        The remote Function name. Defaults to the callable name.
+    input_schema : pyarrow.Schema, optional
+        Explicit input fields in the exact order of the callable parameters.
+        Must be provided together with ``output_schema``.
+    output_schema : pyarrow.DataType, pyarrow.Field, or pyarrow.Schema, optional
+        Explicit scalar or named-struct output. Must be non-nullable and be
+        provided together with ``input_schema``.
+    pip : sequence of str, optional
+        Pip requirements for the remote environment.
+    env : mapping of str to str, optional
+        Non-secret environment variables. Use ``secrets`` for credentials.
+    secrets : sequence of str, optional
+        Names of secrets resolved by the remote service. Secret values are not
+        accepted by this API or included in the registration request.
+    python_version : str, optional
+        Remote Python major/minor version. Defaults to the client version.
+
+    Returns
+    -------
+    UdfDefinition
+        A callable definition accepted by
+        :meth:`lancedb.db.DBConnection.create_function`,
+        :meth:`lancedb.db.AsyncConnection.create_function_async` and
+        :meth:`lancedb.db.DBConnection.create_function_async`.
+
+    Examples
+    --------
+    >>> from lancedb import udf
+    >>> @udf(pip=["numpy==2.2.0"], secrets=["MODEL_TOKEN"])
+    ... def score(value: float) -> float:
+    ...     return value * 2
+    >>> score(1.5)
+    3.0
+    """
+
+    def decorate(target: Callable[..., Any]) -> UdfDefinition:
+        return UdfDefinition(
+            target,
+            name=name,
+            input_schema=input_schema,
+            output_schema=output_schema,
+            pip=tuple(pip),
+            env={} if env is None else env,
+            secrets=tuple(secrets),
+            python_version=python_version,
+        )
+
+    if function is None:
+        return decorate
+    return decorate(function)
+
+
+__all__ = [
+    "ApplicationInput",
+    "FunctionApplication",
+    "FunctionArtifact",
+    "FunctionArtifactContent",
+    "FunctionArtifactRequest",
+    "FunctionBinding",
+    "FunctionOutput",
+    "FunctionParameter",
+    "FunctionRegistrationRequest",
+    "FunctionResultField",
+    "FunctionSignature",
+    "FunctionVersion",
+    "FunctionVersionRef",
+    "InputBinding",
+    "OutputMapping",
+    "PythonEnvironmentSpec",
+    "PythonAdapterSpec",
+    "PythonRuntimeSpec",
+    "RefreshColumnResult",
+    "UdfDefinition",
+    "udf",
+]

python/python/lancedb/job.py

@@ -5,20 +5,23 @@
 
 import asyncio
 from datetime import timedelta
-from typing import Optional
+from typing import Any, Generic, Optional, TypeVar, cast
 
 from lancedb.background_loop import LOOP
 
 from . import _lancedb
+from .functions import FunctionVersion
+
+T = TypeVar("T")
 
 
-class AsyncJob:
+class AsyncJob(Generic[T]):
     """A handle to an operation that may still be running.
 
     The operation may already be complete when the handle is created.
     """
 
-    def __init__(self, inner: Optional["_lancedb.Job"]):
+    def __init__(self, inner: Optional[Any]):
         self._inner = inner
 
     @property
@@ -44,18 +47,20 @@ class AsyncJob:
             return "finished"
         return await self._inner.status()
 
-    async def wait(self, timeout: Optional[timedelta] = None):
+    async def wait(self, timeout: Optional[timedelta] = None) -> T:
         """Wait until the operation reaches a terminal state.
 
         Raises `JobFailedError` if the operation failed, `JobCancelledError`
         if it was cancelled, and `TimeoutError` if `timeout` elapses first.
         """
         if self._inner is None:
-            return
+            return cast(T, None)
         if timeout is None:
-            await self._inner.wait()
-        else:
-            await asyncio.wait_for(self._inner.wait(), timeout.total_seconds())
+            return cast(T, await self._inner.wait())
+        return cast(
+            T,
+            await asyncio.wait_for(self._inner.wait(), timeout.total_seconds()),
+        )
 
     async def cancel(self):
         """Request cancellation. Cancelling a finished operation is a no-op."""
@@ -64,10 +69,10 @@ class AsyncJob:
         await self._inner.cancel()
 
 
-class Job:
+class Job(Generic[T]):
     """Synchronous counterpart of `AsyncJob`."""
 
-    def __init__(self, inner: Optional[AsyncJob]):
+    def __init__(self, inner: Optional[AsyncJob[T]]):
         self._inner = inner
 
     @property
@@ -88,18 +93,64 @@ class Job:
             return "finished"
         return LOOP.run(self._inner.status())
 
-    def wait(self, timeout: Optional[timedelta] = None):
+    def wait(self, timeout: Optional[timedelta] = None) -> T:
         """Block until the operation reaches a terminal state.
 
         Raises `JobFailedError` if the operation failed, `JobCancelledError`
         if it was cancelled, and `TimeoutError` if `timeout` elapses first.
         """
         if self._inner is None:
-            return
-        LOOP.run(self._inner.wait(timeout))
+            return cast(T, None)
+        return LOOP.run(self._inner.wait(timeout))
 
     def cancel(self):
         """Request cancellation. Cancelling a finished operation is a no-op."""
         if self._inner is None:
             return
         LOOP.run(self._inner.cancel())
+
+
+class _FunctionJobAdapter:
+    def __init__(self, inner: "_lancedb.FunctionJob"):
+        self._inner = inner
+
+    @property
+    def id(self) -> Optional[str]:
+        return self._inner.id
+
+    async def status(self) -> str:
+        return await self._inner.status()
+
+    async def wait(self) -> FunctionVersion:
+        return FunctionVersion.from_json(await self._inner.wait())
+
+    async def cancel(self):
+        await self._inner.cancel()
+
+
+def _function_job(inner: "_lancedb.FunctionJob") -> AsyncJob[FunctionVersion]:
+    return AsyncJob(_FunctionJobAdapter(inner))
+
+
+class _RefreshJobAdapter:
+    def __init__(self, inner: "_lancedb.RefreshJob"):
+        self._inner = inner
+
+    @property
+    def id(self) -> Optional[str]:
+        return self._inner.id
+
+    async def status(self) -> str:
+        return await self._inner.status()
+
+    async def wait(self):
+        from .functions import RefreshColumnResult
+
+        return RefreshColumnResult.from_json(await self._inner.wait())
+
+    async def cancel(self):
+        await self._inner.cancel()
+
+
+def _refresh_job(inner: "_lancedb.RefreshJob") -> AsyncJob:
+    return AsyncJob(_RefreshJobAdapter(inner))

python/python/lancedb/pydantic.py

@@ -1,7 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-"""Pydantic (v1 / v2) adapter for LanceDB"""
+"""Pydantic adapter for LanceDB."""
 
 from __future__ import annotations
 
@@ -14,9 +14,6 @@ from enum import Enum
 from typing import (
     TYPE_CHECKING,
     Any,
-    Callable,
-    Dict,
-    Generator,
     List,
     Type,
     Union,
@@ -24,17 +21,9 @@ from typing import (
     GenericAlias,
 )
 
-import numpy as np
 import pyarrow as pa
 import pydantic
-from packaging.version import Version
-
-PYDANTIC_VERSION = Version(pydantic.__version__)
-try:
-    from pydantic_core import CoreSchema, core_schema
-except ImportError:
-    if PYDANTIC_VERSION.major >= 2:
-        raise
+from pydantic_core import CoreSchema, core_schema
 
 if TYPE_CHECKING:
     from pydantic.fields import FieldInfo
@@ -131,25 +120,6 @@ def Vector(
                 ),
             )
 
-        @classmethod
-        def __get_validators__(cls) -> Generator[Callable, None, None]:
-            yield cls.validate
-
-        # For pydantic v1
-        @classmethod
-        def validate(cls, v):
-            if not isinstance(v, (list, range, np.ndarray)) or len(v) != dim:
-                raise TypeError("A list of numbers or numpy.ndarray is needed")
-            return cls(v)
-
-        if PYDANTIC_VERSION.major < 2:
-
-            @classmethod
-            def __modify_schema__(cls, field_schema: Dict[str, Any]):
-                field_schema["items"] = {"type": "number"}
-                field_schema["maxItems"] = dim
-                field_schema["minItems"] = dim
-
     return FixedSizeList
 
 
@@ -157,9 +127,8 @@ def _raise_bare_vector_error(*_args):
     raise TypeError("Vector must be parameterized with a dimension, e.g. Vector(128).")
 
 
-# Pydantic v1 and v2 otherwise treat the bare Vector factory as a field validator
-# and inspect its signature, which produces misleading errors about internal types.
-setattr(Vector, "__get_validators__", _raise_bare_vector_error)
+# Pydantic otherwise inspects the bare factory as a field type and produces
+# misleading errors about its internal annotations.
 setattr(Vector, "__get_pydantic_core_schema__", _raise_bare_vector_error)
 
 
@@ -233,31 +202,6 @@ def MultiVector(
                 ),
             )
 
-        @classmethod
-        def __get_validators__(cls) -> Generator[Callable, None, None]:
-            yield cls.validate
-
-        # For pydantic v1
-        @classmethod
-        def validate(cls, v):
-            if not isinstance(v, (list, range)):
-                raise TypeError("A list of vectors is needed")
-            for vec in v:
-                if not isinstance(vec, (list, range, np.ndarray)) or len(vec) != dim:
-                    raise TypeError(f"Each vector must be a list of {dim} numbers")
-            return cls(v)
-
-        if PYDANTIC_VERSION.major < 2:
-
-            @classmethod
-            def __modify_schema__(cls, field_schema: Dict[str, Any]):
-                field_schema["items"] = {
-                    "type": "array",
-                    "items": {"type": "number"},
-                    "minItems": dim,
-                    "maxItems": dim,
-                }
-
     return MultiVectorList
 
 
@@ -303,20 +247,10 @@ def _py_type_to_arrow_type(py_type: Type[Any], field: FieldInfo) -> pa.DataType:
     )
 
 
-if PYDANTIC_VERSION.major < 2:
-
-    def _pydantic_model_to_fields(model: pydantic.BaseModel) -> List[pa.Field]:
-        return [
-            _pydantic_to_field(name, field) for name, field in model.__fields__.items()
-        ]
-
-else:
-
-    def _pydantic_model_to_fields(model: pydantic.BaseModel) -> List[pa.Field]:
-        return [
-            _pydantic_to_field(name, field)
-            for name, field in model.model_fields.items()
-        ]
+def _pydantic_model_to_fields(model: pydantic.BaseModel) -> List[pa.Field]:
+    return [
+        _pydantic_to_field(name, field) for name, field in model.model_fields.items()
+    ]
 
 
 def _pydantic_type_to_arrow_type(tp: Any, field: FieldInfo) -> pa.DataType:
@@ -509,8 +443,6 @@ class LanceModel(pydantic.BaseModel):
 
     @classmethod
     def safe_get_fields(cls):
-        if PYDANTIC_VERSION.major < 2:
-            return cls.__fields__
         return cls.model_fields
 
     @classmethod
@@ -548,23 +480,9 @@ def get_extras(field_info: FieldInfo, key: str) -> Any:
     """
     Get the extra metadata from a Pydantic FieldInfo.
     """
-    if PYDANTIC_VERSION.major >= 2:
-        return (field_info.json_schema_extra or {}).get(key)
-    return (field_info.field_info.extra or {}).get("json_schema_extra", {}).get(key)
+    return (field_info.json_schema_extra or {}).get(key)
 
 
-if PYDANTIC_VERSION.major < 2:
-
-    def model_to_dict(model: pydantic.BaseModel) -> Dict[str, Any]:
-        """
-        Convert a Pydantic model to a dictionary.
-        """
-        return model.dict()
-
-else:
-
-    def model_to_dict(model: pydantic.BaseModel) -> Dict[str, Any]:
-        """
-        Convert a Pydantic model to a dictionary.
-        """
-        return model.model_dump()
+def model_to_dict(model: pydantic.BaseModel) -> dict[str, Any]:
+    """Convert a Pydantic model to a dictionary."""
+    return model.model_dump()

python/python/lancedb/query.py

@@ -32,8 +32,6 @@ from typing_extensions import Annotated
 
 from lancedb._lancedb import fts_query_to_json
 from lancedb.background_loop import LOOP
-from lancedb.pydantic import PYDANTIC_VERSION
-
 from . import __version__
 from .arrow import AsyncRecordBatchReader
 from .dependencies import pandas as pd
@@ -827,12 +825,7 @@ class Query(pydantic.BaseModel):
 
     # This tells pydantic to allow custom types (needed for the `vector` query since
     # pa.Array wouln't be allowed otherwise)
-    if PYDANTIC_VERSION.major < 2:  # Pydantic 1.x compat
-
-        class Config:
-            arbitrary_types_allowed = True
-    else:
-        model_config = {"arbitrary_types_allowed": True}
+    model_config = pydantic.ConfigDict(arbitrary_types_allowed=True)
 
 
 class LanceQueryBuilder(ABC):
@@ -3251,12 +3244,7 @@ class AsyncStandardQuery(AsyncQueryBase):
         if ordering is None:
             self._inner.order_by(None)
         else:
-            self._inner.order_by(
-                [
-                    o.model_dump() if hasattr(o, "model_dump") else o.dict()
-                    for o in ordering
-                ]
-            )
+            self._inner.order_by([o.model_dump() for o in ordering])
         return self
 
     def fast_search(self) -> Self:

python/python/lancedb/remote/db.py

@@ -23,6 +23,7 @@ import pyarrow as pa
 
 from ..common import DATA
 from ..db import DBConnection, LOOP
+from ..functions import FunctionVersion, UdfDefinition
 from ..job import AsyncJob, Job
 
 if TYPE_CHECKING:
@@ -713,6 +714,14 @@ class RemoteDBConnection(DBConnection):
         """
         return Job(self._conn.job(job_id))
 
+    @override
+    def create_function_async(self, definition: UdfDefinition) -> Job[FunctionVersion]:
+        return Job(LOOP.run(self._conn.create_function_async(definition)))
+
+    @override
+    def get_function(self, name: str, *, version: str) -> FunctionVersion:
+        return LOOP.run(self._conn.get_function(name, version=version))
+
     @override
     def list_jobs(self) -> List["JobInfo"]:
         """List server-side jobs across the database's tables."""

python/python/lancedb/remote/table.py

@@ -49,6 +49,7 @@ from lancedb.index import (
     LabelList,
 )
 from lancedb.job import Job
+from lancedb.functions import FunctionApplication
 from lancedb.remote.db import LOOP
 from lancedb.table import IndexConfigType, KNOWN_METRICS
 import pyarrow as pa
@@ -960,7 +961,9 @@ class RemoteTable(Table):
 
     def add_columns(
         self,
-        transforms: Dict[str, str] | None = None,
+        transforms: Dict[str, str | FunctionApplication]
+        | FunctionApplication
+        | None = None,
         *,
         computed: Dict[str, str] | None = None,
     ) -> AddColumnsResult:

python/python/lancedb/table.py

@@ -72,6 +72,7 @@ from .index import (
     FTS,
 )
 from .expr import Expr
+from .functions import FunctionApplication
 from .merge import LanceMergeInsertBuilder
 from .pydantic import LanceModel, model_to_dict
 from .query import (
@@ -433,6 +434,20 @@ def _cast_to_target_schema(
     return pa.RecordBatchReader.from_batches(reordered_schema, gen())
 
 
+def _field_extension_name(field: pa.Field) -> Optional[str]:
+    extension_name = getattr(field.type, "extension_name", None)
+    if extension_name is not None:
+        return extension_name
+
+    metadata = field.metadata or {}
+    extension_name = metadata.get(b"ARROW:extension:name") or metadata.get(
+        "ARROW:extension:name"
+    )
+    if isinstance(extension_name, bytes):
+        return extension_name.decode()
+    return extension_name
+
+
 def _align_field_types(
     fields: List[pa.Field],
     target_fields: List[pa.Field],
@@ -445,6 +460,16 @@ def _align_field_types(
         target_field = next((f for f in target_fields if f.name == field.name), None)
         if target_field is None:
             raise ValueError(f"Field '{field.name}' not found in target schema")
+        # Preserve arrow.json input until it reaches Lance. LanceDB exposes stored
+        # JSON columns as lance.json (JSONB-backed LargeBinary), but casting the
+        # input to that storage type here merely relabels the raw JSON bytes as
+        # JSONB. Lance must see arrow.json so it can perform the JSONB encoding.
+        if (
+            _field_extension_name(field) == "arrow.json"
+            and _field_extension_name(target_field) == "lance.json"
+        ):
+            new_fields.append(field)
+            continue
         if pa.types.is_struct(target_field.type):
             if pa.types.is_struct(field.type):
                 new_type = pa.struct(
@@ -1918,7 +1943,8 @@ class Table(ABC):
     @abstractmethod
     def add_columns(
         self,
-        transforms: Dict[str, str]
+        transforms: Dict[str, str | FunctionApplication]
+        | FunctionApplication
         | pa.Field
         | List[pa.Field]
         | pa.Schema
@@ -1931,13 +1957,21 @@ class Table(ABC):
 
         Parameters
         ----------
-        transforms: Dict[str, str], pa.Field, List[pa.Field], pa.Schema
+        transforms: Dict[str, str | FunctionApplication], FunctionApplication,
+            pa.Field, List[pa.Field], pa.Schema
             A map of column name to a SQL expression to use to calculate the
             value of the new column. These expressions will be evaluated for
             each row in the table, and can reference existing columns.
             Alternatively, a pyarrow Field or Schema can be provided to add
             new columns with the specified data types. The new columns will
             be initialized with null values.
+
+            A mapping with one ``FunctionApplication`` value keeps its scalar
+            or named-struct result in the named table column. A bare
+            named-struct application expands its ordered result fields as one
+            atomic sibling group; aliases come from ``rename(columns=...)``.
+            Function columns are supported only on LanceDB Cloud and
+            Enterprise.
         computed: Dict[str, str], optional
             A map of column name to a SQL expression defining the column. The
             column's type and inputs are derived from the expression, so no
@@ -4032,9 +4066,10 @@ class LanceTable(Table):
 
     def add_columns(
         self,
-        transforms: Dict[str, str]
-        | pa.field
-        | List[pa.field]
+        transforms: Dict[str, str | FunctionApplication]
+        | FunctionApplication
+        | pa.Field
+        | List[pa.Field]
         | pa.Schema
         | None = None,
         *,
@@ -5968,9 +6003,10 @@ class AsyncTable:
 
     async def add_columns(
         self,
-        transforms: dict[str, str]
-        | pa.field
-        | List[pa.field]
+        transforms: dict[str, str | FunctionApplication]
+        | FunctionApplication
+        | pa.Field
+        | List[pa.Field]
         | pa.Schema
         | None = None,
         *,
@@ -5981,12 +6017,19 @@ class AsyncTable:
 
         Parameters
         ----------
-        transforms: Dict[str, str]
+        transforms: Dict[str, str | FunctionApplication] or FunctionApplication
             A map of column name to a SQL expression to use to calculate the
             value of the new column. These expressions will be evaluated for
             each row in the table, and can reference existing columns.
             Alternatively, you can pass a pyarrow field or schema to add
             new columns with NULLs.
+
+            A mapping with one ``FunctionApplication`` value keeps its scalar
+            or named-struct result in the named table column. A bare
+            named-struct application expands its ordered result fields as one
+            atomic sibling group; aliases come from ``rename(columns=...)``.
+            Function columns are supported only on LanceDB Cloud and
+            Enterprise.
         computed: Dict[str, str], optional
             A map of column name to a SQL expression defining the column. The
             column's type and inputs are derived from the expression.
@@ -6010,6 +6053,32 @@ class AsyncTable:
             version: the new version number of the table after adding columns.
 
