[python] Bump version: 0.6.4 → 0.6.5

changed the error msg for query with wrong vector dim thus need this
change to pass the nodejs tests.

.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.4.12
+current_version = 0.4.13
 commit = True
 message = Bump version: {current_version} → {new_version}
 tag = True

Cargo.toml

@@ -14,10 +14,10 @@ keywords = ["lancedb", "lance", "database", "vector", "search"]
 categories = ["database-implementations"]
 
 [workspace.dependencies]
-lance = { "version" = "=0.10.2", "features" = ["dynamodb"] }
-lance-index = { "version" = "=0.10.2" }
-lance-linalg = { "version" = "=0.10.2" }
-lance-testing = { "version" = "=0.10.2" }
+lance = { "version" = "=0.10.5", "features" = ["dynamodb"] }
+lance-index = { "version" = "=0.10.5" }
+lance-linalg = { "version" = "=0.10.5" }
+lance-testing = { "version" = "=0.10.5" }
 # Note that this one does not include pyarrow
 arrow = { version = "50.0", optional = false }
 arrow-array = "50.0"
@@ -28,13 +28,14 @@ arrow-schema = "50.0"
 arrow-arith = "50.0"
 arrow-cast = "50.0"
 async-trait = "0"
-chrono = "0.4.23"
+chrono = "0.4.35"
 half = { "version" = "=2.3.1", default-features = false, features = [
     "num-traits",
 ] }
 futures = "0"
 log = "0.4"
 object_store = "0.9.0"
+pin-project = "1.0.7"
 snafu = "0.7.4"
 url = "2"
 num-traits = "0.2"

docs/mkdocs.yml

@@ -27,7 +27,6 @@ theme:
     - content.tabs.link
     - content.action.edit
     - toc.follow
-    # - toc.integrate
     - navigation.top
     - navigation.tabs
     - navigation.tabs.sticky
@@ -64,7 +63,7 @@ plugins:
     add_image: True # Automatically add meta image
     add_keywords: True # Add page keywords in the header tag
     add_share_buttons: True # Add social share buttons
-    add_authors: False # Display page authors 
+    add_authors: False # Display page authors
     add_desc: False
     add_dates: False
 
@@ -140,12 +139,14 @@ nav:
       - Serverless Website Chatbot: examples/serverless_website_chatbot.md
       - YouTube Transcript Search: examples/youtube_transcript_bot_with_nodejs.md
       - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
+    - 🦀 Rust:
+      - Overview: examples/examples_rust.md
   - 🔧 CLI & Config: cli_config.md
   - 💭 FAQs: faq.md
   - ⚙️ API reference:
     - 🐍 Python: python/python.md
     - 👾 JavaScript: javascript/modules.md
-    - 🦀 Rust: https://docs.rs/vectordb/latest/vectordb/
+    - 🦀 Rust: https://docs.rs/lancedb/latest/lancedb/
   - ☁️ LanceDB Cloud:
     - Overview: cloud/index.md
     - API reference:
@@ -189,21 +190,21 @@ nav:
   - Pydantic: python/pydantic.md
   - Voxel51: integrations/voxel51.md
   - PromptTools: integrations/prompttools.md
-- Python examples:
+- Examples:
   - examples/index.md
   - YouTube Transcript Search: notebooks/youtube_transcript_search.ipynb
   - Documentation QA Bot using LangChain: notebooks/code_qa_bot.ipynb
   - Multimodal search using CLIP: notebooks/multimodal_search.ipynb
   - Serverless QA Bot with S3 and Lambda: examples/serverless_lancedb_with_s3_and_lambda.md
   - Serverless QA Bot with Modal: examples/serverless_qa_bot_with_modal_and_langchain.md
-- Javascript examples:
-  - Overview: examples/examples_js.md
-  - YouTube Transcript Search: examples/youtube_transcript_bot_with_nodejs.md
+  - YouTube Transcript Search (JS): examples/youtube_transcript_bot_with_nodejs.md
   - Serverless Chatbot from any website: examples/serverless_website_chatbot.md
   - TransformersJS Embedding Search: examples/transformerjs_embedding_search_nodejs.md
 - API reference:
+  - Overview: api_reference.md
   - Python: python/python.md
   - Javascript: javascript/modules.md
+  - Rust: https://docs.rs/lancedb/latest/lancedb/index.html
 - LanceDB Cloud:
     - Overview: cloud/index.md
     - API reference:

docs/src/ann_indexes.md

@@ -19,39 +19,61 @@ Lance supports `IVF_PQ` index type by default.
 
 === "Python"
 
-     Creating indexes is done via the [create_index](https://lancedb.github.io/lancedb/python/#lancedb.table.LanceTable.create_index) method.
+    Creating indexes is done via the [create_index](https://lancedb.github.io/lancedb/python/#lancedb.table.LanceTable.create_index) method.
 
-     ```python
-     import lancedb
-     import numpy as np
-     uri = "data/sample-lancedb"
-     db = lancedb.connect(uri)
+    ```python
+    import lancedb
+    import numpy as np
+    uri = "data/sample-lancedb"
+    db = lancedb.connect(uri)
 
-     # Create 10,000 sample vectors
-     data = [{"vector": row, "item": f"item {i}"}
+    # Create 10,000 sample vectors
+    data = [{"vector": row, "item": f"item {i}"}
         for i, row in enumerate(np.random.random((10_000, 1536)).astype('float32'))]
 
-     # Add the vectors to a table
-     tbl = db.create_table("my_vectors", data=data)
+    # Add the vectors to a table
+    tbl = db.create_table("my_vectors", data=data)
 
-     # Create and train the index - you need to have enough data in the table for an effective training step
-     tbl.create_index(num_partitions=256, num_sub_vectors=96)
-     ```
+    # Create and train the index - you need to have enough data in the table for an effective training step
+    tbl.create_index(num_partitions=256, num_sub_vectors=96)
+    ```
 
 === "Typescript"
 
-     ```typescript
-     --8<--- "docs/src/ann_indexes.ts:import"
+    ```typescript
+    --8<--- "docs/src/ann_indexes.ts:import"
 
-     --8<-- "docs/src/ann_indexes.ts:ingest"
-     ```
+    --8<-- "docs/src/ann_indexes.ts:ingest"
+    ```
 
-- **metric** (default: "L2"): The distance metric to use. By default it uses euclidean distance "`L2`".
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/ivf_pq.rs:create_index"
+    ```
+
+    IVF_PQ index parameters are more fully defined in the [crate docs](https://docs.rs/lancedb/latest/lancedb/index/vector/struct.IvfPqIndexBuilder.html).
+
+The following IVF_PQ paramters can be specified:
+
+- **distance_type**: The distance metric to use. By default it uses euclidean distance "`L2`".
   We also support "cosine" and "dot" distance as well.
-- **num_partitions** (default: 256): The number of partitions of the index.
-- **num_sub_vectors** (default: 96): The number of sub-vectors (M) that will be created during Product Quantization (PQ).
-  For D dimensional vector, it will be divided into `M` of `D/M` sub-vectors, each of which is presented by
-  a single PQ code.
+- **num_partitions**: The number of partitions in the index. The default is the square root
+  of the number of rows.
+
+!!! note
+
+    In the synchronous python SDK and node's `vectordb` the default is 256. This default has
+    changed in the asynchronous python SDK and node's `lancedb`.
+
+- **num_sub_vectors**: The number of sub-vectors (M) that will be created during Product Quantization (PQ).
+  For D dimensional vector, it will be divided into `M` subvectors with dimension `D/M`, each of which is replaced by
+  a single PQ code. The default is the dimension of the vector divided by 16.
+
+!!! note
+
+    In the synchronous python SDK and node's `vectordb` the default is currently 96. This default has
+    changed in the asynchronous python SDK and node's `lancedb`.
 
 <figure markdown>
   ![IVF PQ](./assets/ivf_pq.png)
@@ -114,25 +136,33 @@ There are a couple of parameters that can be used to fine-tune the search:
 
 === "Python"
 
-     ```python
-     tbl.search(np.random.random((1536))) \
-         .limit(2) \
-         .nprobes(20) \
-         .refine_factor(10) \
-         .to_pandas()
-     ```
+    ```python
+    tbl.search(np.random.random((1536))) \
+        .limit(2) \
+        .nprobes(20) \
+        .refine_factor(10) \
+        .to_pandas()
+    ```
 
-     ```text
+    ```text
                                               vector       item       _distance
-     0  [0.44949695, 0.8444449, 0.06281311, 0.23338133...  item 1141  103.575333
-     1  [0.48587373, 0.269207, 0.15095535, 0.65531915,...  item 3953  108.393867
-     ```
+    0  [0.44949695, 0.8444449, 0.06281311, 0.23338133...  item 1141  103.575333
+    1  [0.48587373, 0.269207, 0.15095535, 0.65531915,...  item 3953  108.393867
+    ```
 
 === "Typescript"
 
-     ```typescript
-     --8<-- "docs/src/ann_indexes.ts:search1"
-     ```
+    ```typescript
+    --8<-- "docs/src/ann_indexes.ts:search1"
+    ```
+
+=== "Rust"
+
+    ```rust
+    --8<-- "rust/lancedb/examples/ivf_pq.rs:search1"
+    ```
+
+    Vector search options are more fully defined in the [crate docs](https://docs.rs/lancedb/latest/lancedb/query/struct.Query.html#method.nearest_to).
 
 The search will return the data requested in addition to the distance of each item.
 
@@ -181,7 +211,7 @@ You can select the columns returned by the query using a select clause.
 ### Why do I need to manually create an index?
 
 Currently, LanceDB does _not_ automatically create the ANN index.
-LanceDB is well-optimized for kNN (exhaustive search) via a disk-based  index. For many use-cases,
+LanceDB is well-optimized for kNN (exhaustive search) via a disk-based index. For many use-cases,
 datasets of the order of ~100K vectors don't require index creation. If you can live with up to
 100ms latency, skipping index creation is a simpler workflow while guaranteeing 100% recall.
 

docs/src/api_reference.md

@@ -0,0 +1,7 @@
+# API Reference
+
+The API reference for the LanceDB client SDKs are available at the following locations:
+
+- [Python](python/python.md)
+- [JavaScript](javascript/modules.md)
+- [Rust](https://docs.rs/lancedb/latest/lancedb/index.html)

docs/src/basic.md

@@ -3,7 +3,7 @@
 !!! info "LanceDB can be run in a number of ways:"
 
     * Embedded within an existing backend (like your Django, Flask, Node.js or FastAPI application)
-    * Connected to directly from a client application like a Jupyter notebook for analytical workloads
+    * Directly from a client application like a Jupyter notebook for analytical workloads
     * Deployed as a remote serverless database
 
 ![](assets/lancedb_embedded_explanation.png)
@@ -24,13 +24,11 @@
 
 === "Rust"
 
-    !!! warning "Rust SDK is experimental, might introduce breaking changes in the near future"
-
     ```shell
-    cargo add vectordb
+    cargo add lancedb
     ```
 
-    !!! info "To use the vectordb create, you first need to install protobuf."
+    !!! info "To use the lancedb create, you first need to install protobuf."
 
     === "macOS"
 
@@ -44,7 +42,7 @@
         sudo apt install -y protobuf-compiler libssl-dev
         ```
 
-    !!! info "Please also make sure you're using the same version of Arrow as in the [vectordb crate](https://github.com/lancedb/lancedb/blob/main/Cargo.toml)"
+    !!! info "Please also make sure you're using the same version of Arrow as in the [lancedb crate](https://github.com/lancedb/lancedb/blob/main/Cargo.toml)"
 
 ## Connect to a database
 
@@ -81,10 +79,11 @@ If you need a reminder of the uri, you can call `db.uri()`.
 
 ## Create a table
 
-### Directly insert data to a new table
+### Create a table from initial data
 
-If you have data to insert into the table at creation time, you can simultaneously create a 
-table and insert the data to it.
+If you have data to insert into the table at creation time, you can simultaneously create a
+table and insert the data into it.  The schema of the data will be used as the schema of the
+table.
 
 === "Python"
 
@@ -120,21 +119,27 @@ table and insert the data to it.
 === "Rust"
 
     ```rust
-    use arrow_schema::{DataType, Schema, Field};
-    use arrow_array::{RecordBatch, RecordBatchIterator};
-
     --8<-- "rust/lancedb/examples/simple.rs:create_table"
     ```
 
-    If the table already exists, LanceDB will raise an error by default.
+    If the table already exists, LanceDB will raise an error by default.  See
+    [the mode option](https://docs.rs/lancedb/latest/lancedb/connection/struct.CreateTableBuilder.html#method.mode)
+    for details on how to overwrite (or open) existing tables instead.
 
-!!! info "Under the hood, LanceDB converts the input data into an Apache Arrow table and persists it to disk using the [Lance format](https://www.github.com/lancedb/lance)."
+    !!! Providing table records in Rust
+
+        The Rust SDK currently expects data to be provided as an Arrow
+        [RecordBatchReader](https://docs.rs/arrow-array/latest/arrow_array/trait.RecordBatchReader.html)
+        Support for additional formats (such as serde or polars) is on the roadmap.
+
+!!! info "Under the hood, LanceDB reads in the Apache Arrow data and persists it to disk using the [Lance format](https://www.github.com/lancedb/lance)."
 
 ### Create an empty table
 
 Sometimes you may not have the data to insert into the table at creation time.
 In this case, you can create an empty table and specify the schema, so that you can add
-data to the table at a later time (such that it conforms to the schema).
+data to the table at a later time (as long as it conforms to the schema).  This is
+similar to a `CREATE TABLE` statement in SQL.
 
 === "Python"
 
@@ -175,7 +180,7 @@ Once created, you can open a table as follows:
 === "Rust"
 
     ```rust
-    --8<-- "rust/lancedb/examples/simple.rs:open_with_existing_file"
+    --8<-- "rust/lancedb/examples/simple.rs:open_existing_tbl"
     ```
 
 If you forget the name of your table, you can always get a listing of all table names:
@@ -254,6 +259,14 @@ Once you've embedded the query, you can find its nearest neighbors as follows:
     --8<-- "rust/lancedb/examples/simple.rs:search"
     ```
 
+    !!! Query vectors in Rust
+        Rust does not yet support automatic execution of embedding functions.  You will need to
+        calculate embeddings yourself.  Support for this is on the roadmap and can be tracked at
+        https://github.com/lancedb/lancedb/issues/994
+
+        Query vectors can be provided as Arrow arrays or a Vec/slice of Rust floats.
+        Support for additional formats (e.g. `polars::series::Series`) is on the roadmap.
+
 By default, LanceDB runs a brute-force scan over dataset to find the K nearest neighbours (KNN).
 For tables with more than 50K vectors, creating an ANN index is recommended to speed up search performance.
 LanceDB allows you to create an ANN index on a table as follows:
@@ -277,7 +290,7 @@ LanceDB allows you to create an ANN index on a table as follows:
     ```
 
 !!! note "Why do I need to create an index manually?"
-    LanceDB does not automatically create the ANN index, for two reasons. The first is that it's optimized
+    LanceDB does not automatically create the ANN index for two reasons. The first is that it's optimized
     for really fast retrievals via a disk-based index, and the second is that data and query workloads can
     be very diverse, so there's no one-size-fits-all index configuration. LanceDB provides many parameters
     to fine-tune index size, query latency and accuracy. See the section on
@@ -308,8 +321,9 @@ This can delete any number of rows that match the filter.
     ```
 
 The deletion predicate is a SQL expression that supports the same expressions
-as the `where()` clause on a search. They can be as simple or complex as needed.
-To see what expressions are supported, see the [SQL filters](sql.md) section.
+as the `where()` clause (`only_if()` in Rust) on a search. They can be as
+simple or complex as needed.  To see what expressions are supported, see the
+[SQL filters](sql.md) section.
 
 === "Python"
 
@@ -319,6 +333,10 @@ To see what expressions are supported, see the [SQL filters](sql.md) section.
 
       Read more: [vectordb.Table.delete](javascript/interfaces/Table.md#delete)
 
+=== "Rust"
+
+      Read more: [lancedb::Table::delete](https://docs.rs/lancedb/latest/lancedb/table/struct.Table.html#method.delete)
+
 ## Drop a table
 
 Use the `drop_table()` method on the database to remove a table.

docs/src/concepts/index_ivfpq.md

@@ -31,7 +31,7 @@ As an example, consider starting with 128-dimensional vector consisting of 32-bi
 
 While PQ helps with reducing the size of the index, IVF primarily addresses search performance. The primary purpose of an inverted file index is to facilitate rapid and effective nearest neighbor search by narrowing down the search space.
 
-In IVF, the PQ vector space is divided into *Voronoi cells*, which are essentially partitions that consist of all the points in the space that are within a threshold distance of the given region's seed point. These seed points are used to create an inverted index that correlates each centroid with a list of vectors in the space, allowing a search to be restricted to just a subset of vectors in the index.
+In IVF, the PQ vector space is divided into *Voronoi cells*, which are essentially partitions that consist of all the points in the space that are within a threshold distance of the given region's seed point. These seed points are initialized by running K-means over the stored vectors. The centroids of K-means turn into the seed points which then each define a region. These regions are then are used to create an inverted index that correlates each centroid with a list of vectors in the space, allowing a search to be restricted to just a subset of vectors in the index.
 
 ![](../assets/ivfpq_ivf_desc.webp)
 

docs/src/embeddings/default_embedding_functions.md

@@ -224,7 +224,6 @@ This embedding function supports ingesting images as both bytes and urls. You ca
 !!! info
     LanceDB supports ingesting images directly from accessible links.
 
-
 ```python
 
 db = lancedb.connect(tmp_path)
@@ -290,4 +289,67 @@ print(actual.label)
 
 ```
 
+### Imagebind embeddings
+We have support for [imagebind](https://github.com/facebookresearch/ImageBind) model embeddings. You can download our version of the packaged model via - `pip install imagebind-packaged==0.1.2`.
+
+This function is registered as `imagebind` and supports Audio, Video and Text modalities(extending to Thermal,Depth,IMU data):
+
+| Parameter | Type | Default Value | Description |
+|---|---|---|---|
+| `name` | `str` | `"imagebind_huge"` | Name of the model. |
+| `device` | `str` | `"cpu"` | The device to run the model on. Can be `"cpu"` or `"gpu"`. |
+| `normalize` | `bool` | `False` | set to `True` to normalize your inputs before model ingestion. |
+
+Below is an example demonstrating how the API works:
+
+```python
+db = lancedb.connect(tmp_path)
+registry = EmbeddingFunctionRegistry.get_instance()
+func = registry.get("imagebind").create()
+
+class ImageBindModel(LanceModel):
+    text: str
+    image_uri: str = func.SourceField()
+    audio_path: str
+    vector: Vector(func.ndims()) = func.VectorField()
+
+# add locally accessible image paths
+text_list=["A dog.", "A car", "A bird"]
+image_paths=[".assets/dog_image.jpg", ".assets/car_image.jpg", ".assets/bird_image.jpg"]
+audio_paths=[".assets/dog_audio.wav", ".assets/car_audio.wav", ".assets/bird_audio.wav"]
+
+# Load data
+inputs = [
+    {"text": a, "audio_path": b, "image_uri": c}
+    for a, b, c in zip(text_list, audio_paths, image_paths)
+]
+
+#create table and add data
+table = db.create_table("img_bind", schema=ImageBindModel)
+table.add(inputs)
+```
+
+Now, we can search using any modality:
+
+#### image search
+```python
+query_image = "./assets/dog_image2.jpg" #download an image and enter that path here
+actual = table.search(query_image).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "dog")
+```
+#### audio search
+
+```python
+query_audio = "./assets/car_audio2.wav" #download an audio clip and enter path here
+actual = table.search(query_audio).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "car")
+```
+#### Text search
+You can add any input query and fetch the result as follows:
+```python
+query = "an animal which flies and tweets" 
+actual = table.search(query).limit(1).to_pydantic(ImageBindModel)[0]
+print(actual.text == "bird")
+```
+
 If you have any questions about the embeddings API, supported models, or see a relevant model missing, please raise an issue [on GitHub](https://github.com/lancedb/lancedb/issues).

docs/src/examples/examples_rust.md

@@ -0,0 +1,3 @@
+# Examples: Rust
+
+Our Rust SDK is now stable. Examples are coming soon.

docs/src/examples/index.md

@@ -2,10 +2,11 @@
 
 ## Recipes and example code
 
-LanceDB provides language APIs, allowing you to embed a database in your language of choice. We currently provide Python and Javascript APIs, with the Rust API and examples actively being worked on and will be available soon.
+LanceDB provides language APIs, allowing you to embed a database in your language of choice.
 
 * 🐍 [Python](examples_python.md) examples
-* 👾 [JavaScript](exampled_js.md) examples
+* 👾 [JavaScript](examples_js.md) examples
+* 🦀 Rust examples (coming soon)
 
 ## Applications powered by LanceDB
 

docs/src/faq.md

@@ -16,7 +16,7 @@ As we mention in our talk titled “[Lance, a modern columnar data format](https
 
 ### Why build in Rust? 🦀
 
-We believe that the Rust ecosystem has attained mainstream maturity and that Rust will form the underpinnings of large parts of the data and ML landscape in a few years. Performance, latency and reliability are paramount to a vector DB, and building in Rust allows us to iterate and release updates more rapidly due to Rust’s safety guarantees. Both Lance (the data format) and LanceDB (the database) are written entirely in Rust. We also provide Python and JavaScript client libraries to interact with the database. Our Rust API is a little rough around the edges right now, but is fast becoming on par with the Python and JS APIs.
+We believe that the Rust ecosystem has attained mainstream maturity and that Rust will form the underpinnings of large parts of the data and ML landscape in a few years. Performance, latency and reliability are paramount to a vector DB, and building in Rust allows us to iterate and release updates more rapidly due to Rust’s safety guarantees. Both Lance (the data format) and LanceDB (the database) are written entirely in Rust. We also provide Python, JavaScript, and Rust client libraries to interact with the database.
 
 ### What is the difference between LanceDB OSS and LanceDB Cloud?
 
@@ -44,7 +44,7 @@ For large-scale (>1M) or higher dimension vectors, it is beneficial to create an
 
 ### Does LanceDB support full-text search?
 
-Yes, LanceDB supports full-text search (FTS) via [Tantivy](https://github.com/quickwit-oss/tantivy). Our current FTS integration is Python-only, and our goal is to push it down to the Rust level in future versions to enable much more powerful search capabilities available to our Python, JavaScript and Rust clients.
+Yes, LanceDB supports full-text search (FTS) via [Tantivy](https://github.com/quickwit-oss/tantivy). Our current FTS integration is Python-only, and our goal is to push it down to the Rust level in future versions to enable much more powerful search capabilities available to our Python, JavaScript and Rust clients.  Follow along in the [Github issue](https://github.com/lancedb/lance/issues/1195)
 
 ### How can I speed up data inserts?
 

docs/src/fts.md

@@ -1,6 +1,6 @@
 # Full-text search
 
-LanceDB provides support for full-text search via [Tantivy](https://github.com/quickwit-oss/tantivy) (currently Python only), allowing you to incorporate keyword-based search (based on BM25) in your retrieval solutions. Our goal is to push the FTS integration down to the Rust level in the future, so that it's available for JavaScript users as well.
+LanceDB provides support for full-text search via [Tantivy](https://github.com/quickwit-oss/tantivy) (currently Python only), allowing you to incorporate keyword-based search (based on BM25) in your retrieval solutions. Our goal is to push the FTS integration down to the Rust level in the future, so that it's available for Rust and JavaScript users as well.  Follow along at [this Github issue](https://github.com/lancedb/lance/issues/1195)
 
 A hybrid search solution combining vector and full-text search is also on the way.
 
@@ -75,9 +75,39 @@ applied on top of the full text search results. This can be invoked via the fami
 table.search("puppy").limit(10).where("meta='foo'").to_list()
 ```
 
+## Sorting
+
+You can pre-sort the documents by specifying `ordering_field_names` when
+creating the full-text search index. Once pre-sorted, you can then specify
+`ordering_field_name` while searching to return results sorted by the given
+field. For example, 
+
+```
+table.create_fts_index(["text_field"], ordering_field_names=["sort_by_field"])
+
+(table.search("terms", ordering_field_name="sort_by_field")
+ .limit(20)
+ .to_list())
+```
+
+!!! note
+    If you wish to specify an ordering field at query time, you must also
+    have specified it during indexing time. Otherwise at query time, an
+    error will be raised that looks like `ValueError: The field does not exist: xxx`
+
+!!! note
+    The fields to sort on must be of typed unsigned integer, or else you will see 
+    an error during indexing that looks like 
+    `TypeError: argument 'value': 'float' object cannot be interpreted as an integer`.
+
+!!! note
+    You can specify multiple fields for ordering at indexing time.
+    But at query time only one ordering field is supported.
+
+
 ## Phrase queries vs. terms queries
 
-For full-text search you can specify either a **phrase** query like `"the old man and the sea"`, 
+For full-text search you can specify either a **phrase** query like `"the old man and the sea"`,
 or a **terms** search query like `"(Old AND Man) AND Sea"`. For more details on the terms
 query syntax, see Tantivy's [query parser rules](https://docs.rs/tantivy/latest/tantivy/query/struct.QueryParser.html).
 
@@ -112,7 +142,7 @@ double quotes replaced by single quotes.
 
 ## Configurations
 
-By default, LanceDB configures a 1GB heap size limit for creating the index. You can 
+By default, LanceDB configures a 1GB heap size limit for creating the index. You can
 reduce this if running on a smaller node, or increase this for faster performance while
 indexing a larger corpus.
 
@@ -128,7 +158,6 @@ table.create_fts_index(["text1", "text2"], writer_heap_size=heap, replace=True)
    If you add data after FTS index creation, it won't be reflected
    in search results until you do a full reindex.
 
-2. We currently only support local filesystem paths for the FTS index. 
+2. We currently only support local filesystem paths for the FTS index.
    This is a tantivy limitation. We've implemented an object store plugin
    but there's no way in tantivy-py to specify to use it.
-

docs/src/guides/tables.md

@@ -168,151 +168,151 @@ This guide will show how to create tables, insert data into them, and update the
     --8<-- "docs/src/basic_legacy.ts:create_f16_table"
     ```
 
-    ### From Pydantic Models
+### From Pydantic Models
 
-    When you create an empty table without data, you must specify the table schema.
-    LanceDB supports creating tables by specifying a PyArrow schema or a specialized
-    Pydantic model called `LanceModel`.
+When you create an empty table without data, you must specify the table schema.
+LanceDB supports creating tables by specifying a PyArrow schema or a specialized
+Pydantic model called `LanceModel`.
 
-    For example, the following Content model specifies a table with 5 columns:
-    `movie_id`, `vector`, `genres`, `title`, and `imdb_id`. When you create a table, you can
-    pass the class as the value of the `schema` parameter to `create_table`.
-    The `vector` column is a `Vector` type, which is a specialized Pydantic type that
-    can be configured with the vector dimensions. It is also important to note that
-    LanceDB only understands subclasses of `lancedb.pydantic.LanceModel`
-    (which itself derives from `pydantic.BaseModel`).
+For example, the following Content model specifies a table with 5 columns:
+`movie_id`, `vector`, `genres`, `title`, and `imdb_id`. When you create a table, you can
+pass the class as the value of the `schema` parameter to `create_table`.
+The `vector` column is a `Vector` type, which is a specialized Pydantic type that
+can be configured with the vector dimensions. It is also important to note that
+LanceDB only understands subclasses of `lancedb.pydantic.LanceModel`
+(which itself derives from `pydantic.BaseModel`).
 
-    ```python
-    from lancedb.pydantic import Vector, LanceModel
+```python
+from lancedb.pydantic import Vector, LanceModel
 
-    class Content(LanceModel):
-        movie_id: int
-        vector: Vector(128)
-        genres: str
-        title: str
-        imdb_id: int
+class Content(LanceModel):
+    movie_id: int
+    vector: Vector(128)
+    genres: str
+    title: str
+    imdb_id: int
 
-        @property
-        def imdb_url(self) -> str:
-            return f"https://www.imdb.com/title/tt{self.imdb_id}"
+    @property
+    def imdb_url(self) -> str:
+        return f"https://www.imdb.com/title/tt{self.imdb_id}"
 
-    import pyarrow as pa
-    db = lancedb.connect("~/.lancedb")
-    table_name = "movielens_small"
-    table = db.create_table(table_name, schema=Content)
-    ```
+import pyarrow as pa
+db = lancedb.connect("~/.lancedb")
+table_name = "movielens_small"
+table = db.create_table(table_name, schema=Content)
+```
 
-    #### Nested schemas
+#### Nested schemas
 
-    Sometimes your data model may contain nested objects.
-    For example, you may want to store the document string
-    and the document soure name as a nested Document object:
+Sometimes your data model may contain nested objects.
+For example, you may want to store the document string
+and the document soure name as a nested Document object:
 
-    ```python
-    class Document(BaseModel):
-        content: str
-        source: str
-    ```
+```python
+class Document(BaseModel):
+    content: str
+    source: str
+```
 
-    This can be used as the type of a LanceDB table column:
+This can be used as the type of a LanceDB table column:
 
-    ```python
-    class NestedSchema(LanceModel):
-        id: str
-        vector: Vector(1536)
-        document: Document
+```python
+class NestedSchema(LanceModel):
+    id: str
+    vector: Vector(1536)
+    document: Document
 
-    tbl = db.create_table("nested_table", schema=NestedSchema, mode="overwrite")
-    ```
+tbl = db.create_table("nested_table", schema=NestedSchema, mode="overwrite")
+```
 
-    This creates a struct column called "document" that has two subfields 
-    called "content" and "source":
+This creates a struct column called "document" that has two subfields 
+called "content" and "source":
 
-    ```
-    In [28]: tbl.schema
-    Out[28]:
-    id: string not null
-    vector: fixed_size_list<item: float>[1536] not null
-        child 0, item: float
-    document: struct<content: string not null, source: string not null> not null
-        child 0, content: string not null
-        child 1, source: string not null    
-    ```
+```
+In [28]: tbl.schema
+Out[28]:
+id: string not null
+vector: fixed_size_list<item: float>[1536] not null
+    child 0, item: float
+document: struct<content: string not null, source: string not null> not null
+    child 0, content: string not null
+    child 1, source: string not null    
+```
 
-    #### Validators
+#### Validators
 
-    Note that neither Pydantic nor PyArrow automatically validates that input data
-    is of the correct timezone, but this is easy to add as a custom field validator:
+Note that neither Pydantic nor PyArrow automatically validates that input data
+is of the correct timezone, but this is easy to add as a custom field validator:
 
-    ```python
-    from datetime import datetime
-    from zoneinfo import ZoneInfo
+```python
+from datetime import datetime
+from zoneinfo import ZoneInfo
 
-    from lancedb.pydantic import LanceModel
-    from pydantic import Field, field_validator, ValidationError, ValidationInfo
+from lancedb.pydantic import LanceModel
+from pydantic import Field, field_validator, ValidationError, ValidationInfo
 
-    tzname = "America/New_York"
-    tz = ZoneInfo(tzname)
+tzname = "America/New_York"
+tz = ZoneInfo(tzname)
 
-    class TestModel(LanceModel):
-        dt_with_tz: datetime = Field(json_schema_extra={"tz": tzname})
+class TestModel(LanceModel):
+    dt_with_tz: datetime = Field(json_schema_extra={"tz": tzname})
 
-        @field_validator('dt_with_tz')
-        @classmethod
-        def tz_must_match(cls, dt: datetime) -> datetime:
-            assert dt.tzinfo == tz
-            return dt        
+    @field_validator('dt_with_tz')
+    @classmethod
+    def tz_must_match(cls, dt: datetime) -> datetime:
+        assert dt.tzinfo == tz
+        return dt        
 
-    ok = TestModel(dt_with_tz=datetime.now(tz))
+ok = TestModel(dt_with_tz=datetime.now(tz))
 
-    try:
-        TestModel(dt_with_tz=datetime.now(ZoneInfo("Asia/Shanghai")))
-        assert 0 == 1, "this should raise ValidationError"
-    except ValidationError:
-        print("A ValidationError was raised.")
-        pass
-    ```
+try:
+    TestModel(dt_with_tz=datetime.now(ZoneInfo("Asia/Shanghai")))
+    assert 0 == 1, "this should raise ValidationError"
+except ValidationError:
+    print("A ValidationError was raised.")
+    pass
+```
 
-    When you run this code it should print "A ValidationError was raised."
+When you run this code it should print "A ValidationError was raised."
 
-    #### Pydantic custom types
+#### Pydantic custom types
 
-    LanceDB does NOT yet support converting pydantic custom types. If this is something you need,
-    please file a feature request on the [LanceDB Github repo](https://github.com/lancedb/lancedb/issues/new).
+LanceDB does NOT yet support converting pydantic custom types. If this is something you need,
+please file a feature request on the [LanceDB Github repo](https://github.com/lancedb/lancedb/issues/new).
 
-    ### Using Iterators / Writing Large Datasets
+### Using Iterators / Writing Large Datasets
 
-    It is recommended to use iterators to add large datasets in batches when creating your table in one go. This does not create multiple versions of your dataset unlike manually adding batches using `table.add()`
+It is recommended to use iterators to add large datasets in batches when creating your table in one go. This does not create multiple versions of your dataset unlike manually adding batches using `table.add()`
 
-    LanceDB additionally supports PyArrow's `RecordBatch` Iterators or other generators producing supported data types.
+LanceDB additionally supports PyArrow's `RecordBatch` Iterators or other generators producing supported data types.
 
-    Here's an example using using `RecordBatch` iterator for creating tables.
+Here's an example using using `RecordBatch` iterator for creating tables.
 
