docs(python): fix cross-references that resolved to the wrong page

`mkdocs build --strict` only catches references it cannot resolve. A bare
anchor such as `[limit][]` or `[vector search][search]` is matched by
autorefs against any heading on the site, so six of them silently linked
into the JavaScript reference instead. The relative links in
`permutation.py` and `remote/errors.py` pointed at in-page anchors and
paths that do not exist.

Targets that still exist here or in an imported inventory now use
mkdocstrings references; the guide pages deleted in #2770 use their
lancedb.com URLs.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

AGENTS.md

@@ -92,6 +92,8 @@ Python bindings changes:
     * Should use `LOOP.run()` to call the corresponding `AsyncTable` method.
 6. Add concrete sync method to `RemoteTable` class in `python/python/lancedb/remote/table.py`.
 7. Add unit test in `python/tests/test_table.py`.
+8. If you added a new public class or module-level function (not just a method on an
+   existing class), expose it in the API reference. See "Python API reference" below.
 
 TypeScript bindings changes:
 
@@ -103,6 +105,33 @@ TypeScript bindings changes:
 5. Add test in `nodejs/__test__/table.test.ts`.
 6. Run `npm run docs` to generate TypeScript documentation.
 
+## Python API reference
+
+`docs/src/python/python.md` is the entire Python API reference. It is maintained by
+hand, and anything not listed there is not rendered at all, so new public classes and
+module-level functions have to be added explicitly. How depends on the module:
+
+* `lancedb.index`, `lancedb.embeddings`, `lancedb.remote`, and `lancedb.rerankers` are
+  rendered by a single directive each, driven by the module's `__all__`. Add the new
+  name to `__all__` and it appears; forget, and it is silently omitted.
+* Everything else (`lancedb`, `lancedb.table`, `lancedb.query`, `lancedb.db`, ...) is
+  listed symbol by symbol. Add a `::: lancedb.<module>.<Name>` line to the matching
+  section, and remember that the page separates synchronous and asynchronous APIs.
+
+Deliberately undocumented: concrete implementations reached through an abstract base
+(`LanceTable`, `LanceDBConnection`, `RemoteDBConnection`), query base classes already
+covered by `inherited_members`, and internal helpers.
+
+Cross-references in docstrings use mkdocstrings syntax, `[text][lancedb.table.Table]`.
+Plain relative links such as `[Table](Table)` do not resolve. To check your work:
+
+```shell
+pip install -r docs/requirements.txt
+cd docs && PYTHONPATH=. mkdocs build
+```
+
+The docs site only builds on pushes to `main`, so this is not covered by PR CI.
+
 ## Review Guidelines
 
 Please consider the following when reviewing code contributions.

Cargo.lock

@@ -5379,7 +5379,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb"
-version = "0.37.0-beta.0"
+version = "0.37.1-beta.0"
 dependencies = [
  "ahash",
  "anyhow",
@@ -5467,7 +5467,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb-nodejs"
-version = "0.37.0-beta.0"
+version = "0.37.1-beta.0"
 dependencies = [
  "arrow-array",
  "arrow-buffer",
@@ -5492,7 +5492,7 @@ dependencies = [
 
 [[package]]
 name = "lancedb-python"
-version = "0.37.0-beta.0"
+version = "0.37.1-beta.0"
 dependencies = [
  "arrow",
  "async-trait",

docs/mkdocs.yml

@@ -51,6 +51,11 @@ plugins:
           paths: [../python/python]
           options:
             docstring_style: numpy
+            docstring_options:
+              # Attributes documented in a `Parameters` section, and pydantic
+              # dataclasses whose `__init__` griffe cannot see statically, both
+              # trip this check. It reports nothing actionable here.
+              warn_unknown_params: false
             heading_level: 3
             show_signature_annotations: true
             show_root_heading: true

docs/openapi.yml

@@ -453,6 +453,16 @@ paths:
                     The metric type to use for the index. l2, Cosine, Dot are supported.
                 index_type:
                   type: string
+                custom_stop_words:
+                  type: [array, "null"]
+                  items:
+                    type: string
+                  description: |
+                    The custom stop-word list for an FTS index. A non-null
+                    array replaces the language's built-in stop-word list and is only
+                    applied when remove_stop_words is enabled. Null uses the built-in
+                    language list, while an empty array explicitly replaces it with no
+                    stop words.
       responses:
         "200":
           description: Index successfully created
@@ -510,4 +520,4 @@ paths:
         "401":
           $ref: "#/components/responses/unauthorized"
         "404":
-          $ref: "#/components/responses/not_found"
+          $ref: "#/components/responses/not_found"

docs/src/js/_media/CONTRIBUTING.md

@@ -1,7 +1,7 @@
 # Contributing to LanceDB Typescript
 
 This document outlines the process for contributing to LanceDB Typescript.
-For general contribution guidelines, see [CONTRIBUTING.md](../CONTRIBUTING.md).
+For general contribution guidelines, see [CONTRIBUTING.md](https://github.com/lancedb/lancedb/blob/main/CONTRIBUTING.md).
 
 ## Project layout
 

docs/src/js/interfaces/FtsOptions.md

@@ -56,6 +56,21 @@ the experimental FTS V3 format and may introduce breaking changes.
 
 ***
 
+### customStopWords?
+
+```ts
+optional customStopWords: string[];
+```
+
+Custom stop words that replace the built-in list for `language`.
+
+This option only affects tokenization when `removeStopWords` is true.
+
+`undefined` keeps the built-in language list. An empty array explicitly
+replaces it with no stop words.
+
+***
+
 ### language?
 
 ```ts

docs/src/js/interfaces/TokenizeOptions.md

@@ -30,6 +30,21 @@ The tokenizer to use. The default is "simple".
 
 ***
 
+### customStopWords?
+
+```ts
+optional customStopWords: string[];
+```
+
+Custom stop words that replace the built-in list for `language`.
+
+This option only affects tokenization when `removeStopWords` is true.
+
+`undefined` keeps the built-in language list. An empty array explicitly
+replaces it with no stop words.
+
+***
+
 ### language?
 
 ```ts

docs/src/python/python.md

@@ -26,6 +26,18 @@ is also an [asynchronous API client](#connections-asynchronous).
 
 ::: lancedb.db.DBConnection
 
+::: lancedb.Session
+
+## Namespaces (Synchronous)
+
+A namespace-backed connection resolves tables through a
+[Lance namespace](https://lancedb.github.io/lance-namespace/) service instead of
+listing a storage directory.
+
+::: lancedb.connect_namespace
+
+::: lancedb.namespace.LanceNamespaceDBConnection
+
 ## Tables (Synchronous)
 
 ::: lancedb.table.Table
@@ -34,8 +46,12 @@ is also an [asynchronous API client](#connections-asynchronous).
 
 ::: lancedb.table.FragmentSummaryStats
 
+::: lancedb.table.TableStatistics
+
 ::: lancedb.table.Tags
 
+::: lancedb.table.Branches
+
 ## Expressions
 
 Type-safe expression builder for filters and projections. Use these instead
@@ -62,29 +78,46 @@ of raw SQL strings with [where][lancedb.query.LanceQueryBuilder.where] and
 
 ::: lancedb.query.LanceHybridQueryBuilder
 
+::: lancedb.query.LanceEmptyQueryBuilder
+
+::: lancedb.query.LanceTakeQueryBuilder
+
+## Full text queries
+
+Structured full text queries can be passed to
+[Table.search][lancedb.table.Table.search] or
+[AsyncTable.search][lancedb.table.AsyncTable.search] in place of a query string,
+and combined with [BooleanQuery][lancedb.query.BooleanQuery].
+
+::: lancedb.query.FullTextQuery
+
+::: lancedb.query.MatchQuery
+
+::: lancedb.query.PhraseQuery
+
+::: lancedb.query.BoostQuery
+
+::: lancedb.query.MultiMatchQuery
+
+::: lancedb.query.BooleanQuery
+
+::: lancedb.query.FullTextOperator
+
+::: lancedb.query.Occur
+
 ## Embeddings
 