         """
+        function_application = None
+        function_output_name = None
+        if isinstance(transforms, FunctionApplication):
+            function_application = transforms
+        elif isinstance(transforms, dict) and any(
+            isinstance(value, FunctionApplication) for value in transforms.values()
+        ):
+            if len(transforms) != 1 or not all(
+                isinstance(value, FunctionApplication) for value in transforms.values()
+            ):
+                raise ValueError(
+                    "one add_columns call declares exactly one Function sibling group"
+                )
+            function_output_name, function_application = next(iter(transforms.items()))
+
+        if function_application is not None:
+            if computed:
+                raise ValueError(
+                    "add_columns cannot mix a Function application with SQL "
+                    "computed columns"
+                )
+            function_application._ensure_declarable()
+            return await self._inner.add_function_columns(
+                function_application.to_canonical_json(), function_output_name
+            )
+
         if isinstance(transforms, pa.Field):
             transforms = [transforms]
         if isinstance(transforms, list) and all(
@@ -6079,7 +6148,9 @@ class AsyncTable:
         >>> asyncio.run(refresh_in_background())
         'finished'
         """
-        return AsyncJob(await self._inner.refresh_column_async(column))
+        from .job import _refresh_job
+
+        return _refresh_job(await self._inner.refresh_column_async(column))
 
     async def alter_columns(
         self, *alterations: Iterable[dict[str, Any]]

python/python/tests/test_first_class_function_slice1.py

@@ -0,0 +1,311 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+import json
+from pathlib import Path
+
+import pytest
+
+import lancedb.functions as functions
+from lancedb.functions import (
+    FunctionApplication,
+    FunctionBinding,
+    FunctionVersion,
+    PythonRuntimeSpec,
+    RefreshColumnResult,
+)
+from lancedb.table import AsyncTable
+
+
+FIXTURES = (
+    Path(__file__).parents[3]
+    / "rust"
+    / "lancedb"
+    / "tests"
+    / "fixtures"
+    / "first_class_functions"
+    / "v1"
+)
+
+
+def fixture(name: str) -> str:
+    return (FIXTURES / name).read_text()
+
+
+def job_result(name: str) -> dict:
+    return json.loads(fixture(name))["result"]
+
+
+def assert_no_secret_values(value):
+    if isinstance(value, dict):
+        for key, child in value.items():
+            assert key not in {
+                "secret_value",
+                "secret_values",
+                "resolved_secret",
+                "resolved_secrets",
+            }
+            assert_no_secret_values(child)
+    elif isinstance(value, list):
+        for child in value:
+            assert_no_secret_values(child)
+
+
+def test_public_function_values_are_in_api_reference():
+    docs = Path(__file__).parents[3] / "docs" / "src" / "python" / "python.md"
+    rendered = docs.read_text()
+    for name in functions.__all__:
+        assert f"::: lancedb.functions.{name}" in rendered
+
+
+@pytest.mark.parametrize(
+    ("fixture_name", "canonical_name", "model", "nested_result"),
+    [
+        (
+            "remote_function_job.json",
+            "remote_function_version.canonical.json",
+            FunctionVersion,
+            True,
+        ),
+        (
+            "remote_function_application.json",
+            "remote_function_application.canonical.json",
+            FunctionApplication,
+            False,
+        ),
+        (
+            "remote_function_binding.json",
+            "remote_function_binding.canonical.json",
+            FunctionBinding,
+            False,
+        ),
+        (
+            "remote_refresh_job.json",
+            "remote_refresh_result.canonical.json",
+            RefreshColumnResult,
+            True,
+        ),
+        (
+            "remote_refresh_result_without_published_version.json",
+            "remote_refresh_result_without_published_version.canonical.json",
+            RefreshColumnResult,
+            False,
+        ),
+    ],
+)
+def test_python_and_rust_share_remote_canonical_goldens(
+    fixture_name, canonical_name, model, nested_result
+):
+    value = json.loads(fixture(fixture_name))
+    if nested_result:
+        value = value["result"]
+    decoded = model.from_json(json.dumps(value))
+    assert decoded.to_canonical_json() == fixture(canonical_name).strip()
+
+
+def test_function_version_identity_is_immutable_and_exact():
+    value = job_result("remote_function_job.json")
+    version = FunctionVersion.from_json(json.dumps(value))
+    assert version.name == "embed"
+    assert version.version == "fv_01K3EXACT"
+    assert version.required_secrets == ("HF_TOKEN",)
+
+    with pytest.raises((TypeError, ValueError)):
+        version.version = "fv_changed"
+    with pytest.raises(TypeError, match="immutable"):
+        version.runtime.env["TOKENIZERS_PARALLELISM"] = "true"
+
+    changed = dict(value)
+    changed["version"] = "fv_changed"
+    assert FunctionVersion(**changed) != version
+
+
+def test_unknown_fields_and_discriminators_are_forward_decodable():
+    value = job_result("remote_function_job.json")
+    value["future_version_metadata"] = {"retention_class": "catalog"}
+    value["runtime"] = {"kind": "wasm", "module_digest": "sha256:wasm"}
+    value["signature"]["output"]["kind"] = "future_output_shape"
+
+    version = FunctionVersion.from_json(json.dumps(value))
+    assert version.runtime.kind == "wasm"
+    assert version.runtime.python_version is None
+    assert version.runtime.environment is None
+    assert json.loads(version.to_canonical_json())["runtime"] == {"kind": "wasm"}
+    assert version.signature.output.kind == "future_output_shape"
+
+
+def test_function_application_uses_rename_columns_only():
+    application = FunctionApplication.from_json(
+        fixture("remote_function_application.json")
+    )
+    renamed = application.rename(columns={"normalized_text": "body_normalized"})
+
+    assert application.columns["normalized_text"] == "search_text"
+    assert renamed.columns["normalized_text"] == "body_normalized"
+    assert renamed.function == application.function
+    assert renamed.group_id == application.group_id
+    assert not hasattr(application, "rename_outputs")
+    with pytest.raises(TypeError, match="immutable"):
+        renamed.columns["normalized_text"] = "changed"
+    with pytest.raises(TypeError, match="immutable"):
+        application.inputs[0].value["path"] = "changed"
+
+    with pytest.raises(ValueError, match="unknown Function result fields"):
+        application.rename(columns={"missing": "search_text"})
+    with pytest.raises(ValueError, match="destinations must be unique"):
+        application.rename(columns={"normalized_text": "same", "token_count": "same"})
+
+    bare_value = json.loads(fixture("remote_function_application.json"))
+    bare_value.pop("columns")
+    bare = FunctionApplication(**bare_value)
+    with pytest.raises(ValueError, match="destinations must be unique"):
+        bare.rename(columns={"normalized_text": "token_count"})
+
+
+def test_binding_and_refresh_result_keep_stable_remote_fields():
+    binding = FunctionBinding.from_json(fixture("remote_function_binding.json"))
+    assert binding.revision == 3
+    assert binding.function.version == "fv_01K3TEXT"
+    assert [output.output_ordinal for output in binding.outputs] == [0, 1]
+    assert binding.input_schema is not None
+    assert binding.output_schema is not None
+
+    result = RefreshColumnResult.from_json(
+        json.dumps(job_result("remote_refresh_job.json"))
+    )
+    assert result.rows_filled == result.rows_assigned
+    assert result.version == result.published_version
+
+    result = RefreshColumnResult.from_json(
+        fixture("remote_refresh_result_without_published_version.json")
+    )
+    assert result.published_version is None
+    assert RefreshColumnResult.from_json(result.to_canonical_json()) == result
+
+
+def test_function_literal_numeric_domain_matches_rust():
+    with pytest.raises(ValueError, match="floating-point Function literals"):
+        FunctionApplication.from_json(fixture("remote_function_application_float.json"))
+
+    value = json.loads(fixture("remote_function_application_float.json"))
+    value["inputs"][0]["value"] = 2**64
+    with pytest.raises(ValueError, match="outside the canonical JSON range"):
+        FunctionApplication.from_json(json.dumps(value))
+
+
+def test_empty_default_maps_have_stable_canonical_bytes():
+    runtime = PythonRuntimeSpec(
+        kind="python", python_version="3.12", environment={"kind": "pip"}
+    )
+    assert runtime.to_canonical_json() == (
+        '{"environment":{"kind":"pip"},"kind":"python","python_version":"3.12"}'
+    )
+
+    value = json.loads(fixture("remote_function_application.json"))
+    value.pop("columns")
+    application = FunctionApplication.from_json(json.dumps(value))
+    assert "columns" not in json.loads(application.to_canonical_json())
+
+
+@pytest.mark.parametrize("field", ["rows_assigned", "source_version"])
+def test_refresh_result_rejects_non_u64_values(field):
+    value = job_result("remote_refresh_job.json")
+    value[field] = -1
+    with pytest.raises(ValueError):
+        RefreshColumnResult.from_json(json.dumps(value))
+
+    value[field] = "1"
+    with pytest.raises(ValueError):
+        RefreshColumnResult.from_json(json.dumps(value))
+
+
+def test_canonical_client_values_contain_secret_names_only():
+    version = FunctionVersion.from_json(
+        json.dumps(job_result("remote_function_job.json"))
+    )
+    canonical = json.loads(version.to_canonical_json())
+    assert canonical["required_secrets"] == ["HF_TOKEN"]
+    assert_no_secret_values(canonical)
+
+
+class _FunctionDeclarationInner:
+    def __init__(self):
+        self.calls = []
+
+    async def add_function_columns(self, application_json, output_name):
+        self.calls.append((json.loads(application_json), output_name))
+        return "declared"
+
+
+def known_application() -> FunctionApplication:
+    value = json.loads(fixture("remote_function_application.json"))
+    value.pop("future_application")
+    return FunctionApplication(**value)
+
+
+@pytest.mark.asyncio
+async def test_add_columns_routes_struct_as_one_and_grouped_expansion_atomically():
+    inner = _FunctionDeclarationInner()
+    table = AsyncTable(inner)
+    application = known_application()
+
+    result = await table.add_columns(
+        {"features": application._copy(update={"columns": {}})}
+    )
+    assert result == "declared"
+    assert inner.calls[-1][1] == "features"
+
+    bare = application._copy(update={"columns": {}}).rename(
+        columns={"normalized_text": "search_text"}
+    )
+    result = await table.add_columns(bare)
+    assert result == "declared"
+    assert inner.calls[-1][1] is None
+    assert inner.calls[-1][0]["columns"] == {"normalized_text": "search_text"}
+
+
+@pytest.mark.asyncio
+async def test_add_columns_rejects_mixed_groups_and_unknown_newer_application():
+    inner = _FunctionDeclarationInner()
+    table = AsyncTable(inner)
+    application = known_application()
+
+    with pytest.raises(ValueError, match="exactly one Function sibling group"):
+        await table.add_columns({"a": application, "b": application})
+
+    future = json.loads(fixture("remote_function_application.json"))
+    application = FunctionApplication(**future)
+    with pytest.raises(ValueError, match="newer contract"):
+        await table.add_columns(application)
+
+    future.pop("future_application")
+    future["output"]["assignment"] = "cell_flag"
+    application = FunctionApplication(**future)
+    assert "assignment" not in json.loads(application.to_canonical_json())["output"]
+    with pytest.raises(ValueError, match="output.assignment"):
+        await table.add_columns(application)
+    assert inner.calls == []
+
+
+def test_rename_requires_named_struct_and_keeps_partial_mapping_immutable():
+    scalar = FunctionApplication.from_json(
+        json.dumps(
+            {
+                "function": {"name": "embed", "version": "fv_exact"},
+                "inputs": [],
+                "output": {
+                    "kind": "scalar",
+                    "arrow_type": "list<float32>",
+                    "nullable": False,
+                },
+                "group_id": "fg_scalar",
+            }
+        )
+    )
+    with pytest.raises(ValueError, match="named-struct"):
+        scalar.rename(columns={"value": "embedding"})
+
+    application = known_application()._copy(update={"columns": {}})
+    renamed = application.rename(columns={"normalized_text": "search_text"})
+    assert dict(application.columns) == {}
+    assert dict(renamed.columns) == {"normalized_text": "search_text"}

python/python/tests/test_first_class_function_slice2.py

@@ -0,0 +1,254 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+from __future__ import annotations
+
+import contextlib
+import http.server
+import json
+from pathlib import Path
+import threading
+from typing import Optional
+
+import pyarrow as pa
+import pytest
+
+import lancedb
+from lancedb.functions import UdfDefinition, udf
+
+
+FIXTURES = (
+    Path(__file__).parents[3]
+    / "rust"
+    / "lancedb"
+    / "tests"
+    / "fixtures"
+    / "first_class_functions"
+    / "v1"
+)
+
+
+@udf(
+    pip=["numpy>=2"],
+    env={"MODE": "test"},
+    secrets=["API_TOKEN"],
+    python_version="3.12",
+)
+def normalize_score(value: float) -> float:
+    return value / 100.0
+
+
+def _assert_no_secret_values(value):
+    if isinstance(value, dict):
+        for key, child in value.items():
+            assert key not in {
+                "secret_value",
+                "secret_values",
+                "resolved_secret",
+                "resolved_secrets",
+            }
+            _assert_no_secret_values(child)
+    elif isinstance(value, list):
+        for child in value:
+            _assert_no_secret_values(child)
+
+
+def test_scalar_udf_matches_shared_registration_golden_and_remains_callable():
+    assert isinstance(normalize_score, UdfDefinition)
+    assert normalize_score(25.0) == 0.25
+    assert (
+        normalize_score.registration_request.to_canonical_json()
+        == (FIXTURES / "remote_function_registration_request.canonical.json")
+        .read_text()
+        .strip()
+    )
+    request = json.loads(normalize_score.registration_request.to_canonical_json())
+    assert request["artifact"]["adapter"] == {
+        "kind": "scalar_to_arrow_batch",
+        "version": 1,
+    }
+    assert request["required_secrets"] == ["API_TOKEN"]
+    _assert_no_secret_values(request)
+
+
+def test_explicit_arrow_schema_is_deterministic():
+    input_schema = pa.schema([pa.field("value", pa.float32(), nullable=True)])
+    output_schema = pa.field("embedding", pa.list_(pa.float32(), 3), nullable=False)
+
+    @udf(input_schema=input_schema, output_schema=output_schema)
+    def explicit(value):
+        return [value, value, value]
+
+    signature = explicit.registration_request.signature
+    assert signature.inputs[0].arrow_type == "float32"
+    assert signature.inputs[0].nullable is True
+    assert signature.output.arrow_type == "fixed_size_list<float32>[3]"
+    assert signature.output.nullable is False
+
+
+def test_annotation_and_explicit_schema_validation_fail_closed():
+    with pytest.raises(TypeError, match="missing Function annotations"):
+
+        @udf
+        def missing(value):
+            return value
+
+    with pytest.raises(TypeError, match="unsupported Function annotation"):
+
+        @udf
+        def unsupported(value: set[str]) -> str:
+            return ""
+
+    with pytest.raises(ValueError, match="output must be non-nullable"):
+
+        @udf
+        def nullable_output(value: int) -> Optional[int]:
+            return value
+
+    with pytest.raises(ValueError, match="provided together"):
+
+        @udf(input_schema=pa.schema([pa.field("value", pa.int64())]))
+        def partial_schema(value):
+            return value
+
+    with pytest.raises(ValueError, match="exactly match callable parameters"):
+
+        @udf(
+            input_schema=pa.schema([pa.field("other", pa.int64())]),
+            output_schema=pa.int64(),
+        )
+        def wrong_name(value):
+            return value
+
+    with pytest.raises(ValueError, match="output must be non-nullable"):
+
+        @udf(
+            input_schema=pa.schema([pa.field("value", pa.int64())]),
+            output_schema=pa.field("result", pa.int64(), nullable=True),
+        )
+        def nullable_explicit(value):
+            return value
+
+
+def test_environment_rejects_secret_value_overlap():
+    with pytest.raises(ValueError, match="must be disjoint"):
+
+        @udf(env={"TOKEN": "plaintext"}, secrets=["TOKEN"])
+        def overlapping(value: int) -> int:
+            return value
+
+
+def test_local_function_catalog_operations_are_not_supported(tmp_path):
+    db = lancedb.connect(tmp_path)
+    message = "Function catalog operations are not supported by this database"
+    with pytest.raises(NotImplementedError, match=message):
+        db.create_function(normalize_score)
+    with pytest.raises(NotImplementedError, match=message):
+        db.create_function_async(normalize_score)
+    with pytest.raises(NotImplementedError, match=message):
+        db.get_function("normalize_score", version="fv_exact")
+
+
+@contextlib.contextmanager
+def _mock_remote_function_catalog():
+    state = {"requests": [], "version": None}
+
+    class Handler(http.server.BaseHTTPRequestHandler):
+        def log_message(self, *args):
+            pass
+
+        def do_POST(self):
+            length = int(self.headers.get("Content-Length", "0"))
+            body = json.loads(self.rfile.read(length) or b"{}")
+            state["requests"].append((self.path, body))
+            status = 200
+            if self.path == "/v1/function/create":
+                state["version"] = {
+                    "name": body["name"],
+                    "version": "fv_exact",
+                    "artifact": {
+                        key: body["artifact"][key]
+                        for key in ("kind", "digest", "entrypoint")
+                    },
+                    "signature": body["signature"],
+                    "runtime": body["runtime"],
+                    "runtime_digest": "sha256:runtime",
+                    "environment_digest": "sha256:environment",
+                    "required_secrets": body.get("required_secrets", []),
+                    "created_at": "2026-08-21T00:00:00Z",
+                }
+                response = {"job_id": "job-register"}
+                status = 202
+            elif self.path == "/v1/jobs/describe":
+                assert body == {"job_id": "job-register"}
+                response = {
+                    "job_id": "job-register",
+                    "job_type": "create_function",
+                    "job_state": "DONE",
+                    "result": state["version"],
+                }
+            elif self.path == "/v1/function/describe":
+                assert body == {
+                    "name": "normalize_score",
+                    "version": "fv_exact",
+                }
+                response = state["version"]
+            else:
+                status = 404
+                response = {"error": "not found"}
+            encoded = json.dumps(response).encode()
+            self.send_response(status)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(encoded)))
+            self.end_headers()
+            self.wfile.write(encoded)
+
+    with http.server.HTTPServer(("localhost", 0), Handler) as server:
+        thread = threading.Thread(target=server.serve_forever)
+        thread.start()
+        try:
+            yield f"http://localhost:{server.server_address[1]}", state
+        finally:
+            server.shutdown()
+            thread.join()
+
+
+def test_remote_registration_job_and_exact_version_reopen_round_trip():
+    with _mock_remote_function_catalog() as (host, state):
+        db = lancedb.connect(
+            "db://dev",
+            api_key="fake",
+            host_override=host,
+            client_config={"retry_config": {"retries": 0}},
+        )
+        registration = db.create_function_async(normalize_score)
+        assert registration.id == "job-register"
+        created = registration.wait()
+        reopened = db.get_function("normalize_score", version=created.version)
+
+    assert created == reopened
+    assert reopened.name == "normalize_score"
+    assert reopened.version == "fv_exact"
+    create_request = state["requests"][0][1]
+    assert create_request == json.loads(
+        normalize_score.registration_request.to_canonical_json()
+    )
+    _assert_no_secret_values(create_request)
+
+
+def test_blocking_remote_registration_returns_function_version():
+    with _mock_remote_function_catalog() as (host, state):
+        db = lancedb.connect(
+            "db://dev",
+            api_key="fake",
+            host_override=host,
+            client_config={"retry_config": {"retries": 0}},
+        )
+        created = db.create_function(normalize_score)
+
+    assert created.name == "normalize_score"
+    assert created.version == "fv_exact"
+    assert [path for path, _ in state["requests"]] == [
+        "/v1/function/create",
+        "/v1/jobs/describe",
+    ]