-    ```python
-    import pyarrow as pa
+```python
+import pyarrow as pa
 
-    def make_batches():
-        for i in range(5):
-            yield pa.RecordBatch.from_arrays(
-                [
-                    pa.array([[3.1, 4.1, 5.1, 6.1], [5.9, 26.5, 4.7, 32.8]],
-                            pa.list_(pa.float32(), 4)),
-                    pa.array(["foo", "bar"]),
-                    pa.array([10.0, 20.0]),
-                ],
-                ["vector", "item", "price"],
-            )
+def make_batches():
+    for i in range(5):
+        yield pa.RecordBatch.from_arrays(
+            [
+                pa.array([[3.1, 4.1, 5.1, 6.1], [5.9, 26.5, 4.7, 32.8]],
+                        pa.list_(pa.float32(), 4)),
+                pa.array(["foo", "bar"]),
+                pa.array([10.0, 20.0]),
+            ],
+            ["vector", "item", "price"],
+        )
 
-    schema = pa.schema([
-        pa.field("vector", pa.list_(pa.float32(), 4)),
-        pa.field("item", pa.utf8()),
-        pa.field("price", pa.float32()),
-    ])
+schema = pa.schema([
+    pa.field("vector", pa.list_(pa.float32(), 4)),
+    pa.field("item", pa.utf8()),
+    pa.field("price", pa.float32()),
+])
 
-    db.create_table("batched_tale", make_batches(), schema=schema)
-    ```
+db.create_table("batched_tale", make_batches(), schema=schema)
+```
 
-    You can also use iterators of other types like Pandas DataFrame or Pylists directly in the above example.
+You can also use iterators of other types like Pandas DataFrame or Pylists directly in the above example.
 
 ## Open existing tables
 

docs/src/index.md

@@ -28,7 +28,7 @@ LanceDB **Cloud** is a SaaS (software-as-a-service) solution that runs serverles
 
 * Fast production-scale vector similarity, full-text & hybrid search and a SQL query interface (via [DataFusion](https://github.com/apache/arrow-datafusion))
 
-* Native Python and Javascript/Typescript support
+* Python, Javascript/Typescript, and Rust support
 
 * Store, query & manage multi-modal data (text, images, videos, point clouds, etc.), not just the embeddings and metadata
 
@@ -54,3 +54,4 @@ The following pages go deeper into the internal of LanceDB and how to use it.
 * [Ecosystem Integrations](integrations/index.md): Integrate LanceDB with other tools in the data ecosystem
 * [Python API Reference](python/python.md): Python OSS and Cloud API references
 * [JavaScript API Reference](javascript/modules.md): JavaScript OSS and Cloud API references
+* [Rust API Reference](https://docs.rs/lancedb/latest/lancedb/index.html): Rust API reference

docs/src/search.md

@@ -22,7 +22,7 @@ Currently, LanceDB supports the following metrics:
 ## Exhaustive search (kNN)
 
 If you do not create a vector index, LanceDB exhaustively scans the _entire_ vector space
-and compute the distance to every vector in order to find the exact nearest neighbors. This is effectively a kNN search.
+and computes the distance to every vector in order to find the exact nearest neighbors. This is effectively a kNN search.
 
 <!-- Setup Code
 ```python
@@ -85,7 +85,7 @@ To perform scalable vector retrieval with acceptable latencies, it's common to b
 While the exhaustive search is guaranteed to always return 100% recall, the approximate nature of
 an ANN search means that using an index often involves a trade-off between recall and latency.
 
-See the [IVF_PQ index](./concepts/index_ivfpq.md.md) for a deeper description of how `IVF_PQ`
+See the [IVF_PQ index](./concepts/index_ivfpq.md) for a deeper description of how `IVF_PQ`
 indexes work in LanceDB.
 
 ## Output search results
@@ -184,4 +184,3 @@ Let's create a LanceDB table with a nested schema:
 
     Note that in this case the extra `_distance` field is discarded since
     it's not part of the LanceSchema.
-

node/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "vectordb",
-  "version": "0.4.12",
+  "version": "0.4.13",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "vectordb",
-      "version": "0.4.12",
+      "version": "0.4.13",
       "cpu": [
         "x64",
         "arm64"
@@ -52,11 +52,11 @@
         "uuid": "^9.0.0"
       },
       "optionalDependencies": {
-        "@lancedb/vectordb-darwin-arm64": "0.4.12",
-        "@lancedb/vectordb-darwin-x64": "0.4.12",
-        "@lancedb/vectordb-linux-arm64-gnu": "0.4.12",
-        "@lancedb/vectordb-linux-x64-gnu": "0.4.12",
-        "@lancedb/vectordb-win32-x64-msvc": "0.4.12"
+        "@lancedb/vectordb-darwin-arm64": "0.4.13",
+        "@lancedb/vectordb-darwin-x64": "0.4.13",
+        "@lancedb/vectordb-linux-arm64-gnu": "0.4.13",
+        "@lancedb/vectordb-linux-x64-gnu": "0.4.13",
+        "@lancedb/vectordb-win32-x64-msvc": "0.4.13"
       },
       "peerDependencies": {
         "@apache-arrow/ts": "^14.0.2",
@@ -334,9 +334,9 @@
       }
     },
     "node_modules/@lancedb/vectordb-darwin-arm64": {
-      "version": "0.4.12",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.12.tgz",
-      "integrity": "sha512-38/rkJRlWXkPWXuj9onzvbrhnIWcIUQjgEp5G9v5ixPosBowm7A4j8e2Q8CJMsVSNcVX2JLqwWVldiWegZFuYw==",
+      "version": "0.4.13",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-arm64/-/vectordb-darwin-arm64-0.4.13.tgz",
+      "integrity": "sha512-JfroNCG8yKIU931Y+x8d0Fp8C9DHUSC5j+CjI+e5err7rTWtie4j3JbsXlWAnPFaFEOg0Xk3BWkSikCvhPGJGg==",
       "cpu": [
         "arm64"
       ],
@@ -346,9 +346,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-darwin-x64": {
-      "version": "0.4.12",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.12.tgz",
-      "integrity": "sha512-psE48dztyO450hXWdv9Rl9aayM2HQ1uF9wErfC0gKmDUh1N0NdVq2viDuFpZxnmCis/nvGwKlYiYT9OnYNCJ9g==",
+      "version": "0.4.13",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-darwin-x64/-/vectordb-darwin-x64-0.4.13.tgz",
+      "integrity": "sha512-dG6IMvfpHpnHdbJ0UffzJ7cZfMiC02MjIi6YJzgx+hKz2UNXWNBIfTvvhqli85mZsGRXL1OYDdYv0K1YzNjXlA==",
       "cpu": [
         "x64"
       ],
@@ -358,9 +358,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-arm64-gnu": {
-      "version": "0.4.12",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.12.tgz",
-      "integrity": "sha512-xwkgF6MiF5aAdG9JG8v4ke652YxUJrhs9z4OrsEfrENnvsIQd2C5UyKMepVLdvij4BI/XPFRFWXdjPvP7S9rTA==",
+      "version": "0.4.13",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-arm64-gnu/-/vectordb-linux-arm64-gnu-0.4.13.tgz",
+      "integrity": "sha512-BRR1VzaMviXby7qmLm0axNZM8eUZF3ZqfvnDKdVRpC3LaRueD6pMXHuC2IUKaFkn7xktf+8BlDZb6foFNEj8bQ==",
       "cpu": [
         "arm64"
       ],
@@ -370,9 +370,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-linux-x64-gnu": {
-      "version": "0.4.12",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.12.tgz",
-      "integrity": "sha512-gJqYR0aymrS+C60xc4EQPzmQ5/69XfeFv2ofBvAj7qW+c6BcnoAcfVl+7s1IrcWeGz251sm5cD5Lx4AzJd89dA==",
+      "version": "0.4.13",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-linux-x64-gnu/-/vectordb-linux-x64-gnu-0.4.13.tgz",
+      "integrity": "sha512-WnekZ7ZMlria+NODZ6aBCljCFQSe2bBNUS9ZpyFl/Y1vHduSQPuBxM6V7vp2QubC0daq/rifgjDob89DF+x3xw==",
       "cpu": [
         "x64"
       ],
@@ -382,9 +382,9 @@
       ]
     },
     "node_modules/@lancedb/vectordb-win32-x64-msvc": {
-      "version": "0.4.12",
-      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.12.tgz",
-      "integrity": "sha512-LhCzpyEeBUyO6L2fuVqeP3mW8kYDryyU9PNqcM01m88sZB1Do6AlwiM+GjPRQ0SpzD0LK9oxQqSmJrdcNGqjbw==",
+      "version": "0.4.13",
+      "resolved": "https://registry.npmjs.org/@lancedb/vectordb-win32-x64-msvc/-/vectordb-win32-x64-msvc-0.4.13.tgz",
+      "integrity": "sha512-3NDpMWBL2ksDHXAraXhowiLqQcNWM5bdbeHwze4+InYMD54hyQ2ODNc+4usxp63Nya9biVnFS27yXULqkzIEqQ==",
       "cpu": [
         "x64"
       ],

node/package.json

@@ -1,6 +1,6 @@
 {
   "name": "vectordb",
-  "version": "0.4.12",
+  "version": "0.4.13",
   "description": " Serverless, low-latency vector database for AI applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -88,10 +88,10 @@
     }
   },
   "optionalDependencies": {
-    "@lancedb/vectordb-darwin-arm64": "0.4.12",
-    "@lancedb/vectordb-darwin-x64": "0.4.12",
-    "@lancedb/vectordb-linux-arm64-gnu": "0.4.12",
-    "@lancedb/vectordb-linux-x64-gnu": "0.4.12",
-    "@lancedb/vectordb-win32-x64-msvc": "0.4.12"
+    "@lancedb/vectordb-darwin-arm64": "0.4.13",
+    "@lancedb/vectordb-darwin-x64": "0.4.13",
+    "@lancedb/vectordb-linux-arm64-gnu": "0.4.13",
+    "@lancedb/vectordb-linux-x64-gnu": "0.4.13",
+    "@lancedb/vectordb-win32-x64-msvc": "0.4.13"
   }
 }

node/src/index.ts

@@ -176,16 +176,21 @@ export async function connect (
     opts = { uri: arg }
   } else {
     // opts = { uri: arg.uri, awsCredentials = arg.awsCredentials }
-    opts = Object.assign(
-      {
-        uri: '',
-        awsCredentials: undefined,
-        awsRegion: defaultAwsRegion,
-        apiKey: undefined,
-        region: defaultAwsRegion
-      },
-      arg
-    )
+    const keys = Object.keys(arg)
+    if (keys.length === 1 && keys[0] === 'uri' && typeof arg.uri === 'string') {
+      opts = { uri: arg.uri }
+    } else {
+      opts = Object.assign(
+        {
+          uri: '',
+          awsCredentials: undefined,
+          awsRegion: defaultAwsRegion,
+          apiKey: undefined,
+          region: defaultAwsRegion
+        },
+        arg
+      )
+    }
   }
 
   if (opts.uri.startsWith('db://')) {

node/src/sanitize.ts

@@ -79,7 +79,7 @@ import {
 import type { IntBitWidth, TimeBitWidth } from "apache-arrow/type";
 
 function sanitizeMetadata(
-  metadataLike?: unknown
+  metadataLike?: unknown,
 ): Map<string, string> | undefined {
   if (metadataLike === undefined || metadataLike === null) {
     return undefined;
@@ -90,7 +90,7 @@ function sanitizeMetadata(
   for (const item of metadataLike) {
     if (!(typeof item[0] === "string" || !(typeof item[1] === "string"))) {
       throw Error(
-        "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values"
+        "Expected metadata, if present, to be a Map<string, string> but it had non-string keys or values",
       );
     }
   }
@@ -105,7 +105,7 @@ function sanitizeInt(typeLike: object) {
     typeof typeLike.isSigned !== "boolean"
   ) {
     throw Error(
-      "Expected an Int Type to have a `bitWidth` and `isSigned` property"
+      "Expected an Int Type to have a `bitWidth` and `isSigned` property",
     );
   }
   return new Int(typeLike.isSigned, typeLike.bitWidth as IntBitWidth);
@@ -128,7 +128,7 @@ function sanitizeDecimal(typeLike: object) {
     typeof typeLike.bitWidth !== "number"
   ) {
     throw Error(
-      "Expected a Decimal Type to have `scale`, `precision`, and `bitWidth` properties"
+      "Expected a Decimal Type to have `scale`, `precision`, and `bitWidth` properties",
     );
   }
   return new Decimal(typeLike.scale, typeLike.precision, typeLike.bitWidth);
@@ -149,7 +149,7 @@ function sanitizeTime(typeLike: object) {
     typeof typeLike.bitWidth !== "number"
   ) {
     throw Error(
-      "Expected a Time type to have `unit` and `bitWidth` properties"
+      "Expected a Time type to have `unit` and `bitWidth` properties",
     );
   }
   return new Time(typeLike.unit, typeLike.bitWidth as TimeBitWidth);
@@ -172,7 +172,7 @@ function sanitizeTypedTimestamp(
     | typeof TimestampNanosecond
     | typeof TimestampMicrosecond
     | typeof TimestampMillisecond
-    | typeof TimestampSecond
+    | typeof TimestampSecond,
 ) {
   let timezone = null;
   if ("timezone" in typeLike && typeof typeLike.timezone === "string") {
@@ -191,7 +191,7 @@ function sanitizeInterval(typeLike: object) {
 function sanitizeList(typeLike: object) {
   if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
     throw Error(
-      "Expected a List type to have an array-like `children` property"
+      "Expected a List type to have an array-like `children` property",
     );
   }
   if (typeLike.children.length !== 1) {
@@ -203,7 +203,7 @@ function sanitizeList(typeLike: object) {
 function sanitizeStruct(typeLike: object) {
   if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
     throw Error(
-      "Expected a Struct type to have an array-like `children` property"
+      "Expected a Struct type to have an array-like `children` property",
     );
   }
   return new Struct(typeLike.children.map((child) => sanitizeField(child)));
@@ -216,47 +216,47 @@ function sanitizeUnion(typeLike: object) {
     typeof typeLike.mode !== "number"
   ) {
     throw Error(
-      "Expected a Union type to have `typeIds` and `mode` properties"
+      "Expected a Union type to have `typeIds` and `mode` properties",
     );
   }
   if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
     throw Error(
-      "Expected a Union type to have an array-like `children` property"
+      "Expected a Union type to have an array-like `children` property",
     );
   }
 
   return new Union(
     typeLike.mode,
     typeLike.typeIds as any,
-    typeLike.children.map((child) => sanitizeField(child))
+    typeLike.children.map((child) => sanitizeField(child)),
   );
 }
 
 function sanitizeTypedUnion(
   typeLike: object,
-  UnionType: typeof DenseUnion | typeof SparseUnion
+  UnionType: typeof DenseUnion | typeof SparseUnion,
 ) {
   if (!("typeIds" in typeLike)) {
     throw Error(
-      "Expected a DenseUnion/SparseUnion type to have a `typeIds` property"
+      "Expected a DenseUnion/SparseUnion type to have a `typeIds` property",
     );
   }
   if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
     throw Error(
-      "Expected a DenseUnion/SparseUnion type to have an array-like `children` property"
+      "Expected a DenseUnion/SparseUnion type to have an array-like `children` property",
     );
   }
 
   return new UnionType(
     typeLike.typeIds as any,
-    typeLike.children.map((child) => sanitizeField(child))
+    typeLike.children.map((child) => sanitizeField(child)),
   );
 }
 
 function sanitizeFixedSizeBinary(typeLike: object) {
   if (!("byteWidth" in typeLike) || typeof typeLike.byteWidth !== "number") {
     throw Error(
-      "Expected a FixedSizeBinary type to have a `byteWidth` property"
+      "Expected a FixedSizeBinary type to have a `byteWidth` property",
     );
   }
   return new FixedSizeBinary(typeLike.byteWidth);
@@ -268,7 +268,7 @@ function sanitizeFixedSizeList(typeLike: object) {
   }
   if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
     throw Error(
-      "Expected a FixedSizeList type to have an array-like `children` property"
+      "Expected a FixedSizeList type to have an array-like `children` property",
     );
   }
   if (typeLike.children.length !== 1) {
@@ -276,14 +276,14 @@ function sanitizeFixedSizeList(typeLike: object) {
   }
   return new FixedSizeList(
     typeLike.listSize,
-    sanitizeField(typeLike.children[0])
+    sanitizeField(typeLike.children[0]),
   );
 }
 
 function sanitizeMap(typeLike: object) {
   if (!("children" in typeLike) || !Array.isArray(typeLike.children)) {
     throw Error(
-      "Expected a Map type to have an array-like `children` property"
+      "Expected a Map type to have an array-like `children` property",
     );
   }
   if (!("keysSorted" in typeLike) || typeof typeLike.keysSorted !== "boolean") {
@@ -291,7 +291,7 @@ function sanitizeMap(typeLike: object) {
   }
   return new Map_(
     typeLike.children.map((field) => sanitizeField(field)) as any,
-    typeLike.keysSorted
+    typeLike.keysSorted,
   );
 }
 
@@ -319,7 +319,7 @@ function sanitizeDictionary(typeLike: object) {
     sanitizeType(typeLike.dictionary),
     sanitizeType(typeLike.indices) as any,
     typeLike.id,
-    typeLike.isOrdered
+    typeLike.isOrdered,
   );
 }
 
@@ -454,7 +454,7 @@ function sanitizeField(fieldLike: unknown): Field {
     !("nullable" in fieldLike)
   ) {
     throw Error(
-      "The field passed in is missing a `type`/`name`/`nullable` property"
+      "The field passed in is missing a `type`/`name`/`nullable` property",
     );
   }
   const type = sanitizeType(fieldLike.type);
@@ -473,6 +473,13 @@ function sanitizeField(fieldLike: unknown): Field {
   return new Field(name, type, nullable, metadata);
 }
 
+/**
+ * Convert something schemaLike into a Schema instance
+ *
+ * This method is often needed even when the caller is using a Schema
+ * instance because they might be using a different instance of apache-arrow
+ * than lancedb is using.
+ */
 export function sanitizeSchema(schemaLike: unknown): Schema {
   if (schemaLike instanceof Schema) {
     return schemaLike;
@@ -482,7 +489,7 @@ export function sanitizeSchema(schemaLike: unknown): Schema {
   }
   if (!("fields" in schemaLike)) {
     throw Error(
-      "The schema passed in does not appear to be a schema (no 'fields' property)"
+      "The schema passed in does not appear to be a schema (no 'fields' property)",
     );
   }
   let metadata;
@@ -491,11 +498,11 @@ export function sanitizeSchema(schemaLike: unknown): Schema {
   }
   if (!Array.isArray(schemaLike.fields)) {
     throw Error(
-      "The schema passed in had a 'fields' property but it was not an array"
+      "The schema passed in had a 'fields' property but it was not an array",
     );
   }
   const sanitizedFields = schemaLike.fields.map((field) =>
-    sanitizeField(field)
+    sanitizeField(field),
   );
   return new Schema(sanitizedFields, metadata);
 }

node/src/test/test.ts

@@ -128,6 +128,11 @@ describe('LanceDB client', function () {
       assertResults(results)
       results = await table.where('id % 2 = 0').execute()
       assertResults(results)
+
+      // Should reject a bad filter
+      await expect(table.filter('id % 2 = 0 AND').execute()).to.be.rejectedWith(
+        /.*sql parser error: Expected an expression:, found: EOF.*/
+      )
     })
 
     it('uses a filter / where clause', async function () {
@@ -283,7 +288,8 @@ describe('LanceDB client', function () {
 
     it('create a table from an Arrow Table', async function () {
       const dir = await track().mkdir('lancejs')
-      const con = await lancedb.connect(dir)
+      // Also test the connect function with an object
+      const con = await lancedb.connect({ uri: dir })
 
       const i32s = new Int32Array(new Array<number>(10))
       const i32 = makeVector(i32s)
@@ -745,11 +751,11 @@ describe('LanceDB client', function () {
         num_sub_vectors: 2
       })
       await expect(createIndex).to.be.rejectedWith(
-        /VectorIndex requires the column data type to be fixed size list of float32s/
+        "index cannot be created on the column `name` which has data type Utf8"
       )
     })
 
-    it('it should fail when the column is not a vector', async function () {
+    it('it should fail when num_partitions is invalid', async function () {
       const uri = await createTestDB(32, 300)
       const con = await lancedb.connect(uri)
       const table = await con.openTable('vectors')

nodejs/Cargo.toml

@@ -14,12 +14,10 @@ crate-type = ["cdylib"]
 [dependencies]
 arrow-ipc.workspace = true
 futures.workspace = true
-lance-linalg.workspace = true
-lance.workspace = true
 lancedb = { path = "../rust/lancedb" }
 napi = { version = "2.15", default-features = false, features = [
     "napi7",
-    "async"
+    "async",
 ] }
 napi-derive = "2"
 

nodejs/__test__/table.test.ts

@@ -27,6 +27,7 @@ import {
   Float64,
 } from "apache-arrow";
 import { makeArrowTable } from "../dist/arrow";
+import { Index } from "../dist/indices";
 
 describe("Given a table", () => {
   let tmpDir: tmp.DirResult;
@@ -65,21 +66,36 @@ describe("Given a table", () => {
     expect(table.isOpen()).toBe(false);
     expect(table.countRows()).rejects.toThrow("Table some_table is closed");
   });
+
+  it("should let me update values", async () => {
+    await table.add([{ id: 1 }]);
+    expect(await table.countRows("id == 1")).toBe(1);
+    expect(await table.countRows("id == 7")).toBe(0);
+    await table.update({ id: "7" });
+    expect(await table.countRows("id == 1")).toBe(0);
+    expect(await table.countRows("id == 7")).toBe(1);
+    await table.add([{ id: 2 }]);
+    // Test Map as input
+    await table.update(new Map(Object.entries({ id: "10" })), {
+      where: "id % 2 == 0",
+    });
+    expect(await table.countRows("id == 2")).toBe(0);
+    expect(await table.countRows("id == 7")).toBe(1);
+    expect(await table.countRows("id == 10")).toBe(1);
+  });
 });
 
-describe("Test creating index", () => {
+describe("When creating an index", () => {
   let tmpDir: tmp.DirResult;
   const schema = new Schema([
     new Field("id", new Int32(), true),
     new Field("vec", new FixedSizeList(32, new Field("item", new Float32()))),
   ]);
+  let tbl: Table;
+  let queryVec: number[];
 
-  beforeEach(() => {
+  beforeEach(async () => {
     tmpDir = tmp.dirSync({ unsafeCleanup: true });
-  });
-  afterEach(() => tmpDir.removeCallback());
-
-  test("create vector index with no column", async () => {
     const db = await connect(tmpDir.name);
     const data = makeArrowTable(
       Array(300)
@@ -94,47 +110,80 @@ describe("Test creating index", () => {
         schema,
       },
     );
-    const tbl = await db.createTable("test", data);
-    await tbl.createIndex().build();
+    queryVec = data.toArray()[5].vec.toJSON();
+    tbl = await db.createTable("test", data);
+  });
+  afterEach(() => tmpDir.removeCallback());
+
+  it("should create a vector index on vector columns", async () => {
+    await tbl.createIndex("vec");
 
     // check index directory
     const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
-    // TODO: check index type.
+    const indices = await tbl.listIndices();
+    expect(indices.length).toBe(1);
+    expect(indices[0]).toEqual({
+      indexType: "IvfPq",
+      columns: ["vec"],
+    });
 
     // Search without specifying the column
-    const queryVector = data.toArray()[5].vec.toJSON();
-    const rst = await tbl.query().nearestTo(queryVector).limit(2).toArrow();
+    let rst = await tbl
+      .query()
+      .limit(2)
+      .nearestTo(queryVec)
+      .distanceType("DoT")
+      .toArrow();
+    expect(rst.numRows).toBe(2);
+
+    // Search using `vectorSearch`
+    rst = await tbl.vectorSearch(queryVec).limit(2).toArrow();
     expect(rst.numRows).toBe(2);
 
     // Search with specifying the column
-    const rst2 = await tbl.search(queryVector, "vec").limit(2).toArrow();
+    const rst2 = await tbl
+      .query()
+      .limit(2)
+      .nearestTo(queryVec)
+      .column("vec")
+      .toArrow();
     expect(rst2.numRows).toBe(2);
     expect(rst.toString()).toEqual(rst2.toString());
   });
 
-  test("no vector column available", async () => {
-    const db = await connect(tmpDir.name);
-    const tbl = await db.createTable(
-      "no_vec",
-      makeArrowTable([
-        { id: 1, val: 2 },
-        { id: 2, val: 3 },
-      ]),
-    );
-    await expect(tbl.createIndex().build()).rejects.toThrow(
-      "No vector column found",
-    );
+  it("should allow parameters to be specified", async () => {
+    await tbl.createIndex("vec", {
+      config: Index.ivfPq({
+        numPartitions: 10,
+      }),
+    });
 
-    await tbl.createIndex("val").build();
-    const indexDir = path.join(tmpDir.name, "no_vec.lance", "_indices");
+    // TODO: Verify parameters when we can load index config as part of list indices
+  });
+
+  it("should allow me to replace (or not) an existing index", async () => {
+    await tbl.createIndex("id");
+    // Default is replace=true
+    await tbl.createIndex("id");
+    await expect(tbl.createIndex("id", { replace: false })).rejects.toThrow(
+      "already exists",
+    );
+    await tbl.createIndex("id", { replace: true });
+  });
+
+  test("should create a scalar index on scalar columns", async () => {
+    await tbl.createIndex("id");
+    const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
     expect(fs.readdirSync(indexDir)).toHaveLength(1);
 
-    for await (const r of tbl.query().filter("id > 1").select(["id"])) {
-      expect(r.numRows).toBe(1);
+    for await (const r of tbl.query().where("id > 1").select(["id"])) {
+      expect(r.numRows).toBe(298);
     }
   });
 
+  // TODO: Move this test to the query API test (making sure we can reject queries
+  // when the dimension is incorrect)
   test("two columns with different dimensions", async () => {
     const db = await connect(tmpDir.name);
     const schema = new Schema([
@@ -164,71 +213,48 @@ describe("Test creating index", () => {
     );
 
     // Only build index over v1
-    await expect(tbl.createIndex().build()).rejects.toThrow(
-      /.*More than one vector columns found.*/,
-    );
-    tbl
-      .createIndex("vec")
-      // eslint-disable-next-line @typescript-eslint/naming-convention
-      .ivf_pq({ num_partitions: 2, num_sub_vectors: 2 })
-      .build();
+    await tbl.createIndex("vec", {
+      config: Index.ivfPq({ numPartitions: 2, numSubVectors: 2 }),
+    });
 
     const rst = await tbl
       .query()
+      .limit(2)
       .nearestTo(
         Array(32)
           .fill(1)
           .map(() => Math.random()),
       )
-      .limit(2)
       .toArrow();
     expect(rst.numRows).toBe(2);
 
     // Search with specifying the column
     await expect(
       tbl
-        .search(
+        .query()
+        .limit(2)
+        .nearestTo(
           Array(64)
             .fill(1)
             .map(() => Math.random()),
-          "vec",
         )
-        .limit(2)
+        .column("vec")
         .toArrow(),
-    ).rejects.toThrow(/.*does not match the dimension.*/);
+    ).rejects.toThrow(/.* query dim=64, expected vector dim=32.*/);
 
     const query64 = Array(64)
       .fill(1)
       .map(() => Math.random());
-    const rst64Query = await tbl.query().nearestTo(query64).limit(2).toArrow();
-    const rst64Search = await tbl.search(query64, "vec2").limit(2).toArrow();
+    const rst64Query = await tbl.query().limit(2).nearestTo(query64).toArrow();
+    const rst64Search = await tbl
+      .query()
+      .limit(2)
+      .nearestTo(query64)
+      .column("vec2")
+      .toArrow();
     expect(rst64Query.toString()).toEqual(rst64Search.toString());
     expect(rst64Query.numRows).toBe(2);
   });
-
-  test("create scalar index", async () => {
-    const db = await connect(tmpDir.name);
-    const data = makeArrowTable(
-      Array(300)
-        .fill(1)
-        .map((_, i) => ({
-          id: i,
-          vec: Array(32)
-            .fill(1)
-            .map(() => Math.random()),
-        })),
-      {
-        schema,
-      },
-    );
-    const tbl = await db.createTable("test", data);
-    await tbl.createIndex("id").build();
-
-    // check index directory
-    const indexDir = path.join(tmpDir.name, "test.lance", "_indices");
-    expect(fs.readdirSync(indexDir)).toHaveLength(1);
-    // TODO: check index type.
-  });
 });
 
 describe("Read consistency interval", () => {
@@ -348,3 +374,48 @@ describe("schema evolution", function () {
     expect(await table.schema()).toEqual(expectedSchema);
   });
 });
+
+describe("when dealing with versioning", () => {
+  let tmpDir: tmp.DirResult;
+  beforeEach(() => {
+    tmpDir = tmp.dirSync({ unsafeCleanup: true });
+  });
+  afterEach(() => {
+    tmpDir.removeCallback();
+  });
+
+  it("can travel in time", async () => {
+    // Setup
+    const con = await connect(tmpDir.name);
+    const table = await con.createTable("vectors", [
+      { id: 1n, vector: [0.1, 0.2] },
+    ]);
+    const version = await table.version();
+    await table.add([{ id: 2n, vector: [0.1, 0.2] }]);
+    expect(await table.countRows()).toBe(2);
+    // Make sure we can rewind
+    await table.checkout(version);
+    expect(await table.countRows()).toBe(1);
+    // Can't add data in time travel mode
+    await expect(table.add([{ id: 3n, vector: [0.1, 0.2] }])).rejects.toThrow(
+      "table cannot be modified when a specific version is checked out",
+    );
+    // Can go back to normal mode
+    await table.checkoutLatest();
+    expect(await table.countRows()).toBe(2);
+    // Should be able to add data again
+    await table.add([{ id: 2n, vector: [0.1, 0.2] }]);
+    expect(await table.countRows()).toBe(3);
+    // Now checkout and restore
+    await table.checkout(version);
+    await table.restore();
+    expect(await table.countRows()).toBe(1);
+    // Should be able to add data
+    await table.add([{ id: 2n, vector: [0.1, 0.2] }]);
+    expect(await table.countRows()).toBe(2);
+    // Can't use restore if not checked out
+    await expect(table.restore()).rejects.toThrow(
+      "checkout before running restore",
+    );
+  });
+});

nodejs/eslint.config.js

@@ -4,14 +4,25 @@
 const eslint = require("@eslint/js");
 const tseslint = require("typescript-eslint");
 const eslintConfigPrettier = require("eslint-config-prettier");
+const jsdoc = require("eslint-plugin-jsdoc");
 
 module.exports = tseslint.config(
   eslint.configs.recommended,
+  jsdoc.configs["flat/recommended"],
   eslintConfigPrettier,
   ...tseslint.configs.recommended,
   {
     rules: {
       "@typescript-eslint/naming-convention": "error",
+      "jsdoc/require-returns": "off",
+      "jsdoc/require-param": "off",
+      "jsdoc/require-jsdoc": [
+        "error",
+        {
+          publicOnly: true,
+        },
+      ],
     },
+    plugins: jsdoc,
   },
 );

nodejs/lancedb/arrow.ts

@@ -31,6 +31,7 @@ import {
   DataType,
   Binary,
   Float32,
+  type makeTable,
 } from "apache-arrow";
 import { type EmbeddingFunction } from "./embedding/embedding_function";
 import { sanitizeSchema } from "./sanitize";
@@ -128,14 +129,7 @@ export class MakeArrowTableOptions {
  *  - Buffer => Binary
  *  - Record<String, any> => Struct
  *  - Array<any> => List
- *
- * @param data input data
- * @param options options to control the makeArrowTable call.
- *
  * @example
- *
- * ```ts
- *
  * import { fromTableToBuffer, makeArrowTable } from "../arrow";
  * import { Field, FixedSizeList, Float16, Float32, Int32, Schema } from "apache-arrow";
  *
@@ -307,7 +301,9 @@ export function makeEmptyTable(schema: Schema): ArrowTable {
   return makeArrowTable([], { schema });
 }
 
