## Summary
- lazily initialize built-in OpenAI and Hugging Face providers when
consumers call the public embedding registry API
- choose automatic vector versus FTS search from embedding metadata on a
fresh pinned table revision for every execution
- expose automatic string searches as an `AutoQuery` with only
operations common to both native query families
- keep the registry shared and built-in registration safe across
duplicated module graphs
## Root cause
Nitro treats dependency modules as side-effect-free and removes the bare
OpenAI provider import from its generated route. Registration therefore
never runs, so `getRegistry().get("openai")` remains undefined even when
the registry itself is shared globally. Bundlers may also duplicate the
provider and registry module graphs.
The public embedding entry point now initializes built-in providers only
when `getRegistry()` is explicitly called, keeping initialization on a
live path that Nitro retains. Each terminal automatic-search execution
pins the exact table revision visible at dispatch, reads embedding
metadata and computes an embedding from that snapshot, replays the
builder operations, and constructs and executes the selected native
query against the same snapshot. Pinned native snapshots execute locally
when namespace pushdown cannot carry their revision, while remote
snapshots are seeded directly from one version-and-schema response. The
public `AutoQuery` builder exposes only the operations shared by FTS and
vector search, so runtime class narrowing cannot expose invalid
vector-only methods. Repeated built-in registration replaces stale
constructors from duplicated module graphs while public `register()`
retains its duplicate-alias error.
## Validation
- `cargo fmt --all`
- `cargo check --quiet --features remote --tests --examples`
- `cargo clippy --quiet --features remote --tests --examples`
- `pnpm build`
- `pnpm lint`
- `pnpm run docs`
- `pnpm test --runInBand` (783 passed, 5 skipped)
- serial examples suite with a local OpenAI mock (11 passed), including
`sentence-transformers.test.ts`
- packaged Nitro 2.13.4 server route using the reported imports returned
`{"registered":true}`
- fresh-process FTS fixture initialized both public built-ins and
confirmed automatic string search still returned the indexed row
- schema-consistency regressions cover read-consistency refresh,
checkout, checkoutLatest, restore, runtime class narrowing, concurrent
overwrite during embedding computation, and reused automatic-search
builders
- focused regressions confirm pinned native snapshots bypass unversioned
namespace pushdown and remote snapshots use one describe request
Fixes#2429
<!-- lance-gatekeeper-fix:v1 agent=2adf0f21b8bfb634606ed8897a849e30
generation=1 -->
---------
Co-authored-by: Gatefixer <313497061+lancedb-gatefixer[bot]@users.noreply.github.com>
Co-authored-by: Xuanwo <github@xuanwo.io>
Two bugs in Node's reading of the embedding_functions schema metadata.
First, parseFunctions keyed its result map by function name, so a table
whose metadata configures the same function for two vector columns came
back with only the last one. It now keys by the vector column, the
convention Python's parser already uses.
Second, Node could not read metadata written by the Python bindings at
all, which spell the keys snake_case: configs parsed with both columns
undefined, breaking embedding application on add() and leaving only
query-side embedding working. The parse now accepts both spellings.
Both fixes land in one shared parser used by every reader --
parseFunctions and the makeArrowTable schema validator, which had its
own private camelCase-only parse -- so the wire contract cannot fork
between entry points. A config naming no source or vector column is an
error at the boundary rather than a default downstream, as are two
configs claiming one column. The "vector" fallback remains only on the
optional field of user-supplied configs.
Breaking: parseFunctions is exported and its map keys change from
function name to vector column.
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
Fixes#2716
## Summary
Add support for querying with Float16Array, Float64Array, and Uint8Array
vectors in the Node.js SDK, eliminating precision loss from the previous
\Float32Array.from()\ conversion.
## Implementation
Follows @wjones127's [5-step
plan](https://github.com/lancedb/lancedb/issues/2716#issuecomment-3447750543):
### Rust (\
odejs/src/query.rs\)
1. \ytes_to_arrow_array(data: Uint8Array, dtype: String)\ helper that:
- Creates an Arrow \Buffer\ from the raw bytes
- Wraps it in a typed \ScalarBuffer<T>\ based on the dtype enum
- Constructs a \PrimitiveArray\ and returns \Arc<dyn Array>\
2. \
earest_to_raw(data, dtype)\ and \dd_query_vector_raw(data, dtype)\ NAPI
methods that pass the type-erased array to the core \
earest_to\/\dd_query_vector\ which already accept \impl
IntoQueryVector\ for \Arc<dyn Array>\
### TypeScript (\
odejs/lancedb/query.ts\, \rrow.ts\)
3. Extended \IntoVector\ type to include \Uint8Array\ (and
\Float16Array\ via runtime check for Node 22+)
4. \xtractVectorBuffer()\ helper detects non-Float32 typed arrays and
extracts their underlying byte buffer + dtype string
5. \
earestTo()\ and \ddQueryVector()\ route through the raw NAPI path when
the input is Float16/Float64/Uint8
### Backward compatibility
Existing \Float32Array\ and \
umber[]\ inputs are unchanged -- they still use the original \
earest_to(Float32Array)\ NAPI method. The new raw path is only used when
a non-Float32 typed array is detected.
## Usage
\\\ ypescript
// Float16Array (Node 22+) -- no precision loss
const f16vec = new Float16Array([0.1, 0.2, 0.3]);
const results = await
table.query().nearestTo(f16vec).limit(10).toArray();
// Float64Array -- no precision loss
const f64vec = new Float64Array([0.1, 0.2, 0.3]);
const results = await
table.query().nearestTo(f64vec).limit(10).toArray();
// Uint8Array (binary embeddings)
const u8vec = new Uint8Array([1, 0, 1, 1, 0]);
const results = await
table.query().nearestTo(u8vec).limit(10).toArray();
// Existing usage unchanged
const results = await table.query().nearestTo([0.1, 0.2,
0.3]).limit(10).toArray();
\\\
## Note on dependencies
The Rust side uses \rrow_array\, \rrow_buffer\, and \half\ crates.
These should already be in the dependency tree via \lancedb\ core, but
\Cargo.toml\ may need explicit entries for \half\ and the arrow
sub-crates in the nodejs workspace.
---------
Signed-off-by: Vedant Madane <6527493+VedantMadane@users.noreply.github.com>
Co-authored-by: Will Jones <willjones127@gmail.com>
BREAKING CHANGE: embedding function implementations in Node need to now
call `resolveVariables()` in their constructors and should **not**
implement `toJSON()`.
This tries to address the handling of secrets. In Node, they are
currently lost. In Python, they are currently leaked into the table
schema metadata.
This PR introduces an in-memory variable store on the function registry.
It also allows embedding function definitions to label certain config
values as "sensitive", and the preprocessing logic will raise an error
if users try to pass in hard-coded values.
Closes#2110Closes#521
---------
Co-authored-by: Weston Pace <weston.pace@gmail.com>
* Make `npm run docs` fail if there are any warnings. This will catch
items missing from the API reference.
* Add a check in our CI to make sure `npm run dos` runs without warnings
and doesn't generate any new files (indicating it might be out-of-date.
* Hide constructors that aren't user facing.
* Remove unused enum `WriteMode`.
Closes#2068
* Sets `"useCodeBlocks": true`
* Adds a post-processing script `nodejs/typedoc_post_process.js` that
puts the parameter description on the same line as the parameter name,
like it is in our Python docs. This makes the text hierarchy clearer in
those sections and also makes the sections shorter.
lancedb-gatefixer[bot]
Wyatt Alt
Vedant Madane
Will Jones
Cory Grinstead