python/python/tests/test_lsm_write_spec.py

@@ -21,6 +21,11 @@ SCHEMA = pa.schema(
 )
 
 
+def test_lsm_write_spec_module_metadata():
+    assert lancedb.LsmWriteSpec is LsmWriteSpec
+    assert LsmWriteSpec.__module__ == "lancedb._lancedb"
+
+
 def _batch(ids, vs):
     return pa.RecordBatch.from_arrays(
         [pa.array(ids, type=pa.utf8()), pa.array(vs, type=pa.int32())],

python/python/tests/test_pydantic.py

@@ -10,11 +10,10 @@ import pyarrow as pa
 import pydantic
 import pytest
 from lancedb.pydantic import (
-    PYDANTIC_VERSION,
     LanceModel,
+    MultiVector,
     Vector,
     pydantic_to_schema,
-    MultiVector,
 )
 from pydantic import BaseModel
 from pydantic import Field
@@ -432,16 +431,10 @@ def test_fixed_size_list_field():
         li: List[int]
 
     data = TestModel(vec=list(range(16)), li=[1, 2, 3])
-    if PYDANTIC_VERSION.major >= 2:
-        assert json.loads(data.model_dump_json()) == {
-            "vec": list(range(16)),
-            "li": [1, 2, 3],
-        }
-    else:
-        assert data.dict() == {
-            "vec": list(range(16)),
-            "li": [1, 2, 3],
-        }
+    assert json.loads(data.model_dump_json()) == {
+        "vec": list(range(16)),
+        "li": [1, 2, 3],
+    }
 
     schema = pydantic_to_schema(TestModel)
     assert schema == pa.schema(
@@ -451,10 +444,7 @@ def test_fixed_size_list_field():
         ]
     )
 
-    if PYDANTIC_VERSION.major >= 2:
-        json_schema = TestModel.model_json_schema()
-    else:
-        json_schema = TestModel.schema()
+    json_schema = TestModel.model_json_schema()
 
     assert json_schema == {
         "properties": {

python/python/tests/test_table.py

@@ -2772,6 +2772,56 @@ async def test_merge_insert_async(mem_db_async: AsyncConnection):
     assert (await table.to_arrow()).sort_by("a") == expected
 
 
+@pytest.mark.skipif(not hasattr(pa, "json_"), reason="requires PyArrow JSON type")
+@pytest.mark.asyncio
+async def test_merge_insert_encodes_json(mem_db_async: AsyncConnection):
+    json_type = pa.json_()
+    schema = pa.schema([pa.field("id", pa.string()), pa.field("j", json_type)])
+
+    def json_table(rows):
+        json_values = pa.ExtensionArray.from_storage(
+            json_type,
+            pa.array([value for _, value in rows], type=json_type.storage_type),
+        )
+        return pa.Table.from_arrays(
+            [pa.array([row_id for row_id, _ in rows]), json_values], schema=schema
+        )
+
+    table = await mem_db_async.create_table("json_merge", schema=schema)
+    await table.add(json_table([("a", '{"k": 1}'), ("b", '{"k": 9}')]))
+
+    await (
+        table.merge_insert("id")
+        .when_matched_update_all()
+        .execute(json_table([("a", '{"k": 2}')]))
+    )
+
+    rows = sorted(await table.query().to_list(), key=lambda row: row["id"])
+    assert rows == [
+        {"id": "a", "j": '{"k":2}'},
+        {"id": "b", "j": '{"k":9}'},
+    ]
+    filtered = await table.query().where("json_extract(j, '$.k') = '2'").to_list()
+    assert filtered == [{"id": "a", "j": '{"k":2}'}]
+
+
+@pytest.mark.skipif(not hasattr(pa, "json_"), reason="requires PyArrow JSON type")
+@pytest.mark.asyncio
+async def test_add_sanitization_encodes_json(mem_db_async: AsyncConnection):
+    json_type = pa.json_()
+    schema = pa.schema([pa.field("id", pa.string()), pa.field("j", json_type)])
+    json_values = pa.ExtensionArray.from_storage(
+        json_type, pa.array(['{"k": 3}'], type=json_type.storage_type)
+    )
+    data = pa.Table.from_arrays([pa.array(["c"]), json_values], schema=schema)
+
+    table = await mem_db_async.create_table("json_add", schema=schema)
+    await table.add(data, on_bad_vectors="fill")
+
+    rows = await table.query().where("json_extract(j, '$.k') = '3'").to_list()
+    assert rows == [{"id": "c", "j": '{"k":3}'}]
+
+
 def test_create_with_embedding_function(mem_db: DBConnection):
     class MyTable(LanceModel):
         text: str

python/src/connection.rs

@@ -563,6 +563,38 @@ impl Connection {
         Ok(crate::job::Job::new(inner.job(job_id).infer_error()?))
     }
 
+    pub fn create_function_async(
+        self_: PyRef<'_, Self>,
+        request_json: String,
+    ) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.get_inner()?.clone();
+        let request = lancedb::function::FunctionRegistrationRequest::from_json(&request_json)
+            .infer_error()?;
+        future_into_py(self_.py(), async move {
+            inner
+                .create_function_async(request)
+                .await
+                .infer_error()
+                .map(crate::job::FunctionJob::new)
+        })
+    }
+
+    pub fn get_function(
+        self_: PyRef<'_, Self>,
+        name: String,
+        version: String,
+    ) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.get_inner()?.clone();
+        future_into_py(self_.py(), async move {
+            inner
+                .get_function(name, version)
+                .await
+                .infer_error()?
+                .to_canonical_json()
+                .infer_error()
+        })
+    }
+
     pub fn list_jobs(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
         let inner = self_.get_inner()?.clone();
         future_into_py(self_.py(), async move {

python/src/job.rs

@@ -13,6 +13,29 @@ pub struct Job {
     inner: Arc<lancedb::Job>,
 }
 
+/// Python bridge for a typed remote Function registration job.
+///
+/// The public Python layer decodes the canonical JSON returned by `wait`
+/// into its immutable `FunctionVersion` model.
+#[pyclass]
+pub struct FunctionJob {
+    inner: Arc<lancedb::Job<lancedb::function::FunctionVersion>>,
+}
+
+/// Python bridge for a typed remote Function-column refresh job.
+#[pyclass]
+pub struct RefreshJob {
+    inner: Arc<lancedb::Job>,
+}
+
+impl FunctionJob {
+    pub(crate) fn new(inner: lancedb::Job<lancedb::function::FunctionVersion>) -> Self {
+        Self {
+            inner: Arc::new(inner),
+        }
+    }
+}
+
 impl Job {
     pub(crate) fn new(inner: lancedb::Job) -> Self {
         Self {
@@ -21,6 +44,92 @@ impl Job {
     }
 }
 
+impl RefreshJob {
+    pub(crate) fn new(inner: lancedb::Job) -> Self {
+        Self {
+            inner: Arc::new(inner),
+        }
+    }
+}
+
+#[pymethods]
+impl RefreshJob {
+    #[getter]
+    pub fn id(&self) -> Option<String> {
+        self.inner.id().map(str::to_string)
+    }
+
+    pub fn status(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner.clone();
+        future_into_py(
+            self_.py(),
+            async move { inner.status().await.infer_error() },
+        )
+    }
+
+    pub fn wait(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            let value = inner.wait_raw_json().await.infer_error()?.ok_or_else(|| {
+                pyo3::exceptions::PyValueError::new_err(
+                    "successful refresh job did not contain a result",
+                )
+            })?;
+            let result: lancedb::function::RefreshColumnResult = serde_json::from_value(value)
+                .map_err(|error| {
+                    pyo3::exceptions::PyValueError::new_err(format!(
+                        "invalid refresh job result: {error}"
+                    ))
+                })?;
+            result.to_canonical_json().infer_error()
+        })
+    }
+
+    pub fn cancel(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            inner.cancel().await.infer_error()?;
+            Ok(())
+        })
+    }
+}
+
+#[pymethods]
+impl FunctionJob {
+    #[getter]
+    pub fn id(&self) -> Option<String> {
+        self.inner.id().map(str::to_string)
+    }
+
+    pub fn status(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner.clone();
+        future_into_py(
+            self_.py(),
+            async move { inner.status().await.infer_error() },
+        )
+    }
+
+    pub fn wait(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            inner
+                .wait()
+                .await
+                .infer_error()?
+                .to_canonical_json()
+                .infer_error()
+        })
+    }
+
+    pub fn cancel(self_: PyRef<'_, Self>) -> PyResult<Bound<'_, PyAny>> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            inner.cancel().await.infer_error()?;
+            Ok(())
+        })
+    }
+}
+
 #[pymethods]
 impl Job {
     #[getter]

python/src/lib.rs

@@ -47,6 +47,8 @@ pub fn _lancedb(_py: Python, m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_class::<Session>()?;
     m.add_class::<Table>()?;
     m.add_class::<crate::job::Job>()?;
+    m.add_class::<crate::job::FunctionJob>()?;
+    m.add_class::<crate::job::RefreshJob>()?;
     m.add_class::<crate::job::JobInfo>()?;
     m.add_class::<crate::job::JobDescription>()?;
     m.add_class::<crate::job::JobFailureInfo>()?;

python/src/table.rs

@@ -262,7 +262,7 @@ fn fmt_maintained(maintained: &Option<Vec<String>>) -> String {
 /// classmethods, then optionally chain `with_maintained_indexes(...)` and
 /// `with_writer_config_defaults(...)`. A fresh spec maintains every index the
 /// MemWAL supports, resolved on install.
-#[pyclass(from_py_object)]
+#[pyclass(module = "lancedb._lancedb", from_py_object)]
 #[derive(Clone, Debug)]
 pub struct LsmWriteSpec {
     inner: lancedb::table::LsmWriteSpec,
@@ -1551,6 +1551,24 @@ impl Table {
         })
     }
 
+    pub fn add_function_columns(
+        self_: PyRef<'_, Self>,
+        application_json: String,
+        output_name: Option<String>,
+    ) -> PyResult<Bound<'_, PyAny>> {
+        let application =
+            lancedb::function::FunctionApplication::from_json(&application_json).infer_error()?;
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            let builder = match output_name {
+                Some(name) => inner.add_columns().function_as(name, application),
+                None => inner.add_columns().function(application),
+            };
+            let result = builder.execute().await.infer_error()?;
+            Ok(AddColumnsResult::from(result))
+        })
+    }
+
     pub fn refresh_column(self_: PyRef<'_, Self>, column: String) -> PyResult<Bound<'_, PyAny>> {
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
@@ -1566,7 +1584,7 @@ impl Table {
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
             let job = inner.refresh_column_async(column).await.infer_error()?;
-            Ok(crate::job::Job::new(job))
+            Ok(crate::job::RefreshJob::new(job))
         })
     }
 

python/uv.lock

@@ -2003,7 +2003,7 @@ requires-dist = [
     { name = "pyarrow", specifier = ">=16" },
     { name = "pyarrow", marker = "extra == 'tests'", specifier = "<25" },
     { name = "pyarrow-stubs", marker = "extra == 'tests'", specifier = ">=16.0" },
-    { name = "pydantic", specifier = ">=1.10" },
+    { name = "pydantic", specifier = ">=2.7.4,<3" },
     { name = "pylance", marker = "extra == 'pylance'", specifier = ">=5.0.0b5" },
     { name = "pylance", marker = "extra == 'tests'", specifier = "==9.0.0rc1" },
     { name = "pyright", marker = "extra == 'dev'", specifier = ">=1.1.350" },

rust/lancedb/Cargo.toml

@@ -51,20 +51,20 @@ metrics = { workspace = true, optional = true }
 metrics-util = { workspace = true, optional = true }
 moka = { workspace = true }
 pin-project = { workspace = true }
-tokio = { version = "1.23", features = ["rt-multi-thread", "sync"] }
+tokio = { workspace = true }
 log.workspace = true
-async-trait = "0"
-bytes = "1"
+async-trait = { workspace = true }
+bytes = { workspace = true }
 futures.workspace = true
 num-traits.workspace = true
 url.workspace = true
 rand.workspace = true
 regex.workspace = true
-serde = { version = "^1" }
-serde_json = { version = "1" }
+serde = { workspace = true }
+serde_json = { workspace = true }
 async-openai = { version = "0.20.0", optional = true }
 serde_with = { version = "3.8.1" }
-tempfile = "3.5.0"
+tempfile = { workspace = true }
 aws-sdk-bedrockruntime = { version = "1.27.0", optional = true }
 # For remote feature
 reqwest = { version = "0.12.0", default-features = false, features = [
@@ -79,7 +79,7 @@ reqwest = { version = "0.12.0", default-features = false, features = [
 ], optional = true }
 http = { version = "1", optional = true } # Matching what is in reqwest
 urlencoding = { version = "2", optional = true }
-uuid = { version = "1.7.0", features = ["v4", "v5"] }
+uuid = { workspace = true, features = ["v5"] }
 polars-arrow = { version = ">=0.37,<0.40.0", optional = true }
 polars = { version = ">=0.37,<0.40.0", optional = true }
 hf-hub = { version = "0.4.1", optional = true, default-features = false, features = [
@@ -96,11 +96,11 @@ semver = { workspace = true }
 [dev-dependencies]
 anyhow = "1"
 lance-testing = { workspace = true }
-tempfile = "3.5.0"
+tempfile = { workspace = true }
 random_word = { version = "0.4.3", features = ["en"] }
 roaring = "0.11.4"
-tokio = { version = "1.23", features = ["io-util", "macros", "net", "rt-multi-thread", "sync", "test-util"] }
-uuid = { version = "1.7.0", features = ["v4"] }
+tokio = { workspace = true, features = ["io-util", "macros", "net", "test-util"] }
+uuid = { workspace = true }
 walkdir = "2"
 aws-sdk-dynamodb = { version = "1.55.0" }
 aws-sdk-s3 = { version = "1.55.0" }

rust/lancedb/examples/bench_open_missing_table.rs

@@ -1,14 +1,14 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
-// Release benchmark for opening a missing table as sibling-table cardinality grows.
+// Benchmark for opening a missing table as sibling-table cardinality grows.
 //
 // The fixture uses real `.lance` directories and marker files. Fixture creation is
 // outside the timed section. Defaults intentionally cover 1k, 10k, and 100k siblings
 // with 10 warmups and 100 distinct missing-table opens per scale:
 //
 // ```text
-// cargo run --release -p lancedb --example bench_open_missing_table
+// cargo run --profile release-no-lto -p lancedb --example bench_open_missing_table
 // ```
 //
 // `BENCH_SIBLINGS`, `BENCH_WARMUPS`, and `BENCH_TRIALS` override those defaults.

rust/lancedb/examples/bench_streaming_dataloader.rs

@@ -5,10 +5,10 @@
 //! streaming dataloader.
 //!
 //! Normal sweep:
-//!   cargo run --release --example bench_streaming_dataloader
+//!   cargo run --profile release-with-debug --example bench_streaming_dataloader
 //!
 //! Flamegraph (self-contained, no perf/dtrace needed):
-//!   BENCH_PROFILE=1 BENCH_CHUNK=64 cargo run --release \
+//!   BENCH_PROFILE=1 BENCH_CHUNK=64 cargo run --profile release-with-debug \
 //!       --example bench_streaming_dataloader
 //!   # writes flamegraph.svg in the current directory
 //!