-// Helper function to convert Array<Array<any>> to a variable sized list array
+/**
+ * Helper function to convert Array<Array<any>> to a variable sized list array
+ */
 // @ts-expect-error (Vector<unknown> is not assignable to Vector<any>)
 function makeListVector(lists: unknown[][]): Vector<unknown> {
   if (lists.length === 0 || lists[0].length === 0) {
@@ -333,7 +329,7 @@ function makeListVector(lists: unknown[][]): Vector<unknown> {
   return listBuilder.finish().toVector();
 }
 
-// Helper function to convert an Array of JS values to an Arrow Vector
+/** Helper function to convert an Array of JS values to an Arrow Vector */
 function makeVector(
   values: unknown[],
   type?: DataType,
@@ -374,6 +370,7 @@ function makeVector(
   }
 }
 
+/** Helper function to apply embeddings to an input table */
 async function applyEmbeddings<T>(
   table: ArrowTable,
   embeddings?: EmbeddingFunction<T>,
@@ -466,7 +463,7 @@ async function applyEmbeddings<T>(
   return newTable;
 }
 
-/*
+/**
  * Convert an Array of records into an Arrow Table, optionally applying an
  * embeddings function to it.
  *
@@ -493,7 +490,7 @@ export async function convertToTable<T>(
   return await applyEmbeddings(table, embeddings, makeTableOptions?.schema);
 }
 
-// Creates the Arrow Type for a Vector column with dimension `dim`
+/** Creates the Arrow Type for a Vector column with dimension `dim` */
 function newVectorType<T extends Float>(
   dim: number,
   innerType: T,
@@ -565,6 +562,14 @@ export async function fromTableToBuffer<T>(
   return Buffer.from(await writer.toUint8Array());
 }
 
+/**
+ * Serialize an Arrow Table into a buffer using the Arrow IPC File serialization
+ *
+ * This function will apply `embeddings` to the table in a manner similar to
+ * `convertToTable`.
+ *
+ * `schema` is required if the table is empty
+ */
 export async function fromDataToBuffer<T>(
   data: Data,
   embeddings?: EmbeddingFunction<T>,
@@ -599,6 +604,9 @@ export async function fromTableToStreamBuffer<T>(
   return Buffer.from(await writer.toUint8Array());
 }
 
+/**
+ * Reorder the columns in `batch` so that they agree with the field order in `schema`
+ */
 function alignBatch(batch: RecordBatch, schema: Schema): RecordBatch {
   const alignedChildren = [];
   for (const field of schema.fields) {
@@ -621,6 +629,9 @@ function alignBatch(batch: RecordBatch, schema: Schema): RecordBatch {
   return new RecordBatch(schema, newData);
 }
 
+/**
+ * Reorder the columns in `table` so that they agree with the field order in `schema`
+ */
 function alignTable(table: ArrowTable, schema: Schema): ArrowTable {
   const alignedBatches = table.batches.map((batch) =>
     alignBatch(batch, schema),
@@ -628,7 +639,9 @@ function alignTable(table: ArrowTable, schema: Schema): ArrowTable {
   return new ArrowTable(schema, alignedBatches);
 }
 
-// Creates an empty Arrow Table
+/**
+ * Create an empty table with the given schema
+ */
 export function createEmptyTable(schema: Schema): ArrowTable {
   return new ArrowTable(sanitizeSchema(schema));
 }

nodejs/lancedb/connection.ts

@@ -78,7 +78,8 @@ export class Connection {
     return this.inner.isOpen();
   }
 
-  /** Close the connection, releasing any underlying resources.
+  /**
+   * Close the connection, releasing any underlying resources.
    *
    * It is safe to call this method multiple times.
    *
@@ -93,11 +94,12 @@ export class Connection {
     return this.inner.display();
   }
 
-  /** List all the table names in this database.
+  /**
+   * List all the table names in this database.
    *
    * Tables will be returned in lexicographical order.
-   *
-   * @param options Optional parameters to control the listing.
+   * @param {Partial<TableNamesOptions>} options - options to control the
+   * paging / start point
    */
   async tableNames(options?: Partial<TableNamesOptions>): Promise<string[]> {
     return this.inner.tableNames(options?.startAfter, options?.limit);
@@ -105,9 +107,7 @@ export class Connection {
 
   /**
    * Open a table in the database.
-   *
-   * @param name The name of the table.
-   * @param embeddings An embedding function to use on this table
+   * @param {string} name - The name of the table
    */
   async openTable(name: string): Promise<Table> {
     const innerTable = await this.inner.openTable(name);
@@ -116,9 +116,9 @@ export class Connection {
 
   /**
    * Creates a new Table and initialize it with new data.
-   *
    * @param {string} name - The name of the table.
-   * @param data - Non-empty Array of Records to be inserted into the table
+   * @param {Record<string, unknown>[] | ArrowTable} data - Non-empty Array of Records
+   * to be inserted into the table
    */
   async createTable(
     name: string,
@@ -145,9 +145,8 @@ export class Connection {
 
   /**
    * Creates a new empty Table
-   *
    * @param {string} name - The name of the table.
-   * @param schema - The schema of the table
+   * @param {Schema} schema - The schema of the table
    */
   async createEmptyTable(
     name: string,
@@ -169,7 +168,7 @@ export class Connection {
 
   /**
    * Drop an existing table.
-   * @param name The name of the table to drop.
+   * @param {string} name The name of the table to drop.
    */
   async dropTable(name: string): Promise<void> {
     return this.inner.dropTable(name);

nodejs/lancedb/embedding/embedding_function.ts

@@ -62,6 +62,7 @@ export interface EmbeddingFunction<T> {
   embed: (data: T[]) => Promise<number[][]>;
 }
 
+/** Test if the input seems to be an embedding function */
 export function isEmbeddingFunction<T>(
   value: unknown,
 ): value is EmbeddingFunction<T> {

nodejs/lancedb/index.ts

@@ -18,15 +18,9 @@ import {
   ConnectionOptions,
 } from "./native.js";
 
-export {
-  ConnectionOptions,
-  WriteOptions,
-  Query,
-  MetricType,
-} from "./native.js";
-export { Connection } from "./connection";
-export { Table } from "./table";
-export { IvfPQOptions, IndexBuilder } from "./indexer";
+export { ConnectionOptions, WriteOptions, Query } from "./native.js";
+export { Connection, CreateTableOptions } from "./connection";
+export { Table, AddDataOptions } from "./table";
 
 /**
  * Connect to a LanceDB instance at the given URI.
@@ -36,9 +30,8 @@ export { IvfPQOptions, IndexBuilder } from "./indexer";
  * - `/path/to/database` - local database
  * - `s3://bucket/path/to/database` or `gs://bucket/path/to/database` - database on cloud storage
  * - `db://host:port` - remote database (LanceDB cloud)
- *
- * @param uri The uri of the database. If the database uri starts with `db://` then it connects to a remote database.
- *
+ * @param {string} uri - The uri of the database. If the database uri starts
+ * with `db://` then it connects to a remote database.
  * @see {@link ConnectionOptions} for more details on the URI format.
  */
 export async function connect(

nodejs/lancedb/indexer.ts

@@ -1,105 +0,0 @@
-// Copyright 2024 Lance Developers.
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-// TODO: Re-enable this as part of https://github.com/lancedb/lancedb/pull/1052
-/* eslint-disable @typescript-eslint/naming-convention */
-
-import {
-  MetricType,
-  IndexBuilder as NativeBuilder,
-  Table as NativeTable,
-} from "./native";
-
-/** Options to create `IVF_PQ` index */
-export interface IvfPQOptions {
-  /** Number of IVF partitions. */
-  num_partitions?: number;
-
-  /** Number of sub-vectors in PQ coding. */
-  num_sub_vectors?: number;
-
-  /** Number of bits used for each PQ code.
-   */
-  num_bits?: number;
-
-  /** Metric type to calculate the distance between vectors.
-   *
-   * Supported metrics: `L2`, `Cosine` and `Dot`.
-   */
-  metric_type?: MetricType;
-
-  /** Number of iterations to train K-means.
-   *
-   * Default is 50. The more iterations it usually yield better results,
-   * but it takes longer to train.
-   */
-  max_iterations?: number;
-
-  sample_rate?: number;
-}
-
-/**
- * Building an index on LanceDB {@link Table}
- *
- * @see {@link Table.createIndex} for detailed usage.
- */
-export class IndexBuilder {
-  private inner: NativeBuilder;
-
-  constructor(tbl: NativeTable) {
-    this.inner = tbl.createIndex();
-  }
-
-  /** Instruct the builder to build an `IVF_PQ` index */
-  ivf_pq(options?: IvfPQOptions): IndexBuilder {
-    this.inner.ivfPq(
-      options?.metric_type,
-      options?.num_partitions,
-      options?.num_sub_vectors,
-      options?.num_bits,
-      options?.max_iterations,
-      options?.sample_rate,
-    );
-    return this;
-  }
-
-  /** Instruct the builder to build a Scalar index. */
-  scalar(): IndexBuilder {
-    this.scalar();
-    return this;
-  }
-
-  /** Set the column(s) to create index on top of. */
-  column(col: string): IndexBuilder {
-    this.inner.column(col);
-    return this;
-  }
-
-  /** Set to true to replace existing index. */
-  replace(val: boolean): IndexBuilder {
-    this.inner.replace(val);
-    return this;
-  }
-
-  /** Specify the name of the index. Optional */
-  name(n: string): IndexBuilder {
-    this.inner.name(n);
-    return this;
-  }
-
-  /** Building the index. */
-  async build() {
-    await this.inner.build();
-  }
-}

nodejs/lancedb/indices.ts

@@ -0,0 +1,203 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+import { Index as LanceDbIndex } from "./native";
+
+/**
+ * Options to create an `IVF_PQ` index
+ */
+export interface IvfPqOptions {
+  /**
+   * The number of IVF partitions to create.
+   *
+   * This value should generally scale with the number of rows in the dataset.
+   * By default the number of partitions is the square root of the number of
+   * rows.
+   *
+   * If this value is too large then the first part of the search (picking the
+   * right partition) will be slow.  If this value is too small then the second
+   * part of the search (searching within a partition) will be slow.
+   */
+  numPartitions?: number;
+
+  /**
+   * Number of sub-vectors of PQ.
+   *
+   * This value controls how much the vector is compressed during the quantization step.
+   * The more sub vectors there are the less the vector is compressed.  The default is
+   * the dimension of the vector divided by 16.  If the dimension is not evenly divisible
+   * by 16 we use the dimension divded by 8.
+   *
+   * The above two cases are highly preferred.  Having 8 or 16 values per subvector allows
+   * us to use efficient SIMD instructions.
+   *
+   * If the dimension is not visible by 8 then we use 1 subvector.  This is not ideal and
+   * will likely result in poor performance.
+   */
+  numSubVectors?: number;
+
+  /**
+   * Distance type to use to build the index.
+   *
+   * Default value is "l2".
+   *
+   * This is used when training the index to calculate the IVF partitions
+   * (vectors are grouped in partitions with similar vectors according to this
+   * distance type) and to calculate a subvector's code during quantization.
+   *
+   * The distance type used to train an index MUST match the distance type used
+   * to search the index.  Failure to do so will yield inaccurate results.
+   *
+   * The following distance types are available:
+   *
+   * "l2" - Euclidean distance. This is a very common distance metric that
+   * accounts for both magnitude and direction when determining the distance
+   * between vectors. L2 distance has a range of [0, ∞).
+   *
+   * "cosine" - Cosine distance.  Cosine distance is a distance metric
+   * calculated from the cosine similarity between two vectors. Cosine
+   * similarity is a measure of similarity between two non-zero vectors of an
+   * inner product space. It is defined to equal the cosine of the angle
+   * between them.  Unlike L2, the cosine distance is not affected by the
+   * magnitude of the vectors.  Cosine distance has a range of [0, 2].
+   *
+   * Note: the cosine distance is undefined when one (or both) of the vectors
+   * are all zeros (there is no direction).  These vectors are invalid and may
+   * never be returned from a vector search.
+   *
+   * "dot" - Dot product. Dot distance is the dot product of two vectors. Dot
+   * 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.
+   */
+  distanceType?: "l2" | "cosine" | "dot";
+
+  /**
+   * Max iteration to train IVF kmeans.
+   *
+   * When training an IVF PQ index we use kmeans to calculate the partitions.  This parameter
+   * controls how many iterations of kmeans to run.
+   *
+   * Increasing this might improve the quality of the index but in most cases these extra
+   * iterations have diminishing returns.
+   *
+   * The default value is 50.
+   */
+  maxIterations?: number;
+
+  /**
+   * The number of vectors, per partition, to sample when training IVF kmeans.
+   *
+   * When an IVF PQ index is trained, we need to calculate partitions.  These are groups
+   * of vectors that are similar to each other.  To do this we use an algorithm called kmeans.
+   *
+   * Running kmeans on a large dataset can be slow.  To speed this up we run kmeans on a
+   * random sample of the data.  This parameter controls the size of the sample.  The total
+   * number of vectors used to train the index is `sample_rate * num_partitions`.
+   *
+   * Increasing this value might improve the quality of the index but in most cases the
+   * default should be sufficient.
+   *
+   * The default value is 256.
+   */
+  sampleRate?: number;
+}
+
+export class Index {
+  private readonly inner: LanceDbIndex;
+  private constructor(inner: LanceDbIndex) {
+    this.inner = inner;
+  }
+
+  /**
+   * Create an IvfPq index
+   *
+   * This index stores a compressed (quantized) copy of every vector.  These vectors
+   * are grouped into partitions of similar vectors.  Each partition keeps track of
+   * a centroid which is the average value of all vectors in the group.
+   *
+   * During a query the centroids are compared with the query vector to find the closest
+   * partitions.  The compressed vectors in these partitions are then searched to find
+   * the closest vectors.
+   *
+   * The compression scheme is called product quantization.  Each vector is divided into
+   * subvectors and then each subvector is quantized into a small number of bits.  the
+   * parameters `num_bits` and `num_subvectors` control this process, providing a tradeoff
+   * between index size (and thus search speed) and index accuracy.
+   *
+   * The partitioning process is called IVF and the `num_partitions` parameter controls how
+   * many groups to create.
+   *
+   * Note that training an IVF PQ index on a large dataset is a slow operation and
+   * currently is also a memory intensive operation.
+   */
+  static ivfPq(options?: Partial<IvfPqOptions>) {
+    return new Index(
+      LanceDbIndex.ivfPq(
+        options?.distanceType,
+        options?.numPartitions,
+        options?.numSubVectors,
+        options?.maxIterations,
+        options?.sampleRate,
+      ),
+    );
+  }
+
+  /**
+   * Create a btree index
+   *
+   * A btree index is an index on a scalar columns.  The index stores a copy of the column
+   * in sorted order.  A header entry is created for each block of rows (currently the
+   * block size is fixed at 4096).  These header entries are stored in a separate
+   * cacheable structure (a btree).  To search for data the header is used to determine
+   * which blocks need to be read from disk.
+   *
+   * For example, a btree index in a table with 1Bi rows requires sizeof(Scalar) * 256Ki
+   * bytes of memory and will generally need to read sizeof(Scalar) * 4096 bytes to find
+   * the correct row ids.
+   *
+   * This index is good for scalar columns with mostly distinct values and does best when
+   * the query is highly selective.
+   *
+   * The btree index does not currently have any parameters though parameters such as the
+   * block size may be added in the future.
+   */
+  static btree() {
+    return new Index(LanceDbIndex.btree());
+  }
+}
+
+export interface IndexOptions {
+  /**
+   * Advanced index configuration
+   *
+   * This option allows you to specify a specfic index to create and also
+   * allows you to pass in configuration for training the index.
+   *
+   * See the static methods on Index for details on the various index types.
+   *
+   * If this is not supplied then column data type(s) and column statistics
+   * will be used to determine the most useful kind of index to create.
+   */
+  config?: Index;
+  /**
+   * Whether to replace the existing index
+   *
+   * If this is false, and another index already exists on the same columns
+   * and the same name, then an error will be returned.  This is true even if
+   * that index is out of date.
+   *
+   * The default is true
+   */
+  replace?: boolean;
+}

nodejs/lancedb/native.d.ts

@@ -3,14 +3,17 @@
 
 /* auto-generated by NAPI-RS */
 
-export const enum IndexType {
-  Scalar = 0,
-  IvfPq = 1
-}
-export const enum MetricType {
-  L2 = 0,
-  Cosine = 1,
-  Dot = 2
+/** A description of an index currently configured on a column */
+export interface IndexConfig {
+  /** The type of the index */
+  indexType: string
+  /**
+   * The columns in the index
+   *
+   * Currently this is always an array of size 1.  In the future there may
+   * be more columns to represent composite indices.
+   */
+  columns: Array<string>
 }
 /**
  *  A definition of a column alteration. The alteration changes the column at
@@ -93,28 +96,32 @@ export class Connection {
   /** Drop table with the name. Or raise an error if the table does not exist. */
   dropTable(name: string): Promise<void>
 }
-export class IndexBuilder {
-  replace(v: boolean): void
-  column(c: string): void
-  name(name: string): void
-  ivfPq(metricType?: MetricType | undefined | null, numPartitions?: number | undefined | null, numSubVectors?: number | undefined | null, numBits?: number | undefined | null, maxIterations?: number | undefined | null, sampleRate?: number | undefined | null): void
-  scalar(): void
-  build(): Promise<void>
+export class Index {
+  static ivfPq(distanceType?: string | undefined | null, numPartitions?: number | undefined | null, numSubVectors?: number | undefined | null, maxIterations?: number | undefined | null, sampleRate?: number | undefined | null): Index
+  static btree(): Index
 }
 /** Typescript-style Async Iterator over RecordBatches  */
 export class RecordBatchIterator {
   next(): Promise<Buffer | null>
 }
 export class Query {
-  column(column: string): void
-  filter(filter: string): void
-  select(columns: Array<string>): void
+  onlyIf(predicate: string): void
+  select(columns: Array<[string, string]>): void
   limit(limit: number): void
-  prefilter(prefilter: boolean): void
-  nearestTo(vector: Float32Array): void
+  nearestTo(vector: Float32Array): VectorQuery
+  execute(): Promise<RecordBatchIterator>
+}
+export class VectorQuery {
+  column(column: string): void
+  distanceType(distanceType: string): void
+  postfilter(): void
   refineFactor(refineFactor: number): void
   nprobes(nprobe: number): void
-  executeStream(): Promise<RecordBatchIterator>
+  bypassVectorIndex(): void
+  onlyIf(predicate: string): void
+  select(columns: Array<[string, string]>): void
+  limit(limit: number): void
+  execute(): Promise<RecordBatchIterator>
 }
 export class Table {
   display(): string
@@ -125,9 +132,16 @@ export class Table {
   add(buf: Buffer, mode: string): Promise<void>
   countRows(filter?: string | undefined | null): Promise<number>
   delete(predicate: string): Promise<void>
-  createIndex(): IndexBuilder
+  createIndex(index: Index | undefined | null, column: string, replace?: boolean | undefined | null): Promise<void>
+  update(onlyIf: string | undefined | null, columns: Array<[string, string]>): Promise<void>
   query(): Query
+  vectorSearch(vector: Float32Array): VectorQuery
   addColumns(transforms: Array<AddColumnsSql>): Promise<void>
   alterColumns(alterations: Array<ColumnAlteration>): Promise<void>
   dropColumns(columns: Array<string>): Promise<void>
+  version(): Promise<number>
+  checkout(version: number): Promise<void>
+  checkoutLatest(): Promise<void>
+  restore(): Promise<void>
+  listIndices(): Promise<Array<IndexConfig>>
 }

nodejs/lancedb/native.js

@@ -5,304 +5,325 @@
 /* auto-generated by NAPI-RS */
 
 const { existsSync, readFileSync } = require('fs')
-const { join } = require('path')
+const { join } = require("path");
 
-const { platform, arch } = process
+const { platform, arch } = process;
 
-let nativeBinding = null
-let localFileExisted = false
-let loadError = null
+let nativeBinding = null;
+let localFileExisted = false;
+let loadError = null;
 
 function isMusl() {
   // For Node 10
-  if (!process.report || typeof process.report.getReport !== 'function') {
+  if (!process.report || typeof process.report.getReport !== "function") {
     try {
-      const lddPath = require('child_process').execSync('which ldd').toString().trim()
-      return readFileSync(lddPath, 'utf8').includes('musl')
+      const lddPath = require("child_process")
+        .execSync("which ldd")
+        .toString()
+        .trim();
+      return readFileSync(lddPath, "utf8").includes("musl");
     } catch (e) {
-      return true
+      return true;
     }
   } else {
-    const { glibcVersionRuntime } = process.report.getReport().header
-    return !glibcVersionRuntime
+    const { glibcVersionRuntime } = process.report.getReport().header;
+    return !glibcVersionRuntime;
   }
 }
 
 switch (platform) {
-  case 'android':
+  case "android":
     switch (arch) {
-      case 'arm64':
-        localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.android-arm64.node'))
+      case "arm64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.android-arm64.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.android-arm64.node')
+            nativeBinding = require("./lancedb-nodejs.android-arm64.node");
           } else {
-            nativeBinding = require('lancedb-android-arm64')
+            nativeBinding = require("lancedb-android-arm64");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
-      case 'arm':
-        localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.android-arm-eabi.node'))
+        break;
+      case "arm":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.android-arm-eabi.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.android-arm-eabi.node')
+            nativeBinding = require("./lancedb-nodejs.android-arm-eabi.node");
           } else {
-            nativeBinding = require('lancedb-android-arm-eabi')
+            nativeBinding = require("lancedb-android-arm-eabi");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
+        break;
       default:
-        throw new Error(`Unsupported architecture on Android ${arch}`)
+        throw new Error(`Unsupported architecture on Android ${arch}`);
     }
-    break
-  case 'win32':
+    break;
+  case "win32":
     switch (arch) {
-      case 'x64':
+      case "x64":
         localFileExisted = existsSync(
-          join(__dirname, 'lancedb-nodejs.win32-x64-msvc.node')
-        )
+          join(__dirname, "lancedb-nodejs.win32-x64-msvc.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.win32-x64-msvc.node')
+            nativeBinding = require("./lancedb-nodejs.win32-x64-msvc.node");
           } else {
-            nativeBinding = require('lancedb-win32-x64-msvc')
+            nativeBinding = require("lancedb-win32-x64-msvc");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
-      case 'ia32':
+        break;
+      case "ia32":
         localFileExisted = existsSync(
-          join(__dirname, 'lancedb-nodejs.win32-ia32-msvc.node')
-        )
+          join(__dirname, "lancedb-nodejs.win32-ia32-msvc.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.win32-ia32-msvc.node')
+            nativeBinding = require("./lancedb-nodejs.win32-ia32-msvc.node");
           } else {
-            nativeBinding = require('lancedb-win32-ia32-msvc')
+            nativeBinding = require("lancedb-win32-ia32-msvc");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
-      case 'arm64':
+        break;
+      case "arm64":
         localFileExisted = existsSync(
-          join(__dirname, 'lancedb-nodejs.win32-arm64-msvc.node')
-        )
+          join(__dirname, "lancedb-nodejs.win32-arm64-msvc.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.win32-arm64-msvc.node')
+            nativeBinding = require("./lancedb-nodejs.win32-arm64-msvc.node");
           } else {
-            nativeBinding = require('lancedb-win32-arm64-msvc')
+            nativeBinding = require("lancedb-win32-arm64-msvc");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
+        break;
       default:
-        throw new Error(`Unsupported architecture on Windows: ${arch}`)
+        throw new Error(`Unsupported architecture on Windows: ${arch}`);
     }
-    break
-  case 'darwin':
-    localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.darwin-universal.node'))
+    break;
+  case "darwin":
+    localFileExisted = existsSync(
+      join(__dirname, "lancedb-nodejs.darwin-universal.node"),
+    );
     try {
       if (localFileExisted) {
-        nativeBinding = require('./lancedb-nodejs.darwin-universal.node')
+        nativeBinding = require("./lancedb-nodejs.darwin-universal.node");
       } else {
-        nativeBinding = require('lancedb-darwin-universal')
+        nativeBinding = require("lancedb-darwin-universal");
       }
-      break
+      break;
     } catch {}
     switch (arch) {
-      case 'x64':
-        localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.darwin-x64.node'))
-        try {
-          if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.darwin-x64.node')
-          } else {
-            nativeBinding = require('lancedb-darwin-x64')
-          }
-        } catch (e) {
-          loadError = e
-        }
-        break
-      case 'arm64':
+      case "x64":
         localFileExisted = existsSync(
-          join(__dirname, 'lancedb-nodejs.darwin-arm64.node')
-        )
+          join(__dirname, "lancedb-nodejs.darwin-x64.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.darwin-arm64.node')
+            nativeBinding = require("./lancedb-nodejs.darwin-x64.node");
           } else {
-            nativeBinding = require('lancedb-darwin-arm64')
+            nativeBinding = require("lancedb-darwin-x64");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
+        break;
+      case "arm64":
+        localFileExisted = existsSync(
+          join(__dirname, "lancedb-nodejs.darwin-arm64.node"),
+        );
+        try {
+          if (localFileExisted) {
+            nativeBinding = require("./lancedb-nodejs.darwin-arm64.node");
+          } else {
+            nativeBinding = require("lancedb-darwin-arm64");
+          }
+        } catch (e) {
+          loadError = e;
+        }
+        break;
       default:
-        throw new Error(`Unsupported architecture on macOS: ${arch}`)
+        throw new Error(`Unsupported architecture on macOS: ${arch}`);
     }
-    break
-  case 'freebsd':
-    if (arch !== 'x64') {
-      throw new Error(`Unsupported architecture on FreeBSD: ${arch}`)
+    break;
+  case "freebsd":
+    if (arch !== "x64") {
+      throw new Error(`Unsupported architecture on FreeBSD: ${arch}`);
     }
-    localFileExisted = existsSync(join(__dirname, 'lancedb-nodejs.freebsd-x64.node'))
+    localFileExisted = existsSync(
+      join(__dirname, "lancedb-nodejs.freebsd-x64.node"),
+    );
     try {
       if (localFileExisted) {
-        nativeBinding = require('./lancedb-nodejs.freebsd-x64.node')
+        nativeBinding = require("./lancedb-nodejs.freebsd-x64.node");
       } else {
-        nativeBinding = require('lancedb-freebsd-x64')
+        nativeBinding = require("lancedb-freebsd-x64");
       }
     } catch (e) {
-      loadError = e
+      loadError = e;
     }
-    break
-  case 'linux':
+    break;
+  case "linux":
     switch (arch) {
-      case 'x64':
+      case "x64":
         if (isMusl()) {
           localFileExisted = existsSync(
-            join(__dirname, 'lancedb-nodejs.linux-x64-musl.node')
-          )
+            join(__dirname, "lancedb-nodejs.linux-x64-musl.node"),
+          );
           try {
             if (localFileExisted) {
-              nativeBinding = require('./lancedb-nodejs.linux-x64-musl.node')
+              nativeBinding = require("./lancedb-nodejs.linux-x64-musl.node");
             } else {
-              nativeBinding = require('lancedb-linux-x64-musl')
+              nativeBinding = require("lancedb-linux-x64-musl");
             }
           } catch (e) {
-            loadError = e
+            loadError = e;
           }
         } else {
           localFileExisted = existsSync(
-            join(__dirname, 'lancedb-nodejs.linux-x64-gnu.node')
-          )
+            join(__dirname, "lancedb-nodejs.linux-x64-gnu.node"),
+          );
           try {
             if (localFileExisted) {
-              nativeBinding = require('./lancedb-nodejs.linux-x64-gnu.node')
+              nativeBinding = require("./lancedb-nodejs.linux-x64-gnu.node");
             } else {
-              nativeBinding = require('lancedb-linux-x64-gnu')
+              nativeBinding = require("lancedb-linux-x64-gnu");
             }
           } catch (e) {
-            loadError = e
+            loadError = e;
           }
         }
-        break
-      case 'arm64':
+        break;
+      case "arm64":
         if (isMusl()) {
           localFileExisted = existsSync(
-            join(__dirname, 'lancedb-nodejs.linux-arm64-musl.node')
-          )
+            join(__dirname, "lancedb-nodejs.linux-arm64-musl.node"),
+          );
           try {
             if (localFileExisted) {
-              nativeBinding = require('./lancedb-nodejs.linux-arm64-musl.node')
+              nativeBinding = require("./lancedb-nodejs.linux-arm64-musl.node");
             } else {
-              nativeBinding = require('lancedb-linux-arm64-musl')
+              nativeBinding = require("lancedb-linux-arm64-musl");
             }
           } catch (e) {
-            loadError = e
+            loadError = e;
           }
         } else {
           localFileExisted = existsSync(
-            join(__dirname, 'lancedb-nodejs.linux-arm64-gnu.node')
-          )
+            join(__dirname, "lancedb-nodejs.linux-arm64-gnu.node"),
+          );
           try {
             if (localFileExisted) {
-              nativeBinding = require('./lancedb-nodejs.linux-arm64-gnu.node')
+              nativeBinding = require("./lancedb-nodejs.linux-arm64-gnu.node");
             } else {
-              nativeBinding = require('lancedb-linux-arm64-gnu')
+              nativeBinding = require("lancedb-linux-arm64-gnu");
             }
           } catch (e) {
-            loadError = e
+            loadError = e;
           }
         }
-        break
-      case 'arm':
+        break;
+      case "arm":
         localFileExisted = existsSync(
-          join(__dirname, 'lancedb-nodejs.linux-arm-gnueabihf.node')
-        )
+          join(__dirname, "lancedb-nodejs.linux-arm-gnueabihf.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.linux-arm-gnueabihf.node')
+            nativeBinding = require("./lancedb-nodejs.linux-arm-gnueabihf.node");
           } else {
-            nativeBinding = require('lancedb-linux-arm-gnueabihf')
+            nativeBinding = require("lancedb-linux-arm-gnueabihf");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
-      case 'riscv64':
+        break;
+      case "riscv64":
         if (isMusl()) {
           localFileExisted = existsSync(
-            join(__dirname, 'lancedb-nodejs.linux-riscv64-musl.node')
-          )
+            join(__dirname, "lancedb-nodejs.linux-riscv64-musl.node"),
+          );
           try {
             if (localFileExisted) {
-              nativeBinding = require('./lancedb-nodejs.linux-riscv64-musl.node')
+              nativeBinding = require("./lancedb-nodejs.linux-riscv64-musl.node");
             } else {
-              nativeBinding = require('lancedb-linux-riscv64-musl')
+              nativeBinding = require("lancedb-linux-riscv64-musl");
             }
           } catch (e) {
-            loadError = e
+            loadError = e;
           }
         } else {
           localFileExisted = existsSync(
-            join(__dirname, 'lancedb-nodejs.linux-riscv64-gnu.node')
-          )
+            join(__dirname, "lancedb-nodejs.linux-riscv64-gnu.node"),
+          );
           try {
             if (localFileExisted) {
-              nativeBinding = require('./lancedb-nodejs.linux-riscv64-gnu.node')
+              nativeBinding = require("./lancedb-nodejs.linux-riscv64-gnu.node");
             } else {
-              nativeBinding = require('lancedb-linux-riscv64-gnu')
+              nativeBinding = require("lancedb-linux-riscv64-gnu");
             }
           } catch (e) {
-            loadError = e
+            loadError = e;
           }
         }
-        break
-      case 's390x':
+        break;
+      case "s390x":
         localFileExisted = existsSync(
-          join(__dirname, 'lancedb-nodejs.linux-s390x-gnu.node')
-        )
+          join(__dirname, "lancedb-nodejs.linux-s390x-gnu.node"),
+        );
         try {
           if (localFileExisted) {
-            nativeBinding = require('./lancedb-nodejs.linux-s390x-gnu.node')
+            nativeBinding = require("./lancedb-nodejs.linux-s390x-gnu.node");
           } else {
-            nativeBinding = require('lancedb-linux-s390x-gnu')
+            nativeBinding = require("lancedb-linux-s390x-gnu");
           }
         } catch (e) {
-          loadError = e
+          loadError = e;
         }
-        break
+        break;
       default:
-        throw new Error(`Unsupported architecture on Linux: ${arch}`)
+        throw new Error(`Unsupported architecture on Linux: ${arch}`);
     }
-    break
+    break;
   default:
-    throw new Error(`Unsupported OS: ${platform}, architecture: ${arch}`)
+    throw new Error(`Unsupported OS: ${platform}, architecture: ${arch}`);
 }
 
 if (!nativeBinding) {
   if (loadError) {
-    throw loadError
+    throw loadError;
   }
-  throw new Error(`Failed to load native binding`)
+  throw new Error(`Failed to load native binding`);
 }
 