-::: lancedb.embeddings.registry.EmbeddingFunctionRegistry
-
-::: lancedb.embeddings.base.EmbeddingFunctionConfig
-
-::: lancedb.embeddings.base.EmbeddingFunction
-
-::: lancedb.embeddings.base.TextEmbeddingFunction
-
-::: lancedb.embeddings.sentence_transformers.SentenceTransformerEmbeddings
-
-::: lancedb.embeddings.openai.OpenAIEmbeddings
-
-::: lancedb.embeddings.open_clip.OpenClipEmbeddings
+::: lancedb.embeddings
+    options:
+      show_root_heading: false
+      show_root_toc_entry: false
 
 ## Remote configuration
 
-::: lancedb.remote.ClientConfig
-
-::: lancedb.remote.TimeoutConfig
-
-::: lancedb.remote.RetryConfig
+::: lancedb.remote
+    options:
+      show_root_heading: false
+      show_root_toc_entry: false
 
 ## Context
 
@@ -94,11 +127,50 @@ of raw SQL strings with [where][lancedb.query.LanceQueryBuilder.where] and
 
 ## Full text search
 
-Use [lancedb.table.Table.create_fts_index][] for the synchronous API or
-[lancedb.table.AsyncTable.create_index][] with [lancedb.index.FTS][] for the
-asynchronous API.
+Pass `custom_stop_words` to [lancedb.index.FTS][]:
 
-::: lancedb.index.FTS
+```python
+from lancedb.index import FTS
+
+table.create_index(
+    "text",
+    config=FTS(remove_stop_words=True, custom_stop_words=["acme", "internal"]),
+)
+```
+
+The list replaces the built-in stop words and is used only when
+`remove_stop_words=True`:
+
+- `custom_stop_words=None` uses the built-in list for `language`.
+- `custom_stop_words=[]` removes no words.
+- Values are passed through without trimming, lowercasing, or other rewriting.
+
+The same option is available on `lancedb.tokenize(...)` and the deprecated
+[lancedb.table.Table.create_fts_index][] compatibility helper:
+
+```python
+import lancedb
+
+tokens = list(lancedb.tokenize("acme makes searchable data",
+                               custom_stop_words=["acme"]))
+```
+
+::: lancedb.tokenize
+
+::: lancedb.FtsToken
+
+## Blobs
+
+Blob columns store large binary values out of line so they can be read lazily
+instead of being materialized with the rest of the row.
+
+::: lancedb.blob
+
+::: lancedb.BlobType
+
+::: lancedb._blob.BlobFile
+    options:
+      show_root_full_path: false
 
 ## Utilities
 
@@ -106,6 +178,14 @@ asynchronous API.
 
 ::: lancedb.merge.LanceMergeInsertBuilder
 
+::: lancedb.otel.instrument_lancedb_metrics
+
+## Exceptions
+
+::: lancedb.exceptions.MissingValueError
+
+::: lancedb.exceptions.MissingColumnError
+
 ## Integrations
 
 ## Pydantic
@@ -114,19 +194,30 @@ asynchronous API.
 
 ::: lancedb.pydantic.vector
 
+::: lancedb.pydantic.Vector
+
+::: lancedb.pydantic.MultiVector
+
 ::: lancedb.pydantic.LanceModel
 
+## PyTorch
+
+::: lancedb.streaming.StreamingDataset
+
+::: lancedb.permutation.permutation_builder
+
+::: lancedb.permutation.PermutationBuilder
+
+::: lancedb.permutation.Permutation
+
+::: lancedb.permutation.Transforms
+
 ## Reranking
 
-::: lancedb.rerankers.linear_combination.LinearCombinationReranker
-
-::: lancedb.rerankers.cohere.CohereReranker
-
-::: lancedb.rerankers.colbert.ColbertReranker
-
-::: lancedb.rerankers.cross_encoder.CrossEncoderReranker
-
-::: lancedb.rerankers.openai.OpenaiReranker
+::: lancedb.rerankers
+    options:
+      show_root_heading: false
+      show_root_toc_entry: false
 
 ## Connections (Asynchronous)
 
@@ -137,6 +228,12 @@ can be used to create, list, or open tables.
 
 ::: lancedb.db.AsyncConnection
 
+## Namespaces (Asynchronous)
+
+::: lancedb.connect_namespace_async
+
+::: lancedb.namespace.AsyncLanceNamespaceDBConnection
+
 ## Tables (Asynchronous)
 
 Table hold your actual data as a collection of records / rows.
@@ -145,32 +242,20 @@ Table hold your actual data as a collection of records / rows.
 
 ::: lancedb.table.AsyncTags
 
+::: lancedb.table.AsyncBranches
+
 ## Indices (Asynchronous)
 
 Indices can be created on a table to speed up queries. This section
 lists the indices that LanceDb supports.
 
-::: lancedb.index.BTree
-
-::: lancedb.index.Bitmap
-
-::: lancedb.index.LabelList
-
-::: lancedb.index.FTS
-
-::: lancedb.index.IvfPq
-
-::: lancedb.index.HnswPq
-
-::: lancedb.index.HnswSq
-
-::: lancedb.index.IvfFlat
-
-::: lancedb.index.IvfSq
-
-::: lancedb.index.IvfRq
-
-::: lancedb.index.HnswFlat
+::: lancedb.index
+    options:
+      show_root_heading: false
+      show_root_toc_entry: false
+      # `lang_mapping` is defined in the module rather than imported, so it is
+      # picked up despite not being in `__all__`. It is an internal lookup table.
+      filters: ["!^_", "!^lang_mapping$"]
 
 ::: lancedb.table.IndexStatistics
 
@@ -198,3 +283,7 @@ rows nearest to a query vector and can be created with the
 ::: lancedb.query.AsyncHybridQuery
     options:
       inherited_members: true
+
+::: lancedb.query.AsyncTakeQuery
+    options:
+      inherited_members: true

nodejs/CONTRIBUTING.md

@@ -1,7 +1,7 @@
 # Contributing to LanceDB Typescript
 
 This document outlines the process for contributing to LanceDB Typescript.
-For general contribution guidelines, see [CONTRIBUTING.md](../CONTRIBUTING.md).
+For general contribution guidelines, see [CONTRIBUTING.md](https://github.com/lancedb/lancedb/blob/main/CONTRIBUTING.md).
 
 ## Project layout
 

nodejs/__test__/remote.test.ts

@@ -226,7 +226,7 @@ describe("remote connection", () => {
     );
   });
 
-  it("sends the FTS posting block size to remote tables", async () => {
+  it("sends FTS options to remote tables", async () => {
     let createIndexBody: Record<string, unknown> | undefined;
 
     await withMockDatabase(
@@ -264,7 +264,11 @@ describe("remote connection", () => {
       async (db) => {
         const table = await db.openTable("t");
         await table.createIndex("text", {
-          config: Index.fts({ blockSize: 256 }),
+          config: Index.fts({
+            blockSize: 256,
+            removeStopWords: true,
+            customStopWords: ["the"],
+          }),
         });
       },
     );
@@ -272,6 +276,7 @@ describe("remote connection", () => {
     expect(createIndexBody?.["column"]).toBe("text");
     expect(createIndexBody?.["index_type"]).toBe("FTS");
     expect(createIndexBody?.["block_size"]).toBe(256);
+    expect(createIndexBody?.["custom_stop_words"]).toEqual(["the"]);
   });
 
   it("diffs and merges remote branches", async () => {

nodejs/__test__/table.test.ts

@@ -2769,6 +2769,15 @@ describe.each([arrow15, arrow16, arrow17, arrow18])(
   },
 );
 
+test("tokenize supports custom stop words", async () => {
+  const tokens = await tokenize("the lance data", {
+    stem: false,
+    removeStopWords: true,
+    customStopWords: ["lance"],
+  });
+  expect(tokens.map((token) => token.text)).toEqual(["the", "data"]);
+});
+
 describe("when calling explainPlan", () => {
   let tmpDir: tmp.DirResult;
   let table: Table;

nodejs/examples/full_text_search.test.ts

@@ -29,8 +29,14 @@ test("full text search", async () => {
     const tbl = await db.createTable("myVectors", data, { mode: "overwrite" });
 
     await tbl.createIndex("doc", {
-      config: lancedb.Index.fts(),
+      config: lancedb.Index.fts({
+        stem: false,
+        removeStopWords: true,
+        customStopWords: ["banana"],
+      }),
     });
+    const tokens = await tbl.tokenize("apple banana", { column: "doc" });
+    expect(tokens.map((token) => token.text)).toEqual(["apple"]);
 