rust/lancedb/src/connection.rs

@@ -496,6 +496,33 @@ impl Connection {
         )
     }
 
+    /// Register a Python callable as a new immutable Function version.
+    ///
+    /// Registration is remote-only and always asynchronous. Waiting on the
+    /// returned typed job yields the durable [`crate::function::FunctionVersion`].
+    /// Local databases return [`Error::NotSupported`].
+    pub async fn create_function_async(
+        &self,
+        request: crate::function::FunctionRegistrationRequest,
+    ) -> Result<crate::job::Job<crate::function::FunctionVersion>> {
+        self.internal.create_function_async(request).await
+    }
+
+    /// Look up one exact immutable Function version in the remote catalog.
+    ///
+    /// Both the logical name and server-assigned version id are required;
+    /// mutable aliases and latest-version lookup are intentionally absent.
+    /// Local databases return [`Error::NotSupported`].
+    pub async fn get_function(
+        &self,
+        name: impl AsRef<str>,
+        version: impl AsRef<str>,
+    ) -> Result<crate::function::FunctionVersion> {
+        self.internal
+            .get_function(name.as_ref(), version.as_ref())
+            .await
+    }
+
     /// Rename a table in the database.
     ///
     /// This is only supported in LanceDB Cloud.

rust/lancedb/src/database.rs

@@ -241,6 +241,12 @@ fn job_op_not_supported<T>(what: &str) -> Result<T> {
     })
 }
 
+fn function_catalog_not_supported<T>() -> Result<T> {
+    Err(crate::error::Error::NotSupported {
+        message: "Function catalog operations are not supported by this database".to_string(),
+    })
+}
+
 /// The `Database` trait defines the interface for database implementations.
 ///
 /// A database is responsible for managing tables and their metadata.
@@ -286,6 +292,21 @@ pub trait Database:
     ///
     /// See [`CloneTableRequest`] for detailed documentation and examples.
     async fn clone_table(&self, request: CloneTableRequest) -> Result<Arc<dyn BaseTable>>;
+    /// Register an immutable Function version through the remote catalog.
+    async fn create_function_async(
+        &self,
+        _request: crate::function::FunctionRegistrationRequest,
+    ) -> Result<crate::job::Job<crate::function::FunctionVersion>> {
+        function_catalog_not_supported()
+    }
+    /// Look up one exact immutable Function version.
+    async fn get_function(
+        &self,
+        _name: &str,
+        _version: &str,
+    ) -> Result<crate::function::FunctionVersion> {
+        function_catalog_not_supported()
+    }
     /// A [`crate::job::Job`] handle for a server-side job by id, suitable for
     /// waiting on or cancelling the job. The handle is constructed without a
     /// server round trip; an unknown id surfaces when the handle is used.

rust/lancedb/src/function.rs

@@ -0,0 +1,606 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+//! Canonical values exchanged with the Enterprise Function service.
+//!
+//! This module contains client/wire values only. Catalog persistence,
+//! environment bake, secret resolution, and execution are owned by Sophon.
+
+use std::collections::BTreeMap;
+
+use serde::de::{self, DeserializeOwned};
+use serde::{Deserialize, Deserializer, Serialize, Serializer};
+use serde_json::Value;
+
+use crate::{Error, Result};
+
+fn invalid_json(error: impl std::fmt::Display) -> Error {
+    Error::InvalidInput {
+        message: format!("invalid remote Function JSON: {error}"),
+    }
+}
+
+fn write_canonical_json(value: &Value, output: &mut String) -> serde_json::Result<()> {
+    match value {
+        Value::Object(map) => {
+            output.push('{');
+            let mut entries = map.iter().collect::<Vec<_>>();
+            entries.sort_unstable_by_key(|(key, _)| *key);
+            for (index, (key, value)) in entries.into_iter().enumerate() {
+                if index != 0 {
+                    output.push(',');
+                }
+                output.push_str(&serde_json::to_string(key)?);
+                output.push(':');
+                write_canonical_json(value, output)?;
+            }
+            output.push('}');
+        }
+        Value::Array(values) => {
+            output.push('[');
+            for (index, value) in values.iter().enumerate() {
+                if index != 0 {
+                    output.push(',');
+                }
+                write_canonical_json(value, output)?;
+            }
+            output.push(']');
+        }
+        other => output.push_str(&serde_json::to_string(other)?),
+    }
+    Ok(())
+}
+
+fn canonical_json<T: Serialize>(value: &T) -> Result<String> {
+    let value = serde_json::to_value(value).map_err(invalid_json)?;
+    let mut output = String::new();
+    write_canonical_json(&value, &mut output).map_err(invalid_json)?;
+    Ok(output)
+}
+
+fn from_json<T: DeserializeOwned>(json: &str) -> Result<T> {
+    serde_json::from_str(json).map_err(invalid_json)
+}
+
+fn validate_literal(value: &Value) -> Result<()> {
+    match value {
+        Value::Number(number) if number.is_f64() => Err(Error::InvalidInput {
+            message: "floating-point Function literals are not part of the Slice 1 canonical wire contract"
+                .to_string(),
+        }),
+        Value::Array(values) => values.iter().try_for_each(validate_literal),
+        Value::Object(values) => values.values().try_for_each(validate_literal),
+        _ => Ok(()),
+    }
+}
+
+fn has_unknown_keys(value: &Value, allowed: &[&str]) -> bool {
+    value
+        .as_object()
+        .is_some_and(|object| object.keys().any(|key| !allowed.contains(&key.as_str())))
+}
+
+fn application_has_unknown_nested_fields(value: &Value) -> bool {
+    let Some(application) = value.as_object() else {
+        return false;
+    };
+    if application
+        .get("function")
+        .is_some_and(|value| has_unknown_keys(value, &["name", "version"]))
+    {
+        return true;
+    }
+    if application
+        .get("inputs")
+        .and_then(Value::as_array)
+        .is_some_and(|inputs| {
+            inputs
+                .iter()
+                .any(|input| has_unknown_keys(input, &["parameter", "kind", "value"]))
+        })
+    {
+        return true;
+    }
+    application.get("output").is_some_and(|output| {
+        has_unknown_keys(output, &["kind", "arrow_type", "nullable", "fields"])
+            || output
+                .get("fields")
+                .and_then(Value::as_array)
+                .is_some_and(|fields| {
+                    fields
+                        .iter()
+                        .any(|field| has_unknown_keys(field, &["name", "arrow_type", "nullable"]))
+                })
+    })
+}
+
+macro_rules! impl_json {
+    ($type:ty) => {
+        impl $type {
+            /// Decode a remote value. Unknown fields and discriminator values
+            /// are accepted so newer servers remain readable.
+            pub fn from_json(json: &str) -> Result<Self> {
+                from_json(json)
+            }
+
+            /// Encode the known client contract with bytewise-sorted JSON keys.
+            pub fn to_canonical_json(&self) -> Result<String> {
+                canonical_json(self)
+            }
+        }
+    };
+}
+
+/// Packaged Python artifact identity. Source bytes are never part of this value.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionArtifact {
+    pub kind: String,
+    pub digest: String,
+    pub entrypoint: String,
+}
+
+/// One ordered Arrow input parameter.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionParameter {
+    pub name: String,
+    pub arrow_type: String,
+    pub nullable: bool,
+}
+
+/// One field of an ordered named-struct result.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionResultField {
+    pub name: String,
+    pub arrow_type: String,
+    pub nullable: bool,
+}
+
+/// Scalar or named-struct Function output.
+///
+/// `kind` remains a string so unknown future result shapes can be decoded.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionOutput {
+    pub kind: String,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub arrow_type: Option<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub nullable: Option<bool>,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub fields: Vec<FunctionResultField>,
+}
+
+/// Ordered language-neutral Function signature.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionSignature {
+    pub inputs: Vec<FunctionParameter>,
+    pub output: FunctionOutput,
+}
+
+/// One Python environment source.
+///
+/// The selected source is interpreted by Sophon. `kind` is open for forward
+/// compatible decoding.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct PythonEnvironmentSpec {
+    pub kind: String,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub packages: Vec<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub path: Option<String>,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub modules: Vec<String>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub image: Option<String>,
+}
+
+/// Reproducible Python runtime definition understood by Sophon.
+///
+/// `env` contains non-secret values. Secret values have no client model;
+/// [`FunctionVersion::required_secrets`] contains names only.
+#[derive(Debug, Clone, PartialEq, Eq)]
+#[non_exhaustive]
+pub enum PythonRuntimeSpec {
+    /// The V1 Sophon-managed Python runtime.
+    Python {
+        python_version: String,
+        environment: PythonEnvironmentSpec,
+        env: BTreeMap<String, String>,
+    },
+    /// A runtime kind introduced by a newer server.
+    ///
+    /// Unknown payload fields are intentionally not retained because the
+    /// client does not proxy catalog values.
+    Unrecognized { kind: String },
+}
+
+impl PythonRuntimeSpec {
+    /// The wire discriminator reported by Sophon.
+    pub fn kind(&self) -> &str {
+        match self {
+            Self::Python { .. } => "python",
+            Self::Unrecognized { kind } => kind,
+        }
+    }
+
+    /// The Python version for the V1 runtime, or `None` for an unknown kind.
+    pub fn python_version(&self) -> Option<&str> {
+        match self {
+            Self::Python { python_version, .. } => Some(python_version),
+            Self::Unrecognized { .. } => None,
+        }
+    }
+
+    /// The Python environment for the V1 runtime, or `None` for an unknown kind.
+    pub fn environment(&self) -> Option<&PythonEnvironmentSpec> {
+        match self {
+            Self::Python { environment, .. } => Some(environment),
+            Self::Unrecognized { .. } => None,
+        }
+    }
+
+    /// Non-secret environment variables, or `None` for an unknown kind.
+    pub fn env(&self) -> Option<&BTreeMap<String, String>> {
+        match self {
+            Self::Python { env, .. } => Some(env),
+            Self::Unrecognized { .. } => None,
+        }
+    }
+}
+
+#[derive(Deserialize)]
+struct PythonRuntimeWire {
+    kind: String,
+    #[serde(default)]
+    python_version: Option<String>,
+    #[serde(default)]
+    environment: Option<PythonEnvironmentSpec>,
+    #[serde(default)]
+    env: BTreeMap<String, String>,
+}
+
+impl<'de> Deserialize<'de> for PythonRuntimeSpec {
+    fn deserialize<D: Deserializer<'de>>(deserializer: D) -> std::result::Result<Self, D::Error> {
+        let wire = PythonRuntimeWire::deserialize(deserializer)?;
+        if wire.kind == "python" {
+            Ok(Self::Python {
+                python_version: wire
+                    .python_version
+                    .ok_or_else(|| de::Error::missing_field("python_version"))?,
+                environment: wire
+                    .environment
+                    .ok_or_else(|| de::Error::missing_field("environment"))?,
+                env: wire.env,
+            })
+        } else {
+            Ok(Self::Unrecognized { kind: wire.kind })
+        }
+    }
+}
+
+impl Serialize for PythonRuntimeSpec {
+    fn serialize<S: Serializer>(&self, serializer: S) -> std::result::Result<S::Ok, S::Error> {
+        #[derive(Serialize)]
+        struct PythonRuntimeRef<'a> {
+            kind: &'static str,
+            python_version: &'a str,
+            environment: &'a PythonEnvironmentSpec,
+            #[serde(skip_serializing_if = "BTreeMap::is_empty")]
+            env: &'a BTreeMap<String, String>,
+        }
+
+        #[derive(Serialize)]
+        struct UnrecognizedRuntimeRef<'a> {
+            kind: &'a str,
+        }
+
+        match self {
+            Self::Python {
+                python_version,
+                environment,
+                env,
+            } => PythonRuntimeRef {
+                kind: "python",
+                python_version,
+                environment,
+                env,
+            }
+            .serialize(serializer),
+            Self::Unrecognized { kind } => UnrecognizedRuntimeRef { kind }.serialize(serializer),
+        }
+    }
+}
+
+/// Immutable Function version returned by the Enterprise catalog.
+///
+/// Scheduling resources, priority, concurrency, and retry policy belong to
+/// the submitting Job and are not part of this identity.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionVersion {
+    name: String,
+    version: String,
+    artifact: FunctionArtifact,
+    signature: FunctionSignature,
+    runtime: PythonRuntimeSpec,
+    runtime_digest: String,
+    environment_digest: String,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    required_secrets: Vec<String>,
+    created_at: String,
+}
+
+impl FunctionVersion {
+    pub fn name(&self) -> &str {
+        &self.name
+    }
+
+    pub fn version(&self) -> &str {
+        &self.version
+    }
+
+    pub fn artifact(&self) -> &FunctionArtifact {
+        &self.artifact
+    }
+
+    pub fn signature(&self) -> &FunctionSignature {
+        &self.signature
+    }
+
+    pub fn runtime(&self) -> &PythonRuntimeSpec {
+        &self.runtime
+    }
+
+    pub fn runtime_digest(&self) -> &str {
+        &self.runtime_digest
+    }
+
+    pub fn environment_digest(&self) -> &str {
+        &self.environment_digest
+    }
+
+    /// Required secret names. Resolved values exist only inside Sophon.
+    pub fn required_secrets(&self) -> &[String] {
+        &self.required_secrets
+    }
+
+    pub fn created_at(&self) -> &str {
+        &self.created_at
+    }
+}
+
+impl_json!(FunctionVersion);
+
+/// Encoded artifact bytes uploaded with a Function registration request.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionArtifactContent {
+    /// Encoding of `data`. V1 Python authoring uses `base64`.
+    pub encoding: String,
+    pub data: String,
+}
+
+/// Internal execution adapter selected for a Python callable artifact.
+///
+/// The adapter converts the public scalar callable to the Arrow batch ABI
+/// used by the remote executor. It is part of the request envelope, not a
+/// public batch-UDF authoring mode.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct PythonAdapterSpec {
+    pub kind: String,
+    pub version: u32,
+}
+
+/// Python artifact uploaded while registering a Function.
+///
+/// Unlike [`FunctionArtifact`], which is the durable artifact identity
+/// returned by the catalog, this request value contains the encoded source
+/// bytes that Sophon must durably bake before publishing a FunctionVersion.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionArtifactRequest {
+    pub kind: String,
+    pub digest: String,
+    pub entrypoint: String,
+    pub content: FunctionArtifactContent,
+    pub adapter: PythonAdapterSpec,
+}
+
+/// Stable request envelope for remote immutable Function registration.
+///
+/// Secret values deliberately have no field in this model. The only secret
+/// material the client may send is the ordered set of names Sophon resolves
+/// inside the remote runtime.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionRegistrationRequest {
+    pub name: String,
+    pub artifact: FunctionArtifactRequest,
+    pub signature: FunctionSignature,
+    pub runtime: PythonRuntimeSpec,
+    #[serde(default, skip_serializing_if = "Vec::is_empty")]
+    pub required_secrets: Vec<String>,
+}
+
+impl_json!(FunctionRegistrationRequest);
+
+/// Exact FunctionVersion reference embedded in applications and bindings.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionVersionRef {
+    pub name: String,
+    pub version: String,
+}
+
+/// Parameter binding in a FunctionApplication.
+///
+/// `kind` remains open until Python authoring is added in Slice 2. Slice 1
+/// freezes JSON integers, strings, booleans, nulls, arrays, and objects as
+/// canonical literal values. Floating-point literals are rejected until a
+/// language-neutral numeric representation is defined.
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct ApplicationInput {
+    pub parameter: String,
+    pub kind: String,
+    pub value: Value,
+}
+
+/// Pre-declaration application of an exact FunctionVersion.
+#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct FunctionApplication {
+    function: FunctionVersionRef,
+    inputs: Vec<ApplicationInput>,
+    output: FunctionOutput,
+    group_id: String,
+    #[serde(default, skip_serializing_if = "BTreeMap::is_empty")]
+    columns: BTreeMap<String, String>,
+    #[serde(default, flatten, skip_serializing)]
+    unknown_fields: BTreeMap<String, Value>,
+    #[serde(default, skip)]
+    unknown_nested_fields: bool,
+}
+
+impl FunctionApplication {
+    pub fn function(&self) -> &FunctionVersionRef {
+        &self.function
+    }
+
+    pub fn inputs(&self) -> &[ApplicationInput] {
+        &self.inputs
+    }
+
+    pub fn output(&self) -> &FunctionOutput {
+        &self.output
+    }
+
+    pub fn group_id(&self) -> &str {
+        &self.group_id
+    }
+
+    pub fn columns(&self) -> &BTreeMap<String, String> {
+        &self.columns
+    }
+    /// Whether a newer writer attached application fields this client cannot
+    /// validate. Such applications remain readable but must not be declared.
+    pub fn has_unknown_fields(&self) -> bool {
+        !self.unknown_fields.is_empty() || self.unknown_nested_fields
+    }
+
+    /// Decode a remote application after validating the Slice 1 literal domain.
+    pub fn from_json(json: &str) -> Result<Self> {
+        let value: Value = from_json(json)?;
+        let has_unknown_nested_fields = application_has_unknown_nested_fields(&value);
+        let mut application: Self = serde_json::from_value(value).map_err(invalid_json)?;
+        application.unknown_nested_fields = has_unknown_nested_fields;
+        application
+            .inputs
+            .iter()
+            .try_for_each(|input| validate_literal(&input.value))?;
+        Ok(application)
+    }
+
+    /// Encode the application with bytewise-sorted JSON keys.
+    pub fn to_canonical_json(&self) -> Result<String> {
+        self.inputs
+            .iter()
+            .try_for_each(|input| validate_literal(&input.value))?;
+        canonical_json(self)
+    }
+}
+
+/// Stable table input bound to a registered parameter.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct InputBinding {
+    pub parameter: String,
+    pub field_id: i32,
+    pub field_path: String,
+    pub arrow_type: String,
+    pub nullable: bool,
+}
+
+/// Ordered result-field to table-field mapping for a grouped binding.
+///
+/// Assignment state is not part of the Slice 1 client contract. During the
+/// NULL transition there is no public Lance cell-flag identifier to persist.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct OutputMapping {
+    pub result_field: String,
+    pub output_name: String,
+    pub output_field_id: i32,
+    pub output_ordinal: u32,
+    pub arrow_type: String,
+    pub nullable: bool,
+}
+
+/// Immutable grouped binding persisted by the Enterprise table service.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct FunctionBinding {
+    binding_id: String,
+    revision: u64,
+    function: FunctionVersionRef,
+    group_id: String,
+    inputs: Vec<InputBinding>,
+    outputs: Vec<OutputMapping>,
+    /// Exact Arrow schema presented to the Function, encoded with the Lance
+    /// Namespace Arrow JSON representation.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    input_schema: Option<Value>,
+    /// Exact physical Arrow schema of the grouped table outputs.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    output_schema: Option<Value>,
+}
+
+impl FunctionBinding {
+    pub fn binding_id(&self) -> &str {
+        &self.binding_id
+    }
+
+    pub fn revision(&self) -> u64 {
+        self.revision
+    }
+
+    pub fn function(&self) -> &FunctionVersionRef {
+        &self.function
+    }
+
+    pub fn group_id(&self) -> &str {
+        &self.group_id
+    }
+
+    pub fn inputs(&self) -> &[InputBinding] {
+        &self.inputs
+    }
+
+    pub fn outputs(&self) -> &[OutputMapping] {
+        &self.outputs
+    }
+
+    pub fn input_schema(&self) -> Option<&Value> {
+        self.input_schema.as_ref()
+    }
+
+    pub fn output_schema(&self) -> Option<&Value> {
+        self.output_schema.as_ref()
+    }
+}
+
+impl_json!(FunctionBinding);
+
+/// Stable terminal result of a remote Function-column refresh Job.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
+pub struct RefreshColumnResult {
+    pub rows_assigned: u64,
+    pub rows_failed: u64,
+    pub rows_remaining: u64,
+    pub source_version: u64,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub published_version: Option<u64>,
+}
+
+impl RefreshColumnResult {
+    /// Deprecated compatibility alias for `rows_assigned`.
+    pub fn rows_filled(&self) -> u64 {
+        self.rows_assigned
+    }
+
+    /// Deprecated compatibility alias for `published_version`.
+    pub fn version(&self) -> Option<u64> {
+        self.published_version
+    }
+}
+
+impl_json!(RefreshColumnResult);

rust/lancedb/src/job.rs

@@ -6,6 +6,8 @@
 use std::sync::Arc;
 
 use async_trait::async_trait;
+use serde::de::DeserializeOwned;
+use serde_json::Value;
 use tokio::sync::watch;
 use tokio::task::{AbortHandle, JoinHandle};
 