-const { Connection, IndexType, MetricType, IndexBuilder, RecordBatchIterator, Query, Table, WriteMode, connect } = nativeBinding
+const {
+  Connection,
+  Index,
+  RecordBatchIterator,
+  Query,
+  VectorQuery,
+  Table,
+  WriteMode,
+  connect,
+} = nativeBinding;
 
-module.exports.Connection = Connection
-module.exports.IndexType = IndexType
-module.exports.MetricType = MetricType
-module.exports.IndexBuilder = IndexBuilder
-module.exports.RecordBatchIterator = RecordBatchIterator
-module.exports.Query = Query
-module.exports.Table = Table
-module.exports.WriteMode = WriteMode
-module.exports.connect = connect
+module.exports.Connection = Connection;
+module.exports.Index = Index;
+module.exports.RecordBatchIterator = RecordBatchIterator;
+module.exports.Query = Query;
+module.exports.VectorQuery = VectorQuery;
+module.exports.Table = Table;
+module.exports.WriteMode = WriteMode;
+module.exports.connect = connect;

nodejs/lancedb/query.ts

@@ -17,18 +17,15 @@ import {
   RecordBatchIterator as NativeBatchIterator,
   Query as NativeQuery,
   Table as NativeTable,
+  VectorQuery as NativeVectorQuery,
 } from "./native";
-
+import { type IvfPqOptions } from "./indices";
 class RecordBatchIterator implements AsyncIterator<RecordBatch> {
   private promisedInner?: Promise<NativeBatchIterator>;
   private inner?: NativeBatchIterator;
 
-  constructor(
-    inner?: NativeBatchIterator,
-    promise?: Promise<NativeBatchIterator>,
-  ) {
+  constructor(promise?: Promise<NativeBatchIterator>) {
     // TODO: check promise reliably so we dont need to pass two arguments.
-    this.inner = inner;
     this.promisedInner = promise;
   }
 
@@ -53,82 +50,113 @@ class RecordBatchIterator implements AsyncIterator<RecordBatch> {
 }
 /* eslint-enable */
 
-/** Query executor */
-export class Query implements AsyncIterable<RecordBatch> {
-  private readonly inner: NativeQuery;
+/** Common methods supported by all query types */
+export class QueryBase<
+  NativeQueryType extends NativeQuery | NativeVectorQuery,
+  QueryType,
+> implements AsyncIterable<RecordBatch>
+{
+  protected constructor(protected inner: NativeQueryType) {}
 
-  constructor(tbl: NativeTable) {
-    this.inner = tbl.query();
+  /**
+   * A filter statement to be applied to this query.
+   *
+   * The filter should be supplied as an SQL query string.  For example:
+   * @example
+   * x > 10
+   * y > 0 AND y < 100
+   * x > 5 OR y = 'test'
+   *
+   * Filtering performance can often be improved by creating a scalar index
+   * on the filter column(s).
+   */
+  where(predicate: string): QueryType {
+    this.inner.onlyIf(predicate);
+    return this as unknown as QueryType;
   }
 
-  /** Set the column to run query. */
-  column(column: string): Query {
-    this.inner.column(column);
-    return this;
+  /**
+   * Return only the specified columns.
+   *
+   * By default a query will return all columns from the table.  However, this can have
+   * a very significant impact on latency.  LanceDb stores data in a columnar fashion.  This
+   * means we can finely tune our I/O to select exactly the columns we need.
+   *
+   * As a best practice you should always limit queries to the columns that you need.  If you
+   * pass in an array of column names then only those columns will be returned.
+   *
+   * You can also use this method to create new "dynamic" columns based on your existing columns.
+   * For example, you may not care about "a" or "b" but instead simply want "a + b".  This is often
+   * seen in the SELECT clause of an SQL query (e.g. `SELECT a+b FROM my_table`).
+   *
+   * To create dynamic columns you can pass in a Map<string, string>.  A column will be returned
+   * for each entry in the map.  The key provides the name of the column.  The value is
+   * an SQL string used to specify how the column is calculated.
+   *
+   * For example, an SQL query might state `SELECT a + b AS combined, c`.  The equivalent
+   * input to this method would be:
+   * @example
+   * new Map([["combined", "a + b"], ["c", "c"]])
+   *
+   * Columns will always be returned in the order given, even if that order is different than
+   * the order used when adding the data.
+   *
+   * Note that you can pass in a `Record<string, string>` (e.g. an object literal). This method
+   * uses `Object.entries` which should preserve the insertion order of the object.  However,
+   * object insertion order is easy to get wrong and `Map` is more foolproof.
+   */
+  select(
+    columns: string[] | Map<string, string> | Record<string, string>,
+  ): QueryType {
+    let columnTuples: [string, string][];
+    if (Array.isArray(columns)) {
+      columnTuples = columns.map((c) => [c, c]);
+    } else if (columns instanceof Map) {
+      columnTuples = Array.from(columns.entries());
+    } else {
+      columnTuples = Object.entries(columns);
+    }
+    this.inner.select(columnTuples);
+    return this as unknown as QueryType;
   }
 
-  /** Set the filter predicate, only returns the results that satisfy the filter.
+  /**
+   * Set the maximum number of results to return.
+   *
+   * By default, a plain search has no limit.  If this method is not
+   * called then every valid row from the table will be returned.
+   */
+  limit(limit: number): QueryType {
+    this.inner.limit(limit);
+    return this as unknown as QueryType;
+  }
+
+  protected nativeExecute(): Promise<NativeBatchIterator> {
+    return this.inner.execute();
+  }
+
+  /**
+   * Execute the query and return the results as an @see {@link AsyncIterator}
+   * of @see {@link RecordBatch}.
+   *
+   * By default, LanceDb will use many threads to calculate results and, when
+   * the result set is large, multiple batches will be processed at one time.
+   * This readahead is limited however and backpressure will be applied if this
+   * stream is consumed slowly (this constrains the maximum memory used by a
+   * single query)
    *
    */
-  filter(predicate: string): Query {
-    this.inner.filter(predicate);
-    return this;
+  protected execute(): RecordBatchIterator {
+    return new RecordBatchIterator(this.nativeExecute());
   }
 
-  /**
-   * Select the columns to return. If not set, all columns are returned.
-   */
-  select(columns: string[]): Query {
-    this.inner.select(columns);
-    return this;
+  // eslint-disable-next-line @typescript-eslint/no-explicit-any
+  [Symbol.asyncIterator](): AsyncIterator<RecordBatch<any>> {
+    const promise = this.nativeExecute();
+    return new RecordBatchIterator(promise);
   }
 
-  /**
-   * Set the limit of rows to return.
-   */
-  limit(limit: number): Query {
-    this.inner.limit(limit);
-    return this;
-  }
-
-  prefilter(prefilter: boolean): Query {
-    this.inner.prefilter(prefilter);
-    return this;
-  }
-
-  /**
-   * Set the query vector.
-   */
-  nearestTo(vector: number[]): Query {
-    this.inner.nearestTo(Float32Array.from(vector));
-    return this;
-  }
-
-  /**
-   * Set the number of IVF partitions to use for the query.
-   */
-  nprobes(nprobes: number): Query {
-    this.inner.nprobes(nprobes);
-    return this;
-  }
-
-  /**
-   * Set the refine factor for the query.
-   */
-  refineFactor(refineFactor: number): Query {
-    this.inner.refineFactor(refineFactor);
-    return this;
-  }
-
-  /**
-   * Execute the query and return the results as an AsyncIterator.
-   */
-  async executeStream(): Promise<RecordBatchIterator> {
-    const inner = await this.inner.executeStream();
-    return new RecordBatchIterator(inner);
-  }
-
-  /** Collect the results as an Arrow Table. */
+  /** Collect the results as an Arrow @see {@link ArrowTable}. */
   async toArrow(): Promise<ArrowTable> {
     const batches = [];
     for await (const batch of this) {
@@ -137,18 +165,211 @@ export class Query implements AsyncIterable<RecordBatch> {
     return new ArrowTable(batches);
   }
 
-  /** Returns a JSON Array of All results.
-   *
-   */
+  /** Collect the results as an array of objects. */
   async toArray(): Promise<unknown[]> {
     const tbl = await this.toArrow();
     // eslint-disable-next-line @typescript-eslint/no-unsafe-return
     return tbl.toArray();
   }
+}
 
-  // eslint-disable-next-line @typescript-eslint/no-explicit-any
-  [Symbol.asyncIterator](): AsyncIterator<RecordBatch<any>> {
-    const promise = this.inner.executeStream();
-    return new RecordBatchIterator(undefined, promise);
+/**
+ * An interface for a query that can be executed
+ *
+ * Supported by all query types
+ */
+export interface ExecutableQuery {}
+
+/**
+ * A builder used to construct a vector search
+ *
+ * This builder can be reused to execute the query many times.
+ */
+export class VectorQuery extends QueryBase<NativeVectorQuery, VectorQuery> {
+  constructor(inner: NativeVectorQuery) {
+    super(inner);
+  }
+
+  /**
+   * Set the number of partitions to search (probe)
+   *
+   * This argument is only used when the vector column has an IVF PQ index.
+   * If there is no index then this value is ignored.
+   *
+   * The IVF stage of IVF PQ divides the input into partitions (clusters) of
+   * related values.
+   *
+   * The partition whose centroids are closest to the query vector will be
+   * exhaustiely searched to find matches.  This parameter controls how many
+   * partitions should be searched.
+   *
+   * Increasing this value will increase the recall of your query but will
+   * also increase the latency of your query.  The default value is 20.  This
+   * default is good for many cases but the best value to use will depend on
+   * your data and the recall that you need to achieve.
+   *
+   * For best results we recommend tuning this parameter with a benchmark against
+   * your actual data to find the smallest possible value that will still give
+   * you the desired recall.
+   */
+  nprobes(nprobes: number): VectorQuery {
+    this.inner.nprobes(nprobes);
+    return this;
+  }
+
+  /**
+   * Set the vector column to query
+   *
+   * This controls which column is compared to the query vector supplied in
+   * the call to @see {@link Query#nearestTo}
+   *
+   * This parameter must be specified if the table has more than one column
+   * whose data type is a fixed-size-list of floats.
+   */
+  column(column: string): VectorQuery {
+    this.inner.column(column);
+    return this;
+  }
+
+  /**
+   * Set the distance metric to use
+   *
+   * When performing a vector search we try and find the "nearest" vectors according
+   * to some kind of distance metric.  This parameter controls which distance metric to
+   * use.  See @see {@link IvfPqOptions.distanceType} for more details on the different
+   * distance metrics available.
+   *
+   * Note: if there is a vector index then the distance type used MUST match the distance
+   * type used to train the vector index.  If this is not done then the results will be
+   * invalid.
+   *
+   * By default "l2" is used.
+   */
+  distanceType(distanceType: string): VectorQuery {
+    this.inner.distanceType(distanceType);
+    return this;
+  }
+
+  /**
+   * A multiplier to control how many additional rows are taken during the refine step
+   *
+   * This argument is only used when the vector column has an IVF PQ index.
+   * If there is no index then this value is ignored.
+   *
+   * An IVF PQ index stores compressed (quantized) values.  They query vector is compared
+   * against these values and, since they are compressed, the comparison is inaccurate.
+   *
+   * This parameter can be used to refine the results.  It can improve both improve recall
+   * and correct the ordering of the nearest results.
+   *
+   * To refine results LanceDb will first perform an ANN search to find the nearest
+   * `limit` * `refine_factor` results.  In other words, if `refine_factor` is 3 and
+   * `limit` is the default (10) then the first 30 results will be selected.  LanceDb
+   * then fetches the full, uncompressed, values for these 30 results.  The results are
+   * then reordered by the true distance and only the nearest 10 are kept.
+   *
+   * Note: there is a difference between calling this method with a value of 1 and never
+   * calling this method at all.  Calling this method with any value will have an impact
+   * on your search latency.  When you call this method with a `refine_factor` of 1 then
+   * LanceDb still needs to fetch the full, uncompressed, values so that it can potentially
+   * reorder the results.
+   *
+   * Note: if this method is NOT called then the distances returned in the _distance column
+   * will be approximate distances based on the comparison of the quantized query vector
+   * and the quantized result vectors.  This can be considerably different than the true
+   * distance between the query vector and the actual uncompressed vector.
+   */
+  refineFactor(refineFactor: number): VectorQuery {
+    this.inner.refineFactor(refineFactor);
+    return this;
+  }
+
+  /**
+   * If this is called then filtering will happen after the vector search instead of
+   * before.
+   *
+   * By default filtering will be performed before the vector search.  This is how
+   * filtering is typically understood to work.  This prefilter step does add some
+   * additional latency.  Creating a scalar index on the filter column(s) can
+   * often improve this latency.  However, sometimes a filter is too complex or scalar
+   * indices cannot be applied to the column.  In these cases postfiltering can be
+   * used instead of prefiltering to improve latency.
+   *
+   * Post filtering applies the filter to the results of the vector search.  This means
+   * we only run the filter on a much smaller set of data.  However, it can cause the
+   * query to return fewer than `limit` results (or even no results) if none of the nearest
+   * results match the filter.
+   *
+   * Post filtering happens during the "refine stage" (described in more detail in
+   * @see {@link VectorQuery#refineFactor}).  This means that setting a higher refine
+   * factor can often help restore some of the results lost by post filtering.
+   */
+  postfilter(): VectorQuery {
+    this.inner.postfilter();
+    return this;
+  }
+
+  /**
+   * If this is called then any vector index is skipped
+   *
+   * An exhaustive (flat) search will be performed.  The query vector will
+   * be compared to every vector in the table.  At high scales this can be
+   * expensive.  However, this is often still useful.  For example, skipping
+   * the vector index can give you ground truth results which you can use to
+   * calculate your recall to select an appropriate value for nprobes.
+   */
+  bypassVectorIndex(): VectorQuery {
+    this.inner.bypassVectorIndex();
+    return this;
+  }
+}
+
+/** A builder for LanceDB queries. */
+export class Query extends QueryBase<NativeQuery, Query> {
+  constructor(tbl: NativeTable) {
+    super(tbl.query());
+  }
+
+  /**
+   * Find the nearest vectors to the given query vector.
+   *
+   * This converts the query from a plain query to a vector query.
+   *
+   * This method will attempt to convert the input to the query vector
+   * expected by the embedding model.  If the input cannot be converted
+   * then an error will be thrown.
+   *
+   * By default, there is no embedding model, and the input should be
+   * an array-like object of numbers (something that can be used as input
+   * to Float32Array.from)
+   *
+   * If there is only one vector column (a column whose data type is a
+   * fixed size list of floats) then the column does not need to be specified.
+   * If there is more than one vector column you must use
+   * @see {@link VectorQuery#column}  to specify which column you would like
+   * to compare with.
+   *
+   * If no index has been created on the vector column then a vector query
+   * will perform a distance comparison between the query vector and every
+   * vector in the database and then sort the results.  This is sometimes
+   * called a "flat search"
+   *
+   * For small databases, with a few hundred thousand vectors or less, this can
+   * be reasonably fast.  In larger databases you should create a vector index
+   * on the column.  If there is a vector index then an "approximate" nearest
+   * neighbor search (frequently called an ANN search) will be performed.  This
+   * search is much faster, but the results will be approximate.
+   *
+   * The query can be further parameterized using the returned builder.  There
+   * 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.  @see {@link Query#limit}
+   */
+  nearestTo(vector: unknown): VectorQuery {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const vectorQuery = this.inner.nearestTo(Float32Array.from(vector as any));
+    return new VectorQuery(vectorQuery);
   }
 }

nodejs/lancedb/sanitize.ts

@@ -481,6 +481,13 @@ function sanitizeField(fieldLike: unknown): Field {
   return new Field(name, type, nullable, metadata);
 }
 
+/**
+ * Convert something schemaLike into a Schema instance
+ *
+ * This method is often needed even when the caller is using a Schema
+ * instance because they might be using a different instance of apache-arrow
+ * than lancedb is using.
+ */
 export function sanitizeSchema(schemaLike: unknown): Schema {
   if (schemaLike instanceof Schema) {
     return schemaLike;

nodejs/lancedb/table.ts

@@ -16,23 +16,40 @@ import { Schema, tableFromIPC } from "apache-arrow";
 import {
   AddColumnsSql,
   ColumnAlteration,
+  IndexConfig,
   Table as _NativeTable,
 } from "./native";
-import { Query } from "./query";
-import { IndexBuilder } from "./indexer";
+import { Query, VectorQuery } from "./query";
+import { IndexOptions } from "./indices";
 import { Data, fromDataToBuffer } from "./arrow";
 
+export { IndexConfig } from "./native";
 /**
  * Options for adding data to a table.
  */
 export interface AddDataOptions {
-  /** If "append" (the default) then the new data will be added to the table
+  /**
+   * If "append" (the default) then the new data will be added to the table
    *
    * If "overwrite" then the new data will replace the existing data in the table.
    */
   mode: "append" | "overwrite";
 }
 
+export interface UpdateOptions {
+  /**
+   * A filter that limits the scope of the update.
+   *
+   * This should be an SQL filter expression.
+   *
+   * Only rows that satisfy the expression will be updated.
+   *
+   * For example, this could be 'my_col == 0' to replace all instances
+   * of 0 in a column with some other default value.
+   */
+  where: string;
+}
+
 /**
  * A Table is a collection of Records in a LanceDB Database.
  *
@@ -58,7 +75,8 @@ export class Table {
     return this.inner.isOpen();
   }
 
-  /** Close the table, releasing any underlying resources.
+  /**
+   * Close the table, releasing any underlying resources.
    *
    * It is safe to call this method multiple times.
    *
@@ -82,9 +100,7 @@ export class Table {
 
   /**
    * Insert records into this Table.
-   *
    * @param {Data} data Records to be inserted into the Table
-   * @return The number of rows added to the table
    */
   async add(data: Data, options?: Partial<AddDataOptions>): Promise<void> {
     const mode = options?.mode ?? "append";
@@ -93,6 +109,45 @@ export class Table {
     await this.inner.add(buffer, mode);
   }
 
+  /**
+   * Update existing records in the Table
+   *
+   * An update operation can be used to adjust existing values.  Use the
+   * returned builder to specify which columns to update.  The new value
+   * can be a literal value (e.g. replacing nulls with some default value)
+   * or an expression applied to the old value (e.g. incrementing a value)
+   *
+   * An optional condition can be specified (e.g. "only update if the old
+   * value is 0")
+   *
+   * Note: if your condition is something like "some_id_column == 7" and
+   * you are updating many rows (with different ids) then you will get
+   * better performance with a single [`merge_insert`] call instead of
+   * repeatedly calilng this method.
+   * @param {Map<string, string> | Record<string, string>} updates - the
+   * columns to update
+   *
+   * Keys in the map should specify the name of the column to update.
+   * Values in the map provide the new value of the column.  These can
+   * be SQL literal strings (e.g. "7" or "'foo'") or they can be expressions
+   * based on the row being updated (e.g. "my_col + 1")
+   * @param {Partial<UpdateOptions>} options - additional options to control
+   * the update behavior
+   */
+  async update(
+    updates: Map<string, string> | Record<string, string>,
+    options?: Partial<UpdateOptions>,
+  ) {
+    const onlyIf = options?.where;
+    let columns: [string, string][];
+    if (updates instanceof Map) {
+      columns = Array.from(updates.entries());
+    } else {
+      columns = Object.entries(updates);
+    }
+    await this.inner.update(onlyIf, columns);
+  }
+
   /** Count the total number of rows in the dataset. */
   async countRows(filter?: string): Promise<number> {
     return await this.inner.countRows(filter);
@@ -103,106 +158,105 @@ export class Table {
     await this.inner.delete(predicate);
   }
 
-  /** Create an index over the columns.
-   *
-   * @param {string} column The column to create the index on. If not specified,
-   *                        it will create an index on vector field.
+  /**
+   * Create an index to speed up queries.
    *
+   * Indices can be created on vector columns or scalar columns.
+   * Indices on vector columns will speed up vector searches.
+   * Indices on scalar columns will speed up filtering (in both
+   * vector and non-vector searches)
    * @example
-   *
-   * By default, it creates vector idnex on one vector column.
-   *
-   * ```typescript
+   * // If the column has a vector (fixed size list) data type then
+   * // an IvfPq vector index will be created.
    * const table = await conn.openTable("my_table");
-   * await table.createIndex().build();
-   * ```
-   *
-   * You can specify `IVF_PQ` parameters via `ivf_pq({})` call.
-   * ```typescript
+   * await table.createIndex(["vector"]);
+   * @example
+   * // For advanced control over vector index creation you can specify
+   * // the index type and options.
    * const table = await conn.openTable("my_table");
-   * await table.createIndex("my_vec_col")
+   * await table.createIndex(["vector"], I)
    *   .ivf_pq({ num_partitions: 128, num_sub_vectors: 16 })
    *   .build();
-   * ```
-   *
-   * Or create a Scalar index
-   *
-   * ```typescript
+   * @example
+   * // Or create a Scalar index
    * await table.createIndex("my_float_col").build();
-   * ```
    */
-  createIndex(column?: string): IndexBuilder {
-    let builder = new IndexBuilder(this.inner);
-    if (column !== undefined) {
-      builder = builder.column(column);
-    }
-    return builder;
+  async createIndex(column: string, options?: Partial<IndexOptions>) {
+    // Bit of a hack to get around the fact that TS has no package-scope.
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    const nativeIndex = (options?.config as any)?.inner;
+    await this.inner.createIndex(nativeIndex, column, options?.replace);
   }
 
   /**
-   * Create a generic {@link Query} Builder.
+   * Create a {@link Query} Builder.
+   *
+   * Queries allow you to search your existing data.  By default the query will
+   * return all the data in the table in no particular order.  The builder
+   * returned by this method can be used to control the query using filtering,
+   * vector similarity, sorting, and more.
+   *
+   * Note: By default, all columns are returned.  For best performance, you should
+   * only fetch the columns you need.  See [`Query::select_with_projection`] for
+   * more details.
    *
    * When appropriate, various indices and statistics based pruning will be used to
    * accelerate the query.
-   *
    * @example
-   *
-   * ### Run a SQL-style query
-   * ```typescript
+   * // SQL-style filtering
+   * //
+   * // This query will return up to 1000 rows whose value in the `id` column
+   * // is greater than 5.  LanceDb supports a broad set of filtering functions.
    * for await (const batch of table.query()
    *                          .filter("id > 1").select(["id"]).limit(20)) {
    *  console.log(batch);
    * }
-   * ```
-   *
-   * ### Run Top-10 vector similarity search
-   * ```typescript
+   * @example
+   * // Vector Similarity Search
+   * //
+   * // This example will find the 10 rows whose value in the "vector" column are
+   * // closest to the query vector [1.0, 2.0, 3.0].  If an index has been created
+   * // on the "vector" column then this will perform an ANN search.
+   * //
+   * // The `refine_factor` and `nprobes` methods are used to control the recall /
+   * // latency tradeoff of the search.
    * for await (const batch of table.query()
    *                    .nearestTo([1, 2, 3])
    *                    .refineFactor(5).nprobe(10)
    *                    .limit(10)) {
    *  console.log(batch);
    * }
-   *```
-   *
-   * ### Scan the full dataset
-   * ```typescript
+   * @example
+   * // Scan the full dataset
+   * //
+   * // This query will return everything in the table in no particular order.
    * for await (const batch of table.query()) {
    *   console.log(batch);
    * }
-   *
-   * ### Return the full dataset as Arrow Table
-   * ```typescript
-   * let arrowTbl = await table.query().nearestTo([1.0, 2.0, 0.5, 6.7]).toArrow();
-   * ```
-   *
-   * @returns {@link Query}
+   * @returns {Query} A builder that can be used to parameterize the query
    */
   query(): Query {
     return new Query(this.inner);
   }
 
-  /** Search the table with a given query vector.
+  /**
+   * Search the table with a given query vector.
    *
-   * This is a convenience method for preparing an ANN {@link Query}.
+   * This is a convenience method for preparing a vector query and
+   * is the same thing as calling `nearestTo` on the builder returned
+   * by `query`.  @see {@link Query#nearestTo} for more details.
    */
-  search(vector: number[], column?: string): Query {
-    const q = this.query();
-    q.nearestTo(vector);
-    if (column !== undefined) {
-      q.column(column);
-    }
-    return q;
+  vectorSearch(vector: unknown): VectorQuery {
+    return this.query().nearestTo(vector);
   }
 
   // TODO: Support BatchUDF
   /**
    * Add new columns with defined values.
-   *
-   * @param newColumnTransforms pairs of column names and the 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 in the table.
+   * @param {AddColumnsSql[]} newColumnTransforms pairs of column names and
+   * the 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 in the table.
    */
   async addColumns(newColumnTransforms: AddColumnsSql[]): Promise<void> {
     await this.inner.addColumns(newColumnTransforms);
@@ -210,8 +264,8 @@ export class Table {
 
   /**
    * Alter the name or nullability of columns.
-   *
-   * @param columnAlterations One or more alterations to apply to columns.
+   * @param {ColumnAlteration[]} columnAlterations One or more alterations to
+   * apply to columns.
    */
   async alterColumns(columnAlterations: ColumnAlteration[]): Promise<void> {
     await this.inner.alterColumns(columnAlterations);
@@ -224,12 +278,76 @@ export class Table {
    * underlying storage. In order to remove the data, you must subsequently
    * call ``compact_files`` to rewrite the data without the removed columns and
    * then call ``cleanup_files`` to remove the old files.
-   *
-   * @param columnNames The names of the columns to drop. These can be nested
-   *                    column references (e.g. "a.b.c") or top-level column
-   *                    names (e.g. "a").
+   * @param {string[]} columnNames The names of the columns to drop. These can
+   * be nested column references (e.g. "a.b.c") or top-level column names
+   * (e.g. "a").
    */
   async dropColumns(columnNames: string[]): Promise<void> {
     await this.inner.dropColumns(columnNames);
   }
+
+  /**
+   * Retrieve the version of the table
+   *
+   * LanceDb supports versioning.  Every operation that modifies the table increases
+   * version.  As long as a version hasn't been deleted you can `[Self::checkout]` that
+   * version to view the data at that point.  In addition, you can `[Self::restore]` the
+   * version to replace the current table with a previous version.
+   */
+  async version(): Promise<number> {
+    return await this.inner.version();
+  }
+
+  /**
+   * Checks out a specific version of the Table
+   *
+   * Any read operation on the table will now access the data at the checked out version.
+   * As a consequence, calling this method will disable any read consistency interval
+   * that was previously set.
+   *
+   * This is a read-only operation that turns the table into a sort of "view"
+   * or "detached head".  Other table instances will not be affected.  To make the 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.
+   *
+   * To return the table to a normal state use `[Self::checkout_latest]`
+   */
+  async checkout(version: number): Promise<void> {
+    await this.inner.checkout(version);
+  }
+
+  /**
+   * Ensures the table is pointing at the latest version
+   *
+   * This can be used to manually update a table when the read_consistency_interval is None
+   * It can also be used to undo a `[Self::checkout]` operation
+   */
+  async checkoutLatest(): Promise<void> {
+    await this.inner.checkoutLatest();
+  }
+
+  /**
+   * Restore the table to the currently checked out version
+   *
+   * This operation will fail if checkout has not been called previously
+   *
+   * This operation will overwrite the latest version of the table with a
+   * previous version.  Any changes made since the checked out version will
+   * no longer be visible.
+   *
+   * Once the operation concludes the table will no longer be in a checked
+   * out state and the read_consistency_interval, if any, will apply.
+   */
+  async restore(): Promise<void> {
+    await this.inner.restore();
+  }
+
+  /**
+   * List all indices that have been created with Self::create_index
+   */
+  async listIndices(): Promise<IndexConfig[]> {
+    return await this.inner.listIndices();
+  }
 }