     // --8<-- [start:full_text_search]
     const result = await tbl

nodejs/lancedb/index.ts

@@ -194,6 +194,16 @@ export interface TokenizeOptions {
   /** Whether to remove stop words. */
   removeStopWords?: boolean;
 
+  /**
+   * Custom stop words that replace the built-in list for `language`.
+   *
+   * This option only affects tokenization when `removeStopWords` is true.
+   *
+   * `undefined` keeps the built-in language list. An empty array explicitly
+   * replaces it with no stop words.
+   */
+  customStopWords?: string[];
+
   /** Whether to fold ASCII characters. */
   asciiFolding?: boolean;
 
@@ -225,6 +235,7 @@ export async function tokenize(
     options?.lowercase,
     options?.stem,
     options?.removeStopWords,
+    options?.customStopWords,
     options?.asciiFolding,
     options?.ngramMinLength,
     options?.ngramMaxLength,

nodejs/lancedb/indices.ts

@@ -553,6 +553,16 @@ export interface FtsOptions {
    */
   removeStopWords?: boolean;
 
+  /**
+   * Custom stop words that replace the built-in list for `language`.
+   *
+   * This option only affects tokenization when `removeStopWords` is true.
+   *
+   * `undefined` keeps the built-in language list. An empty array explicitly
+   * replaces it with no stop words.
+   */
+  customStopWords?: string[];
+
   /**
    * whether to remove punctuation
    */
@@ -755,6 +765,7 @@ export class Index {
         options?.lowercase,
         options?.stem,
         options?.removeStopWords,
+        options?.customStopWords,
         options?.asciiFolding,
         options?.ngramMinLength,
         options?.ngramMaxLength,

nodejs/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@lancedb/lancedb",
-  "version": "0.37.0-beta.0",
+  "version": "0.37.1-beta.0",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "@lancedb/lancedb",
-      "version": "0.37.0-beta.0",
+      "version": "0.37.1-beta.0",
       "cpu": [
         "x64",
         "arm64"

nodejs/src/index.rs

@@ -43,6 +43,7 @@ pub fn tokenize(
     lower_case: Option<bool>,
     stem: Option<bool>,
     remove_stop_words: Option<bool>,
+    custom_stop_words: Option<Vec<String>>,
     ascii_folding: Option<bool>,
     ngram_min_length: Option<u32>,
     ngram_max_length: Option<u32>,
@@ -72,6 +73,7 @@ pub fn tokenize(
     if let Some(remove_stop_words) = remove_stop_words {
         opts = opts.remove_stop_words(remove_stop_words);
     }
+    opts = opts.custom_stop_words(custom_stop_words);
     if let Some(ascii_folding) = ascii_folding {
         opts = opts.ascii_folding(ascii_folding);
     }
@@ -222,6 +224,7 @@ impl Index {
         lower_case: Option<bool>,
         stem: Option<bool>,
         remove_stop_words: Option<bool>,
+        custom_stop_words: Option<Vec<String>>,
         ascii_folding: Option<bool>,
         ngram_min_length: Option<u32>,
         ngram_max_length: Option<u32>,
@@ -250,6 +253,7 @@ impl Index {
         if let Some(remove_stop_words) = remove_stop_words {
             opts = opts.remove_stop_words(remove_stop_words);
         }
+        opts = opts.custom_stop_words(custom_stop_words);
         if let Some(ascii_folding) = ascii_folding {
             opts = opts.ascii_folding(ascii_folding);
         }

python/python/lancedb/__init__.py

@@ -258,6 +258,7 @@ def tokenize(
     lower_case: bool = True,
     stem: bool = True,
     remove_stop_words: bool = True,
+    custom_stop_words: Optional[List[str]] = None,
     ascii_folding: bool = True,
     ngram_min_length: int = 3,
     ngram_max_length: int = 3,
@@ -265,9 +266,10 @@ def tokenize(
 ) -> Iterable[FtsToken]:
     """Tokenize a full-text search query using an explicit tokenizer.
 
-    This does not require a table or FTS index. The tokenizer options match
-    :class:`lancedb.index.FTS`.
+    This does not require an FTS index. The tokenizer options match
+    :class:`lancedb.index.FTS`. ``custom_stop_words`` accepts a list of strings.
     """
+
     return _tokenize(
         query,
         base_tokenizer=base_tokenizer,
@@ -276,6 +278,7 @@ def tokenize(
         lower_case=lower_case,
         stem=stem,
         remove_stop_words=remove_stop_words,
+        custom_stop_words=custom_stop_words,
         ascii_folding=ascii_folding,
         ngram_min_length=ngram_min_length,
         ngram_max_length=ngram_max_length,

python/python/lancedb/_lancedb.pyi

@@ -59,6 +59,7 @@ def tokenize(
     lower_case: bool = True,
     stem: bool = True,
     remove_stop_words: bool = True,
+    custom_stop_words: Optional[List[str]] = None,
     ascii_folding: bool = True,
     ngram_min_length: int = 3,
     ngram_max_length: int = 3,

python/python/lancedb/db.py

@@ -359,7 +359,7 @@ class DBConnection(EnforceOverrides):
 
         Data is converted to Arrow before being written to disk. For maximum
         control over how data is saved, either provide the PyArrow schema to
-        convert to or else provide a [PyArrow Table](pyarrow.Table) directly.
+        convert to or else provide a [PyArrow Table][pyarrow.Table] directly.
 
         >>> import pyarrow as pa
         >>> custom_schema = pa.schema([
@@ -1529,7 +1529,7 @@ class AsyncConnection(object):
 
         Data is converted to Arrow before being written to disk. For maximum
         control over how data is saved, either provide the PyArrow schema to
-        convert to or else provide a [PyArrow Table](pyarrow.Table) directly.
+        convert to or else provide a [PyArrow Table][pyarrow.Table] directly.
 
         >>> import pyarrow as pa
         >>> custom_schema = pa.schema([

python/python/lancedb/embeddings/__init__.py

@@ -21,3 +21,32 @@ from .watsonx import WatsonxEmbeddings
 from .voyageai import VoyageAIEmbeddingFunction
 from .colpali import ColPaliEmbeddings
 from .siglip import SigLipEmbeddings
+
+# The API reference renders this package with a single mkdocstrings directive,
+# which only picks up names listed here. New embedding functions must be added
+# to both the imports above and this list, or they will silently go undocumented.
+__all__ = [
+    "EmbeddingFunction",
+    "EmbeddingFunctionConfig",
+    "TextEmbeddingFunction",
+    "EmbeddingFunctionRegistry",
+    "get_registry",
+    "register",
+    "SentenceTransformerEmbeddings",
+    "OpenAIEmbeddings",
+    "OpenClipEmbeddings",
+    "BedRockText",
+    "CohereEmbeddingFunction",
+    "GeminiText",
+    "GteEmbeddings",
+    "InstructorEmbeddingFunction",
+    "JinaEmbeddings",
+    "OllamaEmbeddings",
+    "TransformersEmbeddingFunction",
+    "ColbertEmbeddings",
+    "VoyageAIEmbeddingFunction",
+    "WatsonxEmbeddings",
+    "ColPaliEmbeddings",
+    "ImageBindEmbeddings",
+    "SigLipEmbeddings",
+]

python/python/lancedb/embeddings/bedrock.py

@@ -21,20 +21,20 @@ class BedRockText(TextEmbeddingFunction):
     """
     Parameters
     ----------
-    name: str, default "amazon.titan-embed-text-v1"
+    name : str, default "amazon.titan-embed-text-v1"
         The model ID of the bedrock model to use. Supported models for are:
         - amazon.titan-embed-text-v1
         - cohere.embed-english-v3
         - cohere.embed-multilingual-v3
-    region: str, default "us-east-1"
+    region : str, default "us-east-1"
         Optional name of the AWS Region in which the service should be called.
-    profile_name: str, default None
+    profile_name : str, default None
         Optional name of the AWS profile to use for calling the Bedrock service.
         If not specified, the default profile will be used.
-    assumed_role: str, default None
+    assumed_role : str, default None
         Optional ARN of an AWS IAM role to assume for calling the Bedrock service.
         If not specified, the current active credentials will be used.
-    role_session_name: str, default "lancedb-embeddings"
+    role_session_name : str, default "lancedb-embeddings"
         Optional name of the AWS IAM role session to use for calling the Bedrock
         service. If not specified, "lancedb-embeddings" name will be used.
 