@@ -19,43 +21,135 @@ pub(crate) trait JobHandle: Send + Sync {
         None
     }
     async fn status(&self) -> Result<String>;
-    async fn wait(&self) -> Result<()>;
+    async fn wait(&self) -> Result<TerminalResult>;
     async fn cancel(&self) -> Result<()>;
 }
 
+/// A backend-neutral successful terminal result.
+///
+/// Local operations do not carry a value. Remote operations may carry JSON
+/// that the public [`Job`] decodes according to its result type.
+pub(crate) struct TerminalResult {
+    #[allow(dead_code)] // Typed remote submit endpoints consume this after Slice 1.
+    value: Option<Value>,
+    #[allow(dead_code)] // Preserved so typed decode errors retain request correlation.
+    request_id: Option<String>,
+}
+
+impl TerminalResult {
+    pub(crate) fn local() -> Self {
+        Self {
+            value: None,
+            request_id: None,
+        }
+    }
+
+    pub(crate) fn remote(value: Option<Value>, request_id: String) -> Self {
+        Self {
+            value,
+            request_id: Some(request_id),
+        }
+    }
+
+    #[allow(dead_code)] // Exercised by the remote typed-result fixtures in Slice 1.
+    fn decode<T: DeserializeOwned>(self) -> Result<T> {
+        let request_id = self.request_id.unwrap_or_default();
+        let value = self.value.ok_or_else(|| Error::Http {
+            source: "successful typed job response did not contain a result".into(),
+            request_id: request_id.clone(),
+            status_code: None,
+        })?;
+        serde_json::from_value(value).map_err(|error| Error::Http {
+            source: format!("failed to parse typed job result: {error}").into(),
+            request_id,
+            status_code: None,
+        })
+    }
+}
+
+type ResultDecoder<T> = fn(TerminalResult) -> Result<T>;
+
+enum JobInner<T> {
+    Handle {
+        handle: Box<dyn JobHandle>,
+        decode: ResultDecoder<T>,
+    },
+    Completed(T),
+}
+
 /// A handle to an operation that may still be running.
 ///
 /// The operation may already be complete when the handle is created.
-pub struct Job {
-    handle: Option<Box<dyn JobHandle>>,
+pub struct Job<T = ()>
+where
+    T: Clone + Send + Sync + 'static,
+{
+    inner: JobInner<T>,
 }
 
-impl std::fmt::Debug for Job {
+impl<T> std::fmt::Debug for Job<T>
+where
+    T: Clone + Send + Sync + 'static,
+{
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         f.debug_struct("Job")
             .field("id", &self.id())
-            .field("done", &self.handle.is_none())
+            .field("done", &matches!(self.inner, JobInner::Completed(_)))
             .finish()
     }
 }
 
-impl Job {
+impl Job<()> {
     /// A job whose operation finished before the handle was created.
     pub(crate) fn new_done() -> Self {
-        Self { handle: None }
+        Self {
+            inner: JobInner::Completed(()),
+        }
     }
 
     pub(crate) fn new(handle: Box<dyn JobHandle>) -> Self {
         Self {
-            handle: Some(handle),
+            inner: JobInner::Handle {
+                handle,
+                decode: |_| Ok(()),
+            },
         }
     }
 
-    /// A job running as a task in this process.
+    /// A unit-result job running as a task in this process.
     pub(crate) fn spawned(task: JoinHandle<Result<()>>) -> Self {
         Self::new(Box::new(SpawnedJob::new(task)))
     }
 
+    /// Wait for an untyped job while preserving a remote terminal payload.
+    #[doc(hidden)]
+    pub async fn wait_raw_json(&self) -> Result<Option<Value>> {
+        match &self.inner {
+            JobInner::Handle { handle, .. } => Ok(handle.wait().await?.value),
+            JobInner::Completed(()) => Ok(None),
+        }
+    }
+}
+
+impl<T> Job<T>
+where
+    T: Clone + DeserializeOwned + Send + Sync + 'static,
+{
+    /// Construct a typed remote Job for a result-specific submit API.
+    pub(crate) fn new_typed(handle: Box<dyn JobHandle>) -> Self {
+        Self {
+            inner: JobInner::Handle {
+                handle,
+                decode: TerminalResult::decode::<T>,
+            },
+        }
+    }
+}
+
+impl<T> Job<T>
+where
+    T: Clone + Send + Sync + 'static,
+{
     /// Identifies the operation on the server that is running it.
     ///
     /// Returned for correlating with server logs or the jobs API. Operations
@@ -63,7 +157,10 @@ impl Job {
     /// value is opaque: parsing it or storing it to resume the job later is
     /// not supported.
     pub fn id(&self) -> Option<&str> {
-        self.handle.as_ref().and_then(|handle| handle.id())
+        match &self.inner {
+            JobInner::Handle { handle, .. } => handle.id(),
+            JobInner::Completed(_) => None,
+        }
     }
 
     /// The operation's current lifecycle state: "running", "finished",
@@ -73,9 +170,9 @@ impl Job {
     /// terminal failure state, or retry. States a newer server reports that
     /// this client version does not know pass through as-is.
     pub async fn status(&self) -> Result<String> {
-        match &self.handle {
-            None => Ok("finished".to_string()),
-            Some(handle) => handle.status().await,
+        match &self.inner {
+            JobInner::Handle { handle, .. } => handle.status().await,
+            JobInner::Completed(_) => Ok("finished".to_string()),
         }
     }
 
@@ -83,10 +180,10 @@ impl Job {
     ///
     /// Returns [`crate::Error::JobFailed`] if the operation failed and
     /// [`crate::Error::JobCancelled`] if it was cancelled.
-    pub async fn wait(&self) -> Result<()> {
-        match &self.handle {
-            None => Ok(()),
-            Some(handle) => handle.wait().await,
+    pub async fn wait(&self) -> Result<T> {
+        match &self.inner {
+            JobInner::Handle { handle, decode } => decode(handle.wait().await?),
+            JobInner::Completed(result) => Ok(result.clone()),
         }
     }
 
@@ -94,9 +191,9 @@ impl Job {
     ///
     /// Cancelling an operation that already finished is a no-op.
     pub async fn cancel(&self) -> Result<()> {
-        match &self.handle {
-            None => Ok(()),
-            Some(handle) => handle.cancel().await,
+        match &self.inner {
+            JobInner::Handle { handle, .. } => handle.cancel().await,
+            JobInner::Completed(_) => Ok(()),
         }
     }
 }
@@ -162,7 +259,7 @@ impl JobHandle for SpawnedJob {
         Ok(label.to_string())
     }
 
-    async fn wait(&self) -> Result<()> {
+    async fn wait(&self) -> Result<TerminalResult> {
         let mut outcome = self.outcome.clone();
         let settled = outcome
             .wait_for(|outcome| outcome.is_some())
@@ -172,7 +269,8 @@ impl JobHandle for SpawnedJob {
             })?
             .clone()
             .expect("wait_for returns once an outcome is set");
-        settled.into_result()
+        settled.into_result()?;
+        Ok(TerminalResult::local())
     }
 
     async fn cancel(&self) -> Result<()> {

rust/lancedb/src/lib.rs

@@ -181,6 +181,7 @@ pub mod dataloader;
 pub mod embeddings;
 pub mod error;
 pub mod expr;
+pub mod function;
 pub mod index;
 pub mod io;
 pub mod ipc;
@@ -205,6 +206,7 @@ use serde::{Deserialize, Serialize};
 pub use blob::{BlobRangeRequest, blob, is_blob};
 pub use connection::{ConnectNamespaceBuilder, Connection};
 pub use error::{Error, JobFailure, Result};
+pub use function::FunctionVersion;
 pub use job::Job;
 use lance_index::vector::ApproxMode as LanceApproxMode;
 use lance_linalg::distance::DistanceType as LanceDistanceType;

rust/lancedb/src/remote/db.rs

@@ -24,8 +24,9 @@ use crate::database::{
     JobDescription, JobInfo, OpenTableRequest, ReadConsistency, TableNamesRequest,
 };
 use crate::error::Result;
+use crate::function::{FunctionRegistrationRequest, FunctionVersion};
 use crate::job::Job;
-use crate::remote::job::RemoteJob;
+use crate::remote::job::{DescribeJobResponse, RemoteJob, job_state_to_client};
 use crate::remote::util::stream_as_body;
 use crate::table::BaseTable;
 
@@ -472,48 +473,6 @@ struct RemoteListJobsResponse {
     page_token: Option<String>,
 }
 
-/// The server's account of why a job failed. Absent from older servers,
-/// which report only the terminal state.
-#[derive(serde::Deserialize)]
-struct RemoteReportedFailure {
-    #[serde(default)]
-    phase: Option<String>,
-    #[serde(default)]
-    message: Option<String>,
-    #[serde(default)]
-    retryable: Option<bool>,
-}
-
-#[derive(serde::Deserialize)]
-struct RemoteDescribeJobResponse {
-    job_id: String,
-    #[serde(default)]
-    job_type: String,
-    job_state: String,
-    #[serde(default)]
-    creation_ms: i64,
-    #[serde(default)]
-    spec: serde_json::Value,
-    #[serde(default)]
-    failure: Option<RemoteReportedFailure>,
-}
-
-/// Server job states -> the client vocabulary ("running" / "finished" /
-/// "failed" / "cancelled"). Covers both the describe enum (IN_PROGRESS /
-/// DONE / FAILED / CANCELLED) and the registry's lowercase list-row states
-/// (in_progress / succeeded / failed / canceled / timed_out). States this
-/// client version does not know (e.g. created, queued) pass through as-is.
-fn job_state_to_client(state: &str) -> String {
-    match state {
-        "IN_PROGRESS" | "in_progress" => "running",
-        "DONE" | "done" | "succeeded" => "finished",
-        "FAILED" | "failed" | "TIMED_OUT" | "timed_out" => "failed",
-        "CANCELLED" | "cancelled" | "canceled" => "cancelled",
-        other => other,
-    }
-    .to_string()
-}
-
 /// Bound on `list_jobs` page walking; a warning is logged when the listing
 /// is truncated at this many pages.
 const MAX_LIST_JOBS_PAGES: usize = 100;
@@ -531,6 +490,39 @@ impl<S: HttpSend> Database for RemoteDatabase<S> {
         })
     }
 
+    async fn create_function_async(
+        &self,
+        request: FunctionRegistrationRequest,
+    ) -> Result<Job<FunctionVersion>> {
+        let req = self.client.post("/v1/functions/create").json(&request);
+        let (request_id, response) = self.client.send(req).await?;
+        let response = self.client.check_response(&request_id, response).await?;
+        let status = response.status();
+        let body = response.text().await.err_to_http(request_id.clone())?;
+        let job_id = extract_job_id(&body).ok_or_else(|| Error::Http {
+            source: "Function registration response did not contain a valid job_id".into(),
+            request_id,
+            status_code: Some(status),
+        })?;
+        Ok(Job::new_typed(Box::new(RemoteJob::new(
+            self.client.clone(),
+            job_id,
+        ))))
+    }
+
+    async fn get_function(&self, name: &str, version: &str) -> Result<FunctionVersion> {
+        let req = self
+            .client
+            .post("/v1/functions/get")
+            .json(&serde_json::json!({
+                "name": name,
+                "version": version,
+            }));
+        let (request_id, response) = self.client.send(req).await?;
+        let response = self.client.check_response(&request_id, response).await?;
+        response.json().await.err_to_http(request_id)
+    }
+
     fn job(&self, job_id: &str) -> Result<crate::job::Job> {
         Ok(crate::job::Job::new(Box::new(super::job::RemoteJob::new(
             self.client.clone(),
@@ -586,19 +578,14 @@ impl<S: HttpSend> Database for RemoteDatabase<S> {
             }) => return Ok(None),
             Err(err) => return Err(err),
         };
-        let body: RemoteDescribeJobResponse = rsp.json().await.err_to_http(request_id)?;
+        let body: DescribeJobResponse = rsp.json().await.err_to_http(request_id)?;
         Ok(Some(JobDescription {
             job_id: body.job_id,
             job_type: body.job_type,
             state: job_state_to_client(&body.job_state),
             creation_ms: body.creation_ms,
             spec: body.spec,
-            failure: body.failure.map(|reported| crate::error::JobFailure {
-                phase: reported.phase,
-                message: reported.message,
-                retryable: reported.retryable,
-                source: None,
-            }),
+            failure: body.failure.map(|reported| reported.into_job_failure()),
         }))
     }
 
@@ -2493,6 +2480,60 @@ mod tests {
         assert_eq!(batches[0].num_rows(), 2);
     }
 
+    #[tokio::test]
+    async fn test_create_function_async_sends_canonical_request_and_decodes_typed_job() {
+        const REQUEST: &str = include_str!(
+            "../../tests/fixtures/first_class_functions/v1/remote_function_registration_request.json"
+        );
+        const FUNCTION_JOB: &str =
+            include_str!("../../tests/fixtures/first_class_functions/v1/remote_function_job.json");
+        let expected: serde_json::Value = serde_json::from_str(REQUEST).unwrap();
+        let conn = Connection::new_with_handler(move |request| match request.url().path() {
+            "/v1/functions/create" => {
+                assert_eq!(request.method(), &reqwest::Method::POST);
+                let body: serde_json::Value =
+                    serde_json::from_slice(request.body().unwrap().as_bytes().unwrap()).unwrap();
+                assert_eq!(body, expected);
+                http::Response::builder()
+                    .status(202)
+                    .body(r#"{"job_id":"job-function-1"}"#)
+                    .unwrap()
+            }
+            "/v1/jobs/describe" => http::Response::builder()
+                .status(200)
+                .body(FUNCTION_JOB)
+                .unwrap(),
+            path => panic!("unexpected path: {path}"),
+        });
+        let request = crate::function::FunctionRegistrationRequest::from_json(REQUEST).unwrap();
+        let job = conn.create_function_async(request).await.unwrap();
+        assert_eq!(job.id(), Some("job-function-1"));
+        let version = job.wait().await.unwrap();
+        assert_eq!(version.name(), "embed");
+        assert_eq!(version.version(), "fv_01K3EXACT");
+    }
+
+    #[tokio::test]
+    async fn test_get_function_requires_and_sends_exact_version() {
+        const VERSION: &str = include_str!(
+            "../../tests/fixtures/first_class_functions/v1/remote_function_version.canonical.json"
+        );
+        let conn = Connection::new_with_handler(|request| {
+            assert_eq!(request.method(), &reqwest::Method::POST);
+            assert_eq!(request.url().path(), "/v1/functions/get");
+            let body: serde_json::Value =
+                serde_json::from_slice(request.body().unwrap().as_bytes().unwrap()).unwrap();
+            assert_eq!(
+                body,
+                serde_json::json!({"name": "embed", "version": "fv_01K3EXACT"})
+            );
+            http::Response::builder().status(200).body(VERSION).unwrap()
+        });
+        let version = conn.get_function("embed", "fv_01K3EXACT").await.unwrap();
+        assert_eq!(version.name(), "embed");
+        assert_eq!(version.version(), "fv_01K3EXACT");
+    }
+
     #[tokio::test]
     async fn test_conn_job_waits_to_done() {
         let polls = Arc::new(AtomicUsize::new(0));
@@ -2507,7 +2548,7 @@ mod tests {
             http::Response::builder()
                 .status(200)
                 .body(format!(
-                    r#"{{"job_id": "job-1", "job_type": "create_index", "job_state": "{}", "creation_ms": 1}}"#,
+                    r#"{{"job_id": "job-1", "job_type": "create_function", "job_state": "{}", "creation_ms": 1, "result": {{"name": "embed", "version": "fv_1"}}}}"#,
                     state
                 ))
                 .unwrap()

rust/lancedb/src/remote/job.rs

@@ -8,10 +8,10 @@ use std::time::Duration;
 use async_trait::async_trait;
 use tokio::time::sleep;
 