nodejs/package-lock.json

@@ -26,6 +26,7 @@
         "apache-arrow-old": "npm:apache-arrow@13.0.0",
         "eslint": "^8.57.0",
         "eslint-config-prettier": "^9.1.0",
+        "eslint-plugin-jsdoc": "^48.2.1",
         "jest": "^29.7.0",
         "prettier": "^3.1.0",
         "tmp": "^0.2.3",
@@ -755,6 +756,20 @@
       "integrity": "sha512-0hYQ8SB4Db5zvZB4axdMHGwEaQjkZzFjQiN9LVYvIFB2nSUHW9tYpxWriPrWDASIxiaXax83REcLxuSdnGPZtw==",
       "dev": true
     },
+    "node_modules/@es-joy/jsdoccomment": {
+      "version": "0.42.0",
+      "resolved": "https://registry.npmjs.org/@es-joy/jsdoccomment/-/jsdoccomment-0.42.0.tgz",
+      "integrity": "sha512-R1w57YlVA6+YE01wch3GPYn6bCsrOV3YW/5oGGE2tmX6JcL9Nr+b5IikrjMPF+v9CV3ay+obImEdsDhovhJrzw==",
+      "dev": true,
+      "dependencies": {
+        "comment-parser": "1.4.1",
+        "esquery": "^1.5.0",
+        "jsdoc-type-pratt-parser": "~4.0.0"
+      },
+      "engines": {
+        "node": ">=16"
+      }
+    },
     "node_modules/@eslint-community/eslint-utils": {
       "version": "4.4.0",
       "resolved": "https://registry.npmjs.org/@eslint-community/eslint-utils/-/eslint-utils-4.4.0.tgz",
@@ -1948,6 +1963,15 @@
       "integrity": "sha512-cumHmIAf6On83X7yP+LrsEyUOf/YlociZelmpRYaGFydoaPdxdt80MAbu6vWerQT2COCp2nPvHdsbD7tHn/YlQ==",
       "dev": true
     },
+    "node_modules/are-docs-informative": {
+      "version": "0.0.2",
+      "resolved": "https://registry.npmjs.org/are-docs-informative/-/are-docs-informative-0.0.2.tgz",
+      "integrity": "sha512-ixiS0nLNNG5jNQzgZJNoUpBKdo9yTYZMGJ+QgT2jmjR7G7+QHRCc4v6LQ3NgE7EBJq+o0ams3waJwkrlBom8Ig==",
+      "dev": true,
+      "engines": {
+        "node": ">=14"
+      }
+    },
     "node_modules/argparse": {
       "version": "1.0.10",
       "resolved": "https://registry.npmjs.org/argparse/-/argparse-1.0.10.tgz",
@@ -2189,6 +2213,18 @@
       "integrity": "sha512-E+XQCRwSbaaiChtv6k6Dwgc+bx+Bs6vuKJHHl5kox/BaKbhiXzqQOwK4cO22yElGp2OCmjwVhT3HmxgyPGnJfQ==",
       "dev": true
     },
+    "node_modules/builtin-modules": {
+      "version": "3.3.0",
+      "resolved": "https://registry.npmjs.org/builtin-modules/-/builtin-modules-3.3.0.tgz",
+      "integrity": "sha512-zhaCDicdLuWN5UbN5IMnFqNMhNfo919sH85y2/ea+5Yg9TsTkeZxpL+JLbp6cgYFS4sRLp3YV4S6yDuqVWHYOw==",
+      "dev": true,
+      "engines": {
+        "node": ">=6"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
     "node_modules/camelcase": {
       "version": "5.3.1",
       "resolved": "https://registry.npmjs.org/camelcase/-/camelcase-5.3.1.tgz",
@@ -2373,6 +2409,15 @@
         "node": ">=12.17"
       }
     },
+    "node_modules/comment-parser": {
+      "version": "1.4.1",
+      "resolved": "https://registry.npmjs.org/comment-parser/-/comment-parser-1.4.1.tgz",
+      "integrity": "sha512-buhp5kePrmda3vhc5B9t7pUQXAb2Tnd0qgpkIhPhkHXxJpiPJ11H0ZEU0oBpJ2QztSbzG/ZxMj/CHsYJqRHmyg==",
+      "dev": true,
+      "engines": {
+        "node": ">= 12.0.0"
+      }
+    },
     "node_modules/concat-map": {
       "version": "0.0.1",
       "resolved": "https://registry.npmjs.org/concat-map/-/concat-map-0.0.1.tgz",
@@ -2660,6 +2705,29 @@
         "eslint": ">=7.0.0"
       }
     },
+    "node_modules/eslint-plugin-jsdoc": {
+      "version": "48.2.1",
+      "resolved": "https://registry.npmjs.org/eslint-plugin-jsdoc/-/eslint-plugin-jsdoc-48.2.1.tgz",
+      "integrity": "sha512-iUvbcyDZSO/9xSuRv2HQBw++8VkV/pt3UWtX9cpPH0l7GKPq78QC/6+PmyQHHvNZaTjAce6QVciEbnc6J/zH5g==",
+      "dev": true,
+      "dependencies": {
+        "@es-joy/jsdoccomment": "~0.42.0",
+        "are-docs-informative": "^0.0.2",
+        "comment-parser": "1.4.1",
+        "debug": "^4.3.4",
+        "escape-string-regexp": "^4.0.0",
+        "esquery": "^1.5.0",
+        "is-builtin-module": "^3.2.1",
+        "semver": "^7.6.0",
+        "spdx-expression-parse": "^4.0.0"
+      },
+      "engines": {
+        "node": ">=18"
+      },
+      "peerDependencies": {
+        "eslint": "^7.0.0 || ^8.0.0 || ^9.0.0"
+      }
+    },
     "node_modules/eslint-scope": {
       "version": "7.2.2",
       "resolved": "https://registry.npmjs.org/eslint-scope/-/eslint-scope-7.2.2.tgz",
@@ -3299,6 +3367,21 @@
       "integrity": "sha512-NcdALwpXkTm5Zvvbk7owOUSvVvBKDgKP5/ewfXEznmQFfs4ZRmanOeKBTjRVjka3QFoN6XJ+9F3USqfHqTaU5w==",
       "optional": true
     },
+    "node_modules/is-builtin-module": {
+      "version": "3.2.1",
+      "resolved": "https://registry.npmjs.org/is-builtin-module/-/is-builtin-module-3.2.1.tgz",
+      "integrity": "sha512-BSLE3HnV2syZ0FK0iMA/yUGplUeMmNz4AW5fnTunbCIqZi4vG3WjJT9FHMy5D69xmAYBHXQhJdALdpwVxV501A==",
+      "dev": true,
+      "dependencies": {
+        "builtin-modules": "^3.3.0"
+      },
+      "engines": {
+        "node": ">=6"
+      },
+      "funding": {
+        "url": "https://github.com/sponsors/sindresorhus"
+      }
+    },
     "node_modules/is-core-module": {
       "version": "2.13.1",
       "resolved": "https://registry.npmjs.org/is-core-module/-/is-core-module-2.13.1.tgz",
@@ -4172,6 +4255,15 @@
         "js-yaml": "bin/js-yaml.js"
       }
     },
+    "node_modules/jsdoc-type-pratt-parser": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/jsdoc-type-pratt-parser/-/jsdoc-type-pratt-parser-4.0.0.tgz",
+      "integrity": "sha512-YtOli5Cmzy3q4dP26GraSOeAhqecewG04hoO8DY56CH4KJ9Fvv5qKWUCCo3HZob7esJQHCv6/+bnTy72xZZaVQ==",
+      "dev": true,
+      "engines": {
+        "node": ">=12.0.0"
+      }
+    },
     "node_modules/jsesc": {
       "version": "2.5.2",
       "resolved": "https://registry.npmjs.org/jsesc/-/jsesc-2.5.2.tgz",
@@ -5018,9 +5110,9 @@
       }
     },
     "node_modules/semver": {
-      "version": "7.5.4",
-      "resolved": "https://registry.npmjs.org/semver/-/semver-7.5.4.tgz",
-      "integrity": "sha512-1bCSESV6Pv+i21Hvpxp3Dx+pSD8lIPt8uVjRrxAUt/nbswYc+tK6Y2btiULjd4+fnq15PX+nqQDC7Oft7WkwcA==",
+      "version": "7.6.0",
+      "resolved": "https://registry.npmjs.org/semver/-/semver-7.6.0.tgz",
+      "integrity": "sha512-EnwXhrlwXMk9gKu5/flx5sv/an57AkRplG3hTK68W7FRDN+k+OWBj65M7719OkA82XLBxrcX0KSHj+X5COhOVg==",
       "dev": true,
       "dependencies": {
         "lru-cache": "^6.0.0"
@@ -5105,6 +5197,28 @@
         "source-map": "^0.6.0"
       }
     },
+    "node_modules/spdx-exceptions": {
+      "version": "2.5.0",
+      "resolved": "https://registry.npmjs.org/spdx-exceptions/-/spdx-exceptions-2.5.0.tgz",
+      "integrity": "sha512-PiU42r+xO4UbUS1buo3LPJkjlO7430Xn5SVAhdpzzsPHsjbYVflnnFdATgabnLude+Cqu25p6N+g2lw/PFsa4w==",
+      "dev": true
+    },
+    "node_modules/spdx-expression-parse": {
+      "version": "4.0.0",
+      "resolved": "https://registry.npmjs.org/spdx-expression-parse/-/spdx-expression-parse-4.0.0.tgz",
+      "integrity": "sha512-Clya5JIij/7C6bRR22+tnGXbc4VKlibKSVj2iHvVeX5iMW7s1SIQlqu699JkODJJIhh/pUu8L0/VLh8xflD+LQ==",
+      "dev": true,
+      "dependencies": {
+        "spdx-exceptions": "^2.1.0",
+        "spdx-license-ids": "^3.0.0"
+      }
+    },
+    "node_modules/spdx-license-ids": {
+      "version": "3.0.17",
+      "resolved": "https://registry.npmjs.org/spdx-license-ids/-/spdx-license-ids-3.0.17.tgz",
+      "integrity": "sha512-sh8PWc/ftMqAAdFiBu6Fy6JUOYjqDJBJvIhpfDMyHrr0Rbp5liZqd4TjtQ/RgfLjKFZb+LMx5hpml5qOWy0qvg==",
+      "dev": true
+    },
     "node_modules/sprintf-js": {
       "version": "1.0.3",
       "resolved": "https://registry.npmjs.org/sprintf-js/-/sprintf-js-1.0.3.tgz",

nodejs/package.json

@@ -25,6 +25,7 @@
     "apache-arrow-old": "npm:apache-arrow@13.0.0",
     "eslint": "^8.57.0",
     "eslint-config-prettier": "^9.1.0",
+    "eslint-plugin-jsdoc": "^48.2.1",
     "jest": "^29.7.0",
     "prettier": "^3.1.0",
     "tmp": "^0.2.3",

nodejs/src/connection.rs

@@ -124,7 +124,7 @@ impl Connection {
         let mode = Self::parse_create_mode_str(&mode)?;
         let tbl = self
             .get_inner()?
-            .create_table(&name, Box::new(batches))
+            .create_table(&name, batches)
             .mode(mode)
             .execute()
             .await

nodejs/src/error.rs

@@ -0,0 +1,12 @@
+pub type Result<T> = napi::Result<T>;
+
+pub trait NapiErrorExt<T> {
+    /// Convert to a napi error using from_reason(err.to_string())
+    fn default_error(self) -> Result<T>;
+}
+
+impl<T> NapiErrorExt<T> for std::result::Result<T, lancedb::Error> {
+    fn default_error(self) -> Result<T> {
+        self.map_err(|err| napi::Error::from_reason(err.to_string()))
+    }
+}

nodejs/src/index.rs

@@ -14,126 +14,66 @@
 
 use std::sync::Mutex;
 
-use lance_linalg::distance::MetricType as LanceMetricType;
-use lancedb::index::IndexBuilder as LanceDbIndexBuilder;
-use lancedb::Table as LanceDbTable;
+use lancedb::index::scalar::BTreeIndexBuilder;
+use lancedb::index::vector::IvfPqIndexBuilder;
+use lancedb::index::Index as LanceDbIndex;
 use napi_derive::napi;
 
-#[napi]
-pub enum IndexType {
-    Scalar,
-    IvfPq,
-}
+use crate::util::parse_distance_type;
 
 #[napi]
-pub enum MetricType {
-    L2,
-    Cosine,
-    Dot,
+pub struct Index {
+    inner: Mutex<Option<LanceDbIndex>>,
 }
 
-impl From<MetricType> for LanceMetricType {
-    fn from(metric: MetricType) -> Self {
-        match metric {
-            MetricType::L2 => Self::L2,
-            MetricType::Cosine => Self::Cosine,
-            MetricType::Dot => Self::Dot,
-        }
+impl Index {
+    pub fn consume(&self) -> napi::Result<LanceDbIndex> {
+        self.inner
+            .lock()
+            .unwrap()
+            .take()
+            .ok_or(napi::Error::from_reason(
+                "attempt to use an index more than once",
+            ))
     }
 }
 
 #[napi]
-pub struct IndexBuilder {
-    inner: Mutex<Option<LanceDbIndexBuilder>>,
-}
-
-impl IndexBuilder {
-    fn modify(
-        &self,
-        mod_fn: impl Fn(LanceDbIndexBuilder) -> LanceDbIndexBuilder,
-    ) -> napi::Result<()> {
-        let mut inner = self.inner.lock().unwrap();
-        let inner_builder = inner.take().ok_or_else(|| {
-            napi::Error::from_reason("IndexBuilder has already been consumed".to_string())
-        })?;
-        let inner_builder = mod_fn(inner_builder);
-        inner.replace(inner_builder);
-        Ok(())
-    }
-}
-
-#[napi]
-impl IndexBuilder {
-    pub fn new(tbl: &LanceDbTable) -> Self {
-        let inner = tbl.create_index(&[]);
-        Self {
-            inner: Mutex::new(Some(inner)),
-        }
-    }
-
-    #[napi]
-    pub fn replace(&self, v: bool) -> napi::Result<()> {
-        self.modify(|b| b.replace(v))
-    }
-
-    #[napi]
-    pub fn column(&self, c: String) -> napi::Result<()> {
-        self.modify(|b| b.columns(&[c.as_str()]))
-    }
-
-    #[napi]
-    pub fn name(&self, name: String) -> napi::Result<()> {
-        self.modify(|b| b.name(name.as_str()))
-    }
-
-    #[napi]
+impl Index {
+    #[napi(factory)]
     pub fn ivf_pq(
-        &self,
-        metric_type: Option<MetricType>,
+        distance_type: Option<String>,
         num_partitions: Option<u32>,
         num_sub_vectors: Option<u32>,
-        num_bits: Option<u32>,
         max_iterations: Option<u32>,
         sample_rate: Option<u32>,
-    ) -> napi::Result<()> {
-        self.modify(|b| {
-            let mut b = b.ivf_pq();
-            if let Some(metric_type) = metric_type {
-                b = b.metric_type(metric_type.into());
-            }
-            if let Some(num_partitions) = num_partitions {
-                b = b.num_partitions(num_partitions);
-            }
-            if let Some(num_sub_vectors) = num_sub_vectors {
-                b = b.num_sub_vectors(num_sub_vectors);
-            }
-            if let Some(num_bits) = num_bits {
-                b = b.num_bits(num_bits);
-            }
-            if let Some(max_iterations) = max_iterations {
-                b = b.max_iterations(max_iterations);
-            }
-            if let Some(sample_rate) = sample_rate {
-                b = b.sample_rate(sample_rate);
-            }
-            b
+    ) -> napi::Result<Self> {
+        let mut ivf_pq_builder = IvfPqIndexBuilder::default();
+        if let Some(distance_type) = distance_type {
+            let distance_type = parse_distance_type(distance_type)?;
+            ivf_pq_builder = ivf_pq_builder.distance_type(distance_type);
+        }
+        if let Some(num_partitions) = num_partitions {
+            ivf_pq_builder = ivf_pq_builder.num_partitions(num_partitions);
+        }
+        if let Some(num_sub_vectors) = num_sub_vectors {
+            ivf_pq_builder = ivf_pq_builder.num_sub_vectors(num_sub_vectors);
+        }
+        if let Some(max_iterations) = max_iterations {
+            ivf_pq_builder = ivf_pq_builder.max_iterations(max_iterations);
+        }
+        if let Some(sample_rate) = sample_rate {
+            ivf_pq_builder = ivf_pq_builder.sample_rate(sample_rate);
+        }
+        Ok(Self {
+            inner: Mutex::new(Some(LanceDbIndex::IvfPq(ivf_pq_builder))),
         })
     }
 
-    #[napi]
-    pub fn scalar(&self) -> napi::Result<()> {
-        self.modify(|b| b.scalar())
-    }
-
-    #[napi]
-    pub async fn build(&self) -> napi::Result<()> {
-        let inner = self.inner.lock().unwrap().take().ok_or_else(|| {
-            napi::Error::from_reason("IndexBuilder has already been consumed".to_string())
-        })?;
-        inner
-            .build()
-            .await
-            .map_err(|e| napi::Error::from_reason(format!("Failed to build index: {}", e)))?;
-        Ok(())
+    #[napi(factory)]
+    pub fn btree() -> Self {
+        Self {
+            inner: Mutex::new(Some(LanceDbIndex::BTree(BTreeIndexBuilder::default()))),
+        }
     }
 }

nodejs/src/iterator.rs

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 use futures::StreamExt;
-use lance::io::RecordBatchStream;
+use lancedb::arrow::SendableRecordBatchStream;
 use lancedb::ipc::batches_to_ipc_file;
 use napi::bindgen_prelude::*;
 use napi_derive::napi;
@@ -21,12 +21,12 @@ use napi_derive::napi;
 /** Typescript-style Async Iterator over RecordBatches */
 #[napi]
 pub struct RecordBatchIterator {
-    inner: Box<dyn RecordBatchStream + Unpin>,
+    inner: SendableRecordBatchStream,
 }
 
 #[napi]
 impl RecordBatchIterator {
-    pub(crate) fn new(inner: Box<dyn RecordBatchStream + Unpin>) -> Self {
+    pub(crate) fn new(inner: SendableRecordBatchStream) -> Self {
         Self { inner }
     }
 

nodejs/src/lib.rs

@@ -16,10 +16,12 @@ use connection::Connection;
 use napi_derive::*;
 
 mod connection;
+mod error;
 mod index;
 mod iterator;
 mod query;
 mod table;
+mod util;
 
 #[napi(object)]
 #[derive(Debug)]

nodejs/src/query.rs

@@ -12,36 +12,38 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use lancedb::query::Query as LanceDBQuery;
+use lancedb::query::ExecutableQuery;
+use lancedb::query::Query as LanceDbQuery;
+use lancedb::query::QueryBase;
+use lancedb::query::Select;
+use lancedb::query::VectorQuery as LanceDbVectorQuery;
 use napi::bindgen_prelude::*;
 use napi_derive::napi;
 
+use crate::error::NapiErrorExt;
 use crate::iterator::RecordBatchIterator;
+use crate::util::parse_distance_type;
 
 #[napi]
 pub struct Query {
-    inner: LanceDBQuery,
+    inner: LanceDbQuery,
 }
 
 #[napi]
 impl Query {
-    pub fn new(query: LanceDBQuery) -> Self {
+    pub fn new(query: LanceDbQuery) -> Self {
         Self { inner: query }
     }
 
+    // We cannot call this r#where because NAPI gets confused by the r#
     #[napi]
-    pub fn column(&mut self, column: String) {
-        self.inner = self.inner.clone().column(&column);
+    pub fn only_if(&mut self, predicate: String) {
+        self.inner = self.inner.clone().only_if(predicate);
     }
 
     #[napi]
-    pub fn filter(&mut self, filter: String) {
-        self.inner = self.inner.clone().filter(filter);
-    }
-
-    #[napi]
-    pub fn select(&mut self, columns: Vec<String>) {
-        self.inner = self.inner.clone().select(&columns);
+    pub fn select(&mut self, columns: Vec<(String, String)>) {
+        self.inner = self.inner.clone().select(Select::dynamic(&columns));
     }
 
     #[napi]
@@ -50,13 +52,46 @@ impl Query {
     }
 
     #[napi]
-    pub fn prefilter(&mut self, prefilter: bool) {
-        self.inner = self.inner.clone().prefilter(prefilter);
+    pub fn nearest_to(&mut self, vector: Float32Array) -> Result<VectorQuery> {
+        let inner = self
+            .inner
+            .clone()
+            .nearest_to(vector.as_ref())
+            .default_error()?;
+        Ok(VectorQuery { inner })
     }
 
     #[napi]
-    pub fn nearest_to(&mut self, vector: Float32Array) {
-        self.inner = self.inner.clone().nearest_to(&vector);
+    pub async fn execute(&self) -> napi::Result<RecordBatchIterator> {
+        let inner_stream = self.inner.execute().await.map_err(|e| {
+            napi::Error::from_reason(format!("Failed to execute query stream: {}", e))
+        })?;
+        Ok(RecordBatchIterator::new(inner_stream))
+    }
+}
+
+#[napi]
+pub struct VectorQuery {
+    inner: LanceDbVectorQuery,
+}
+
+#[napi]
+impl VectorQuery {
+    #[napi]
+    pub fn column(&mut self, column: String) {
+        self.inner = self.inner.clone().column(&column);
+    }
+
+    #[napi]
+    pub fn distance_type(&mut self, distance_type: String) -> napi::Result<()> {
+        let distance_type = parse_distance_type(distance_type)?;
+        self.inner = self.inner.clone().distance_type(distance_type);
+        Ok(())
+    }
+
+    #[napi]
+    pub fn postfilter(&mut self) {
+        self.inner = self.inner.clone().postfilter();
     }
 
     #[napi]
@@ -70,10 +105,30 @@ impl Query {
     }
 
     #[napi]
-    pub async fn execute_stream(&self) -> napi::Result<RecordBatchIterator> {
-        let inner_stream = self.inner.execute_stream().await.map_err(|e| {
+    pub fn bypass_vector_index(&mut self) {
+        self.inner = self.inner.clone().bypass_vector_index()
+    }
+
+    #[napi]
+    pub fn only_if(&mut self, predicate: String) {
+        self.inner = self.inner.clone().only_if(predicate);
+    }
+
+    #[napi]
+    pub fn select(&mut self, columns: Vec<(String, String)>) {
+        self.inner = self.inner.clone().select(Select::dynamic(&columns));
+    }
+
+    #[napi]
+    pub fn limit(&mut self, limit: u32) {
+        self.inner = self.inner.clone().limit(limit as usize);
+    }
+
+    #[napi]
+    pub async fn execute(&self) -> napi::Result<RecordBatchIterator> {
+        let inner_stream = self.inner.execute().await.map_err(|e| {
             napi::Error::from_reason(format!("Failed to execute query stream: {}", e))
         })?;
-        Ok(RecordBatchIterator::new(Box::new(inner_stream)))
+        Ok(RecordBatchIterator::new(inner_stream))
     }
 }

nodejs/src/table.rs

@@ -13,14 +13,17 @@
 // limitations under the License.
 
 use arrow_ipc::writer::FileWriter;
-use lance::dataset::ColumnAlteration as LanceColumnAlteration;
 use lancedb::ipc::ipc_file_to_batches;
-use lancedb::table::{AddDataMode, Table as LanceDbTable};
+use lancedb::table::{
+    AddDataMode, ColumnAlteration as LanceColumnAlteration, NewColumnTransform,
+    Table as LanceDbTable,
+};
 use napi::bindgen_prelude::*;
 use napi_derive::napi;
 
-use crate::index::IndexBuilder;
-use crate::query::Query;
+use crate::error::NapiErrorExt;
+use crate::index::Index;
+use crate::query::{Query, VectorQuery};
 
 #[napi]
 pub struct Table {
@@ -86,7 +89,7 @@ impl Table {
     pub async fn add(&self, buf: Buffer, mode: String) -> napi::Result<()> {
         let batches = ipc_file_to_batches(buf.to_vec())
             .map_err(|e| napi::Error::from_reason(format!("Failed to read IPC file: {}", e)))?;
-        let mut op = self.inner_ref()?.add(Box::new(batches));
+        let mut op = self.inner_ref()?.add(batches);
 
         op = if mode == "append" {
             op.mode(AddDataMode::Append)
@@ -129,8 +132,38 @@ impl Table {
     }
 
     #[napi]
-    pub fn create_index(&self) -> napi::Result<IndexBuilder> {
-        Ok(IndexBuilder::new(self.inner_ref()?))
+    pub async fn create_index(
+        &self,
+        index: Option<&Index>,
+        column: String,
+        replace: Option<bool>,
+    ) -> napi::Result<()> {
+        let lancedb_index = if let Some(index) = index {
+            index.consume()?
+        } else {
+            lancedb::index::Index::Auto
+        };
+        let mut builder = self.inner_ref()?.create_index(&[column], lancedb_index);
+        if let Some(replace) = replace {
+            builder = builder.replace(replace);
+        }
+        builder.execute().await.default_error()
+    }
+
+    #[napi]
+    pub async fn update(
+        &self,
+        only_if: Option<String>,
+        columns: Vec<(String, String)>,
+    ) -> napi::Result<()> {
+        let mut op = self.inner_ref()?.update();
+        if let Some(only_if) = only_if {
+            op = op.only_if(only_if);
+        }
+        for (column_name, value) in columns {
+            op = op.column(column_name, value);
+        }
+        op.execute().await.default_error()
     }
 
     #[napi]
@@ -138,13 +171,18 @@ impl Table {
         Ok(Query::new(self.inner_ref()?.query()))
     }
 
+    #[napi]
+    pub fn vector_search(&self, vector: Float32Array) -> napi::Result<VectorQuery> {
+        self.query()?.nearest_to(vector)
+    }
+
     #[napi]
     pub async fn add_columns(&self, transforms: Vec<AddColumnsSql>) -> napi::Result<()> {
         let transforms = transforms
             .into_iter()
             .map(|sql| (sql.name, sql.value_sql))
             .collect::<Vec<_>>();
-        let transforms = lance::dataset::NewColumnTransform::SqlExpressions(transforms);
+        let transforms = NewColumnTransform::SqlExpressions(transforms);
         self.inner_ref()?
             .add_columns(transforms, None)
             .await
@@ -197,6 +235,67 @@ impl Table {
             })?;
         Ok(())
     }
+
+    #[napi]
+    pub async fn version(&self) -> napi::Result<i64> {
+        self.inner_ref()?
+            .version()
+            .await
+            .map(|val| val as i64)
+            .default_error()
+    }
+
+    #[napi]
+    pub async fn checkout(&self, version: i64) -> napi::Result<()> {
+        self.inner_ref()?
+            .checkout(version as u64)
+            .await
+            .default_error()
+    }
+
+    #[napi]
+    pub async fn checkout_latest(&self) -> napi::Result<()> {
+        self.inner_ref()?.checkout_latest().await.default_error()
+    }
+
+    #[napi]
+    pub async fn restore(&self) -> napi::Result<()> {
+        self.inner_ref()?.restore().await.default_error()
+    }
+
+    #[napi]
+    pub async fn list_indices(&self) -> napi::Result<Vec<IndexConfig>> {
+        Ok(self
+            .inner_ref()?
+            .list_indices()
+            .await
+            .default_error()?
+            .into_iter()
+            .map(IndexConfig::from)
+            .collect::<Vec<_>>())
+    }
+}
+
+#[napi(object)]
+/// A description of an index currently configured on a column
+pub struct IndexConfig {
+    /// The type of the index
+    pub index_type: String,
+    /// The columns in the index
+    ///
+    /// Currently this is always an array of size 1.  In the future there may
+    /// be more columns to represent composite indices.
+    pub columns: Vec<String>,
+}
+
+impl From<lancedb::index::IndexConfig> for IndexConfig {
+    fn from(value: lancedb::index::IndexConfig) -> Self {
+        let index_type = format!("{:?}", value.index_type);
+        Self {
+            index_type,
+            columns: value.columns,
+        }
+    }
 }
 
 ///  A definition of a column alteration. The alteration changes the column at

nodejs/src/util.rs

@@ -0,0 +1,13 @@
+use lancedb::DistanceType;
+
+pub fn parse_distance_type(distance_type: impl AsRef<str>) -> napi::Result<DistanceType> {
+    match distance_type.as_ref().to_lowercase().as_str() {
+        "l2" => Ok(DistanceType::L2),
+        "cosine" => Ok(DistanceType::Cosine),
+        "dot" => Ok(DistanceType::Dot),
+        _ => Err(napi::Error::from_reason(format!(
+            "Invalid distance type '{}'.  Must be one of l2, cosine, or dot",
+            distance_type.as_ref()
+        ))),
+    }
+}

python/.bumpversion.cfg

@@ -1,5 +1,5 @@
 [bumpversion]
-current_version = 0.6.3
+current_version = 0.6.5
 commit = True
 message = [python] Bump version: {current_version} → {new_version}
 tag = True

python/Cargo.toml

@@ -22,6 +22,9 @@ pyo3-asyncio = { version = "0.20", features = ["attributes", "tokio-runtime"] }
 
 # Prevent dynamic linking of lzma, which comes from datafusion
 lzma-sys = { version = "*", features = ["static"] }
+pin-project = "1.1.5"
+futures.workspace = true
+tokio = { version = "1.36.0", features = ["sync"] }
 
 [build-dependencies]
 pyo3-build-config = { version = "0.20.3", features = [

python/pyproject.toml

@@ -1,9 +1,9 @@
 [project]
 name = "lancedb"
-version = "0.6.3"
+version = "0.6.5"
 dependencies = [
     "deprecation",
-    "pylance==0.10.2",
+    "pylance==0.10.5",
     "ratelimiter~=1.0",
     "retry>=0.9.2",
     "tqdm>=4.27.0",
@@ -81,6 +81,7 @@ embeddings = [
     "awscli>=1.29.57",
     "botocore>=1.31.57",
 ]
+azure = ["adlfs>=2024.2.0"]
 
 [tool.maturin]
 python-source = "python"
@@ -93,13 +94,11 @@ lancedb = "lancedb.cli.cli:cli"
 requires = ["maturin>=1.4"]
 build-backend = "maturin"
 
-
 [tool.ruff.lint]
 select = ["F", "E", "W", "I", "G", "TCH", "PERF"]
 
 [tool.pytest.ini_options]
 addopts = "--strict-markers --ignore-glob=lancedb/embeddings/*.py"
-
 markers = [
     "slow: marks tests as slow (deselect with '-m \"not slow\"')",
     "asyncio",

python/python/lancedb/__init__.py

@@ -23,8 +23,9 @@ from ._lancedb import connect as lancedb_connect
 from .common import URI, sanitize_uri
 from .db import AsyncConnection, DBConnection, LanceDBConnection
 from .remote.db import RemoteDBConnection
-from .schema import vector  # noqa: F401
-from .utils import sentry_log  # noqa: F401
+from .schema import vector
+from .table import AsyncTable
+from .utils import sentry_log
 
 
 def connect(
@@ -188,3 +189,19 @@ async def connect_async(
             read_consistency_interval_secs,
         )
     )
+
+
+__all__ = [
+    "connect",
+    "connect_async",
+    "AsyncConnection",
+    "AsyncTable",
+    "URI",
+    "sanitize_uri",
+    "sentry_log",
+    "vector",
+    "DBConnection",
+    "LanceDBConnection",
+    "RemoteDBConnection",
+    "__version__",
+]

python/python/lancedb/_lancedb.pyi

@@ -1,7 +1,19 @@
-from typing import Optional
+from typing import Dict, List, Optional, Tuple
 
 import pyarrow as pa
 
+class Index:
+    @staticmethod
+    def ivf_pq(
+        distance_type: Optional[str],
+        num_partitions: Optional[int],
+        num_sub_vectors: Optional[int],
+        max_iterations: Optional[int],
+        sample_rate: Optional[int],
+    ) -> Index: ...
+    @staticmethod
+    def btree() -> Index: ...
+
 class Connection(object):
     async def table_names(
         self, start_after: Optional[str], limit: Optional[int]
@@ -13,10 +25,27 @@ class Connection(object):
         self, name: str, mode: str, schema: pa.Schema
     ) -> Table: ...
 
-class Table(object):
+class Table:
     def name(self) -> str: ...
     def __repr__(self) -> str: ...
     async def schema(self) -> pa.Schema: ...
+    async def add(self, data: pa.RecordBatchReader, mode: str) -> None: ...
+    async def update(self, updates: Dict[str, str], where: Optional[str]) -> None: ...
+    async def count_rows(self, filter: Optional[str]) -> int: ...
+    async def create_index(
+        self, column: str, config: Optional[Index], replace: Optional[bool]
+    ): ...
+    async def version(self) -> int: ...
+    async def checkout(self, version): ...
+    async def checkout_latest(self): ...
+    async def restore(self): ...
+    async def list_indices(self) -> List[IndexConfig]: ...
+    def query(self) -> Query: ...
+    def vector_search(self) -> VectorQuery: ...
+
+class IndexConfig:
+    index_type: str
+    columns: List[str]
 
 async def connect(
     uri: str,
@@ -25,3 +54,27 @@ async def connect(
     host_override: Optional[str],
     read_consistency_interval: Optional[float],
 ) -> Connection: ...
+
+class RecordBatchStream:
+    def schema(self) -> pa.Schema: ...
+    async def next(self) -> Optional[pa.RecordBatch]: ...
+
+class Query:
+    def where(self, filter: str): ...
+    def select(self, columns: Tuple[str, str]): ...
+    def limit(self, limit: int): ...
+    def nearest_to(self, query_vec: pa.Array) -> VectorQuery: ...
+    async def execute(self) -> RecordBatchStream: ...
+
+class VectorQuery:
+    async def execute(self) -> RecordBatchStream: ...
+    def where(self, filter: str): ...
+    def select(self, columns: List[str]): ...
+    def select_with_projection(self, columns: Tuple[str, str]): ...
+    def limit(self, limit: int): ...
+    def column(self, column: str): ...
+    def distance_type(self, distance_type: str): ...
+    def postfilter(self): ...
+    def refine_factor(self, refine_factor: int): ...
+    def nprobes(self, nprobes: int): ...
+    def bypass_vector_index(self): ...

python/python/lancedb/arrow.py

@@ -0,0 +1,44 @@
+from typing import List
+
+import pyarrow as pa
+
+from ._lancedb import RecordBatchStream
+
+
+class AsyncRecordBatchReader:
+    """
+    An async iterator over a stream of RecordBatches.
+
+    Also allows access to the schema of the stream
+    """
+
+    def __init__(self, inner: RecordBatchStream):
+        self.inner_ = inner
+
+    @property
+    def schema(self) -> pa.Schema:
+        """
+        Get the schema of the batches produced by the stream
+
+        Accessing the schema does not consume any data from the stream
+        """
+        return self.inner_.schema()
+
+    async def read_all(self) -> List[pa.RecordBatch]:
+        """
+        Read all the record batches from the stream
+
+        This consumes the entire stream and returns a list of record batches
+
+        If there are a lot of results this may consume a lot of memory
+        """
+        return [batch async for batch in self]
+
+    def __aiter__(self):
+        return self
+
+    async def __anext__(self) -> pa.RecordBatch:
+        next = await self.inner_.next()
+        if next is None:
+            raise StopAsyncIteration
+        return next

python/python/lancedb/db.py

@@ -529,7 +529,7 @@ class AsyncConnection(object):
         on_bad_vectors: Optional[str] = None,
         fill_value: Optional[float] = None,
         embedding_functions: Optional[List[EmbeddingFunctionConfig]] = None,
-    ) -> Table:
+    ) -> AsyncTable:
         """Create a [Table][lancedb.table.Table] in the database.
 