python/python/lancedb/embeddings/cohere.py

@@ -22,7 +22,7 @@ class CohereEmbeddingFunction(TextEmbeddingFunction):
 
     Parameters
     ----------
-    name: str, default "embed-multilingual-v2.0"
+    name : str, default "embed-multilingual-v2.0"
         The name of the model to use. List of acceptable models:
 
             * embed-english-v3.0
@@ -33,12 +33,14 @@ class CohereEmbeddingFunction(TextEmbeddingFunction):
             * embed-english-light-v2.0
             * embed-multilingual-v2.0
 
-    source_input_type: str, default "search_document"
+    source_input_type : str, default "search_document"
         The input type for the source column in the database
 
-    query_input_type: str, default "search_query"
+    query_input_type : str, default "search_query"
         The input type for the query column in the database
 
+    Notes
+    -----
     Cohere supports following input types:
 
     | Input Type               | Description                          |

python/python/lancedb/embeddings/colpali.py

@@ -44,7 +44,7 @@ class ColPaliEmbeddings(EmbeddingFunction):
         The token pooling strategy to use, by default "hierarchical".
         - "hierarchical": Progressively pools tokens to reduce sequence length.
         - "lambda": A simpler pooling that uses a custom `pooling_func`.
-    pooling_func: typing.Callable, optional
+    pooling_func : typing.Callable, optional
         A function to use for pooling when `pooling_strategy` is "lambda".
     pool_factor : int
         Factor to reduce sequence length if token pooling is enabled (default 2).
@@ -52,7 +52,7 @@ class ColPaliEmbeddings(EmbeddingFunction):
         Quantization configuration for the model. (default None, bitsandbytes needed)
     batch_size : int
         Batch size for processing inputs (default 2).
-    offload_folder: str, optional
+    offload_folder : str, optional
         Folder to offload model weights if using CPU offloading (default None). This is
         useful for large models that do not fit in memory.
     """

python/python/lancedb/embeddings/gemini_text.py

@@ -48,16 +48,16 @@ class GeminiText(TextEmbeddingFunction):
 
     Parameters
     ----------
-    name: str, default "gemini-embedding-001"
+    name : str, default "gemini-embedding-001"
         The name of the model to use. Supported models include:
         - "gemini-embedding-001" (768 dimensions)
 
         Note: The legacy "models/embedding-001" format is also supported but
         "gemini-embedding-001" is recommended.
 
-    query_task_type: str, default "retrieval_query"
+    query_task_type : str, default "retrieval_query"
         Sets the task type for the queries.
-    source_task_type: str, default "retrieval_document"
+    source_task_type : str, default "retrieval_document"
         Sets the task type for ingestion.
 
     Examples

python/python/lancedb/embeddings/gte.py

@@ -26,13 +26,13 @@ class GteEmbeddings(TextEmbeddingFunction):
 
     Parameters
     ----------
-    name: str, default "thenlper/gte-large"
+    name : str, default "thenlper/gte-large"
         The name of the model to use.
-    device: str, default "cpu"
+    device : str, default "cpu"
         Sets the device type for the model.
-    normalize: str, default "True"
+    normalize : str, default "True"
         Controls normalize param in encode function for the transformer.
-    mlx: bool, default False
+    mlx : bool, default False
         Controls which model to use. False for gte-large,True for the mlx version.
 
     Examples

python/python/lancedb/embeddings/instructor.py

@@ -35,23 +35,23 @@ class InstructorEmbeddingFunction(TextEmbeddingFunction):
 
     Parameters
     ----------
-    name: str
+    name : str
         The name of the model to use. Available models are listed at
         https://github.com/xlang-ai/instructor-embedding#model-list;
         The default model is hkunlp/instructor-base
-    batch_size: int, default 32
+    batch_size : int, default 32
         The batch size to use when generating embeddings
-    device: str, default "cpu"
+    device : str, default "cpu"
         The device to use when generating embeddings
-    show_progress_bar: bool, default True
+    show_progress_bar : bool, default True
         Whether to show a progress bar when generating embeddings
-    normalize_embeddings: bool, default True
+    normalize_embeddings : bool, default True
         Whether to normalize the embeddings
-    quantize: bool, default False
+    quantize : bool, default False
         Whether to quantize the model
-    source_instruction: str, default "represent the document for retrieval"
+    source_instruction : str, default "represent the document for retrieval"
         The instruction for the source column
-    query_instruction: str, default "represent the document for retrieving the most
+    query_instruction : str, default "represent the document for retrieving the most
         similar documents"
         The instruction for the query
 

python/python/lancedb/embeddings/jinaai.py

@@ -40,10 +40,10 @@ class JinaEmbeddings(EmbeddingFunction):
 
     Parameters
     ----------
-    name: str, default "jina-clip-v1". Note that some models support both image
+    name : str, default "jina-clip-v1". Note that some models support both image
         and text embeddings and some just text embedding
 
-    api_key: str, default None
+    api_key : str, default None
         The api key to access Jina API. If you pass None, you can set JINA_API_KEY
         environment variable
 

python/python/lancedb/embeddings/sentence_transformers.py

@@ -21,13 +21,13 @@ class SentenceTransformerEmbeddings(TextEmbeddingFunction):
 
     Parameters
     ----------
-    name: str, default "all-MiniLM-L6-v2"
+    name : str, default "all-MiniLM-L6-v2"
         The name of the model to use.
-    device: str, default "cpu"
+    device : str, default "cpu"
         The device to use for the model
-    normalize: bool, default True
+    normalize : bool, default True
         Whether to normalize the embeddings
-    trust_remote_code: bool, default True
+    trust_remote_code : bool, default True
         Whether to trust the remote code
     """
 

python/python/lancedb/embeddings/voyageai.py

@@ -167,7 +167,7 @@ class VoyageAIEmbeddingFunction(EmbeddingFunction):
 
     Parameters
     ----------
-    name: str
+    name : str
         The name of the model to use. List of acceptable models:
 
             * voyage-4 (1024 dims, general-purpose and multilingual retrieval)
@@ -185,7 +185,7 @@ class VoyageAIEmbeddingFunction(EmbeddingFunction):
             * voyage-law-2
             * voyage-code-2
 
-    output_dimension: int, optional
+    output_dimension : int, optional
         The output dimension for models that support flexible dimensions.
         Currently only voyage-multimodal-3.5 supports this feature.
         Valid options: 256, 512, 1024 (default), 2048.

python/python/lancedb/index.py

@@ -2,7 +2,7 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 from dataclasses import dataclass
-from typing import Literal, Optional
+from typing import List, Literal, Optional
 
 from ._lancedb import (
     IndexConfig,
@@ -151,6 +151,11 @@ class FTS:
     remove_stop_words : bool, default True
         Whether to remove stop words. Stop words are common words that are often
         removed from text before indexing. For example, in English "the" and "and".
+    custom_stop_words : list of str, optional
+        Custom words replace the built-in language stop words
+        and only take effect when ``remove_stop_words`` is True. ``None`` uses
+        the built-in language list, while an empty list explicitly uses no
+        stop words.
     ascii_folding : bool, default True
         Whether to fold ASCII characters. This converts accented characters to
         their ASCII equivalent. For example, "café" would be converted to "cafe".
@@ -179,6 +184,7 @@ class FTS:
     ngram_max_length: int = 3
     prefix_only: bool = False
     block_size: int = 128
+    custom_stop_words: Optional[List[str]] = None
 
 
 @dataclass
@@ -213,7 +219,7 @@ class HnswPq:
         distance has a range of (-∞, ∞). If the vectors are normalized (i.e. their
         l2 norm is 1), then dot distance is equivalent to the cosine distance.
 
-    num_partitions, default sqrt(num_rows)
+    num_partitions: int, default sqrt(num_rows)
 
         The number of IVF partitions to create.
 
@@ -222,7 +228,7 @@ class HnswPq:
         will require too much memory. Each partition becomes its own HNSW graph, so
         setting this value higher reduces the peak memory use of training.
 
-    num_sub_vectors, default is vector dimension / 16
+    num_sub_vectors: int, default is vector dimension / 16
 
         Number of sub-vectors of PQ.
 
@@ -238,13 +244,13 @@ class HnswPq:
         If the dimension is not visible by 8 then we use 1 subvector.  This is not
         ideal and will likely result in poor performance.
 