-use serde::{Deserialize, Deserializer};
+use serde::Deserialize;
 
 use crate::error::{Error, JobFailure, Result};
-use crate::job::JobHandle;
+use crate::job::{JobHandle, TerminalResult};
 use crate::remote::client::{HttpSend, RequestResultExt, RestfulLanceDbClient};
 
 /// Delay before the second job-state poll; doubles up to [`MAX_POLL_INTERVAL`].
@@ -29,12 +29,6 @@ enum JobState {
     Other(String),
 }
 
-impl<'de> Deserialize<'de> for JobState {
-    fn deserialize<D: Deserializer<'de>>(deserializer: D) -> std::result::Result<Self, D::Error> {
-        Ok(Self::from(String::deserialize(deserializer)?.as_str()))
-    }
-}
-
 impl JobState {
     /// The client vocabulary label for this state.
     fn client_label(&self) -> String {
@@ -51,22 +45,26 @@ impl JobState {
 impl From<&str> for JobState {
     fn from(state: &str) -> Self {
         match state {
-            "IN_PROGRESS" => Self::InProgress,
-            "CANCELLED" => Self::Cancelled,
+            "IN_PROGRESS" | "in_progress" => Self::InProgress,
+            "CANCELLED" | "cancelled" | "canceled" => Self::Cancelled,
             // The server reports a timed-out job as FAILED on describe;
             // accept the raw registry state too in case a future server
             // stops folding it.
-            "FAILED" | "TIMED_OUT" => Self::Failed,
-            "DONE" => Self::Done,
+            "FAILED" | "failed" | "TIMED_OUT" | "timed_out" => Self::Failed,
+            "DONE" | "done" | "succeeded" => Self::Done,
             other => Self::Other(other.to_string()),
         }
     }
 }
 
+pub(super) fn job_state_to_client(state: &str) -> String {
+    JobState::from(state).client_label()
+}
+
 /// The server's account of why a job failed. Absent from older servers, which
 /// report only the terminal state.
 #[derive(Deserialize)]
-struct ReportedFailure {
+pub(super) struct ReportedFailure {
     #[serde(default)]
     phase: Option<String>,
     #[serde(default)]
@@ -75,11 +73,43 @@ struct ReportedFailure {
     retryable: Option<bool>,
 }
 
+/// Forward-compatible `/v1/jobs/describe` wire envelope.
 #[derive(Deserialize)]
-struct DescribeJobResponse {
-    job_state: JobState,
+pub(super) struct DescribeJobResponse {
     #[serde(default)]
-    failure: Option<ReportedFailure>,
+    pub(super) job_id: String,
+    #[serde(default)]
+    pub(super) job_type: String,
+    pub(super) job_state: String,
+    #[serde(default)]
+    pub(super) creation_ms: i64,
+    #[serde(default)]
+    pub(super) spec: serde_json::Value,
+    #[serde(default)]
+    result: Option<serde_json::Value>,
+    #[serde(default)]
+    pub(super) failure: Option<ReportedFailure>,
+}
+
+impl ReportedFailure {
+    pub(super) fn into_job_failure(self) -> JobFailure {
+        JobFailure {
+            phase: self.phase,
+            message: self.message,
+            retryable: self.retryable,
+            source: None,
+        }
+    }
+}
+
+impl DescribeJobResponse {
+    fn state(&self) -> JobState {
+        JobState::from(self.job_state.as_str())
+    }
+
+    fn into_terminal_result(self, request_id: String) -> TerminalResult {
+        TerminalResult::remote(self.result, request_id)
+    }
 }
 
 pub struct RemoteJob<S: HttpSend> {
@@ -93,7 +123,7 @@ impl<S: HttpSend> RemoteJob<S> {
     }
 
     /// One `/v1/jobs/describe` round trip.
-    async fn describe(&self) -> Result<DescribeJobResponse> {
+    async fn describe(&self) -> Result<(String, DescribeJobResponse)> {
         let request = self
             .client
             .post("/v1/jobs/describe")
@@ -104,10 +134,10 @@ impl<S: HttpSend> RemoteJob<S> {
         let description: DescribeJobResponse =
             serde_json::from_str(&body).map_err(|e| Error::Http {
                 source: format!("failed to parse job description: {}", e).into(),
-                request_id,
+                request_id: request_id.clone(),
                 status_code: None,
             })?;
-        Ok(description)
+        Ok((request_id, description))
     }
 }
 
@@ -118,26 +148,21 @@ impl<S: HttpSend> JobHandle for RemoteJob<S> {
     }
 
     async fn status(&self) -> Result<String> {
-        Ok(self.describe().await?.job_state.client_label())
+        Ok(self.describe().await?.1.state().client_label())
     }
 
-    async fn wait(&self) -> Result<()> {
+    async fn wait(&self) -> Result<TerminalResult> {
         let mut interval = INITIAL_POLL_INTERVAL;
         loop {
-            let description = self.describe().await?;
-            match description.job_state {
-                JobState::Done => return Ok(()),
+            let (request_id, description) = self.describe().await?;
+            match description.state() {
+                JobState::Done => return Ok(description.into_terminal_result(request_id)),
                 JobState::Failed => {
                     return Err(Error::JobFailed {
                         job_id: Some(self.job_id.clone()),
                         failure: description
                             .failure
-                            .map(|reported| JobFailure {
-                                phase: reported.phase,
-                                message: reported.message,
-                                retryable: reported.retryable,
-                                source: None,
-                            })
+                            .map(ReportedFailure::into_job_failure)
                             .unwrap_or_default(),
                     });
                 }
@@ -168,3 +193,78 @@ impl<S: HttpSend> JobHandle for RemoteJob<S> {
             .map(|_| ())
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use async_trait::async_trait;
+
+    use crate::Result;
+    use crate::function::{FunctionVersion, RefreshColumnResult};
+    use crate::job::{Job, JobHandle, TerminalResult};
+
+    use super::DescribeJobResponse;
+
+    const FUNCTION_JOB: &str =
+        include_str!("../../tests/fixtures/first_class_functions/v1/remote_function_job.json");
+    const REFRESH_JOB: &str =
+        include_str!("../../tests/fixtures/first_class_functions/v1/remote_refresh_job.json");
+    const UNIT_JOB: &str =
+        include_str!("../../tests/fixtures/first_class_functions/v1/remote_unit_job.json");
+    const MISSING_RESULT_JOB: &str = r#"{"job_state":"DONE"}"#;
+
+    struct FixtureRemoteJob(&'static str);
+
+    #[async_trait]
+    impl JobHandle for FixtureRemoteJob {
+        async fn status(&self) -> Result<String> {
+            Ok("finished".to_string())
+        }
+
+        async fn wait(&self) -> Result<TerminalResult> {
+            let description: DescribeJobResponse =
+                serde_json::from_str(self.0).expect("remote job fixture");
+            Ok(description.into_terminal_result("fixture-request".to_string()))
+        }
+
+        async fn cancel(&self) -> Result<()> {
+            Ok(())
+        }
+    }
+
+    #[tokio::test]
+    async fn typed_remote_job_fixtures_decode_terminal_results() {
+        let function = Job::<FunctionVersion>::new_typed(Box::new(FixtureRemoteJob(FUNCTION_JOB)));
+        let result = function.wait().await.expect("typed FunctionVersion result");
+        assert_eq!(result.version(), "fv_01K3EXACT");
+
+        let refresh =
+            Job::<RefreshColumnResult>::new_typed(Box::new(FixtureRemoteJob(REFRESH_JOB)));
+        let result = refresh.wait().await.expect("typed RefreshColumnResult");
+        assert_eq!(result.rows_assigned, 999_998_800);
+        assert_eq!(result.rows_filled(), result.rows_assigned);
+
+        let unit = Job::new(Box::new(FixtureRemoteJob(UNIT_JOB)));
+        unit.wait()
+            .await
+            .expect("unit result ignores additive remote payloads");
+    }
+
+    #[tokio::test]
+    async fn typed_remote_job_requires_a_terminal_result() {
+        let typed =
+            Job::<RefreshColumnResult>::new_typed(Box::new(FixtureRemoteJob(MISSING_RESULT_JOB)));
+        let error = typed.wait().await.unwrap_err();
+        assert!(
+            error
+                .to_string()
+                .contains("successful typed job response did not contain a result")
+        );
+    }
+
+    #[test]
+    fn remote_wire_unknown_fields_are_forward_decodable() {
+        let response: DescribeJobResponse =
+            serde_json::from_str(FUNCTION_JOB).expect("function job fixture");
+        assert_eq!(response.job_state, "DONE");
+    }
+}

rust/lancedb/src/remote/table.rs

@@ -162,13 +162,13 @@ impl<S: HttpSend> crate::job::JobHandle for FreshnessJob<S> {
         crate::job::JobHandle::status(&self.inner).await
     }
 
-    async fn wait(&self) -> Result<()> {
-        crate::job::JobHandle::wait(&self.inner).await?;
+    async fn wait(&self) -> Result<crate::job::TerminalResult> {
+        let result = crate::job::JobHandle::wait(&self.inner).await?;
         let version = self.version.read().await;
         if version.is_none() {
             self.freshness.lock().unwrap().checkout_baseline = Some(SystemTime::now());
         }
-        Ok(())
+        Ok(result)
     }
 
     async fn cancel(&self) -> Result<()> {
@@ -2147,6 +2147,7 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
         self.check_mutable().await?;
 
         let table_schema = self.schema().await?;
+        crate::table::computed_columns::ensure_supported_function_metadata(table_schema.as_ref())?;
         let table_def = TableDefinition::try_from_rich_schema(table_schema.clone())?;
 
         let num_partitions = if self.server_version.support_multipart_write() {
@@ -2698,6 +2699,10 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
         _read_columns: Option<Vec<String>>,
     ) -> Result<AddColumnsResult> {
         self.check_mutable().await?;
+        crate::table::computed_columns::ensure_no_function_bindings_for_mutation(
+            self.schema().await?.as_ref(),
+            "schema evolution",
+        )?;
         match transforms {
             NewColumnTransform::SqlExpressions(expressions) => {
                 let body = expressions
@@ -2746,6 +2751,10 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
 
     async fn add_computed_columns(&self, columns: &[(String, String)]) -> Result<AddColumnsResult> {
         self.check_mutable().await?;
+        crate::table::computed_columns::ensure_no_function_bindings_for_mutation(
+            self.schema().await?.as_ref(),
+            "schema evolution",
+        )?;
         // The server plans the declaration: expression validation, type
         // inference and the persisted binding all happen there.
         let entries = columns
@@ -2785,6 +2794,63 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
         Ok(result)
     }
 
+    async fn add_function_columns(
+        &self,
+        application: &crate::function::FunctionApplication,
+        output_name: Option<&str>,
+    ) -> Result<AddColumnsResult> {
+        self.check_mutable().await?;
+        let schema = self.schema().await?;
+        let plan = crate::table::computed_columns::plan_function_application(
+            schema.as_ref(),
+            application,
+            output_name,
+        )?;
+        let new_columns = plan
+            .outputs
+            .iter()
+            .map(|output| {
+                serde_json::json!({
+                    "name": output.output_name,
+                    "all_null": true,
+                })
+            })
+            .collect::<Vec<_>>();
+        let mut body = serde_json::json!({
+            "new_columns": new_columns,
+            "function": {
+                "application": plan.application,
+                "binding_metadata_version": plan.binding_metadata_version,
+                "input_bindings": plan.input_bindings,
+                "input_schema": plan.input_schema,
+                "output_schema": plan.output_schema,
+                "outputs": plan.outputs,
+            },
+        });
+        self.apply_branch_body(&mut body);
+        let request = self
+            .client
+            .post(&format!("/v1/table/{}/add_columns/", 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())?;
+
+        if body.trim().is_empty() {
+            return Ok(AddColumnsResult { version: 0 });
+        }
+
+        let result: AddColumnsResult = serde_json::from_str(&body).map_err(|e| Error::Http {
+            source: format!("Failed to parse add Function columns response: {e}").into(),
+            request_id,
+            status_code: None,
+        })?;
+
+        self.invalidate_schema_cache();
+        self.track_write_version(result.version);
+        Ok(result)
+    }
+
     async fn refresh_column(&self, _column: &str) -> Result<RefreshColumnResult> {
         // The server runs a refresh as a job and does not report a fill
         // count, so the blocking form has no honest result to return.
@@ -2800,8 +2866,7 @@ impl<S: HttpSend> BaseTable for RemoteTable<S> {
         let mut body = serde_json::json!({ "column": column });
         self.apply_branch_body(&mut body);
         let request = self
-            .client
-            .post(&format!("/v1/table/{}/backfill_column", self.identifier))
+            .post_read(&format!("/v1/table/{}/backfill_column", self.identifier))
             .json(&body);
         let (request_id, response) = self.send(request, true).await?;
         let response = self.check_table_response(&request_id, response).await?;
@@ -3810,11 +3875,14 @@ mod tests {
                 assert_eq!(rename, "y");
 
                 if old_server {
-                    http::Response::builder().status(200).body("{}").unwrap()
+                    http::Response::builder()
+                        .status(200)
+                        .body("{}".to_string())
+                        .unwrap()
                 } else {
                     http::Response::builder()
                         .status(200)
-                        .body(r#"{"version": 43}"#)
+                        .body(r#"{"version": 43}"#.to_string())
                         .unwrap()
                 }
             } else {
@@ -3945,11 +4013,14 @@ mod tests {
                 assert_eq!(predicate, "id in (1, 2, 3)");
 
                 if old_server {
-                    http::Response::builder().status(200).body("{}").unwrap()
+                    http::Response::builder()
+                        .status(200)
+                        .body("{}".to_string())
+                        .unwrap()
                 } else {
                     http::Response::builder()
                         .status(200)
-                        .body(r#"{"version": 43}"#)
+                        .body(r#"{"version": 43}"#.to_string())
                         .unwrap()
                 }
             } else {
@@ -6516,7 +6587,9 @@ mod tests {
     #[tokio::test]
     async fn test_add_columns(#[case] old_server: bool) {
         let table = Table::new_with_handler("my_table", move |request| {
-            if request.url().path() == "/v1/table/my_table/add_columns/" {
+            if request.url().path() == "/v1/table/my_table/describe/" {
+                simple_describe_response()
+            } else if request.url().path() == "/v1/table/my_table/add_columns/" {
                 assert_eq!(request.method(), "POST");
                 assert_eq!(
                     request.headers().get("Content-Type").unwrap(),
@@ -6540,11 +6613,14 @@ mod tests {
                 assert_eq!(expression, "cast(NULL as int32)");
 
                 if old_server {
-                    http::Response::builder().status(200).body("{}").unwrap()
+                    http::Response::builder()
+                        .status(200)
+                        .body("{}".to_string())
+                        .unwrap()
                 } else {
                     http::Response::builder()
                         .status(200)
-                        .body(r#"{"version": 43}"#)
+                        .body(r#"{"version": 43}"#.to_string())
                         .unwrap()
                 }
             } else {
@@ -6569,19 +6645,22 @@ mod tests {
     /// plan; the client never types the expression itself.
     #[tokio::test]
     async fn test_add_computed_columns_sends_the_expression() {
-        let table = Table::new_with_handler("my_table", |request| {
-            assert_eq!(request.method(), "POST");
-            assert_eq!(request.url().path(), "/v1/table/my_table/add_columns/");
-            let body = request.body().unwrap().as_bytes().unwrap();
-            let value: serde_json::Value = serde_json::from_slice(body).unwrap();
-            assert_eq!(
-                value["new_columns"],
-                serde_json::json!([{"name": "doubled", "computed": "x * 2"}])
-            );
-            http::Response::builder()
-                .status(200)
-                .body(r#"{"version": 7}"#)
-                .unwrap()
+        let table = Table::new_with_handler("my_table", |request| match request.url().path() {
+            "/v1/table/my_table/describe/" => simple_describe_response(),
+            "/v1/table/my_table/add_columns/" => {
+                assert_eq!(request.method(), "POST");
+                let body = request.body().unwrap().as_bytes().unwrap();
+                let value: serde_json::Value = serde_json::from_slice(body).unwrap();
+                assert_eq!(
+                    value["new_columns"],
+                    serde_json::json!([{"name": "doubled", "computed": "x * 2"}])
+                );
+                http::Response::builder()
+                    .status(200)
+                    .body(r#"{"version": 7}"#.to_string())
+                    .unwrap()
+            }
+            path => panic!("Unexpected path: {path}"),
         });
 
         let result = table
@@ -6593,6 +6672,129 @@ mod tests {
         assert_eq!(result.version, 7);
     }
 