         Parameters

python/python/lancedb/embeddings/imagebind.py

@@ -31,7 +31,7 @@ class ImageBindEmbeddings(EmbeddingFunction):
     six different modalities: images, text, audio, depth, thermal, and IMU data
 
     to download package, run :
-        `pip install imagebind@git+https://github.com/raghavdixit99/ImageBind`
+        `pip install imagebind-packaged==0.1.2`
     """
 
     name: str = "imagebind_huge"

python/python/lancedb/embeddings/openai.py

@@ -113,5 +113,5 @@ class OpenAIEmbeddings(TextEmbeddingFunction):
         if self.organization:
             kwargs["organization"] = self.organization
         if self.api_key:
-            kwargs["api_key"] = self
+            kwargs["api_key"] = self.api_key
         return openai.OpenAI(**kwargs)

python/python/lancedb/fts.py

@@ -28,7 +28,9 @@ except ImportError:
 from .table import LanceTable
 
 
-def create_index(index_path: str, text_fields: List[str]) -> tantivy.Index:
+def create_index(
+    index_path: str, text_fields: List[str], ordering_fields: List[str] = None
+) -> tantivy.Index:
     """
     Create a new Index (not populated)
 
@@ -38,12 +40,16 @@ def create_index(index_path: str, text_fields: List[str]) -> tantivy.Index:
         Path to the index directory
     text_fields : List[str]
         List of text fields to index
+    ordering_fields: List[str]
+        List of unsigned type fields to order by at search time
 
     Returns
     -------
     index : tantivy.Index
         The index object (not yet populated)
     """
+    if ordering_fields is None:
+        ordering_fields = []
     # Declaring our schema.
     schema_builder = tantivy.SchemaBuilder()
     # special field that we'll populate with row_id
@@ -51,6 +57,9 @@ def create_index(index_path: str, text_fields: List[str]) -> tantivy.Index:
     # data fields
     for name in text_fields:
         schema_builder.add_text_field(name, stored=True)
+    if ordering_fields:
+        for name in ordering_fields:
+            schema_builder.add_unsigned_field(name, fast=True)
     schema = schema_builder.build()
     os.makedirs(index_path, exist_ok=True)
     index = tantivy.Index(schema, path=index_path)
@@ -62,6 +71,7 @@ def populate_index(
     table: LanceTable,
     fields: List[str],
     writer_heap_size: int = 1024 * 1024 * 1024,
+    ordering_fields: List[str] = None,
 ) -> int:
     """
     Populate an index with data from a LanceTable
@@ -82,8 +92,11 @@ def populate_index(
     int
         The number of rows indexed
     """
+    if ordering_fields is None:
+        ordering_fields = []
     # first check the fields exist and are string or large string type
     nested = []
+
     for name in fields:
         try:
             f = table.schema.field(name)  # raises KeyError if not found
@@ -104,7 +117,7 @@ def populate_index(
     if len(nested) > 0:
         max_nested_level = max([len(name.split(".")) for name in nested])
 
-    for b in dataset.to_batches(columns=fields):
+    for b in dataset.to_batches(columns=fields + ordering_fields):
         if max_nested_level > 0:
             b = pa.Table.from_batches([b])
             for _ in range(max_nested_level - 1):
@@ -115,6 +128,10 @@ def populate_index(
                 value = b[name][i].as_py()
                 if value is not None:
                     doc.add_text(name, value)
+            for name in ordering_fields:
+                value = b[name][i].as_py()
+                if value is not None:
+                    doc.add_unsigned(name, value)
             if not doc.is_empty:
                 doc.add_integer("doc_id", row_id)
                 writer.add_document(doc)
@@ -149,7 +166,7 @@ def resolve_path(schema, field_name: str) -> pa.Field:
 
 
 def search_index(
-    index: tantivy.Index, query: str, limit: int = 10
+    index: tantivy.Index, query: str, limit: int = 10, ordering_field=None
 ) -> Tuple[Tuple[int], Tuple[float]]:
     """
     Search an index for a query
@@ -172,7 +189,10 @@ def search_index(
     searcher = index.searcher()
     query = index.parse_query(query)
     # get top results
-    results = searcher.search(query, limit)
+    if ordering_field:
+        results = searcher.search(query, limit, order_by_field=ordering_field)
+    else:
+        results = searcher.search(query, limit)
     if results.count == 0:
         return tuple(), tuple()
     return tuple(

python/python/lancedb/index.py

@@ -0,0 +1,163 @@
+from typing import Optional
+
+from ._lancedb import (
+    Index as LanceDbIndex,
+)
+from ._lancedb import (
+    IndexConfig,
+)
+
+
+class BTree(object):
+    """Describes a btree index configuration
+
+    A btree index is an index on scalar columns.  The index stores a copy of the
+    column in sorted order.  A header entry is created for each block of rows
+    (currently the block size is fixed at 4096).  These header entries are stored
+    in a separate cacheable structure (a btree).  To search for data the header is
+    used to determine which blocks need to be read from disk.
+
+    For example, a btree index in a table with 1Bi rows requires
+    sizeof(Scalar) * 256Ki bytes of memory and will generally need to read
+    sizeof(Scalar) * 4096 bytes to find the correct row ids.
+
+    This index is good for scalar columns with mostly distinct values and does best
+    when the query is highly selective.
+
+    The btree index does not currently have any parameters though parameters such as
+    the block size may be added in the future.
+    """
+
+    def __init__(self):
+        self._inner = LanceDbIndex.btree()
+
+
+class IvfPq(object):
+    """Describes an IVF PQ Index
+
+    This index stores a compressed (quantized) copy of every vector.  These vectors
+    are grouped into partitions of similar vectors.  Each partition keeps track of
+    a centroid which is the average value of all vectors in the group.
+
+    During a query the centroids are compared with the query vector to find the
+    closest partitions.  The compressed vectors in these partitions are then
+    searched to find the closest vectors.
+
+    The compression scheme is called product quantization.  Each vector is divide
+    into subvectors and then each subvector is quantized into a small number of
+    bits.  the parameters `num_bits` and `num_subvectors` control this process,
+    providing a tradeoff between index size (and thus search speed) and index
+    accuracy.
+
+    The partitioning process is called IVF and the `num_partitions` parameter
+    controls how many groups to create.
+
+    Note that training an IVF PQ index on a large dataset is a slow operation and
+    currently is also a memory intensive operation.
+    """
+
+    def __init__(
+        self,
+        *,
+        distance_type: Optional[str] = None,
+        num_partitions: Optional[int] = None,
+        num_sub_vectors: Optional[int] = None,
+        max_iterations: Optional[int] = None,
+        sample_rate: Optional[int] = None,
+    ):
+        """
+        Create an IVF PQ index config
+
+        Parameters
+        ----------
+        distance_type: str, default "L2"
+            The distance metric used to train the index
+
+            This is used when training the index to calculate the IVF partitions
+            (vectors are grouped in partitions with similar vectors according to this
+            distance type) and to calculate a subvector's code during quantization.
+
+            The distance type used to train an index MUST match the distance type used
+            to search the index.  Failure to do so will yield inaccurate results.
+
+            The following distance types are available:
+
+            "l2" - Euclidean distance. This is a very common distance metric that
+            accounts for both magnitude and direction when determining the distance
+            between vectors. L2 distance has a range of [0, ∞).
+
+            "cosine" - Cosine distance.  Cosine distance is a distance metric
+            calculated from the cosine similarity between two vectors. Cosine
+            similarity is a measure of similarity between two non-zero vectors of an
+            inner product space. It is defined to equal the cosine of the angle
+            between them.  Unlike L2, the cosine distance is not affected by the
+            magnitude of the vectors.  Cosine distance has a range of [0, 2].
+
+            Note: the cosine distance is undefined when one (or both) of the vectors
+            are all zeros (there is no direction).  These vectors are invalid and may
+            never be returned from a vector search.
+
+            "dot" - Dot product. Dot distance is the dot product of two vectors. Dot
+            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: int, default sqrt(num_rows)
+            The number of IVF partitions to create.
+
+            This value should generally scale with the number of rows in the dataset.
+            By default the number of partitions is the square root of the number of
+            rows.
+
+            If this value is too large then the first part of the search (picking the
+            right partition) will be slow.  If this value is too small then the second
+            part of the search (searching within a partition) will be slow.
+        num_sub_vectors: int, default is vector dimension / 16
+            Number of sub-vectors of PQ.
+
+            This value controls how much the vector is compressed during the
+            quantization step.  The more sub vectors there are the less the vector is
+            compressed.  The default is the dimension of the vector divided by 16.  If
+            the dimension is not evenly divisible by 16 we use the dimension divded by
+            8.
+
+            The above two cases are highly preferred.  Having 8 or 16 values per
+            subvector allows us to use efficient SIMD instructions.
+
+            If the dimension is not visible by 8 then we use 1 subvector.  This is not
+            ideal and will likely result in poor performance.
+        max_iterations: int, default 50
+            Max iteration to train kmeans.
+
+            When training an IVF PQ index we use kmeans to calculate the partitions.
+            This parameter controls how many iterations of kmeans to run.
+
+            Increasing this might improve the quality of the index but in most cases
+            these extra iterations have diminishing returns.
+
+            The default value is 50.
+        sample_rate: int, default 256
+            The rate used to calculate the number of training vectors for kmeans.
+
+            When an IVF PQ index is trained, we need to calculate partitions.  These
+            are groups of vectors that are similar to each other.  To do this we use an
+            algorithm called kmeans.
+
+            Running kmeans on a large dataset can be slow.  To speed this up we run
+            kmeans on a random sample of the data.  This parameter controls the size of
+            the sample.  The total number of vectors used to train the index is
+            `sample_rate * num_partitions`.
+
+            Increasing this value might improve the quality of the index but in most
+            cases the default should be sufficient.
+
+            The default value is 256.
+        """
+        self._inner = LanceDbIndex.ivf_pq(
+            distance_type=distance_type,
+            num_partitions=num_partitions,
+            num_sub_vectors=num_sub_vectors,
+            max_iterations=max_iterations,
+            sample_rate=sample_rate,
+        )
+
+
+__all__ = ["BTree", "IvfPq", "IndexConfig"]

python/python/lancedb/query.py

@@ -16,7 +16,16 @@ from __future__ import annotations
 from abc import ABC, abstractmethod
 from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
-from typing import TYPE_CHECKING, Dict, List, Literal, Optional, Tuple, Type, Union
+from typing import (
+    TYPE_CHECKING,
+    Dict,
+    List,
+    Literal,
+    Optional,
+    Tuple,
+    Type,
+    Union,
+)
 
 import deprecation
 import numpy as np
@@ -24,6 +33,7 @@ import pyarrow as pa
 import pydantic
 
 from . import __version__
+from .arrow import AsyncRecordBatchReader
 from .common import VEC
 from .rerankers.base import Reranker
 from .rerankers.linear_combination import LinearCombinationReranker
@@ -33,6 +43,8 @@ if TYPE_CHECKING:
     import PIL
     import polars as pl
 
+    from ._lancedb import Query as LanceQuery
+    from ._lancedb import VectorQuery as LanceVectorQuery
     from .pydantic import LanceModel
     from .table import Table
 
@@ -117,6 +129,7 @@ class LanceQueryBuilder(ABC):
         query: Optional[Union[np.ndarray, str, "PIL.Image.Image", Tuple]],
         query_type: str,
         vector_column_name: str,
+        ordering_field_name: str = None,
     ) -> LanceQueryBuilder:
         """
         Create a query builder based on the given query and query type.
@@ -141,6 +154,9 @@ class LanceQueryBuilder(ABC):
             # hybrid fts and vector query
             return LanceHybridQueryBuilder(table, query, vector_column_name)
 
+        # remember the string query for reranking purpose
+        str_query = query if isinstance(query, str) else None
+
         # convert "auto" query_type to "vector", "fts"
         # or "hybrid" and convert the query to vector if needed
         query, query_type = cls._resolve_query(
@@ -152,7 +168,9 @@ class LanceQueryBuilder(ABC):
 
         if isinstance(query, str):
             # fts
-            return LanceFtsQueryBuilder(table, query)
+            return LanceFtsQueryBuilder(
+                table, query, ordering_field_name=ordering_field_name
+            )
 
         if isinstance(query, list):
             query = np.array(query, dtype=np.float32)
@@ -161,7 +179,7 @@ class LanceQueryBuilder(ABC):
         else:
             raise TypeError(f"Unsupported query type: {type(query)}")
 
-        return LanceVectorQueryBuilder(table, query, vector_column_name)
+        return LanceVectorQueryBuilder(table, query, vector_column_name, str_query)
 
     @classmethod
     def _resolve_query(cls, table, query, query_type, vector_column_name):
@@ -425,6 +443,7 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         table: "Table",
         query: Union[np.ndarray, list, "PIL.Image.Image"],
         vector_column: str,
+        str_query: Optional[str] = None,
     ):
         super().__init__(table)
         self._query = query
@@ -433,6 +452,8 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         self._refine_factor = None
         self._vector_column = vector_column
         self._prefilter = False
+        self._reranker = None
+        self._str_query = str_query
 
     def metric(self, metric: Literal["L2", "cosine"]) -> LanceVectorQueryBuilder:
         """Set the distance metric to use.
@@ -503,6 +524,21 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         and also the "_distance" column which is the distance between the query
         vector and the returned vectors.
         """
+        return self.to_batches().read_all()
+
+    def to_batches(self, /, batch_size: Optional[int] = None) -> pa.RecordBatchReader:
+        """
+        Execute the query and return the result as a RecordBatchReader object.
+
+        Parameters
+        ----------
+        batch_size: int
+            The maximum number of selected records in a RecordBatch object.
+
+        Returns
+        -------
+        pa.RecordBatchReader
+        """
         vector = self._query if isinstance(self._query, list) else self._query.tolist()
         if isinstance(vector[0], np.ndarray):
             vector = [v.tolist() for v in vector]
@@ -518,7 +554,16 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
             vector_column=self._vector_column,
             with_row_id=self._with_row_id,
         )
-        return self._table._execute_query(query)
+        result_set = self._table._execute_query(query, batch_size)
+        if self._reranker is not None:
+            rs_table = result_set.read_all()
+            result_set = self._reranker.rerank_vector(self._str_query, rs_table)
+            # convert result_set back to RecordBatchReader
+            result_set = pa.RecordBatchReader.from_batches(
+                result_set.schema, result_set.to_batches()
+            )
+
+        return result_set
 
     def where(self, where: str, prefilter: bool = False) -> LanceVectorQueryBuilder:
         """Set the where clause.
@@ -544,14 +589,52 @@ class LanceVectorQueryBuilder(LanceQueryBuilder):
         self._prefilter = prefilter
         return self
 
+    def rerank(
+        self, reranker: Reranker, query_string: Optional[str] = None
+    ) -> LanceVectorQueryBuilder:
+        """Rerank the results using the specified reranker.
+
+        Parameters
+        ----------
+        reranker: Reranker
+            The reranker to use.
+
+        query_string: Optional[str]
+            The query to use for reranking. This needs to be specified explicitly here
+            as the query used for vector search may already be vectorized and the
+            reranker requires a string query.
+            This is only required if the query used for vector search is not a string.
+            Note: This doesn't yet support the case where the query is multimodal or a
+            list of vectors.
+
+        Returns
+        -------
+        LanceVectorQueryBuilder
+            The LanceQueryBuilder object.
+        """
+        self._reranker = reranker
+        if self._str_query is None and query_string is None:
+            raise ValueError(
+                """
+                The query used for vector search is not a string.
+                In this case, the reranker query needs to be specified explicitly.
+                """
+            )
+        if query_string is not None and not isinstance(query_string, str):
+            raise ValueError("Reranking currently only supports string queries")
+        self._str_query = query_string if query_string is not None else self._str_query
+        return self
+
 
 class LanceFtsQueryBuilder(LanceQueryBuilder):
     """A builder for full text search for LanceDB."""
 
-    def __init__(self, table: "Table", query: str):
+    def __init__(self, table: "Table", query: str, ordering_field_name: str = None):
         super().__init__(table)
         self._query = query
         self._phrase_query = False
+        self.ordering_field_name = ordering_field_name
+        self._reranker = None
 
     def phrase_query(self, phrase_query: bool = True) -> LanceFtsQueryBuilder:
         """Set whether to use phrase query.
@@ -596,7 +679,9 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
         if self._phrase_query:
             query = query.replace('"', "'")
             query = f'"{query}"'
-        row_ids, scores = search_index(index, query, self._limit)
+        row_ids, scores = search_index(
+            index, query, self._limit, ordering_field=self.ordering_field_name
+        )
         if len(row_ids) == 0:
             empty_schema = pa.schema([pa.field("score", pa.float32())])
             return pa.Table.from_pylist([], schema=empty_schema)
@@ -638,8 +723,27 @@ class LanceFtsQueryBuilder(LanceQueryBuilder):
 
         if self._with_row_id:
             output_tbl = output_tbl.append_column("_rowid", row_ids)
+
+        if self._reranker is not None:
+            output_tbl = self._reranker.rerank_fts(self._query, output_tbl)
         return output_tbl
 
+    def rerank(self, reranker: Reranker) -> LanceFtsQueryBuilder:
+        """Rerank the results using the specified reranker.
+
+        Parameters
+        ----------
+        reranker: Reranker
+            The reranker to use.
+
+        Returns
+        -------
+        LanceFtsQueryBuilder
+            The LanceQueryBuilder object.
+        """
+        self._reranker = reranker
+        return self
+
 
 class LanceEmptyQueryBuilder(LanceQueryBuilder):
     def to_arrow(self) -> pa.Table:
@@ -921,3 +1025,334 @@ class LanceHybridQueryBuilder(LanceQueryBuilder):
         """
         self._vector_query.refine_factor(refine_factor)
         return self
+
+
+class AsyncQueryBase(object):
+    def __init__(self, inner: Union[LanceQuery | LanceVectorQuery]):
+        """
+        Construct an AsyncQueryBase
+
+        This method is not intended to be called directly.  Instead, use the
+        [Table.query][] method to create a query.
+        """
+        self._inner = inner
+
+    def where(self, predicate: str) -> AsyncQuery:
+        """
+        Only return rows matching the given predicate
+
+        The predicate should be supplied as an SQL query string.  For example:
+
+        >>> predicate = "x > 10"
+        >>> predicate = "y > 0 AND y < 100"
+        >>> predicate = "x > 5 OR y = 'test'"
+
+        Filtering performance can often be improved by creating a scalar index
+        on the filter column(s).
+        """
+        self._inner.where(predicate)
+        return self
+
+    def select(self, columns: Union[List[str], dict[str, str]]) -> AsyncQuery:
+        """
+        Return only the specified columns.
+
+        By default a query will return all columns from the table.  However, this can
+        have a very significant impact on latency.  LanceDb stores data in a columnar
+        fashion.  This
+        means we can finely tune our I/O to select exactly the columns we need.
+
+        As a best practice you should always limit queries to the columns that you need.
+        If you pass in a list of column names then only those columns will be
+        returned.
+
+        You can also use this method to create new "dynamic" columns based on your
+        existing columns. For example, you may not care about "a" or "b" but instead
+        simply want "a + b".  This is often seen in the SELECT clause of an SQL query
+        (e.g. `SELECT a+b FROM my_table`).
+
+        To create dynamic columns you can pass in a dict[str, str].  A column will be
+        returned for each entry in the map.  The key provides the name of the column.
+        The value is an SQL string used to specify how the column is calculated.
+
+        For example, an SQL query might state `SELECT a + b AS combined, c`.  The
+        equivalent input to this method would be `{"combined": "a + b", "c": "c"}`.
+
+        Columns will always be returned in the order given, even if that order is
+        different than the order used when adding the data.
+        """
+        if isinstance(columns, dict):
+            column_tuples = list(columns.items())
+        else:
+            try:
+                column_tuples = [(c, c) for c in columns]
+            except TypeError:
+                raise TypeError("columns must be a list of column names or a dict")
+        self._inner.select(column_tuples)
+        return self
+
+    def limit(self, limit: int) -> AsyncQuery:
+        """
+        Set the maximum number of results to return.
+
+        By default, a plain search has no limit.  If this method is not
+        called then every valid row from the table will be returned.
+        """
+        self._inner.limit(limit)
+        return self
+
+    async def to_batches(self) -> AsyncRecordBatchReader:
+        """
+        Execute the query and return the results as an Apache Arrow RecordBatchReader.
+        """
+        return AsyncRecordBatchReader(await self._inner.execute())
+
+    async def to_arrow(self) -> pa.Table:
+        """
+        Execute the query and collect the results into an Apache Arrow Table.
+
+        This method will collect all results into memory before returning.  If
+        you expect a large number of results, you may want to use [to_batches][]
+        """
+        batch_iter = await self.to_batches()
+        return pa.Table.from_batches(
+            await batch_iter.read_all(), schema=batch_iter.schema
+        )
+
+    async def to_pandas(self) -> "pd.DataFrame":
+        """
+        Execute the query and collect the results into a pandas DataFrame.
+
+        This method will collect all results into memory before returning.  If
+        you expect a large number of results, you may want to use [to_batches][]
+        and convert each batch to pandas separately.
+
+        Example
+        -------
+
+        >>> import asyncio
+        >>> from lancedb import connect_async
+        >>> async def doctest_example():
+        ...     conn = await connect_async("./.lancedb")
+        ...     table = await conn.create_table("my_table", data=[{"a": 1, "b": 2}])
+        ...     async for batch in await table.query().to_batches():
+        ...         batch_df = batch.to_pandas()
+        >>> asyncio.run(doctest_example())
+        """
+        return (await self.to_arrow()).to_pandas()
+
+
+class AsyncQuery(AsyncQueryBase):
+    def __init__(self, inner: LanceQuery):
+        """
+        Construct an AsyncQuery
+
+        This method is not intended to be called directly.  Instead, use the
+        [Table.query][] method to create a query.
+        """
+        super().__init__(inner)
+        self._inner = inner
+
+    @classmethod
+    def _query_vec_to_array(self, vec: Union[VEC, Tuple]):
+        if isinstance(vec, list):
+            return pa.array(vec)
+        if isinstance(vec, np.ndarray):
+            return pa.array(vec)
+        if isinstance(vec, pa.Array):
+            return vec
+        if isinstance(vec, pa.ChunkedArray):
+            return vec.combine_chunks()
+        if isinstance(vec, tuple):
+            return pa.array(vec)
+        # We've checked everything we formally support in our typings
+        # but, as a fallback, let pyarrow try and convert it anyway.
+        # This can allow for some more exotic things like iterables
+        return pa.array(vec)
+
+    def nearest_to(
+        self, query_vector: Optional[Union[VEC, Tuple]] = None
+    ) -> AsyncVectorQuery:
+        """
+        Find the nearest vectors to the given query vector.
+
+        This converts the query from a plain query to a vector query.
+
+        This method will attempt to convert the input to the query vector
+        expected by the embedding model.  If the input cannot be converted
+        then an error will be thrown.
+
+        By default, there is no embedding model, and the input should be
+        something that can be converted to a pyarrow array of floats.  This
+        includes lists, numpy arrays, and tuples.
+
+        If there is only one vector column (a column whose data type is a
+        fixed size list of floats) then the column does not need to be specified.
+        If there is more than one vector column you must use
+        [AsyncVectorQuery::column][] to specify which column you would like to
+        compare with.
+
+        If no index has been created on the vector column then a vector query
+        will perform a distance comparison between the query vector and every
+        vector in the database and then sort the results.  This is sometimes
+        called a "flat search"
+
+        For small databases, with tens of thousands of vectors or less, this can
+        be reasonably fast.  In larger databases you should create a vector index
+        on the column.  If there is a vector index then an "approximate" nearest
+        neighbor search (frequently called an ANN search) will be performed.  This
+        search is much faster, but the results will be approximate.
+
+        The query can be further parameterized using the returned builder.  There
+        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.
+        """
+        return AsyncVectorQuery(
+            self._inner.nearest_to(AsyncQuery._query_vec_to_array(query_vector))
+        )
+
+
+class AsyncVectorQuery(AsyncQueryBase):
+    def __init__(self, inner: LanceVectorQuery):
+        """
+        Construct an AsyncVectorQuery
+
+        This method is not intended to be called directly.  Instead, create
+        a query first with [Table.query][] and then use [AsyncQuery.nearest_to][]
+        to convert to a vector query.
+        """
+        super().__init__(inner)
+        self._inner = inner
+
+    def column(self, column: str) -> AsyncVectorQuery:
+        """
+        Set the vector column to query
+
+        This controls which column is compared to the query vector supplied in
+        the call to [Query.nearest_to][].
+
+        This parameter must be specified if the table has more than one column
+        whose data type is a fixed-size-list of floats.
+        """
+        self._inner.column(column)
+        return self
+
+    def nprobes(self, nprobes: int) -> AsyncVectorQuery:
+        """
+        Set the number of partitions to search (probe)
+
+        This argument is only used when the vector column has an IVF PQ index.
+        If there is no index then this value is ignored.
+
+        The IVF stage of IVF PQ divides the input into partitions (clusters) of
+        related values.
+
+        The partition whose centroids are closest to the query vector will be
+        exhaustiely searched to find matches.  This parameter controls how many
+        partitions should be searched.
+
+        Increasing this value will increase the recall of your query but will
+        also increase the latency of your query.  The default value is 20.  This
+        default is good for many cases but the best value to use will depend on
+        your data and the recall that you need to achieve.
+
+        For best results we recommend tuning this parameter with a benchmark against
+        your actual data to find the smallest possible value that will still give
+        you the desired recall.
+        """
+        self._inner.nprobes(nprobes)
+        return self
+
+    def refine_factor(self, refine_factor: int) -> AsyncVectorQuery:
+        """
+        A multiplier to control how many additional rows are taken during the refine
+        step
+
+        This argument is only used when the vector column has an IVF PQ index.
+        If there is no index then this value is ignored.
+
+        An IVF PQ index stores compressed (quantized) values.  They query vector is
+        compared against these values and, since they are compressed, the comparison is
+        inaccurate.
+
+        This parameter can be used to refine the results.  It can improve both improve
+        recall and correct the ordering of the nearest results.
+
+        To refine results LanceDb will first perform an ANN search to find the nearest
+        `limit` * `refine_factor` results.  In other words, if `refine_factor` is 3 and
+        `limit` is the default (10) then the first 30 results will be selected.  LanceDb
+        then fetches the full, uncompressed, values for these 30 results.  The results
+        are then reordered by the true distance and only the nearest 10 are kept.
+
+        Note: there is a difference between calling this method with a value of 1 and
+        never calling this method at all.  Calling this method with any value will have
+        an impact on your search latency.  When you call this method with a
+        `refine_factor` of 1 then LanceDb still needs to fetch the full, uncompressed,
+        values so that it can potentially reorder the results.
+
+        Note: if this method is NOT called then the distances returned in the _distance
+        column will be approximate distances based on the comparison of the quantized
+        query vector and the quantized result vectors.  This can be considerably
+        different than the true distance between the query vector and the actual
+        uncompressed vector.
+        """
+        self._inner.refine_factor(refine_factor)
+        return self
+
+    def distance_type(self, distance_type: str) -> AsyncVectorQuery:
+        """
+        Set the distance metric to use
+
+        When performing a vector search we try and find the "nearest" vectors according
+        to some kind of distance metric.  This parameter controls which distance metric
+        to use.  See @see {@link IvfPqOptions.distanceType} for more details on the
+        different distance metrics available.
+
+        Note: if there is a vector index then the distance type used MUST match the
+        distance type used to train the vector index.  If this is not done then the
+        results will be invalid.
+
+        By default "l2" is used.
+        """
+        self._inner.distance_type(distance_type)
+        return self
+
+    def postfilter(self) -> AsyncVectorQuery:
+        """
+        If this is called then filtering will happen after the vector search instead of
+        before.
+
+        By default filtering will be performed before the vector search.  This is how
+        filtering is typically understood to work.  This prefilter step does add some
+        additional latency.  Creating a scalar index on the filter column(s) can
+        often improve this latency.  However, sometimes a filter is too complex or
+        scalar indices cannot be applied to the column.  In these cases postfiltering
+        can be used instead of prefiltering to improve latency.
+
+        Post filtering applies the filter to the results of the vector search.  This
+        means we only run the filter on a much smaller set of data.  However, it can
+        cause the query to return fewer than `limit` results (or even no results) if
+        none of the nearest results match the filter.
+
+        Post filtering happens during the "refine stage" (described in more detail in
+        @see {@link VectorQuery#refineFactor}).  This means that setting a higher refine
+        factor can often help restore some of the results lost by post filtering.
+        """
+        self._inner.postfilter()
+        return self
+
+    def bypass_vector_index(self) -> AsyncVectorQuery:
+        """
+        If this is called then any vector index is skipped
+
+        An exhaustive (flat) search will be performed.  The query vector will
+        be compared to every vector in the table.  At high scales this can be
+        expensive.  However, this is often still useful.  For example, skipping
+        the vector index can give you ground truth results which you can use to
+        calculate your recall to select an appropriate value for nprobes.
+        """
+        self._inner.bypass_vector_index()
+        return self

python/python/lancedb/remote/table.py

@@ -68,10 +68,16 @@ class RemoteTable(Table):
 
     def list_indices(self):
         """List all the indices on the table"""
-        print(self._name)
         resp = self._conn._client.post(f"/v1/table/{self._name}/index/list/")
         return resp
 
+    def index_stats(self, index_uuid: str):
+        """List all the indices on the table"""
+        resp = self._conn._client.post(
+            f"/v1/table/{self._name}/index/{index_uuid}/stats/"
+        )
+        return resp
+
     def create_scalar_index(
         self,
         column: str,
@@ -289,7 +295,9 @@ class RemoteTable(Table):
             vector_column_name = inf_vector_column_query(self.schema)
         return LanceVectorQueryBuilder(self, query, vector_column_name)
 
-    def _execute_query(self, query: Query) -> pa.Table:
+    def _execute_query(
+        self, query: Query, batch_size: Optional[int] = None
+    ) -> pa.RecordBatchReader:
         if (
             query.vector is not None
             and len(query.vector) > 0
@@ -315,13 +323,12 @@ class RemoteTable(Table):
                 q = query.copy()
                 q.vector = v
                 results.append(submit(self._name, q))
-
             return pa.concat_tables(
                 [add_index(r.result().to_arrow(), i) for i, r in enumerate(results)]
-            )
+            ).to_reader()
         else:
             result = self._conn._client.query(self._name, query)
-            return result.to_arrow()
+            return result.to_arrow().to_reader()
 
     def _do_merge(
         self,

python/python/lancedb/rerankers/base.py

@@ -24,8 +24,59 @@ class Reranker(ABC):
             raise ValueError("score must be either 'relevance' or 'all'")
         self.score = return_score
 