-     num_bits: int, default 8
+    num_bits: int, default 8
         Number of bits to encode each sub-vector.
 
         This value controls how much the sub-vectors are compressed.  The more bits
         the more accurate the index but the slower search. Only 4 and 8 are supported.
 
-    max_iterations, default 50
+    max_iterations: int, default 50
 
         Max iterations to train kmeans.
 
@@ -257,7 +263,7 @@ class HnswPq:
         those cases it is unlikely that setting this larger will lead to the index
         converging anyways.
 
-    sample_rate, default 256
+    sample_rate: int, default 256
 
         The rate used to calculate the number of training vectors for kmeans.
 
@@ -273,14 +279,14 @@ class HnswPq:
         Increasing this value might improve the quality of the index but in
         most cases the default should be sufficient.
 
-    m, default 20
+    m: int, default 20
 
         The number of neighbors to select for each vector in the HNSW graph.
 
         This value controls the tradeoff between search speed and accuracy.
         The higher the value the more accurate the search but the slower it will be.
 
-    ef_construction, default 300
+    ef_construction: int, default 300
 
         The number of candidates to evaluate during the construction of the HNSW graph.
 
@@ -291,7 +297,7 @@ class HnswPq:
         This value should be set to a value that is not less than `ef` in the
         search phase.
 
-    target_partition_size, default is 1,048,576
+    target_partition_size: int, default is 1,048,576
 
         The target size of each partition.
 
@@ -345,7 +351,7 @@ class HnswSq:
         distance has a range of (-∞, ∞). If the vectors are normalized (i.e. their
         l2 norm is 1), then dot distance is equivalent to the cosine distance.
 
-    num_partitions, default sqrt(num_rows)
+    num_partitions: int, default sqrt(num_rows)
 
         The number of IVF partitions to create.
 
@@ -354,7 +360,7 @@ class HnswSq:
         will require too much memory. Each partition becomes its own HNSW graph, so
         setting this value higher reduces the peak memory use of training.
 
-    max_iterations, default 50
+    max_iterations: int, default 50
 
         Max iterations to train kmeans.
 
@@ -367,7 +373,7 @@ class HnswSq:
         In those cases it is unlikely that setting this larger will lead to
         the index converging anyways.
 
-    sample_rate, default 256
+    sample_rate: int, default 256
 
         The rate used to calculate the number of training vectors for kmeans.
 
@@ -383,14 +389,14 @@ class HnswSq:
         Increasing this value might improve the quality of the index but in
         most cases the default should be sufficient.
 
-    m, default 20
+    m: int, default 20
 
         The number of neighbors to select for each vector in the HNSW graph.
 
         This value controls the tradeoff between search speed and accuracy.
         The higher the value the more accurate the search but the slower it will be.
 
-    ef_construction, default 300
+    ef_construction: int, default 300
 
         The number of candidates to evaluate during the construction of the HNSW graph.
 
@@ -401,7 +407,7 @@ class HnswSq:
         This value should be set to a value that is not less than `ef` in the search
         phase.
 
-    target_partition_size, default is 1,048,576
+    target_partition_size: int, default is 1,048,576
 
         The target size of each partition.
 
@@ -454,7 +460,7 @@ class HnswFlat:
         distance has a range of (-∞, ∞). If the vectors are normalized (i.e. their
         l2 norm is 1), then dot distance is equivalent to the cosine distance.
 
-    num_partitions, default sqrt(num_rows)
+    num_partitions: int, default sqrt(num_rows)
 
         The number of IVF partitions to create.
 
@@ -464,18 +470,18 @@ class HnswFlat:
         graph, so setting this value higher reduces the peak memory use of
         training.
 
-    max_iterations, default 50
+    max_iterations: int, default 50
 
         Max iterations to train kmeans.
 
         When training an IVF index we use kmeans to calculate the partitions.
         This parameter controls how many iterations of kmeans to run.
 
-    sample_rate, default 256
+    sample_rate: int, default 256
 
         The rate used to calculate the number of training vectors for kmeans.
 
-    m, default 20
+    m: int, default 20
 
         The number of neighbors to select for each vector in the HNSW graph.
 
@@ -483,7 +489,7 @@ class HnswFlat:
         The higher the value the more accurate the search but the slower it
         will be.
 
-    ef_construction, default 300
+    ef_construction: int, default 300
 
         The number of candidates to evaluate during the construction of the HNSW
         graph.
@@ -495,7 +501,7 @@ class HnswFlat:
         than 500.  This value should be set to a value that is not less than `ef`
         in the search phase.
 
-    target_partition_size, default is 1,048,576
+    target_partition_size: int, default is 1,048,576
 
         The target size of each partition.
     """
@@ -599,7 +605,7 @@ class IvfFlat:
 
         The default value is 256.
 
-    target_partition_size, default is 8192
+    target_partition_size: int, default is 8192
 
         The target size of each partition.
 
@@ -763,7 +769,7 @@ class IvfPq:
 
         The default value is 256.
 
-    target_partition_size, default is 8192
+    target_partition_size: int, default is 8192
 
         The target size of each partition.
 
@@ -824,7 +830,7 @@ class IvfRq:
     sample_rate: int, default 256
         Controls the number of training vectors: sample_rate * num_partitions.
 
-    target_partition_size, default is 8192
+    target_partition_size: int, default is 8192
         Target size of each partition.
     """
 
@@ -839,6 +845,9 @@ class IvfRq:
     accelerator: Optional[str] = None
 
 
+# The API reference renders this module with a single mkdocstrings directive,
+# which only picks up names listed here. New public names must be added to this
+# list, or they will silently go undocumented.
 __all__ = [
     "BTree",
     "IvfPq",

python/python/lancedb/permutation.py

@@ -438,7 +438,8 @@ class Permutation:
         _reader: Optional[PermutationReader] = None,
     ):
         """
-        Internal constructor.  Use [from_tables](#from_tables) instead.
+        Internal constructor.  Use
+        [from_tables][lancedb.permutation.Permutation.from_tables] instead.
         """
         assert base_table is not None, "base_table is required"
         assert selection is not None, "selection is required"
@@ -985,8 +986,9 @@ class Permutation:
         types.  Conversion of strings, lists, and structs will require creating python
         objects and this is not zero-copy.
 
-        For custom formatting, use [with_transform](#with_transform) which overrides
-        this method.
+        For custom formatting, use
+        [with_transform][lancedb.permutation.Permutation.with_transform] which
+        overrides this method.
         """
         assert format is not None, "format is required"
         if format == "python":
@@ -1061,7 +1063,8 @@ class Permutation:
         Note: this method returns a new permutation and does not modify `self`
         It is provided for compatibility with the huggingface Dataset API.
 
-        Use [with_skip](#with_skip) instead to avoid confusion.
+        Use [with_skip][lancedb.permutation.Permutation.with_skip] instead to
+        avoid confusion.
         """
         return self.with_skip(skip)
 
@@ -1084,7 +1087,8 @@ class Permutation:
         Note: this method returns a new permutation and does not modify `self`
         It is provided for compatibility with the huggingface Dataset API.
 
-        Use [with_take](#with_take) instead to avoid confusion.
+        Use [with_take][lancedb.permutation.Permutation.with_take] instead to
+        avoid confusion.
         """
         return self.with_take(limit)
 
@@ -1107,7 +1111,8 @@ class Permutation:
         Note: this method returns a new permutation and does not modify `self`
         It is provided for compatibility with the huggingface Dataset API.
 
-        Use [with_repeat](#with_repeat) instead to avoid confusion.
+        Use [with_repeat][lancedb.permutation.Permutation.with_repeat] instead
+        to avoid confusion.
         """
         return self.with_repeat(times)
 

python/python/lancedb/query.py

@@ -651,7 +651,8 @@ class Query(pydantic.BaseModel):
     distance_type : Optional[str]
         the distance type to use for vector search
 
-        This can be l2 (default), cosine and dot.  See [metric definitions][search] for
+        This can be l2 (default), cosine and dot.  See
+        [metric definitions](https://lancedb.com/docs/search/vector-search/) for
         more details.
 
         If this is not a vector search this will be None.
@@ -664,8 +665,9 @@ class Query(pydantic.BaseModel):
 
         - A higher number makes search more accurate but also slower.
 