+    #[tokio::test]
+    async fn test_add_scalar_function_column_sends_atomic_null_declaration() {
+        let table = Table::new_with_handler("my_table", |request| {
+            match request.url().path() {
+            "/v1/table/my_table/describe/" => http::Response::builder()
+                .status(200)
+                .body(
+                    r#"{"version":1,"schema":{"fields":[{"name":"description","nullable":true,"type":{"type":"string"}}]}}"#,
+                )
+                .unwrap(),
+            "/v1/table/my_table/add_columns/" => {
+                let actual: serde_json::Value = serde_json::from_slice(
+                    request.body().unwrap().as_bytes().unwrap(),
+                )
+                .unwrap();
+                let expected: serde_json::Value = serde_json::from_str(include_str!(
+                    "../../tests/fixtures/first_class_functions/v1/remote_scalar_declaration_request.json"
+                ))
+                .unwrap();
+                assert_eq!(actual, expected);
+                http::Response::builder()
+                    .status(200)
+                    .body(r#"{"version":8}"#)
+                    .unwrap()
+            }
+            path => panic!("Unexpected path: {path}"),
+        }
+        });
+        let application = crate::function::FunctionApplication::from_json(
+            r#"{
+                "function":{"name":"embed","version":"fv_01K3EXACT"},
+                "inputs":[{"parameter":"text","kind":"column","value":{"path":"description"}}],
+                "output":{"kind":"scalar","arrow_type":"list<float32>","nullable":false},
+                "group_id":"fg_scalar"
+            }"#,
+        )
+        .unwrap();
+
+        let result = table
+            .add_columns()
+            .function_as("embedding", application)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(result.version, 8);
+    }
+
+    #[tokio::test]
+    async fn test_add_named_struct_function_expands_one_atomic_sibling_group() {
+        let table = Table::new_with_handler("my_table", |request| match request.url().path() {
+            "/v1/table/my_table/describe/" => http::Response::builder()
+                .status(200)
+                .body(
+                    r#"{"version":1,"schema":{"fields":[
+                        {"name":"title","nullable":true,"type":{"type":"string"}},
+                        {"name":"body","nullable":true,"type":{"type":"string"}}
+                    ]}}"#,
+                )
+                .unwrap(),
+            "/v1/table/my_table/add_columns/" => {
+                let actual: serde_json::Value =
+                    serde_json::from_slice(request.body().unwrap().as_bytes().unwrap()).unwrap();
+                let expected: serde_json::Value = serde_json::from_str(include_str!(
+                    "../../tests/fixtures/first_class_functions/v1/remote_grouped_declaration_request.json"
+                ))
+                .unwrap();
+                assert_eq!(actual, expected);
+                http::Response::builder()
+                    .status(200)
+                    .body(r#"{"version":9}"#)
+                    .unwrap()
+            }
+            path => panic!("Unexpected path: {path}"),
+        });
+        let application = crate::function::FunctionApplication::from_json(
+            r#"{
+                "function":{"name":"text_features","version":"fv_01K3TEXT"},
+                "inputs":[
+                    {"parameter":"title","kind":"column","value":{"path":"title"}},
+                    {"parameter":"body","kind":"column","value":{"path":"body"}}
+                ],
+                "output":{"kind":"named_struct","fields":[
+                    {"name":"normalized_text","arrow_type":"utf8","nullable":false},
+                    {"name":"token_count","arrow_type":"int64","nullable":false}
+                ]},
+                "group_id":"fg_01K3TEXT",
+                "columns":{"normalized_text":"search_text"}
+            }"#,
+        )
+        .unwrap();
+
+        let result = table
+            .add_columns()
+            .function(application)
+            .execute()
+            .await
+            .unwrap();
+        assert_eq!(result.version, 9);
+    }
+
+    #[tokio::test]
+    async fn test_add_columns_fails_closed_on_newer_function_binding_metadata() {
+        let table = Table::new_with_handler("my_table", |request| {
+            match request.url().path() {
+            "/v1/table/my_table/describe/" => http::Response::builder()
+                .status(200)
+                .body(
+                    r#"{"version":1,"schema":{"fields":[{"name":"x","nullable":true,"type":{"type":"int32"}}],"metadata":{"lancedb::function_bindings":"{\"version\":2,\"bindings\":[]}"}}}"#,
+                )
+                .unwrap(),
+            path => panic!("mutation request must not be sent: {path}"),
+        }
+        });
+
+        let err = table
+            .add_columns()
+            .computed("doubled", "x * 2")
+            .execute()
+            .await
+            .unwrap_err();
+        assert!(matches!(err, Error::NotSupported { .. }));
+    }
+
     /// A remote refresh is a server job: the async form returns its handle,
     /// and the blocking form refuses rather than invent a fill count.
     #[tokio::test]
@@ -10911,6 +11113,7 @@ mod tests {
                     .status(200)
                     .body("{}".to_string())
                     .unwrap(),
+                "/v1/table/my_table/describe/" => simple_describe_response(),
                 "/v1/table/my_table/add_columns/"
                 | "/v1/table/my_table/alter_columns/"
                 | "/v1/table/my_table/drop_columns/" => {

rust/lancedb/src/table.rs

@@ -760,6 +760,16 @@ pub trait BaseTable: std::fmt::Display + std::fmt::Debug + Send + Sync {
             message: "computed columns are not supported on this table type".into(),
         })
     }
+    /// Declare one immutable registered-Function output group.
+    async fn add_function_columns(
+        &self,
+        _application: &crate::function::FunctionApplication,
+        _output_name: Option<&str>,
+    ) -> Result<AddColumnsResult> {
+        Err(Error::NotSupported {
+            message: "Function columns are supported only on LanceDB Cloud and Enterprise".into(),
+        })
+    }
     /// Fill a computed column's unfilled rows.
     ///
     /// The default returns `NotSupported`; Lance-backed tables override it.
@@ -3158,6 +3168,7 @@ impl BaseTable for NativeTable {
         let ds = self.dataset.get().await?;
 
         let table_schema = Schema::from(&ds.schema().clone());
+        computed_columns::ensure_supported_function_metadata(&table_schema)?;
         computed_columns::ensure_not_written(
             &table_schema,
             add.data.schema().fields().iter().map(|f| f.name().as_str()),

rust/lancedb/src/table/add_columns.rs

@@ -9,6 +9,7 @@ use lance::dataset::NewColumnTransform;
 
 use super::BaseTable;
 use super::schema_evolution::AddColumnsResult;
+use crate::function::FunctionApplication;
 use crate::{Error, Result};
 
 /// Adds columns to a table. See [`Table::add_columns`](super::Table::add_columns).
@@ -16,6 +17,7 @@ pub struct AddColumnsBuilder {
     parent: Arc<dyn BaseTable>,
     transform: Option<NewColumnTransform>,
     computed: Vec<(String, String)>,
+    function: Option<(FunctionApplication, Option<String>)>,
     read_columns: Option<Vec<String>>,
 }
 
@@ -25,6 +27,7 @@ impl std::fmt::Debug for AddColumnsBuilder {
             .field("parent", &self.parent)
             .field("has_transform", &self.transform.is_some())
             .field("computed", &self.computed)
+            .field("has_function", &self.function.is_some())
             .field("read_columns", &self.read_columns)
             .finish()
     }
@@ -36,6 +39,7 @@ impl AddColumnsBuilder {
             parent,
             transform: None,
             computed: Vec::new(),
+            function: None,
             read_columns: None,
         }
     }
@@ -83,6 +87,48 @@ impl AddColumnsBuilder {
         self
     }
 
+    /// Declare every field of a named-struct Function result as one atomic
+    /// sibling group. Result-field aliases come from
+    /// [`FunctionApplication::columns`](crate::function::FunctionApplication::columns).
+    ///
+    /// ```
+    /// # use lancedb::Table;
+    /// # use lancedb::function::FunctionApplication;
+    /// # async fn declare(table: &Table, application: FunctionApplication) -> lancedb::Result<()> {
+    /// table.add_columns().function(application).execute().await?;
+    /// # Ok(())
+    /// # }
+    /// ```
+    pub fn function(mut self, application: FunctionApplication) -> Self {
+        self.function = Some((application, None));
+        self
+    }
+
+    /// Declare a scalar or entire named-struct Function result as one table
+    /// column. The physical column starts all-null and is materialized by the
+    /// remote Function refresh path.
+    ///
+    /// ```
+    /// # use lancedb::Table;
+    /// # use lancedb::function::FunctionApplication;
+    /// # async fn declare(table: &Table, application: FunctionApplication) -> lancedb::Result<()> {
+    /// table
+    ///     .add_columns()
+    ///     .function_as("embedding", application)
+    ///     .execute()
+    ///     .await?;
+    /// # Ok(())
+    /// # }
+    /// ```
+    pub fn function_as(
+        mut self,
+        name: impl Into<String>,
+        application: FunctionApplication,
+    ) -> Self {
+        self.function = Some((application, Some(name.into())));
+        self
+    }
+
     /// Limit which existing columns a [`NewColumnTransform::BatchUDF`] mapper
     /// receives. Every other transform, and a computed column, determines what
     /// it reads, so setting this alongside one is an error rather than a silent
@@ -98,21 +144,23 @@ impl AddColumnsBuilder {
             parent,
             transform,
             computed,
+            function,
             read_columns,
         } = self;
 
-        match (transform, computed.is_empty()) {
-            (None, true) => Err(Error::InvalidInput {
+        let declaration_count = usize::from(!computed.is_empty()) + usize::from(function.is_some());
+        if transform.is_some() && declaration_count != 0 || declaration_count > 1 {
+            return Err(Error::InvalidInput {
+                message: "add_columns cannot mix transforms, SQL computed columns, and a Function application; they cannot be added atomically in one call"
+                    .into(),
+            });
+        }
+
+        match (transform, computed.is_empty(), function) {
+            (None, true, None) => Err(Error::InvalidInput {
                 message: "add_columns requires a transform or a computed column".into(),
             }),
-            // The two commit through different transforms, so one call covering
-            // both would be two commits and could half-apply.
-            (Some(_), false) => Err(Error::InvalidInput {
-                message: "add_columns cannot mix a transform with computed columns; \
-                          they cannot be added atomically in one call"
-                    .into(),
-            }),
-            (Some(transform), true) => {
+            (Some(transform), true, None) => {
                 if read_columns.is_some() && !matches!(transform, NewColumnTransform::BatchUDF(_)) {
                     return Err(Error::InvalidInput {
                         message: "read_columns applies only to a BatchUDF transform; \
@@ -122,7 +170,7 @@ impl AddColumnsBuilder {
                 }
                 parent.add_columns(transform, read_columns).await
             }
-            (None, false) => {
+            (None, false, None) => {
                 if read_columns.is_some() {
                     return Err(Error::InvalidInput {
                         message: "read_columns applies only to a BatchUDF transform; \
@@ -132,6 +180,18 @@ impl AddColumnsBuilder {
                 }
                 parent.add_computed_columns(&computed).await
             }
+            (None, true, Some((application, output_name))) => {
+                if read_columns.is_some() {
+                    return Err(Error::InvalidInput {
+                        message: "read_columns does not apply to a Function application; its inputs are already bound"
+                            .into(),
+                    });
+                }
+                parent
+                    .add_function_columns(&application, output_name.as_deref())
+                    .await
+            }
+            _ => unreachable!("mixed add_columns modes were rejected above"),
         }
     }
 }

rust/lancedb/src/table/merge.rs

@@ -233,6 +233,10 @@ pub(crate) async fn execute_merge_insert(
     params: MergeInsertBuilder,
     new_data: Box<dyn RecordBatchReader + Send>,
 ) -> Result<MergeResult> {
+    super::computed_columns::ensure_no_function_bindings_for_mutation(
+        table.schema().await?.as_ref(),
+        "merge_insert",
+    )?;
     match lsm::lsm_dispatch_decision(table, &params).await? {
         lsm::LsmDispatch::Lsm(plan) => {
             let future =

rust/lancedb/src/table/refresh.rs

@@ -169,6 +169,9 @@ fn declared_expression(dataset: &Dataset, column: &str) -> Result<String> {
         })?;
     match declaration.kind {
         ComputedColumnKind::Sql { expression } => Ok(expression),
+        ComputedColumnKind::Function { .. } => Err(Error::NotSupported {
+            message: "registered Function columns are refreshed only by a remote server Job".into(),
+        }),
         ComputedColumnKind::Unrecognized { kind } => Err(Error::NotSupported {
             message: format!(
                 "computed column '{column}' is defined by '{kind}', which this version of \

rust/lancedb/src/table/schema_evolution.rs

@@ -101,6 +101,10 @@ pub(crate) async fn execute_add_columns(
     transforms: NewColumnTransform,
     read_columns: Option<Vec<String>>,
 ) -> Result<AddColumnsResult> {
+    computed_columns::ensure_no_function_bindings_for_mutation(
+        table.schema().await?.as_ref(),
+        "schema evolution",
+    )?;
     // Declarations are admitted only through [`execute_declare`].
     match &transforms {
         NewColumnTransform::AllNulls(schema) => {
@@ -124,6 +128,10 @@ pub(crate) async fn execute_declare(
     // checked against latest committed state, not this handle's snapshot.
     // The catch-up flag outlives unset and marks retained SSTable rows.
     table.checkout_latest().await?;
+    computed_columns::ensure_no_function_bindings_for_mutation(
+        table.schema().await?.as_ref(),
+        "schema evolution",
+    )?;
     let catchup = table.dataset.get().await?.manifest().reader_feature_flags
         & lance_table::feature_flags::FLAG_MEM_WAL_INDEX_CATCHUP
         != 0;
@@ -163,6 +171,10 @@ pub(crate) async fn execute_alter_columns(
     // Nullability is not part of what an expression resolves against, so only
     // a rename or a retype can invalidate a binding.
     let schema = std::sync::Arc::new(ArrowSchema::from(dataset.schema()));
+    computed_columns::ensure_no_function_bindings_for_mutation(
+        schema.as_ref(),
+        "schema evolution",
+    )?;
     let rebinding = alterations
         .iter()
         .filter(|alteration| alteration.rename.is_some() || alteration.data_type.is_some())
@@ -190,6 +202,10 @@ pub(crate) async fn execute_drop_columns(
 ) -> Result<DropColumnsResult> {
     table.dataset.ensure_mutable()?;
     let mut dataset = (*table.dataset.get().await?).clone();
+    computed_columns::ensure_no_function_bindings_for_mutation(
+        &ArrowSchema::from(dataset.schema()),
+        "schema evolution",
+    )?;
     computed_columns::ensure_not_an_input(
         &std::sync::Arc::new(ArrowSchema::from(dataset.schema())),
         columns,
@@ -215,6 +231,7 @@ pub(crate) async fn execute_update_field_metadata(
     // binding out from under a refresh. A replace on a declared column would
     // silently erase it.
     let schema = ArrowSchema::from(dataset.schema());
+    computed_columns::ensure_no_function_bindings_for_mutation(&schema, "schema evolution")?;
     let declared: Vec<String> = computed_columns::computed_columns(&schema)
         .into_iter()
         .map(|declaration| declaration.name)

rust/lancedb/src/table/update.rs

@@ -82,6 +82,10 @@ pub(crate) async fn execute_update(
 
     // 1. Snapshot the current dataset
     let dataset = table.dataset.get().await?;
+    super::computed_columns::ensure_no_function_bindings_for_mutation(
+        &arrow_schema::Schema::from(dataset.schema()),
+        "update",
+    )?;
     super::computed_columns::ensure_not_written(
         &arrow_schema::Schema::from(dataset.schema()),
         update.columns.iter().map(|(name, _)| name.as_str()),

rust/lancedb/tests/first_class_function_slice1.rs

@@ -0,0 +1,183 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+use std::fs;
+use std::path::PathBuf;
+
+use lancedb::function::{
+    FunctionApplication, FunctionBinding, FunctionVersion, RefreshColumnResult,
+};
+use serde_json::Value;
+
+fn fixture(name: &str) -> String {
+    let path = PathBuf::from(env!("CARGO_MANIFEST_DIR"))
+        .join("tests/fixtures/first_class_functions/v1")
+        .join(name);
+    fs::read_to_string(path).expect("fixture must be readable")
+}
+
+fn job_result(name: &str) -> Value {
+    serde_json::from_str::<Value>(&fixture(name)).expect("remote Job fixture")["result"].clone()
+}
+
+fn assert_no_secret_values(value: &Value) {
+    match value {
+        Value::Object(values) => {
+            for (key, value) in values {
+                assert!(
+                    !matches!(
+                        key.as_str(),
+                        "secret_value" | "secret_values" | "resolved_secret" | "resolved_secrets"
+                    ),
+                    "client canonical value must not model resolved secret material"
+                );
+                assert_no_secret_values(value);
+            }
+        }
+        Value::Array(values) => values.iter().for_each(assert_no_secret_values),
+        _ => {}
+    }
+}
+
+#[test]
+fn function_version_job_result_matches_shared_canonical_golden() {
+    let result = job_result("remote_function_job.json");
+    let version = FunctionVersion::from_json(&result.to_string()).expect("FunctionVersion result");
+
+    assert_eq!(version.name(), "embed");
+    assert_eq!(version.version(), "fv_01K3EXACT");
+    assert_eq!(version.runtime_digest(), "sha256:runtime");
+    assert_eq!(version.required_secrets(), &["HF_TOKEN"]);
+    assert_eq!(
+        version.to_canonical_json().expect("canonical JSON"),
+        fixture("remote_function_version.canonical.json").trim()
+    );
+}
+
+#[test]
+fn version_identity_is_immutable_and_exact() {
+    let original = job_result("remote_function_job.json");
+    let version =
+        FunctionVersion::from_json(&original.to_string()).expect("FunctionVersion result");
+    let reopened = version.clone();
+    assert_eq!(reopened, version);
+    assert_eq!(reopened.name(), version.name());
+    assert_eq!(reopened.version(), version.version());
+
+    let mut changed = original;
+    changed["version"] = Value::String("fv_01K3DIFFERENT".to_string());
+    let changed = FunctionVersion::from_json(&changed.to_string()).expect("changed version");
+    assert_ne!(changed, version);
+}
+
+#[test]
+fn application_and_binding_match_shared_remote_goldens() {
+    let application = FunctionApplication::from_json(&fixture("remote_function_application.json"))
+        .expect("application fixture");
+    assert_eq!(application.function().version, "fv_01K3TEXT");
+    assert_eq!(application.output().kind, "named_struct");
+    assert_eq!(application.inputs().len(), 2);
+    assert_eq!(
+        application.to_canonical_json().expect("canonical JSON"),
+        fixture("remote_function_application.canonical.json").trim()
+    );
+
+    let binding = FunctionBinding::from_json(&fixture("remote_function_binding.json"))
+        .expect("binding fixture");
+    assert_eq!(binding.revision(), 3);
+    assert_eq!(binding.function().version, "fv_01K3TEXT");
+    assert_eq!(binding.outputs()[0].output_ordinal, 0);
+    assert_eq!(binding.outputs()[1].output_ordinal, 1);
+    assert!(binding.input_schema().is_some());
+    assert!(binding.output_schema().is_some());
+    assert_eq!(
+        binding.to_canonical_json().expect("canonical JSON"),
+        fixture("remote_function_binding.canonical.json").trim()
+    );
+}
+
+#[test]
+fn refresh_job_result_matches_shared_canonical_golden() {
+    let result = job_result("remote_refresh_job.json");
+    let result = RefreshColumnResult::from_json(&result.to_string()).expect("refresh result");
+    assert_eq!(result.rows_assigned, 999_998_800);
+    assert_eq!(result.rows_filled(), result.rows_assigned);
+    assert_eq!(result.version(), result.published_version);
+    assert_eq!(
+        result.to_canonical_json().expect("canonical JSON"),
+        fixture("remote_refresh_result.canonical.json").trim()
+    );
+
+    let result = RefreshColumnResult::from_json(&fixture(
+        "remote_refresh_result_without_published_version.json",
+    ))
+    .expect("optional version");
+    assert_eq!(result.published_version, None);
+    assert_eq!(
+        result
+            .to_canonical_json()
+            .expect("canonical result without version"),
+        fixture("remote_refresh_result_without_published_version.canonical.json").trim()
+    );
+    assert_eq!(
+        RefreshColumnResult::from_json(
+            &result
+                .to_canonical_json()
+                .expect("canonical result without version")
+        )
+        .expect("round-trip result without version"),
+        result
+    );
+}
+
+#[test]
+fn unknown_fields_and_discriminators_are_forward_decodable() {
+    let mut result = job_result("remote_function_job.json");
+    result["future_version_metadata"] = serde_json::json!({"retention_class": "catalog"});
+    result["runtime"] = serde_json::json!({
+        "kind": "wasm",
+        "module_digest": "sha256:wasm"
+    });
+    result["signature"]["output"]["kind"] = Value::String("future_output_shape".to_string());
+
+    let version = FunctionVersion::from_json(&result.to_string()).expect("future remote value");
+    assert_eq!(version.runtime().kind(), "wasm");
+    assert_eq!(version.runtime().python_version(), None);
+    assert_eq!(version.signature().output.kind, "future_output_shape");
+    assert_eq!(
+        serde_json::from_str::<Value>(
+            &version.to_canonical_json().expect("canonical future value")
+        )
+        .expect("canonical JSON")["runtime"],
+        serde_json::json!({"kind": "wasm"})
+    );
+}
+
+#[test]
+fn floating_point_application_literals_are_rejected_consistently() {
+    let error = FunctionApplication::from_json(&fixture("remote_function_application_float.json"))
+        .unwrap_err();
+    assert!(
+        error
+            .to_string()
+            .contains("floating-point Function literals")
+    );
+}
+
+#[test]
+fn canonical_client_values_contain_secret_names_only() {
+    let result = job_result("remote_function_job.json");
+    let version = FunctionVersion::from_json(&result.to_string()).expect("FunctionVersion result");
+    let canonical: Value = serde_json::from_str(
+        &version
+            .to_canonical_json()
+            .expect("canonical FunctionVersion"),
+    )
+    .expect("canonical JSON");
+
+    assert_eq!(
+        canonical["required_secrets"],
+        serde_json::json!(["HF_TOKEN"])
+    );
+    assert_no_secret_values(&canonical);
+}