+    def rerank_vector(
+        self,
+        query: str,
+        vector_results: pa.Table,
+    ):
+        """
+        Rerank function receives the result from the vector search.
+        This isn't mandatory to implement
+
+        Parameters
+        ----------
+        query : str
+            The input query
+        vector_results : pa.Table
+            The results from the vector search
+
+        Returns
+        -------
+        pa.Table
+            The reranked results
+        """
+        raise NotImplementedError(
+            f"{self.__class__.__name__} does not implement rerank_vector"
+        )
+
+    def rerank_fts(
+        self,
+        query: str,
+        fts_results: pa.Table,
+    ):
+        """
+        Rerank function receives the result from the FTS search.
+        This isn't mandatory to implement
+
+        Parameters
+        ----------
+        query : str
+            The input query
+        fts_results : pa.Table
+            The results from the FTS search
+
+        Returns
+        -------
+        pa.Table
+            The reranked results
+        """
+        raise NotImplementedError(
+            f"{self.__class__.__name__} does not implement rerank_fts"
+        )
+
     @abstractmethod
     def rerank_hybrid(
+        self,
         query: str,
         vector_results: pa.Table,
         fts_results: pa.Table,
@@ -43,6 +94,11 @@ class Reranker(ABC):
             The results from the vector search
         fts_results : pa.Table
             The results from the FTS search
+
+        Returns
+        -------
+        pa.Table
+            The reranked results
         """
         pass
 

python/python/lancedb/rerankers/cohere.py

@@ -49,14 +49,8 @@ class CohereReranker(Reranker):
             )
         return cohere.Client(os.environ.get("COHERE_API_KEY") or self.api_key)
 
-    def rerank_hybrid(
-        self,
-        query: str,
-        vector_results: pa.Table,
-        fts_results: pa.Table,
-    ):
-        combined_results = self.merge_results(vector_results, fts_results)
-        docs = combined_results[self.column].to_pylist()
+    def _rerank(self, result_set: pa.Table, query: str):
+        docs = result_set[self.column].to_pylist()
         results = self._client.rerank(
             query=query,
             documents=docs,
@@ -66,12 +60,22 @@ class CohereReranker(Reranker):
         indices, scores = list(
             zip(*[(result.index, result.relevance_score) for result in results])
         )  # tuples
-        combined_results = combined_results.take(list(indices))
+        result_set = result_set.take(list(indices))
         # add the scores
-        combined_results = combined_results.append_column(
+        result_set = result_set.append_column(
             "_relevance_score", pa.array(scores, type=pa.float32())
         )
 
+        return result_set
+
+    def rerank_hybrid(
+        self,
+        query: str,
+        vector_results: pa.Table,
+        fts_results: pa.Table,
+    ):
+        combined_results = self.merge_results(vector_results, fts_results)
+        combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
             combined_results = combined_results.drop_columns(["score", "_distance"])
         elif self.score == "all":
@@ -79,3 +83,25 @@ class CohereReranker(Reranker):
                 "return_score='all' not implemented for cohere reranker"
             )
         return combined_results
+
+    def rerank_vector(
+        self,
+        query: str,
+        vector_results: pa.Table,
+    ):
+        result_set = self._rerank(vector_results, query)
+        if self.score == "relevance":
+            result_set = result_set.drop_columns(["_distance"])
+
+        return result_set
+
+    def rerank_fts(
+        self,
+        query: str,
+        fts_results: pa.Table,
+    ):
+        result_set = self._rerank(fts_results, query)
+        if self.score == "relevance":
+            result_set = result_set.drop_columns(["score"])
+
+        return result_set

python/python/lancedb/rerankers/colbert.py

@@ -33,14 +33,8 @@ class ColbertReranker(Reranker):
             "torch"
         )  # import here for faster ops later
 
-    def rerank_hybrid(
-        self,
-        query: str,
-        vector_results: pa.Table,
-        fts_results: pa.Table,
-    ):
-        combined_results = self.merge_results(vector_results, fts_results)
-        docs = combined_results[self.column].to_pylist()
+    def _rerank(self, result_set: pa.Table, query: str):
+        docs = result_set[self.column].to_pylist()
 
         tokenizer, model = self._model
 
@@ -59,14 +53,25 @@ class ColbertReranker(Reranker):
             scores.append(score.item())
 
         # replace the self.column column with the docs
-        combined_results = combined_results.drop(self.column)
-        combined_results = combined_results.append_column(
+        result_set = result_set.drop(self.column)
+        result_set = result_set.append_column(
             self.column, pa.array(docs, type=pa.string())
         )
         # add the scores
-        combined_results = combined_results.append_column(
+        result_set = result_set.append_column(
             "_relevance_score", pa.array(scores, type=pa.float32())
         )
+
+        return result_set
+
+    def rerank_hybrid(
+        self,
+        query: str,
+        vector_results: pa.Table,
+        fts_results: pa.Table,
+    ):
+        combined_results = self.merge_results(vector_results, fts_results)
+        combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
             combined_results = combined_results.drop_columns(["score", "_distance"])
         elif self.score == "all":
@@ -80,6 +85,32 @@ class ColbertReranker(Reranker):
 
         return combined_results
 
+    def rerank_vector(
+        self,
+        query: str,
+        vector_results: pa.Table,
+    ):
+        result_set = self._rerank(vector_results, query)
+        if self.score == "relevance":
+            result_set = result_set.drop_columns(["_distance"])
+
+        result_set = result_set.sort_by([("_relevance_score", "descending")])
+
+        return result_set
+
+    def rerank_fts(
+        self,
+        query: str,
+        fts_results: pa.Table,
+    ):
+        result_set = self._rerank(fts_results, query)
+        if self.score == "relevance":
+            result_set = result_set.drop_columns(["score"])
+
+        result_set = result_set.sort_by([("_relevance_score", "descending")])
+
+        return result_set
+
     @cached_property
     def _model(self):
         transformers = attempt_import_or_raise("transformers")

python/python/lancedb/rerankers/cross_encoder.py

@@ -46,6 +46,16 @@ class CrossEncoderReranker(Reranker):
 
         return cross_encoder
 
+    def _rerank(self, result_set: pa.Table, query: str):
+        passages = result_set[self.column].to_pylist()
+        cross_inp = [[query, passage] for passage in passages]
+        cross_scores = self.model.predict(cross_inp)
+        result_set = result_set.append_column(
+            "_relevance_score", pa.array(cross_scores, type=pa.float32())
+        )
+
+        return result_set
+
     def rerank_hybrid(
         self,
         query: str,
@@ -53,13 +63,7 @@ class CrossEncoderReranker(Reranker):
         fts_results: pa.Table,
     ):
         combined_results = self.merge_results(vector_results, fts_results)
-        passages = combined_results[self.column].to_pylist()
-        cross_inp = [[query, passage] for passage in passages]
-        cross_scores = self.model.predict(cross_inp)
-        combined_results = combined_results.append_column(
-            "_relevance_score", pa.array(cross_scores, type=pa.float32())
-        )
-
+        combined_results = self._rerank(combined_results, query)
         # sort the results by _score
         if self.score == "relevance":
             combined_results = combined_results.drop_columns(["score", "_distance"])
@@ -72,3 +76,27 @@ class CrossEncoderReranker(Reranker):
         )
 
         return combined_results
+
+    def rerank_vector(
+        self,
+        query: str,
+        vector_results: pa.Table,
+    ):
+        vector_results = self._rerank(vector_results, query)
+        if self.score == "relevance":
+            vector_results = vector_results.drop_columns(["_distance"])
+
+        vector_results = vector_results.sort_by([("_relevance_score", "descending")])
+        return vector_results
+
+    def rerank_fts(
+        self,
+        query: str,
+        fts_results: pa.Table,
+    ):
+        fts_results = self._rerank(fts_results, query)
+        if self.score == "relevance":
+            fts_results = fts_results.drop_columns(["score"])
+
+        fts_results = fts_results.sort_by([("_relevance_score", "descending")])
+        return fts_results

python/python/lancedb/rerankers/openai.py

@@ -39,14 +39,8 @@ class OpenaiReranker(Reranker):
         self.column = column
         self.api_key = api_key
 
-    def rerank_hybrid(
-        self,
-        query: str,
-        vector_results: pa.Table,
-        fts_results: pa.Table,
-    ):
-        combined_results = self.merge_results(vector_results, fts_results)
-        docs = combined_results[self.column].to_pylist()
+    def _rerank(self, result_set: pa.Table, query: str):
+        docs = result_set[self.column].to_pylist()
         response = self._client.chat.completions.create(
             model=self.model_name,
             response_format={"type": "json_object"},
@@ -70,14 +64,25 @@ class OpenaiReranker(Reranker):
             zip(*[(result["content"], result["relevance_score"]) for result in results])
         )  # tuples
         # replace the self.column column with the docs
-        combined_results = combined_results.drop(self.column)
-        combined_results = combined_results.append_column(
+        result_set = result_set.drop(self.column)
+        result_set = result_set.append_column(
             self.column, pa.array(docs, type=pa.string())
         )
         # add the scores
-        combined_results = combined_results.append_column(
+        result_set = result_set.append_column(
             "_relevance_score", pa.array(scores, type=pa.float32())
         )
+
+        return result_set
+
+    def rerank_hybrid(
+        self,
+        query: str,
+        vector_results: pa.Table,
+        fts_results: pa.Table,
+    ):
+        combined_results = self.merge_results(vector_results, fts_results)
+        combined_results = self._rerank(combined_results, query)
         if self.score == "relevance":
             combined_results = combined_results.drop_columns(["score", "_distance"])
         elif self.score == "all":
@@ -91,6 +96,24 @@ class OpenaiReranker(Reranker):
 
         return combined_results
 
+    def rerank_vector(self, query: str, vector_results: pa.Table):
+        vector_results = self._rerank(vector_results, query)
+        if self.score == "relevance":
+            vector_results = vector_results.drop_columns(["_distance"])
+
+        vector_results = vector_results.sort_by([("_relevance_score", "descending")])
+
+        return vector_results
+
+    def rerank_fts(self, query: str, fts_results: pa.Table):
+        fts_results = self._rerank(fts_results, query)
+        if self.score == "relevance":
+            fts_results = fts_results.drop_columns(["score"])
+
+        fts_results = fts_results.sort_by([("_relevance_score", "descending")])
+
+        return fts_results
+
     @cached_property
     def _client(self):
         openai = attempt_import_or_raise(

python/python/lancedb/table.py

@@ -37,13 +37,14 @@ import pyarrow as pa
 import pyarrow.compute as pc
 import pyarrow.fs as pa_fs
 from lance import LanceDataset
+from lance.dependencies import _check_for_hugging_face
 from lance.vector import vec_to_table
 
 from .common import DATA, VEC, VECTOR_COLUMN_NAME
 from .embeddings import EmbeddingFunctionConfig, EmbeddingFunctionRegistry
 from .merge import LanceMergeInsertBuilder
 from .pydantic import LanceModel, model_to_dict
-from .query import LanceQueryBuilder, Query
+from .query import AsyncQuery, AsyncVectorQuery, LanceQueryBuilder, Query
 from .util import (
     fs_from_uri,
     inf_vector_column_query,
@@ -60,6 +61,7 @@ if TYPE_CHECKING:
 
     from ._lancedb import Table as LanceDBTable
     from .db import LanceDBConnection
+    from .index import BTree, IndexConfig, IvfPq
 
 
 pd = safe_import_pandas()
@@ -73,6 +75,27 @@ def _sanitize_data(
     on_bad_vectors: str,
     fill_value: Any,
 ):
+    if _check_for_hugging_face(data):
+        # Huggingface datasets
+        from lance.dependencies import datasets
+
+        if isinstance(data, datasets.dataset_dict.DatasetDict):
+            if schema is None:
+                schema = _schema_from_hf(data, schema)
+            data = _to_record_batch_generator(
+                _to_batches_with_split(data),
+                schema,
+                metadata,
+                on_bad_vectors,
+                fill_value,
+            )
+        elif isinstance(data, datasets.Dataset):
+            if schema is None:
+                schema = data.features.arrow_schema
+            data = _to_record_batch_generator(
+                data.data.to_batches(), schema, metadata, on_bad_vectors, fill_value
+            )
+
     if isinstance(data, list):
         # convert to list of dict if data is a bunch of LanceModels
         if isinstance(data[0], LanceModel):
@@ -109,6 +132,37 @@ def _sanitize_data(
     return data
 
 
+def _schema_from_hf(data, schema):
+    """
+    Extract pyarrow schema from HuggingFace DatasetDict
+    and validate that they're all the same schema between
+    splits
+    """
+    for dataset in data.values():
+        if schema is None:
+            schema = dataset.features.arrow_schema
+        elif schema != dataset.features.arrow_schema:
+            msg = "All datasets in a HuggingFace DatasetDict must have the same schema"
+            raise TypeError(msg)
+    return schema
+
+
+def _to_batches_with_split(data):
+    """
+    Return a generator of RecordBatches from a HuggingFace DatasetDict
+    with an extra `split` column
+    """
+    for key, dataset in data.items():
+        for batch in dataset.data.to_batches():
+            table = pa.Table.from_batches([batch])
+            if "split" not in table.column_names:
+                table = table.append_column(
+                    "split", pa.array([key] * batch.num_rows, pa.string())
+                )
+            for b in table.to_batches():
+                yield b
+
+
 def _append_vector_col(data: pa.Table, metadata: dict, schema: Optional[pa.Schema]):
     """
     Use the embedding function to automatically embed the source column and add the
@@ -117,7 +171,8 @@ def _append_vector_col(data: pa.Table, metadata: dict, schema: Optional[pa.Schem
     functions = EmbeddingFunctionRegistry.get_instance().parse_functions(metadata)
     for vector_column, conf in functions.items():
         func = conf.function
-        if vector_column not in data.column_names:
+        no_vector_column = vector_column not in data.column_names
+        if no_vector_column or pc.all(pc.is_null(data[vector_column])).as_py():
             col_data = func.compute_source_embeddings_with_retry(
                 data[conf.source_column]
             )
@@ -125,9 +180,16 @@ def _append_vector_col(data: pa.Table, metadata: dict, schema: Optional[pa.Schem
                 dtype = schema.field(vector_column).type
             else:
                 dtype = pa.list_(pa.float32(), len(col_data[0]))
-            data = data.append_column(
-                pa.field(vector_column, type=dtype), pa.array(col_data, type=dtype)
-            )
+            if no_vector_column:
+                data = data.append_column(
+                    pa.field(vector_column, type=dtype), pa.array(col_data, type=dtype)
+                )
+            else:
+                data = data.set_column(
+                    data.column_names.index(vector_column),
+                    pa.field(vector_column, type=dtype),
+                    pa.array(col_data, type=dtype),
+                )
     return data
 
 
@@ -135,12 +197,13 @@ def _to_record_batch_generator(
     data: Iterable, schema, metadata, on_bad_vectors, fill_value
 ):
     for batch in data:
-        if not isinstance(batch, pa.RecordBatch):
-            table = _sanitize_data(batch, schema, metadata, on_bad_vectors, fill_value)
-            for batch in table.to_batches():
-                yield batch
-        else:
-            yield batch
+        # always convert to table because we need to sanitize the data
+        # and do things like add the vector column etc
+        if isinstance(batch, pa.RecordBatch):
+            batch = pa.Table.from_batches([batch])
+        batch = _sanitize_data(batch, schema, metadata, on_bad_vectors, fill_value)
+        for b in batch.to_batches():
+            yield b
 
 
 class Table(ABC):
@@ -505,7 +568,9 @@ class Table(ABC):
         raise NotImplementedError
 
     @abstractmethod
-    def _execute_query(self, query: Query) -> pa.Table:
+    def _execute_query(
+        self, query: Query, batch_size: Optional[int] = None
+    ) -> pa.RecordBatchReader:
         pass
 
     @abstractmethod
@@ -1096,6 +1161,7 @@ class LanceTable(Table):
     def create_fts_index(
         self,
         field_names: Union[str, List[str]],
+        ordering_field_names: Union[str, List[str]] = None,
         *,
         replace: bool = False,
         writer_heap_size: Optional[int] = 1024 * 1024 * 1024,
@@ -1114,12 +1180,18 @@ class LanceTable(Table):
             not yet an atomic operation; the index will be temporarily
             unavailable while the new index is being created.
         writer_heap_size: int, default 1GB
+        ordering_field_names:
+            A list of unsigned type fields to index to optionally order
+            results on at search time
         """
         from .fts import create_index, populate_index
 
         if isinstance(field_names, str):
             field_names = [field_names]
 
+        if isinstance(ordering_field_names, str):
+            ordering_field_names = [ordering_field_names]
+
         fs, path = fs_from_uri(self._get_fts_index_path())
         index_exists = fs.get_file_info(path).type != pa_fs.FileType.NotFound
         if index_exists:
@@ -1127,8 +1199,18 @@ class LanceTable(Table):
                 raise ValueError("Index already exists. Use replace=True to overwrite.")
             fs.delete_dir(path)
 
-        index = create_index(self._get_fts_index_path(), field_names)
-        populate_index(index, self, field_names, writer_heap_size=writer_heap_size)
+        index = create_index(
+            self._get_fts_index_path(),
+            field_names,
+            ordering_fields=ordering_field_names,
+        )
+        populate_index(
+            index,
+            self,
+            field_names,
+            ordering_fields=ordering_field_names,
+            writer_heap_size=writer_heap_size,
+        )
         register_event("create_fts_index")
 
     def _get_fts_index_path(self):
@@ -1262,6 +1344,7 @@ class LanceTable(Table):
         query: Optional[Union[VEC, str, "PIL.Image.Image", Tuple]] = None,
         vector_column_name: Optional[str] = None,
         query_type: str = "auto",
+        ordering_field_name: Optional[str] = None,
     ) -> LanceQueryBuilder:
         """Create a search query to find the nearest neighbors
         of the given query vector. We currently support [vector search][search]
@@ -1329,7 +1412,11 @@ class LanceTable(Table):
             vector_column_name = inf_vector_column_query(self.schema)
         register_event("search_table")
         return LanceQueryBuilder.create(
-            self, query, query_type, vector_column_name=vector_column_name
+            self,
+            query,
+            query_type,
+            vector_column_name=vector_column_name,
+            ordering_field_name=ordering_field_name,
         )
 
     @classmethod
@@ -1511,10 +1598,11 @@ class LanceTable(Table):
         self._dataset_mut.update(values_sql, where)
         register_event("update")
 
-    def _execute_query(self, query: Query) -> pa.Table:
+    def _execute_query(
+        self, query: Query, batch_size: Optional[int] = None
+    ) -> pa.RecordBatchReader:
         ds = self.to_lance()
-
-        return ds.to_table(
+        return ds.scanner(
             columns=query.columns,
             filter=query.filter,
             prefilter=query.prefilter,
@@ -1527,7 +1615,8 @@ class LanceTable(Table):
                 "refine_factor": query.refine_factor,
             },
             with_row_id=query.with_row_id,
-        )
+            batch_size=batch_size,
+        ).to_reader()
 
     def _do_merge(
         self,
@@ -1898,6 +1987,9 @@ class AsyncTable:
         """
         return await self._inner.count_rows(filter)
 
+    def query(self) -> AsyncQuery:
+        return AsyncQuery(self._inner.query())
+
     async def to_pandas(self) -> "pd.DataFrame":
         """Return the table as a pandas DataFrame.
 
@@ -1905,7 +1997,7 @@ class AsyncTable:
         -------
         pd.DataFrame
         """
-        return self.to_arrow().to_pandas()
+        return (await self.to_arrow()).to_pandas()
 
     async def to_arrow(self) -> pa.Table:
         """Return the table as a pyarrow Table.
@@ -1914,115 +2006,51 @@ class AsyncTable:
         -------
         pa.Table
         """
-        raise NotImplementedError
+        return await self.query().to_arrow()
 
     async def create_index(
-        self,
-        metric="L2",
-        num_partitions=256,
-        num_sub_vectors=96,
-        vector_column_name: str = VECTOR_COLUMN_NAME,
-        replace: bool = True,
-        accelerator: Optional[str] = None,
-        index_cache_size: Optional[int] = None,
-    ):
-        """Create an index on the table.
-
-        Parameters
-        ----------
-        metric: str, default "L2"
-            The distance metric to use when creating the index.
-            Valid values are "L2", "cosine", or "dot".
-            L2 is euclidean distance.
-        num_partitions: int, default 256
-            The number of IVF partitions to use when creating the index.
-            Default is 256.
-        num_sub_vectors: int, default 96
-            The number of PQ sub-vectors to use when creating the index.
-            Default is 96.
-        vector_column_name: str, default "vector"
-            The vector column name to create the index.
-        replace: bool, default True
-            - If True, replace the existing index if it exists.
-
-            - If False, raise an error if duplicate index exists.
-        accelerator: str, default None
-            If set, use the given accelerator to create the index.
-            Only support "cuda" for now.
-        index_cache_size : int, optional
-            The size of the index cache in number of entries. Default value is 256.
-        """
-        raise NotImplementedError
-
-    async def create_scalar_index(
         self,
         column: str,
         *,
-        replace: bool = True,
+        replace: Optional[bool] = None,
+        config: Optional[Union[IvfPq, BTree]] = None,
     ):
-        """Create a scalar index on a column.
+        """Create an index to speed up queries
 
-        Scalar indices, like vector indices, can be used to speed up scans.  A scalar
-        index can speed up scans that contain filter expressions on the indexed column.
-        For example, the following scan will be faster if the column ``my_col`` has
-        a scalar index:
-
-        .. code-block:: python
-
-            import lancedb
-
-            db = lancedb.connect("/data/lance")
-            img_table = db.open_table("images")
-            my_df = img_table.search().where("my_col = 7", prefilter=True).to_pandas()
-
-        Scalar indices can also speed up scans containing a vector search and a
-        prefilter:
-
-        .. code-block::python
-
-            import lancedb
-
-            db = lancedb.connect("/data/lance")
-            img_table = db.open_table("images")
-            img_table.search([1, 2, 3, 4], vector_column_name="vector")
-                .where("my_col != 7", prefilter=True)
-                .to_pandas()
-
-        Scalar indices can only speed up scans for basic filters using
-        equality, comparison, range (e.g. ``my_col BETWEEN 0 AND 100``), and set
-        membership (e.g. `my_col IN (0, 1, 2)`)
-
-        Scalar indices can be used if the filter contains multiple indexed columns and
-        the filter criteria are AND'd or OR'd together
-        (e.g. ``my_col < 0 AND other_col> 100``)
-
-        Scalar indices may be used if the filter contains non-indexed columns but,
-        depending on the structure of the filter, they may not be usable.  For example,
-        if the column ``not_indexed`` does not have a scalar index then the filter
-        ``my_col = 0 OR not_indexed = 1`` will not be able to use any scalar index on
-        ``my_col``.
-
-        **Experimental API**
+        Indices can be created on vector columns or scalar columns.
+        Indices on vector columns will speed up vector searches.
+        Indices on scalar columns will speed up filtering (in both
+        vector and non-vector searches)
 
         Parameters
         ----------
-        column : str
-            The column to be indexed.  Must be a boolean, integer, float,
-            or string column.
-        replace : bool, default True
-            Replace the existing index if it exists.
+        index: Index
+            The index to create.
 
-        Examples
-        --------
+            LanceDb supports multiple types of indices.  See the static methods on
+            the Index class for more details.
+        column: str, default None
+            The column to index.
 
-        .. code-block:: python
+            When building a scalar index this must be set.
 
-            import lance
+            When building a vector index, this is optional.  The default will look
+            for any columns of type fixed-size-list with floating point values.  If
+            there is only one column of this type then it will be used.  Otherwise
+            an error will be returned.
+        replace: bool, default True
+            Whether to replace the existing index
 
-            dataset = lance.dataset("./images.lance")
-            dataset.create_scalar_index("category")
+            If this is false, and another index already exists on the same columns
+            and the same name, then an error will be returned.  This is true even if
+            that index is out of date.
+
+            The default is True
         """
-        raise NotImplementedError
+        index = None
+        if config is not None:
+            index = config._inner
+        await self._inner.create_index(column, index=index, replace=replace)
 
     async def add(
         self,
@@ -2066,6 +2094,8 @@ class AsyncTable:
             on_bad_vectors=on_bad_vectors,
             fill_value=fill_value,
         )
+        if isinstance(data, pa.Table):
+            data = pa.RecordBatchReader.from_batches(data.schema, data.to_batches())
         await self._inner.add(data, mode)
         register_event("add")
 
@@ -2129,89 +2159,21 @@ class AsyncTable:
 
         return LanceMergeInsertBuilder(self, on)
 
-    async def search(
+    def vector_search(
         self,
-        query: Optional[Union[VEC, str, "PIL.Image.Image", Tuple]] = None,
-        vector_column_name: Optional[str] = None,
-        query_type: str = "auto",
-    ) -> 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].
-
-        All query options are defined in [Query][lancedb.query.Query].
-
-        Examples
-        --------
-        >>> import lancedb
-        >>> db = lancedb.connect("./.lancedb")
-        >>> data = [
-        ...    {"original_width": 100, "caption": "bar", "vector": [0.1, 2.3, 4.5]},
-        ...    {"original_width": 2000, "caption": "foo",  "vector": [0.5, 3.4, 1.3]},
-        ...    {"original_width": 3000, "caption": "test", "vector": [0.3, 6.2, 2.6]}
-        ... ]
-        >>> table = db.create_table("my_table", data)
-        >>> query = [0.4, 1.4, 2.4]
-        >>> (table.search(query)
-        ...     .where("original_width > 1000", prefilter=True)
-        ...     .select(["caption", "original_width", "vector"])
-        ...     .limit(2)
-        ...     .to_pandas())
-          caption  original_width           vector  _distance
-        0     foo            2000  [0.5, 3.4, 1.3]   5.220000
-        1    test            3000  [0.3, 6.2, 2.6]  23.089996
-
-        Parameters
-        ----------
-        query: list/np.ndarray/str/PIL.Image.Image, default None
-            The targetted vector to search for.
-
-            - *default None*.
-            Acceptable types are: list, np.ndarray, PIL.Image.Image
-
-            - If None then the select/where/limit clauses are applied to filter
-            the table
-        vector_column_name: str, optional
-            The name of the vector column to search.
-
-            The vector column needs to be a pyarrow fixed size list type
-
-            - If not specified then the vector column is inferred from
-            the table schema
-
-            - If the table has multiple vector columns then the *vector_column_name*
-            needs to be specified. Otherwise, an error is raised.
-        query_type: str
-            *default "auto"*.
-            Acceptable types are: "vector", "fts", "hybrid", or "auto"
-
-            - If "auto" then the query type is inferred from the query;
-
-                - If `query` is a list/np.ndarray then the query type is
-                "vector";
-
-                - If `query` is a PIL.Image.Image then either do vector search,
-                or raise an error if no corresponding embedding function is found.
-
-            - If `query` is a string, then the query type is "vector" if the
-            table has embedding functions else the query type is "fts"
-
-        Returns
-        -------
-        LanceQueryBuilder
-            A query builder object representing the query.
-            Once executed, the query returns
-
-            - selected columns
-
-            - the vector
-
-            - and also the "_distance" column which is the distance between the query
-            vector and the returned vector.
+        query_vector: Optional[Union[VEC, Tuple]] = None,
+    ) -> AsyncVectorQuery:
         """
-        raise NotImplementedError
+        Search the table with a given query vector.
+        This is a convenience method for preparing a vector query and
+        is the same thing as calling `nearestTo` on the builder returned
+        by `query`.  Seer [nearest_to][AsyncQuery.nearest_to] for more details.
+        """
+        return self.query().nearest_to(query_vector)
 
-    async def _execute_query(self, query: Query) -> pa.Table:
+    async def _execute_query(
+        self, query: Query, batch_size: Optional[int] = None
+    ) -> pa.RecordBatchReader:
         pass
 
     async def _do_merge(
@@ -2275,58 +2237,57 @@ class AsyncTable:
 
     async def update(
         self,
-        where: Optional[str] = None,
-        values: Optional[dict] = None,
+        updates: Optional[Dict[str, Any]] = None,
         *,
-        values_sql: Optional[Dict[str, str]] = None,
+        where: Optional[str] = None,
+        updates_sql: Optional[Dict[str, str]] = None,
     ):
         """
-        This can be used to update zero to all rows depending on how many
-        rows match the where clause. If no where clause is provided, then
-        all rows will be updated.
+        This can be used to update zero to all rows in the table.
 
-        Either `values` or `values_sql` must be provided. You cannot provide
-        both.
+        If a filter is provided with `where` then only rows matching the
+        filter will be updated.  Otherwise all rows will be updated.
 