-        - See discussion in [Querying an ANN Index][querying-an-ann-index] for
-          tuning advice.
+        - See discussion in
+          [Querying an ANN Index](https://lancedb.com/docs/indexing/)
+          for tuning advice.
 
         Will be None if this is not a vector search.
     refine_factor : Optional[int]
@@ -673,8 +675,9 @@ class Query(pydantic.BaseModel):
 
         - A higher number makes search more accurate but also slower.
 
-        - See discussion in [Querying an ANN Index][querying-an-ann-index] for
-          tuning advice.
+        - See discussion in
+          [Querying an ANN Index](https://lancedb.com/docs/indexing/)
+          for tuning advice.
 
         Will be None if this is not a vector search.
     lower_bound : Optional[float]
@@ -1651,8 +1654,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         Higher values will yield better recall (more likely to find vectors if
         they exist) at the expense of latency.
 
-        See discussion in [Querying an ANN Index][querying-an-ann-index] for
-        tuning advice.
+        See discussion in [Querying an ANN Index](https://lancedb.com/docs/indexing/)
+        for tuning advice.
 
         This method sets both the minimum and maximum number of probes to the same
         value. See `minimum_nprobes` and `maximum_nprobes` for more fine-grained
@@ -1752,8 +1755,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         As an example, a refine factor of 2 will sample 2x as many vectors as
         requested, re-ranks them, and returns the top half most relevant results.
 
-        See discussion in [Querying an ANN Index][querying-an-ann-index] for
-        tuning advice.
+        See discussion in [Querying an ANN Index](https://lancedb.com/docs/indexing/)
+        for tuning advice.
 
         Parameters
         ----------
@@ -3379,8 +3382,9 @@ class AsyncQuery(AsyncStandardQuery):
         are various ANN search parameters that will let you fine tune your recall
         accuracy vs search latency.
 
-        Vector searches always have a [limit][].  If `limit` has not been called then
-        a default `limit` of 10 will be used.
+        Vector searches always have a
+        [limit][lancedb.query.AsyncVectorQuery.limit].  If `limit` has not been
+        called then a default `limit` of 10 will be used.
 
         Typically, a single vector is passed in as the query. However, you can also
         pass in multiple vectors. When multiple vectors are passed in, if the vector
@@ -3511,8 +3515,9 @@ class AsyncFTSQuery(AsyncStandardQuery):
         are various ANN search parameters that will let you fine tune your recall
         accuracy vs search latency.
 
-        Hybrid searches always have a [limit][].  If `limit` has not been called then
-        a default `limit` of 10 will be used.
+        Hybrid searches always have a
+        [limit][lancedb.query.AsyncHybridQuery.limit].  If `limit` has not been
+        called then a default `limit` of 10 will be used.
 
         Typically, a single vector is passed in as the query. However, you can also
         pass in multiple vectors.  This can be useful if you want to find the nearest

python/python/lancedb/remote/__init__.py

@@ -11,6 +11,9 @@ from lancedb import __version__
 from .header import HeaderProvider
 from .oauth import OAuthConfig, OAuthFlowType
 
+# The API reference renders this module with a single mkdocstrings directive,
+# which only picks up names listed here. New public names must be added to this
+# list, or they will silently go undocumented.
 __all__ = [
     "TimeoutConfig",
     "RetryConfig",

python/python/lancedb/remote/errors.py

@@ -53,9 +53,9 @@ class RetryError(LanceDBClientError):
     """An error that occurs when the client has exceeded the maximum number of retries.
 
     The retry strategy can be adjusted by setting the
-    [retry_config](lancedb.remote.ClientConfig.retry_config) in the client
+    [retry_config][lancedb.remote.ClientConfig.retry_config] in the client
     configuration. This is passed in the `client_config` argument of
-    [connect](lancedb.connect) and [connect_async](lancedb.connect_async).
+    [connect][lancedb.connect] and [connect_async][lancedb.connect_async].
 
     The __cause__ attribute of this exception will be the last exception that
     caused the retry to fail. It will be an

python/python/lancedb/remote/table.py

@@ -340,6 +340,7 @@ class RemoteTable(Table):
         lower_case: bool = True,
         stem: bool = True,
         remove_stop_words: bool = True,
+        custom_stop_words: Optional[List[str]] = None,
         ascii_folding: bool = True,
         ngram_min_length: int = 3,
         ngram_max_length: int = 3,
@@ -361,6 +362,7 @@ class RemoteTable(Table):
             lower_case=lower_case,
             stem=stem,
             remove_stop_words=remove_stop_words,
+            custom_stop_words=custom_stop_words,
             ascii_folding=ascii_folding,
             ngram_min_length=ngram_min_length,
             ngram_max_length=ngram_max_length,
@@ -578,8 +580,9 @@ class RemoteTable(Table):
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
     ) -> AddResult:
-        """Add more data to the [Table](Table). It has the same API signature as
-        the OSS version.
+        """Add more data to the [Table][lancedb.table.Table].
+
+        It has the same API signature as the OSS version.
 
         Parameters
         ----------
@@ -639,7 +642,8 @@ class RemoteTable(Table):
         fast_search: bool = False,
     ) -> LanceVectorQueryBuilder:
         """Create a search query to find the nearest neighbors
-        of the given query vector. We currently support [vector search][search]
+        of the given query vector. We currently support
+        [vector search](https://lancedb.com/docs/search/vector-search/)
 
         All query options are defined in
         [LanceVectorQueryBuilder][lancedb.query.LanceVectorQueryBuilder].

python/python/lancedb/rerankers/__init__.py

@@ -14,6 +14,9 @@ from .answerdotai import AnswerdotaiRerankers
 from .voyageai import VoyageAIReranker
 from .watsonx import WatsonxReranker
 
+# The API reference renders this module with a single mkdocstrings directive,
+# which only picks up names listed here. New public names must be added to this
+# list, or they will silently go undocumented.
 __all__ = [
     "Reranker",
     "CrossEncoderReranker",

python/python/lancedb/table.py

@@ -1103,6 +1103,7 @@ class Table(ABC):
         lower_case: bool = True,
         stem: bool = True,
         remove_stop_words: bool = True,
+        custom_stop_words: Optional[List[str]] = None,
         ascii_folding: bool = True,
         ngram_min_length: int = 3,
         ngram_max_length: int = 3,
@@ -1170,6 +1171,9 @@ class Table(ABC):
         remove_stop_words : bool, default True
             Whether to remove stop words. Stop words are common words that are often
             removed from text before indexing. For example, in English "the" and "and".
+        custom_stop_words : list of str, optional
+            Custom words that replace the built-in language stop words. ``None``
+            uses the built-in list; an empty list explicitly uses no stop words.
         ascii_folding : bool, default True
             Whether to fold ASCII characters. This converts accented characters to
             their ASCII equivalent. For example, "café" would be converted to "cafe".
@@ -1207,7 +1211,7 @@ class Table(ABC):
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
     ) -> AddResult:
-        """Add more data to the [Table](Table).
+        """Add more data to the [Table][lancedb.table.Table].
 
         Parameters
         ----------
@@ -1339,8 +1343,8 @@ class Table(ABC):
         fts_columns: Optional[Union[str, List[str]]] = None,
     ) -> LanceQueryBuilder:
         """Create a search query to find the nearest neighbors
-        of the given query vector. We currently support [vector search][search]
-        and [full-text search][experimental-full-text-search].
+        of the given query vector. We currently support [vector search](https://lancedb.com/docs/search/vector-search/)
+        and [full-text search](https://lancedb.com/docs/search/full-text-search/).
 
         All query options are defined in
         [LanceQueryBuilder][lancedb.query.LanceQueryBuilder].
@@ -1774,7 +1778,7 @@ class Table(ABC):
         for faster reads.
 
         Arguments are passed onto Lance's
-        [compact_files][lance.dataset.DatasetOptimizer.compact_files].
+        `lance.dataset.DatasetOptimizer.compact_files`.
         For most cases, the default should be fine.
 
         See Also
@@ -1828,6 +1832,8 @@ class Table(ABC):
         retrain: bool, default False
             This parameter is no longer used and is deprecated.
 
+        Notes
+        -----
         The frequency an application should call optimize is based on the frequency of
         data modifications.  If data is frequently added, deleted, or updated then
         optimize should be run frequently.  A good rule of thumb is to run optimize if
@@ -1982,15 +1988,14 @@ class Table(ABC):
         change permanent you can use the `[Self::restore]` method.
 