rust/lancedb/tests/first_class_function_slice2.rs

@@ -0,0 +1,81 @@
+// SPDX-License-Identifier: Apache-2.0
+// SPDX-FileCopyrightText: Copyright The LanceDB Authors
+
+use std::fs;
+use std::path::PathBuf;
+
+use lancedb::Error;
+use lancedb::function::FunctionRegistrationRequest;
+use serde_json::Value;
+
+fn fixture(name: &str) -> String {
+    let path = PathBuf::from(env!("CARGO_MANIFEST_DIR"))
+        .join("tests/fixtures/first_class_functions/v1")
+        .join(name);
+    fs::read_to_string(path).expect("fixture must be readable")
+}
+
+fn assert_no_secret_values(value: &Value) {
+    match value {
+        Value::Object(values) => {
+            for (key, value) in values {
+                assert!(
+                    !matches!(
+                        key.as_str(),
+                        "secret_value" | "secret_values" | "resolved_secret" | "resolved_secrets"
+                    ),
+                    "registration requests must not model resolved secret material"
+                );
+                assert_no_secret_values(value);
+            }
+        }
+        Value::Array(values) => values.iter().for_each(assert_no_secret_values),
+        _ => {}
+    }
+}
+
+#[test]
+fn registration_request_matches_shared_canonical_golden() {
+    let request = FunctionRegistrationRequest::from_json(&fixture(
+        "remote_function_registration_request.json",
+    ))
+    .expect("registration request");
+    assert_eq!(request.name, "normalize_score");
+    assert_eq!(request.artifact.adapter.kind, "scalar_to_arrow_batch");
+    assert_eq!(request.required_secrets, ["API_TOKEN"]);
+    assert_eq!(
+        request.to_canonical_json().expect("canonical request"),
+        fixture("remote_function_registration_request.canonical.json").trim()
+    );
+
+    let value: Value =
+        serde_json::from_str(&request.to_canonical_json().expect("canonical request"))
+            .expect("request JSON");
+    assert_no_secret_values(&value);
+}
+
+#[tokio::test]
+async fn local_function_catalog_operations_return_stable_not_supported() {
+    let directory = tempfile::tempdir().unwrap();
+    let connection = lancedb::connect(directory.path().to_str().unwrap())
+        .execute()
+        .await
+        .unwrap();
+    let request = FunctionRegistrationRequest::from_json(&fixture(
+        "remote_function_registration_request.json",
+    ))
+    .unwrap();
+
+    let create_error = connection.create_function_async(request).await.unwrap_err();
+    let lookup_error = connection
+        .get_function("normalize_score", "fv_exact")
+        .await
+        .unwrap_err();
+    for error in [create_error, lookup_error] {
+        assert!(matches!(
+            error,
+            Error::NotSupported { message }
+                if message == "Function catalog operations are not supported by this database"
+        ));
+    }
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_application.canonical.json

@@ -0,0 +1 @@
+{"columns":{"normalized_text":"search_text","token_count":"search_token_count"},"function":{"name":"text_features","version":"fv_01K3TEXT"},"group_id":"fg_01K3TEXT","inputs":[{"kind":"column","parameter":"title","value":{"path":"title"}},{"kind":"column","parameter":"body","value":{"path":"body"}}],"output":{"fields":[{"arrow_type":"utf8","name":"normalized_text","nullable":false},{"arrow_type":"int64","name":"token_count","nullable":false}],"kind":"named_struct"}}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_application.json

@@ -0,0 +1,20 @@
+{
+  "function": {"name": "text_features", "version": "fv_01K3TEXT"},
+  "inputs": [
+    {"parameter": "title", "kind": "column", "value": {"path": "title"}},
+    {"parameter": "body", "kind": "column", "value": {"path": "body"}}
+  ],
+  "output": {
+    "kind": "named_struct",
+    "fields": [
+      {"name": "normalized_text", "arrow_type": "utf8", "nullable": false},
+      {"name": "token_count", "arrow_type": "int64", "nullable": false}
+    ]
+  },
+  "group_id": "fg_01K3TEXT",
+  "columns": {
+    "normalized_text": "search_text",
+    "token_count": "search_token_count"
+  },
+  "future_application": {"declaration_mode": "managed"}
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_application_float.json

@@ -0,0 +1,8 @@
+{
+  "function": {"name": "score", "version": "fv_01K3FLOAT"},
+  "inputs": [
+    {"parameter": "threshold", "kind": "literal", "value": 1e-7}
+  ],
+  "output": {"kind": "scalar", "arrow_type": "bool", "nullable": false},
+  "group_id": "fg_01K3FLOAT"
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_binding.canonical.json

@@ -0,0 +1 @@
+{"binding_id":"fb_01K3TEXT","function":{"name":"text_features","version":"fv_01K3TEXT"},"group_id":"fg_01K3TEXT","input_schema":{"fields":[{"name":"title","nullable":true,"type":{"type":"utf8"}},{"name":"body","nullable":true,"type":{"type":"utf8"}}]},"inputs":[{"arrow_type":"utf8","field_id":11,"field_path":"title","nullable":true,"parameter":"title"},{"arrow_type":"utf8","field_id":12,"field_path":"body","nullable":true,"parameter":"body"}],"output_schema":{"fields":[{"name":"search_text","nullable":true,"type":{"type":"utf8"}},{"name":"search_token_count","nullable":true,"type":{"type":"int64"}}]},"outputs":[{"arrow_type":"utf8","nullable":false,"output_field_id":21,"output_name":"search_text","output_ordinal":0,"result_field":"normalized_text"},{"arrow_type":"int64","nullable":false,"output_field_id":22,"output_name":"search_token_count","output_ordinal":1,"result_field":"token_count"}],"revision":3}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_binding.json

@@ -0,0 +1,27 @@
+{
+  "binding_id": "fb_01K3TEXT",
+  "revision": 3,
+  "function": {"name": "text_features", "version": "fv_01K3TEXT"},
+  "group_id": "fg_01K3TEXT",
+  "inputs": [
+    {"parameter": "title", "field_id": 11, "field_path": "title", "arrow_type": "utf8", "nullable": true},
+    {"parameter": "body", "field_id": 12, "field_path": "body", "arrow_type": "utf8", "nullable": true}
+  ],
+  "outputs": [
+    {"result_field": "normalized_text", "output_name": "search_text", "output_field_id": 21, "output_ordinal": 0, "arrow_type": "utf8", "nullable": false},
+    {"result_field": "token_count", "output_name": "search_token_count", "output_field_id": 22, "output_ordinal": 1, "arrow_type": "int64", "nullable": false}
+  ],
+  "input_schema": {
+    "fields": [
+      {"name": "title", "nullable": true, "type": {"type": "utf8"}},
+      {"name": "body", "nullable": true, "type": {"type": "utf8"}}
+    ]
+  },
+  "output_schema": {
+    "fields": [
+      {"name": "search_text", "nullable": true, "type": {"type": "utf8"}},
+      {"name": "search_token_count", "nullable": true, "type": {"type": "int64"}}
+    ]
+  },
+  "future_binding": {"metadata_revision": 1}
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_job.json

@@ -0,0 +1,31 @@
+{
+  "job_id": "job_function_01K3",
+  "job_type": "create_function",
+  "job_state": "DONE",
+  "creation_ms": 1787270400000,
+  "spec": {"name": "embed"},
+  "result": {
+    "name": "embed",
+    "version": "fv_01K3EXACT",
+    "artifact": {
+      "kind": "python_callable",
+      "digest": "sha256:code",
+      "entrypoint": "embed"
+    },
+    "signature": {
+      "inputs": [{"name": "text", "arrow_type": "utf8", "nullable": true}],
+      "output": {"kind": "scalar", "arrow_type": "list<float32>", "nullable": false}
+    },
+    "runtime": {
+      "kind": "python",
+      "python_version": "3.12",
+      "environment": {"kind": "pip", "packages": ["sentence-transformers>=3"]},
+      "env": {"TOKENIZERS_PARALLELISM": "false"}
+    },
+    "runtime_digest": "sha256:runtime",
+    "environment_digest": "sha256:environment",
+    "required_secrets": ["HF_TOKEN"],
+    "created_at": "2026-08-21T00:00:00Z"
+  },
+  "future_job": {"trace_id": "trace-1"}
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_registration_request.canonical.json

@@ -0,0 +1 @@
+{"artifact":{"adapter":{"kind":"scalar_to_arrow_batch","version":1},"content":{"data":"ZnJvbSBfX2Z1dHVyZV9fIGltcG9ydCBhbm5vdGF0aW9ucwoKZGVmIG5vcm1hbGl6ZV9zY29yZSh2YWx1ZTogZmxvYXQpIC0+IGZsb2F0OgogICAgcmV0dXJuIHZhbHVlIC8gMTAwLjAK","encoding":"base64"},"digest":"sha256:760784bdcef57b802f389b97804cc0b618aae39e86044733451bcd0b13089a7f","entrypoint":"normalize_score","kind":"python_callable"},"name":"normalize_score","required_secrets":["API_TOKEN"],"runtime":{"env":{"MODE":"test"},"environment":{"kind":"pip","packages":["numpy>=2"]},"kind":"python","python_version":"3.12"},"signature":{"inputs":[{"arrow_type":"float64","name":"value","nullable":false}],"output":{"arrow_type":"float64","kind":"scalar","nullable":false}}}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_registration_request.json

@@ -0,0 +1,46 @@
+{
+  "name": "normalize_score",
+  "artifact": {
+    "kind": "python_callable",
+    "digest": "sha256:760784bdcef57b802f389b97804cc0b618aae39e86044733451bcd0b13089a7f",
+    "entrypoint": "normalize_score",
+    "content": {
+      "encoding": "base64",
+      "data": "ZnJvbSBfX2Z1dHVyZV9fIGltcG9ydCBhbm5vdGF0aW9ucwoKZGVmIG5vcm1hbGl6ZV9zY29yZSh2YWx1ZTogZmxvYXQpIC0+IGZsb2F0OgogICAgcmV0dXJuIHZhbHVlIC8gMTAwLjAK"
+    },
+    "adapter": {
+      "kind": "scalar_to_arrow_batch",
+      "version": 1
+    }
+  },
+  "signature": {
+    "inputs": [
+      {
+        "name": "value",
+        "arrow_type": "float64",
+        "nullable": false
+      }
+    ],
+    "output": {
+      "kind": "scalar",
+      "arrow_type": "float64",
+      "nullable": false
+    }
+  },
+  "runtime": {
+    "kind": "python",
+    "python_version": "3.12",
+    "environment": {
+      "kind": "pip",
+      "packages": [
+        "numpy>=2"
+      ]
+    },
+    "env": {
+      "MODE": "test"
+    }
+  },
+  "required_secrets": [
+    "API_TOKEN"
+  ]
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_function_version.canonical.json

@@ -0,0 +1 @@
+{"artifact":{"digest":"sha256:code","entrypoint":"embed","kind":"python_callable"},"created_at":"2026-08-21T00:00:00Z","environment_digest":"sha256:environment","name":"embed","required_secrets":["HF_TOKEN"],"runtime":{"env":{"TOKENIZERS_PARALLELISM":"false"},"environment":{"kind":"pip","packages":["sentence-transformers>=3"]},"kind":"python","python_version":"3.12"},"runtime_digest":"sha256:runtime","signature":{"inputs":[{"arrow_type":"utf8","name":"text","nullable":true}],"output":{"arrow_type":"list<float32>","kind":"scalar","nullable":false}},"version":"fv_01K3EXACT"}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_grouped_declaration_request.json

@@ -0,0 +1,45 @@
+{
+  "new_columns": [
+    {"name": "search_text", "all_null": true},
+    {"name": "token_count", "all_null": true}
+  ],
+  "function": {
+    "application": {
+      "function": {"name": "text_features", "version": "fv_01K3TEXT"},
+      "inputs": [
+        {"parameter": "title", "kind": "column", "value": {"path": "title"}},
+        {"parameter": "body", "kind": "column", "value": {"path": "body"}}
+      ],
+      "output": {
+        "kind": "named_struct",
+        "fields": [
+          {"name": "normalized_text", "arrow_type": "utf8", "nullable": false},
+          {"name": "token_count", "arrow_type": "int64", "nullable": false}
+        ]
+      },
+      "group_id": "fg_01K3TEXT",
+      "columns": {"normalized_text": "search_text"}
+    },
+    "binding_metadata_version": 1,
+    "input_bindings": [
+      {"parameter": "title", "field_path": "title", "arrow_type": "utf8", "nullable": true},
+      {"parameter": "body", "field_path": "body", "arrow_type": "utf8", "nullable": true}
+    ],
+    "input_schema": {
+      "fields": [
+        {"name": "title", "nullable": true, "type": {"type": "utf8"}},
+        {"name": "body", "nullable": true, "type": {"type": "utf8"}}
+      ]
+    },
+    "output_schema": {
+      "fields": [
+        {"name": "search_text", "nullable": true, "type": {"type": "utf8"}},
+        {"name": "token_count", "nullable": true, "type": {"type": "int64"}}
+      ]
+    },
+    "outputs": [
+      {"result_field": "normalized_text", "output_name": "search_text", "output_ordinal": 0},
+      {"result_field": "token_count", "output_name": "token_count", "output_ordinal": 1}
+    ]
+  }
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_refresh_job.json

@@ -0,0 +1,15 @@
+{
+  "job_id": "job_refresh_01K3",
+  "job_type": "refresh_function_columns",
+  "job_state": "DONE",
+  "creation_ms": 1787270400001,
+  "spec": {"table": "documents", "binding_revision": 3},
+  "result": {
+    "rows_assigned": 999998800,
+    "rows_failed": 0,
+    "rows_remaining": 0,
+    "source_version": 812,
+    "published_version": 919,
+    "future_result": {"committed_fragment_groups": 100}
+  }
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_refresh_result.canonical.json

@@ -0,0 +1 @@
+{"published_version":919,"rows_assigned":999998800,"rows_failed":0,"rows_remaining":0,"source_version":812}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_refresh_result_without_published_version.canonical.json

@@ -0,0 +1 @@
+{"rows_assigned":120,"rows_failed":0,"rows_remaining":0,"source_version":812}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_refresh_result_without_published_version.json

@@ -0,0 +1,6 @@
+{
+  "rows_assigned": 120,
+  "rows_failed": 0,
+  "rows_remaining": 0,
+  "source_version": 812
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_scalar_declaration_request.json

@@ -0,0 +1,41 @@
+{
+  "new_columns": [
+    {"name": "embedding", "all_null": true}
+  ],
+  "function": {
+    "application": {
+      "function": {"name": "embed", "version": "fv_01K3EXACT"},
+      "inputs": [
+        {"parameter": "text", "kind": "column", "value": {"path": "description"}}
+      ],
+      "output": {"kind": "scalar", "arrow_type": "list<float32>", "nullable": false},
+      "group_id": "fg_scalar"
+    },
+    "binding_metadata_version": 1,
+    "input_bindings": [
+      {"parameter": "text", "field_path": "description", "arrow_type": "utf8", "nullable": true}
+    ],
+    "input_schema": {
+      "fields": [
+        {"name": "text", "nullable": true, "type": {"type": "utf8"}}
+      ]
+    },
+    "output_schema": {
+      "fields": [
+        {
+          "name": "embedding",
+          "nullable": true,
+          "type": {
+            "type": "list",
+            "fields": [
+              {"name": "item", "nullable": false, "type": {"type": "float32"}}
+            ]
+          }
+        }
+      ]
+    },
+    "outputs": [
+      {"result_field": "$value", "output_name": "embedding", "output_ordinal": 0}
+    ]
+  }
+}

rust/lancedb/tests/fixtures/first_class_functions/v1/remote_unit_job.json

@@ -0,0 +1,9 @@
+{
+  "job_id": "job_index_01K3",
+  "job_type": "create_index",
+  "job_state": "DONE",
+  "creation_ms": 1787270400002,
+  "spec": {"column": "vector"},
+  "result": {"future_information": "ignored by Job<()>"},
+  "future_job": {"trace_id": "trace-2"}
+}