         Any operation that modifies the table will fail while the table is in a checked
-        out state.
+        out state. To return the table to a normal state use
+        `[Self::checkout_latest]`.
 
         Parameters
         ----------
         version: int | str,
             The version to check out. A version number (`int`) or a tag
             (`str`) can be provided.
-
-        To return the table to a normal state use `[Self::checkout_latest]`
         """
 
     @abstractmethod
@@ -3055,6 +3060,7 @@ class LanceTable(Table):
         lower_case: bool = True,
         stem: bool = True,
         remove_stop_words: bool = True,
+        custom_stop_words: Optional[List[str]] = None,
         ascii_folding: bool = True,
         ngram_min_length: int = 3,
         ngram_max_length: int = 3,
@@ -3101,6 +3107,7 @@ class LanceTable(Table):
                 "lower_case": lower_case,
                 "stem": stem,
                 "remove_stop_words": remove_stop_words,
+                "custom_stop_words": custom_stop_words,
                 "ascii_folding": ascii_folding,
                 "ngram_min_length": ngram_min_length,
                 "ngram_max_length": ngram_max_length,
@@ -3108,6 +3115,7 @@ class LanceTable(Table):
             }
         else:
             tokenizer_configs = self.infer_tokenizer_configs(tokenizer_name)
+            tokenizer_configs["custom_stop_words"] = custom_stop_words
 
         config = FTS(block_size=block_size, **tokenizer_configs)
 
@@ -3380,8 +3388,8 @@ class LanceTable(Table):
         fts_columns: Optional[Union[str, List[str]]] = None,
     ) -> LanceQueryBuilder:
         """Create a search query to find the nearest neighbors
-        of the given query vector. We currently support [vector search][search]
-        and [full-text search][search].
+        of the given query vector. We currently support [vector search](https://lancedb.com/docs/search/vector-search/)
+        and [full-text search](https://lancedb.com/docs/search/full-text-search/).
 
         Examples
         --------
@@ -3411,8 +3419,9 @@ class LanceTable(Table):
             - *default None*.
             Acceptable types are: list, np.ndarray, PIL.Image.Image
 
-            - If None then the select/[where][sql]/limit clauses are applied
-            to filter the table
+            - If None then the
+            select/[where][lancedb.query.LanceQueryBuilder.where]/limit clauses
+            are applied to filter the table
         vector_column_name: str, optional
             The name of the vector column to search.
 
@@ -3806,6 +3815,8 @@ class LanceTable(Table):
         retrain: bool, default False
             This parameter is no longer used and is deprecated.
 
+        Notes
+        -----
         The frequency an application should call optimize is based on the frequency of
         data modifications.  If data is frequently added, deleted, or updated then
         optimize should be run frequently.  A good rule of thumb is to run optimize if
@@ -4684,7 +4695,7 @@ class AsyncTable:
         Parameters
         ----------
         **kwargs
-            Forwarded to [`lance.dataset`][lance.dataset].
+            Forwarded to `lance.dataset`.
 
         Returns
         -------
@@ -5003,7 +5014,7 @@ class AsyncTable:
         progress: Optional[Union[bool, Callable, Any]] = None,
         write_parallelism: Optional[int] = None,
     ) -> AddResult:
-        """Add more data to the [Table](Table).
+        """Add more data to the [AsyncTable][lancedb.table.AsyncTable].
 
         Parameters
         ----------
@@ -5205,8 +5216,8 @@ class AsyncTable:
         fts_columns: Optional[Union[str, List[str]]] = None,
     ) -> Union[AsyncHybridQuery, AsyncFTSQuery, AsyncVectorQuery]:
         """Create a search query to find the nearest neighbors
-        of the given query vector. We currently support [vector search][search]
-        and [full-text search][experimental-full-text-search].
+        of the given query vector. We currently support [vector search](https://lancedb.com/docs/search/vector-search/)
+        and [full-text search](https://lancedb.com/docs/search/full-text-search/).
 
         All query options are defined in [AsyncQuery][lancedb.query.AsyncQuery].
 
@@ -5767,15 +5778,14 @@ class AsyncTable:
         change permanent you can use the `[Self::restore]` method.
 
         Any operation that modifies the table will fail while the table is in a checked
-        out state.
+        out state. To return the table to a normal state use
+        `[Self::checkout_latest]`.
 
         Parameters
         ----------
         version: int | str,
             The version to check out. A version number (`int`) or a tag
             (`str`) can be provided.
-
-        To return the table to a normal state use `[Self::checkout_latest]`
         """
         try:
             await self._inner.checkout(version)
@@ -5959,6 +5969,8 @@ class AsyncTable:
         retrain: bool, default False
             This parameter is no longer used and is deprecated.
 
+        Notes
+        -----
         The frequency an application should call optimize is based on the frequency of
         data modifications.  If data is frequently added, deleted, or updated then
         optimize should be run frequently.  A good rule of thumb is to run optimize if
@@ -6339,6 +6351,8 @@ class Branches:
         dry_run: bool, default False
             When True, only preview. When False, attempt the merge.
 
+        Notes
+        -----
         A rejected merge returns ``status="rejected"`` instead of raising.
         """
         return LOOP.run(self._table.branches.merge(from_branch, dry_run))

python/python/tests/test_fts.py

@@ -219,11 +219,13 @@ def test_create_inverted_index(table, with_position):
         table.create_fts_index(
             "text",
             with_position=with_position,
+            custom_stop_words=["puppy"],
             name="custom_fts_index",
         )
     indices = table.list_indices()
     fts_indices = [i for i in indices if i.index_type == "FTS"]
     assert any(i.name == "custom_fts_index" for i in fts_indices)
+    assert fts_indices[0].index_details["custom_stop_words"] == ["puppy"]
 
 
 @pytest.mark.parametrize("block_size", [128, 256])
@@ -243,6 +245,24 @@ def test_create_inverted_index_rejects_invalid_block_size(table):
         table.create_index("text", config=FTS(block_size=129))
 
 
+def test_custom_stop_words_list(table):
+    table.create_index(
+        "text",
+        config=FTS(stem=False, custom_stop_words=["lance"]),
+    )
+
+    assert table.list_indices()[0].index_details["custom_stop_words"] == ["lance"]
+    tokens = table.tokenize("the lance data", column="text")
+    assert [token.text for token in tokens] == ["the", "data"]
+    empty_tokens = ldb.tokenize("the lance data", stem=False, custom_stop_words=[])
+    assert [token.text for token in empty_tokens] == ["the", "lance", "data"]
+    with pytest.raises(TypeError, match=r"custom_stop_words.*int"):
+        ldb.tokenize(
+            "the lance data",
+            custom_stop_words=["lance", 42],
+        )
+
+
 def test_search_fts(table):
     table.create_fts_index("text")
     results = table.search("puppy").select(["id", "text"]).limit(5).to_list()

python/python/tests/test_remote_db.py

@@ -771,6 +771,7 @@ def test_table_create_indices():
                 "text",
                 wait_timeout=timedelta(seconds=2),
                 block_size=256,
+                custom_stop_words=["cloud"],
                 name="custom_fts_idx",
             )
 
@@ -795,6 +796,7 @@ def test_table_create_indices():
         assert "name" in fts_req
         assert fts_req["name"] == "custom_fts_idx"
         assert fts_req["block_size"] == 256
+        assert fts_req["custom_stop_words"] == ["cloud"]
 
         # Check vector index request has custom name
         vector_req = received_requests[2]

python/src/index.rs

@@ -59,7 +59,8 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                     .ascii_folding(params.ascii_folding)
                     .ngram_min_length(params.ngram_min_length)
                     .ngram_max_length(params.ngram_max_length)
-                    .ngram_prefix_only(params.prefix_only);
+                    .ngram_prefix_only(params.prefix_only)
+                    .custom_stop_words(params.custom_stop_words);
                 let inner_opts = inner_opts
                     .block_size(params.block_size)
                     .map_err(|err| PyValueError::new_err(err.to_string()))?;
@@ -206,6 +207,7 @@ struct FtsParams {
     lower_case: bool,
     stem: bool,
     remove_stop_words: bool,
+    custom_stop_words: Option<Vec<String>>,
     ascii_folding: bool,
     ngram_min_length: u32,
     ngram_max_length: u32,

python/src/table.rs

@@ -520,6 +520,7 @@ impl From<LanceDbFtsToken> for FtsToken {
     lower_case = true,
     stem = true,
     remove_stop_words = true,
+    custom_stop_words = None,
     ascii_folding = true,
     ngram_min_length = 3,
     ngram_max_length = 3,
@@ -534,6 +535,7 @@ pub fn tokenize(
     lower_case: bool,
     stem: bool,
     remove_stop_words: bool,
+    custom_stop_words: Option<Vec<String>>,
     ascii_folding: bool,
     ngram_min_length: u32,
     ngram_max_length: u32,
@@ -555,7 +557,8 @@ pub fn tokenize(
         .ascii_folding(ascii_folding)
         .ngram_min_length(ngram_min_length)
         .ngram_max_length(ngram_max_length)
-        .ngram_prefix_only(prefix_only);
+        .ngram_prefix_only(prefix_only)
+        .custom_stop_words(custom_stop_words);
     let tokens = lancedb_tokenize(&query, &params).infer_error()?;
     Ok(tokens.into_iter().map(FtsToken::from).collect())
 }

rust/lancedb/examples/full_text_search.rs

@@ -76,7 +76,12 @@ async fn create_table(db: &Connection) -> Result<Table> {
 
 async fn create_index(table: &Table) -> Result<()> {
     table
-        .create_index(&["doc"], Index::FTS(FtsIndexBuilder::default()))
+        .create_index(
+            &["doc"],
+            Index::FTS(
+                FtsIndexBuilder::default().custom_stop_words(Some(vec!["example".to_owned()])),
+            ),
+        )
         .execute()
         .await?;
     Ok(())

rust/lancedb/src/remote/table.rs

@@ -4553,6 +4553,19 @@ mod tests {
                 },
                 Index::FTS(InvertedIndexParams::default().block_size(256).unwrap()),
             ),
+            (
+                "FTS",
+                {
+                    let mut body = serde_json::to_value(InvertedIndexParams::default()).unwrap();
+                    body["custom_stop_words"] = json!(["cat", " cat ", "CAT"]);
+                    body
+                },
+                Index::FTS(InvertedIndexParams::default().custom_stop_words(Some(vec![
+                    "cat".to_string(),
+                    " cat ".to_string(),
+                    "CAT".to_string(),
+                ]))),
+            ),
         ];
 
         for (index_type, expected_body, index) in cases {
@@ -5084,8 +5097,9 @@ mod tests {
             "max_token_length": 40,
             "lower_case": true,
             "stem": false,
-            "remove_stop_words": false,
+            "remove_stop_words": true,
             "ascii_folding": true,
+            "custom_stop_words": ["hello"],
         })
         .to_string();
         let table = Table::new_with_handler("my_table", move |request| {
@@ -5123,10 +5137,6 @@ mod tests {
         assert_eq!(
             tokens,
             vec![
-                FtsToken {
-                    text: "hello".to_string(),
-                    position: 0,
-                },
                 FtsToken {
                     text: "こんにちは".to_string(),
                     position: 1,

rust/lancedb/src/table.rs

@@ -21,6 +21,7 @@ use lance::dataset::WriteMode;
 use lance::dataset::builder::DatasetBuilder;
 use lance::dataset::{InsertBuilder, WriteParams};
 use lance::index::DatasetIndexExt;
+use lance::index::scalar::load_segment_params;
 use lance::io::{ObjectStoreParams, WrappingObjectStore};
 use lance_datafusion::utils::StreamingWriteSource;
 use lance_index::IndexCriteria;
@@ -3193,10 +3194,9 @@ impl BaseTable for NativeTable {
     async fn list_indices(&self) -> Result<Vec<IndexConfig>> {
         let dataset = self.dataset.get().await?;
         let total_rows = dataset.count_rows(None).await? as u64;
-        let indices = dataset
-            .describe_indices(None)
-            .await?
-            .into_iter()
+        let descriptions = dataset.describe_indices(None).await?;
+        let mut indices: Vec<IndexConfig> = descriptions
+            .iter()
             .filter_map(|idx_desc| {
                 let index_type: crate::index::IndexType = idx_desc
                     .index_type()
@@ -3254,6 +3254,31 @@ impl BaseTable for NativeTable {
                 })
             })
             .collect();
+
+        for index in indices
+            .iter_mut()
+            .filter(|index| index.index_type == crate::index::IndexType::FTS)
+        {
+            let Some(description) = descriptions
+                .iter()
+                .find(|description| description.name() == index.name)
+            else {
+                continue;
+            };
+            let segments = description.segments();
+            let Some(segment) = segments.first() else {
+                continue;
+            };
+            let params = load_segment_params(&dataset, segment).await?;
+            let details = serde_json::to_string(&params).map_err(|source| Error::Other {
+                message: format!(
+                    "Failed to serialize full text search configuration for index '{}'",
+                    index.name
+                ),
+                source: Some(Box::new(source)),
+            })?;
+            index.index_details = Some(details);
+        }
         Ok(indices)
     }
 
@@ -4111,10 +4136,10 @@ mod tests {
         Box::new(RecordBatchIterator::new(vec![Ok(batch)], schema))
     }
 
-    // Windows does not support precise sleep durations due to timer resolution limitations.
-    #[cfg(not(target_os = "windows"))]
     #[tokio::test]
     async fn test_read_consistency_interval() {
+        use crate::utils::background_cache::clock;
+
         let intervals = vec![
             None,
             Some(0),
@@ -4141,6 +4166,12 @@ mod tests {
             let conn2 = conn2.execute().await.unwrap();
             let table2 = conn2.open_table("my_table").execute().await.unwrap();
 
+            // Freeze the consistency clock now that `table2` has seeded its cache, so the
+            // interval only elapses when this test advances it. Otherwise the write and
+            // count_rows calls below race the real 100ms interval, which a loaded CI
+            // runner loses. Must come after open_table: creating the cache clears the mock.
+            clock::pin();
+
             assert_eq!(table1.count_rows(None).await.unwrap(), 0);
             assert_eq!(table2.count_rows(None).await.unwrap(), 0);
 
@@ -4158,7 +4189,7 @@ mod tests {
                 }
                 Some(100) => {
                     assert_eq!(table2.count_rows(None).await.unwrap(), 0);
-                    tokio::time::sleep(Duration::from_millis(100)).await;
+                    clock::advance_by(Duration::from_millis(100));
                     assert_eq!(table2.count_rows(None).await.unwrap(), 1);
                 }
                 _ => unreachable!(),

rust/lancedb/src/table/create_index.rs

@@ -1366,11 +1366,19 @@ mod tests {
         table
             .create_index(
                 &["text"],
-                Index::FTS(FtsIndexBuilder::default().block_size(256).unwrap()),
+                Index::FTS(
+                    FtsIndexBuilder::default()
+                        .stem(false)
+                        .custom_stop_words(Some(vec!["cat".to_string()]))
+                        .block_size(256)
+                        .unwrap(),
+                ),
             )
             .execute()
             .await
             .unwrap();
+        drop(table);
+        let table = conn.open_table("test_bitmap").execute().await.unwrap();
         let index_configs = table.list_indices().await.unwrap();
         assert_eq!(index_configs.len(), 1);
         let index = index_configs.into_iter().next().unwrap();
@@ -1381,6 +1389,32 @@ mod tests {
         let index_params: FtsIndexBuilder =
             serde_json::from_str(index.index_details.as_deref().unwrap()).unwrap();
         assert_eq!(index_params.posting_block_size(), 256);
+        assert_eq!(
+            serde_json::to_value(&index_params).unwrap()["custom_stop_words"],
+            serde_json::json!(["cat"])
+        );
+        assert_eq!(
+            table
+                .tokenize("cat dog", "text_idx")
+                .await
+                .unwrap()
+                .into_iter()
+                .map(|token| token.text)
+                .collect::<Vec<_>>(),
+            vec!["dog"]
+        );
+
+        let batches = table
+            .query()
+            .full_text_search(FullTextSearchQuery::new("cat dog".to_string()))
+            .limit(120)
+            .execute()
+            .await
+            .unwrap()
+            .try_collect::<Vec<_>>()
+            .await
+            .unwrap();
+        assert_eq!(batches.iter().map(RecordBatch::num_rows).sum::<usize>(), 40);
 
         let num_rows = 120;
         let stats = table.index_stats("text_idx").await.unwrap().unwrap();