         Parameters
         ----------
+        updates: dict, optional
+            The updates to apply.  The keys should be the name of the column to
+            update.  The values should be the new values to assign.  This is
+            required unless updates_sql is supplied.
         where: str, optional
-            The SQL where clause to use when updating rows. For example, 'x = 2'
-            or 'x IN (1, 2, 3)'. The filter must not be empty, or it will error.
-        values: dict, optional
-            The values to update. The keys are the column names and the values
-            are the values to set.
-        values_sql: dict, optional
-            The values to update, expressed as SQL expression strings. These can
-            reference existing columns. For example, {"x": "x + 1"} will increment
-            the x column by 1.
+            An SQL filter that controls which rows are updated. For example, 'x = 2'
+            or 'x IN (1, 2, 3)'.  Only rows that satisfy this filter will be udpated.
+        updates_sql: dict, optional
+            The updates to apply, expressed as SQL expression strings.  The keys should
+            be column names. The values should be SQL expressions.  These can be SQL
+            literals (e.g. "7" or "'foo'") or they can be expressions based on the
+            previous value of the row (e.g. "x + 1" to increment the x column by 1)
 
         Examples
         --------
+        >>> import asyncio
         >>> import lancedb
         >>> import pandas as pd
-        >>> data = pd.DataFrame({"x": [1, 2, 3], "vector": [[1, 2], [3, 4], [5, 6]]})
-        >>> db = lancedb.connect("./.lancedb")
-        >>> table = db.create_table("my_table", data)
-        >>> table.to_pandas()
-           x      vector
-        0  1  [1.0, 2.0]
-        1  2  [3.0, 4.0]
-        2  3  [5.0, 6.0]
-        >>> table.update(where="x = 2", values={"vector": [10, 10]})
-        >>> table.to_pandas()
-           x        vector
-        0  1    [1.0, 2.0]
-        1  3    [5.0, 6.0]
-        2  2  [10.0, 10.0]
-        >>> table.update(values_sql={"x": "x + 1"})
-        >>> table.to_pandas()
-           x        vector
-        0  2    [1.0, 2.0]
-        1  4    [5.0, 6.0]
-        2  3  [10.0, 10.0]
+        >>> async def demo_update():
+        ...     data = pd.DataFrame({"x": [1, 2], "vector": [[1, 2], [3, 4]]})
+        ...     db = await lancedb.connect_async("./.lancedb")
+        ...     table = await db.create_table("my_table", data)
+        ...     # x is [1, 2], vector is [[1, 2], [3, 4]]
+        ...     await table.update({"vector": [10, 10]}, where="x = 2")
+        ...     # x is [1, 2], vector is [[1, 2], [10, 10]]
+        ...     await table.update(updates_sql={"x": "x + 1"})
+        ...     # x is [2, 3], vector is [[1, 2], [10, 10]]
+        >>> asyncio.run(demo_update())
         """
-        raise NotImplementedError
+        if updates is not None and updates_sql is not None:
+            raise ValueError("Only one of updates or updates_sql can be provided")
+        if updates is None and updates_sql is None:
+            raise ValueError("Either updates or updates_sql must be provided")
+
+        if updates is not None:
+            updates_sql = {k: value_to_sql(v) for k, v in updates.items()}
+
+        return await self._inner.update(updates_sql, where)
 
     async def cleanup_old_versions(
         self,
@@ -2423,3 +2384,65 @@ class AsyncTable:
             The names of the columns to drop.
         """
         raise NotImplementedError
+
+    async def version(self) -> int:
+        """
+        Retrieve the version of the table
+
+        LanceDb supports versioning.  Every operation that modifies the table increases
+        version.  As long as a version hasn't been deleted you can `[Self::checkout]`
+        that version to view the data at that point.  In addition, you can
+        `[Self::restore]` the version to replace the current table with a previous
+        version.
+        """
+        return await self._inner.version()
+
+    async def checkout(self, version):
+        """
+        Checks out a specific version of the Table
+
+        Any read operation on the table will now access the data at the checked out
+        version. As a consequence, calling this method will disable any read consistency
+        interval that was previously set.
+
+        This is a read-only operation that turns the table into a sort of "view"
+        or "detached head".  Other table instances will not be affected.  To make the
+        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.
+
+        To return the table to a normal state use `[Self::checkout_latest]`
+        """
+        await self._inner.checkout(version)
+
+    async def checkout_latest(self):
+        """
+        Ensures the table is pointing at the latest version
+
+        This can be used to manually update a table when the read_consistency_interval
+        is None
+        It can also be used to undo a `[Self::checkout]` operation
+        """
+        await self._inner.checkout_latest()
+
+    async def restore(self):
+        """
+        Restore the table to the currently checked out version
+
+        This operation will fail if checkout has not been called previously
+
+        This operation will overwrite the latest version of the table with a
+        previous version.  Any changes made since the checked out version will
+        no longer be visible.
+
+        Once the operation concludes the table will no longer be in a checked
+        out state and the read_consistency_interval, if any, will apply.
+        """
+        await self._inner.restore()
+
+    async def list_indices(self) -> IndexConfig:
+        """
+        List all indices that have been created with Self::create_index
+        """
+        return await self._inner.list_indices()

python/python/lancedb/util.py

@@ -26,6 +26,18 @@ import pyarrow as pa
 import pyarrow.fs as pa_fs
 
 
+def safe_import_adlfs():
+    try:
+        import adlfs
+
+        return adlfs
+    except ImportError:
+        return None
+
+
+adlfs = safe_import_adlfs()
+
+
 def get_uri_scheme(uri: str) -> str:
     """
     Get the scheme of a URI. If the URI does not have a scheme, assume it is a file URI.
@@ -92,6 +104,17 @@ def fs_from_uri(uri: str) -> Tuple[pa_fs.FileSystem, str]:
         path = get_uri_location(uri)
         return fs, path
 
+    elif get_uri_scheme(uri) == "az" and adlfs is not None:
+        az_blob_fs = adlfs.AzureBlobFileSystem(
+            account_name=os.environ.get("AZURE_STORAGE_ACCOUNT_NAME"),
+            account_key=os.environ.get("AZURE_STORAGE_ACCOUNT_KEY"),
+        )
+
+        fs = pa_fs.PyFileSystem(pa_fs.FSSpecHandler(az_blob_fs))
+
+        path = get_uri_location(uri)
+        return fs, path
+
     return pa_fs.FileSystem.from_uri(uri)
 
 

python/python/lancedb/utils/events.py

@@ -69,7 +69,7 @@ class _Events:
         self.throttled_event_names = ["search_table"]
         self.throttled_events = set()
         self.max_events = 5  # max events to store in memory
-        self.rate_limit = 60.0 * 5  # rate limit (seconds)
+        self.rate_limit = 60.0 * 60.0  # rate limit (seconds)
         self.time = 0.0
 
         if is_git_dir():

python/python/tests/test_embeddings.py

@@ -11,6 +11,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 import sys
+from typing import List, Union
 
 import lance
 import lancedb
@@ -23,6 +24,8 @@ from lancedb.embeddings import (
     EmbeddingFunctionRegistry,
     with_embeddings,
 )
+from lancedb.embeddings.base import TextEmbeddingFunction
+from lancedb.embeddings.registry import get_registry, register
 from lancedb.pydantic import LanceModel, Vector
 
 
@@ -112,3 +115,34 @@ def test_embedding_function_rate_limit(tmp_path):
     table.add([{"text": "hello world"}])
     table.add([{"text": "hello world"}])
     assert len(table) == 2
+
+
+def test_add_optional_vector(tmp_path):
+    @register("mock-embedding")
+    class MockEmbeddingFunction(TextEmbeddingFunction):
+        def ndims(self):
+            return 128
+
+        def generate_embeddings(
+            self, texts: Union[List[str], np.ndarray]
+        ) -> List[np.array]:
+            """
+            Generate the embeddings for the given texts
+            """
+            return [np.random.randn(self.ndims()).tolist() for _ in range(len(texts))]
+
+    registry = get_registry()
+    model = registry.get("mock-embedding").create()
+
+    class LanceSchema(LanceModel):
+        id: str
+        vector: Vector(model.ndims()) = model.VectorField(default=None)
+        text: str = model.SourceField()
+
+    db = lancedb.connect(tmp_path)
+    tbl = db.create_table("optional_vector", schema=LanceSchema)
+
+    # add works
+    expected = LanceSchema(id="id", text="text")
+    tbl.add([expected])
+    assert not (np.abs(tbl.to_pandas()["vector"][0]) < 1e-6).all()

python/python/tests/test_fts.py

@@ -43,6 +43,7 @@ def table(tmp_path) -> ldb.table.LanceTable:
         )
         for _ in range(100)
     ]
+    count = [random.randint(1, 10000) for _ in range(100)]
     table = db.create_table(
         "test",
         data=pd.DataFrame(
@@ -52,6 +53,7 @@ def table(tmp_path) -> ldb.table.LanceTable:
                 "text": text,
                 "text2": text,
                 "nested": [{"text": t} for t in text],
+                "count": count,
             }
         ),
     )
@@ -79,6 +81,39 @@ def test_search_index(tmp_path, table):
     assert len(results[1]) == 10  # _distance
 
 
+def test_search_ordering_field_index_table(tmp_path, table):
+    table.create_fts_index("text", ordering_field_names=["count"])
+    rows = (
+        table.search("puppy", ordering_field_name="count")
+        .limit(20)
+        .select(["text", "count"])
+        .to_list()
+    )
+    for r in rows:
+        assert "puppy" in r["text"]
+    assert sorted(rows, key=lambda x: x["count"], reverse=True) == rows
+
+
+def test_search_ordering_field_index(tmp_path, table):
+    index = ldb.fts.create_index(
+        str(tmp_path / "index"), ["text"], ordering_fields=["count"]
+    )
+
+    ldb.fts.populate_index(index, table, ["text"], ordering_fields=["count"])
+    index.reload()
+    results = ldb.fts.search_index(
+        index, query="puppy", limit=10, ordering_field="count"
+    )
+    assert len(results) == 2
+    assert len(results[0]) == 10  # row_ids
+    assert len(results[1]) == 10  # _distance
+    rows = table.to_lance().take(results[0]).to_pylist()
+
+    for r in rows:
+        assert "puppy" in r["text"]
+    assert sorted(rows, key=lambda x: x["count"], reverse=True) == rows
+
+
 def test_create_index_from_table(tmp_path, table):
     table.create_fts_index("text")
     df = table.search("puppy").limit(10).select(["text"]).to_pandas()
@@ -94,6 +129,7 @@ def test_create_index_from_table(tmp_path, table):
                 "text": "gorilla",
                 "text2": "gorilla",
                 "nested": {"text": "gorilla"},
+                "count": 10,
             }
         ]
     )
@@ -166,6 +202,7 @@ def test_null_input(table):
                 "text": None,
                 "text2": None,
                 "nested": {"text": None},
+                "count": 7,
             }
         ]
     )

python/python/tests/test_huggingface.py

@@ -0,0 +1,126 @@
+#  Copyright 2024 Lance Developers
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+
+from pathlib import Path
+
+import lancedb
+import numpy as np
+import pyarrow as pa
+import pytest
+from lancedb.embeddings import get_registry
+from lancedb.embeddings.base import TextEmbeddingFunction
+from lancedb.embeddings.registry import register
+from lancedb.pydantic import LanceModel, Vector
+
+datasets = pytest.importorskip("datasets")
+
+
+@pytest.fixture(scope="session")
+def mock_embedding_function():
+    @register("random")
+    class MockTextEmbeddingFunction(TextEmbeddingFunction):
+        def generate_embeddings(self, texts):
+            return [np.random.randn(128).tolist() for _ in range(len(texts))]
+
+        def ndims(self):
+            return 128
+
+
+@pytest.fixture
+def mock_hf_dataset():
+    # Create pyarrow table with `text` and `label` columns
+    train = datasets.Dataset(
+        pa.table(
+            {
+                "text": ["foo", "bar"],
+                "label": [0, 1],
+            }
+        ),
+        split="train",
+    )
+
+    test = datasets.Dataset(
+        pa.table(
+            {
+                "text": ["fizz", "buzz"],
+                "label": [0, 1],
+            }
+        ),
+        split="test",
+    )
+    return datasets.DatasetDict({"train": train, "test": test})
+
+
+@pytest.fixture
+def hf_dataset_with_split():
+    # Create pyarrow table with `text` and `label` columns
+    train = datasets.Dataset(
+        pa.table(
+            {"text": ["foo", "bar"], "label": [0, 1], "split": ["train", "train"]}
+        ),
+        split="train",
+    )
+
+    test = datasets.Dataset(
+        pa.table(
+            {"text": ["fizz", "buzz"], "label": [0, 1], "split": ["test", "test"]}
+        ),
+        split="test",
+    )
+    return datasets.DatasetDict({"train": train, "test": test})
+
+
+def test_write_hf_dataset(tmp_path: Path, mock_embedding_function, mock_hf_dataset):
+    db = lancedb.connect(tmp_path)
+    emb = get_registry().get("random").create()
+
+    class Schema(LanceModel):
+        text: str = emb.SourceField()
+        label: int
+        vector: Vector(emb.ndims()) = emb.VectorField()
+
+    train_table = db.create_table("train", schema=Schema)
+    train_table.add(mock_hf_dataset["train"])
+
+    class WithSplit(LanceModel):
+        text: str = emb.SourceField()
+        label: int
+        vector: Vector(emb.ndims()) = emb.VectorField()
+        split: str
+
+    full_table = db.create_table("full", schema=WithSplit)
+    full_table.add(mock_hf_dataset)
+
+    assert len(train_table) == mock_hf_dataset["train"].num_rows
+    assert len(full_table) == sum(ds.num_rows for ds in mock_hf_dataset.values())
+
+    rt_train_table = full_table.to_lance().to_table(
+        columns=["text", "label"], filter="split='train'"
+    )
+    assert rt_train_table.to_pylist() == mock_hf_dataset["train"].data.to_pylist()
+
+
+def test_bad_hf_dataset(tmp_path: Path, mock_embedding_function, hf_dataset_with_split):
+    db = lancedb.connect(tmp_path)
+    emb = get_registry().get("random").create()
+
+    class Schema(LanceModel):
+        text: str = emb.SourceField()
+        label: int
+        vector: Vector(emb.ndims()) = emb.VectorField()
+        split: str
+
+    train_table = db.create_table("train", schema=Schema)
+    # this should still work because we don't add the split column
+    # if it already exists
+    train_table.add(hf_dataset_with_split)

python/python/tests/test_index.py

@@ -0,0 +1,69 @@
+from datetime import timedelta
+
+import pyarrow as pa
+import pytest
+import pytest_asyncio
+from lancedb import AsyncConnection, AsyncTable, connect_async
+from lancedb.index import BTree, IvfPq
+
+
+@pytest_asyncio.fixture
+async def db_async(tmp_path) -> AsyncConnection:
+    return await connect_async(tmp_path, read_consistency_interval=timedelta(seconds=0))
+
+
+def sample_fixed_size_list_array(nrows, dim):
+    vector_data = pa.array([float(i) for i in range(dim * nrows)], pa.float32())
+    return pa.FixedSizeListArray.from_arrays(vector_data, dim)
+
+
+DIM = 8
+NROWS = 256
+
+
+@pytest_asyncio.fixture
+async def some_table(db_async):
+    data = pa.Table.from_pydict(
+        {
+            "id": list(range(256)),
+            "vector": sample_fixed_size_list_array(NROWS, DIM),
+        }
+    )
+    return await db_async.create_table(
+        "some_table",
+        data,
+    )
+
+
+@pytest.mark.asyncio
+async def test_create_scalar_index(some_table: AsyncTable):
+    # Can create
+    await some_table.create_index("id")
+    # Can recreate if replace=True
+    await some_table.create_index("id", replace=True)
+    indices = await some_table.list_indices()
+    assert len(indices) == 1
+    assert indices[0].index_type == "BTree"
+    assert indices[0].columns == ["id"]
+    # Can't recreate if replace=False
+    with pytest.raises(RuntimeError, match="already exists"):
+        await some_table.create_index("id", replace=False)
+    # can also specify index type
+    await some_table.create_index("id", config=BTree())
+
+
+@pytest.mark.asyncio
+async def test_create_vector_index(some_table: AsyncTable):
+    # Can create
+    await some_table.create_index("vector")
+    # Can recreate if replace=True
+    await some_table.create_index("vector", replace=True)
+    # Can't recreate if replace=False
+    with pytest.raises(RuntimeError, match="already exists"):
+        await some_table.create_index("vector", replace=False)
+    # Can also specify index type
+    await some_table.create_index("vector", config=IvfPq(num_partitions=100))
+    indices = await some_table.list_indices()
+    assert len(indices) == 1
+    assert indices[0].index_type == "IvfPq"
+    assert indices[0].columns == ["vector"]

python/python/tests/test_io.py

@@ -16,16 +16,35 @@ import os
 import lancedb
 import pytest
 
+# AWS:
 # You need to setup AWS credentials an a base path to run this test. Example
 #    AWS_PROFILE=default TEST_S3_BASE_URL=s3://my_bucket/dataset pytest tests/test_io.py
+#
+# Azure:
+# You need to setup Azure credentials an a base path to run this test. Example
+#   export AZURE_STORAGE_ACCOUNT_NAME="<account>"
+#   export AZURE_STORAGE_ACCOUNT_KEY="<key>"
+#   export REMOTE_BASE_URL=az://my_blob/dataset
+#   pytest tests/test_io.py
+
+
+@pytest.fixture(autouse=True, scope="module")
+def setup():
+    yield
+
+    if remote_url := os.environ.get("REMOTE_BASE_URL"):
+        db = lancedb.connect(remote_url)
+
+        for table in db.table_names():
+            db.drop_table(table)
 
 
 @pytest.mark.skipif(
-    (os.environ.get("TEST_S3_BASE_URL") is None),
-    reason="please setup s3 base url",
+    (os.environ.get("REMOTE_BASE_URL") is None),
+    reason="please setup remote base url",
 )
-def test_s3_io():
-    db = lancedb.connect(os.environ.get("TEST_S3_BASE_URL"))
+def test_remote_io():
+    db = lancedb.connect(os.environ.get("REMOTE_BASE_URL"))
     assert db.table_names() == []
 
     table = db.create_table(

python/python/tests/test_query.py

@@ -12,16 +12,20 @@
 #  limitations under the License.
 
 import unittest.mock as mock
+from datetime import timedelta
+from typing import Optional
 
 import lance
+import lancedb
 import numpy as np
 import pandas.testing as tm
 import pyarrow as pa
 import pytest
+import pytest_asyncio
 from lancedb.db import LanceDBConnection
 from lancedb.pydantic import LanceModel, Vector
-from lancedb.query import LanceVectorQueryBuilder, Query
-from lancedb.table import LanceTable
+from lancedb.query import AsyncQueryBase, LanceVectorQueryBuilder, Query
+from lancedb.table import AsyncTable, LanceTable
 
 
 class MockTable:
@@ -32,9 +36,9 @@ class MockTable:
     def to_lance(self):
         return lance.dataset(self.uri)
 
-    def _execute_query(self, query):
+    def _execute_query(self, query, batch_size: Optional[int] = None):
         ds = self.to_lance()
-        return ds.to_table(
+        return ds.scanner(
             columns=query.columns,
             filter=query.filter,
             prefilter=query.prefilter,
@@ -46,7 +50,8 @@ class MockTable:
                 "nprobes": query.nprobes,
                 "refine_factor": query.refine_factor,
             },
-        )
+            batch_size=batch_size,
+        ).to_reader()
 
 
 @pytest.fixture
@@ -65,6 +70,24 @@ def table(tmp_path) -> MockTable:
     return MockTable(tmp_path)
 
 
+@pytest_asyncio.fixture
+async def table_async(tmp_path) -> AsyncTable:
+    conn = await lancedb.connect_async(
+        tmp_path, read_consistency_interval=timedelta(seconds=0)
+    )
+    data = pa.table(
+        {
+            "vector": pa.array(
+                [[1, 2], [3, 4]], type=pa.list_(pa.float32(), list_size=2)
+            ),
+            "id": pa.array([1, 2]),
+            "str_field": pa.array(["a", "b"]),
+            "float_field": pa.array([1.0, 2.0]),
+        }
+    )
+    return await conn.create_table("test", data)
+
+
 def test_cast(table):
     class TestModel(LanceModel):
         vector: Vector(2)
@@ -94,6 +117,25 @@ def test_query_builder(table):
     assert all(np.array(rs[0]["vector"]) == [1, 2])
 
 
+def test_query_builder_batches(table):
+    rs = (
+        LanceVectorQueryBuilder(table, [0, 0], "vector")
+        .limit(2)
+        .select(["id", "vector"])
+        .to_batches(1)
+    )
+    rs_list = []
+    for item in rs:
+        rs_list.append(item)
+        assert isinstance(item, pa.RecordBatch)
+    assert len(rs_list) == 1
+    assert len(rs_list[0]["id"]) == 2
+    assert all(rs_list[0].to_pandas()["vector"][0] == [1.0, 2.0])
+    assert rs_list[0].to_pandas()["id"][0] == 1
+    assert all(rs_list[0].to_pandas()["vector"][1] == [3.0, 4.0])
+    assert rs_list[0].to_pandas()["id"][1] == 2
+
+
 def test_dynamic_projection(table):
     rs = (
         LanceVectorQueryBuilder(table, [0, 0], "vector")
@@ -178,9 +220,116 @@ def test_query_builder_with_different_vector_column():
             nprobes=20,
             refine_factor=None,
             vector_column="foo_vector",
-        )
+        ),
+        None,
     )
 
 
 def cosine_distance(vec1, vec2):
     return 1 - np.dot(vec1, vec2) / (np.linalg.norm(vec1) * np.linalg.norm(vec2))
+
+
+async def check_query(
+    query: AsyncQueryBase, *, expected_num_rows=None, expected_columns=None
+):
+    num_rows = 0
+    results = await query.to_batches()
+    async for batch in results:
+        if expected_columns is not None:
+            assert batch.schema.names == expected_columns
+        num_rows += batch.num_rows
+    if expected_num_rows is not None:
+        assert num_rows == expected_num_rows
+
+
+@pytest.mark.asyncio
+async def test_query_async(table_async: AsyncTable):
+    await check_query(
+        table_async.query(),
+        expected_num_rows=2,
+        expected_columns=["vector", "id", "str_field", "float_field"],
+    )
+    await check_query(table_async.query().where("id = 2"), expected_num_rows=1)
+    await check_query(
+        table_async.query().select(["id", "vector"]), expected_columns=["id", "vector"]
+    )
+    await check_query(
+        table_async.query().select({"foo": "id", "bar": "id + 1"}),
+        expected_columns=["foo", "bar"],
+    )
+    await check_query(table_async.query().limit(1), expected_num_rows=1)
+    await check_query(
+        table_async.query().nearest_to(pa.array([1, 2])), expected_num_rows=2
+    )
+    # Support different types of inputs for the vector query
+    for vector_query in [
+        [1, 2],
+        [1.0, 2.0],
+        np.array([1, 2]),
+        (1, 2),
+    ]:
+        await check_query(
+            table_async.query().nearest_to(vector_query), expected_num_rows=2
+        )
+
+    # No easy way to check these vector query parameters are doing what they say.  We
+    # just check that they don't raise exceptions and assume this is tested at a lower
+    # level.
+    await check_query(
+        table_async.query().where("id = 2").nearest_to(pa.array([1, 2])).postfilter(),
+        expected_num_rows=1,
+    )
+    await check_query(
+        table_async.query().nearest_to(pa.array([1, 2])).refine_factor(1),
+        expected_num_rows=2,
+    )
+    await check_query(
+        table_async.query().nearest_to(pa.array([1, 2])).nprobes(10),
+        expected_num_rows=2,
+    )
+    await check_query(
+        table_async.query().nearest_to(pa.array([1, 2])).bypass_vector_index(),
+        expected_num_rows=2,
+    )
+    await check_query(
+        table_async.query().nearest_to(pa.array([1, 2])).distance_type("dot"),
+        expected_num_rows=2,
+    )
+    await check_query(
+        table_async.query().nearest_to(pa.array([1, 2])).distance_type("DoT"),
+        expected_num_rows=2,
+    )
+
+    # Make sure we can use a vector query as a base query (e.g. call limit on it)
+    # Also make sure `vector_search` works
+    await check_query(table_async.vector_search([1, 2]).limit(1), expected_num_rows=1)
+
+    # Also check an empty query
+    await check_query(table_async.query().where("id < 0"), expected_num_rows=0)
+
+
+@pytest.mark.asyncio
+async def test_query_to_arrow_async(table_async: AsyncTable):
+    table = await table_async.to_arrow()
+    assert table.num_rows == 2
+    assert table.num_columns == 4
+
+    table = await table_async.query().to_arrow()
+    assert table.num_rows == 2
+    assert table.num_columns == 4
+
+    table = await table_async.query().where("id < 0").to_arrow()
+    assert table.num_rows == 0
+    assert table.num_columns == 4
+
+
+@pytest.mark.asyncio
+async def test_query_to_pandas_async(table_async: AsyncTable):
+    df = await table_async.to_pandas()
+    assert df.shape == (2, 4)
+
+    df = await table_async.query().to_pandas()
+    assert df.shape == (2, 4)
+
+    df = await table_async.query().where("id < 0").to_pandas()
+    assert df.shape == (0, 4)

python/python/tests/test_rerankers.py

@@ -124,8 +124,9 @@ def test_linear_combination(tmp_path):
 )
 def test_cohere_reranker(tmp_path):
     pytest.importorskip("cohere")
+    reranker = CohereReranker()
     table, schema = get_test_table(tmp_path)
-    # The default reranker
+    # Hybrid search setting
     result1 = (
         table.search("Our father who art in heaven", query_type="hybrid")
         .rerank(normalize="score", reranker=CohereReranker())
@@ -133,7 +134,7 @@ def test_cohere_reranker(tmp_path):
     )
     result2 = (
         table.search("Our father who art in heaven", query_type="hybrid")
-        .rerank(reranker=CohereReranker())
+        .rerank(reranker=reranker)
         .to_pydantic(schema)
     )
     assert result1 == result2
@@ -143,64 +144,120 @@ def test_cohere_reranker(tmp_path):
     result = (
         table.search((query_vector, query))
         .limit(30)
-        .rerank(reranker=CohereReranker())
+        .rerank(reranker=reranker)
         .to_arrow()
     )
 
     assert len(result) == 30
-
-    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), (
+    err = (
         "The _relevance_score column of the results returned by the reranker "
         "represents the relevance of the result to the query & should "
         "be descending."
     )
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+
+    # Vector search setting
+    query = "Our father who art in heaven"
+    result = table.search(query).rerank(reranker=reranker).limit(30).to_arrow()
+    assert len(result) == 30
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+    result_explicit = (
+        table.search(query_vector)
+        .rerank(reranker=reranker, query_string=query)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result_explicit) == 30
+    with pytest.raises(
+        ValueError
+    ):  # This raises an error because vector query is provided without reanking query
+        table.search(query_vector).rerank(reranker=reranker).limit(30).to_arrow()
+
+    # FTS search setting
+    result = (
+        table.search(query, query_type="fts")
+        .rerank(reranker=reranker)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result) > 0
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
 
 
 def test_cross_encoder_reranker(tmp_path):
     pytest.importorskip("sentence_transformers")
+    reranker = CrossEncoderReranker()
     table, schema = get_test_table(tmp_path)
     result1 = (
         table.search("Our father who art in heaven", query_type="hybrid")
-        .rerank(normalize="score", reranker=CrossEncoderReranker())
+        .rerank(normalize="score", reranker=reranker)
         .to_pydantic(schema)
     )
     result2 = (
         table.search("Our father who art in heaven", query_type="hybrid")
-        .rerank(reranker=CrossEncoderReranker())
+        .rerank(reranker=reranker)
         .to_pydantic(schema)
     )
     assert result1 == result2
 
-    # test explicit hybrid query
     query = "Our father who art in heaven"
     query_vector = table.to_pandas()["vector"][0]
     result = (
         table.search((query_vector, query), query_type="hybrid")
         .limit(30)
-        .rerank(reranker=CrossEncoderReranker())
+        .rerank(reranker=reranker)
         .to_arrow()
     )
 
     assert len(result) == 30
 
-    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), (
+    err = (
         "The _relevance_score column of the results returned by the reranker "
         "represents the relevance of the result to the query & should "
         "be descending."
     )
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+
+    # Vector search setting
+    result = table.search(query).rerank(reranker=reranker).limit(30).to_arrow()
+    assert len(result) == 30
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+
+    result_explicit = (
+        table.search(query_vector)
+        .rerank(reranker=reranker, query_string=query)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result_explicit) == 30
+    with pytest.raises(
+        ValueError
+    ):  # This raises an error because vector query is provided without reanking query
+        table.search(query_vector).rerank(reranker=reranker).limit(30).to_arrow()
+
+    # FTS search setting
+    result = (
+        table.search(query, query_type="fts")
+        .rerank(reranker=reranker)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result) > 0
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
 
 
 def test_colbert_reranker(tmp_path):
     pytest.importorskip("transformers")
+    reranker = ColbertReranker()
     table, schema = get_test_table(tmp_path)
     result1 = (
         table.search("Our father who art in heaven", query_type="hybrid")
-        .rerank(normalize="score", reranker=ColbertReranker())
+        .rerank(normalize="score", reranker=reranker)
         .to_pydantic(schema)
     )
     result2 = (
         table.search("Our father who art in heaven", query_type="hybrid")
-        .rerank(reranker=ColbertReranker())
+        .rerank(reranker=reranker)
         .to_pydantic(schema)
     )
     assert result1 == result2
@@ -211,17 +268,43 @@ def test_colbert_reranker(tmp_path):
     result = (
         table.search((query_vector, query))
         .limit(30)
-        .rerank(reranker=ColbertReranker())
+        .rerank(reranker=reranker)
         .to_arrow()
     )
 
     assert len(result) == 30
-
-    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), (
+    err = (
         "The _relevance_score column of the results returned by the reranker "
         "represents the relevance of the result to the query & should "
         "be descending."
     )
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+
+    # Vector search setting
+    result = table.search(query).rerank(reranker=reranker).limit(30).to_arrow()
+    assert len(result) == 30
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+    result_explicit = (
+        table.search(query_vector)
+        .rerank(reranker=reranker, query_string=query)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result_explicit) == 30
+    with pytest.raises(
+        ValueError
+    ):  # This raises an error because vector query is provided without reanking query
+        table.search(query_vector).rerank(reranker=reranker).limit(30).to_arrow()
+
+    # FTS search setting
+    result = (
+        table.search(query, query_type="fts")
+        .rerank(reranker=reranker)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result) > 0
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
 
 
 @pytest.mark.skipif(
@@ -230,9 +313,10 @@ def test_colbert_reranker(tmp_path):
 def test_openai_reranker(tmp_path):
     pytest.importorskip("openai")
     table, schema = get_test_table(tmp_path)
+    reranker = OpenaiReranker()
     result1 = (
         table.search("Our father who art in heaven", query_type="hybrid")
-        .rerank(normalize="score", reranker=OpenaiReranker())
+        .rerank(normalize="score", reranker=reranker)
         .to_pydantic(schema)
     )
     result2 = (
@@ -248,14 +332,40 @@ def test_openai_reranker(tmp_path):
     result = (
         table.search((query_vector, query))
         .limit(30)
-        .rerank(reranker=OpenaiReranker())
+        .rerank(reranker=reranker)
         .to_arrow()
     )
 
     assert len(result) == 30
 
-    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), (
+    err = (
         "The _relevance_score column of the results returned by the reranker "
         "represents the relevance of the result to the query & should "
         "be descending."
     )
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+
+    # Vector search setting
+    result = table.search(query).rerank(reranker=reranker).limit(30).to_arrow()
+    assert len(result) == 30
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err
+    result_explicit = (
+        table.search(query_vector)
+        .rerank(reranker=reranker, query_string=query)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result_explicit) == 30
+    with pytest.raises(
+        ValueError
+    ):  # This raises an error because vector query is provided without reanking query
+        table.search(query_vector).rerank(reranker=reranker).limit(30).to_arrow()
+    # FTS search setting
+    result = (
+        table.search(query, query_type="fts")
+        .rerank(reranker=reranker)
+        .limit(30)
+        .to_arrow()
+    )
+    assert len(result) > 0
+    assert np.all(np.diff(result.column("_relevance_score").to_numpy()) <= 0), err

python/python/tests/test_table.py

@@ -85,6 +85,23 @@ async def test_close(db_async: AsyncConnection):
     assert str(table) == "ClosedTable(some_table)"
 
 
+@pytest.mark.asyncio
+async def test_update_async(db_async: AsyncConnection):
+    table = await db_async.create_table("some_table", data=[{"id": 0}])
+    assert await table.count_rows("id == 0") == 1
+    assert await table.count_rows("id == 7") == 0
+    await table.update({"id": 7})
+    assert await table.count_rows("id == 7") == 1
+    assert await table.count_rows("id == 0") == 0
+    await table.add([{"id": 2}])
+    await table.update(where="id % 2 == 0", updates_sql={"id": "5"})
+    assert await table.count_rows("id == 7") == 1
+    assert await table.count_rows("id == 2") == 0
+    assert await table.count_rows("id == 5") == 1
+    await table.update({"id": 10}, where="id == 5")
+    assert await table.count_rows("id == 10") == 1
+
+
 def test_create_table(db):
     schema = pa.schema(
         [
@@ -974,3 +991,37 @@ def test_drop_columns(tmp_path):
     table = LanceTable.create(db, "my_table", data=data)
     table.drop_columns(["category"])
     assert table.to_arrow().column_names == ["id"]
+
+
+@pytest.mark.asyncio
+async def test_time_travel(db_async: AsyncConnection):
+    # Setup
+    table = await db_async.create_table("some_table", data=[{"id": 0}])
+    version = await table.version()
+    await table.add([{"id": 1}])
+    assert await table.count_rows() == 2
+    # Make sure we can rewind
+    await table.checkout(version)
+    assert await table.count_rows() == 1
+    # Can't add data in time travel mode
+    with pytest.raises(
+        ValueError,
+        match="table cannot be modified when a specific version is checked out",
+    ):
+        await table.add([{"id": 2}])
+    # Can go back to normal mode
+    await table.checkout_latest()
+    assert await table.count_rows() == 2
+    # Should be able to add data again
+    await table.add([{"id": 3}])
+    assert await table.count_rows() == 3
+    # Now checkout and restore
+    await table.checkout(version)
+    await table.restore()
+    assert await table.count_rows() == 1
+    # Should be able to add data
+    await table.add([{"id": 4}])
+    assert await table.count_rows() == 2
+    # Can't use restore if not checked out
+    with pytest.raises(ValueError, match="checkout before running restore"):
+        await table.restore()

python/src/arrow.rs

@@ -0,0 +1,51 @@
+// use arrow::datatypes::SchemaRef;
+// use lancedb::arrow::SendableRecordBatchStream;
+
+use std::sync::Arc;
+
+use arrow::{
+    datatypes::SchemaRef,
+    pyarrow::{IntoPyArrow, ToPyArrow},
+};
+use futures::stream::StreamExt;
+use lancedb::arrow::SendableRecordBatchStream;
+use pyo3::{pyclass, pymethods, PyAny, PyObject, PyRef, PyResult, Python};
+use pyo3_asyncio::tokio::future_into_py;
+
+use crate::error::PythonErrorExt;
+
+#[pyclass]
+pub struct RecordBatchStream {
+    schema: SchemaRef,
+    inner: Arc<tokio::sync::Mutex<SendableRecordBatchStream>>,
+}
+
+impl RecordBatchStream {
+    pub fn new(inner: SendableRecordBatchStream) -> Self {
+        let schema = inner.schema().clone();
+        Self {
+            schema,
+            inner: Arc::new(tokio::sync::Mutex::new(inner)),
+        }
+    }
+}
+
+#[pymethods]
+impl RecordBatchStream {
+    pub fn schema(&self, py: Python) -> PyResult<PyObject> {
+        (*self.schema).clone().into_pyarrow(py)
+    }
+
+    pub fn next(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            let inner_next = inner.lock().await.next().await;
+            inner_next
+                .map(|item| {
+                    let item = item.infer_error()?;
+                    Python::with_gil(|py| item.to_pyarrow(py))
+                })
+                .transpose()
+        })
+    }
+}

python/src/connection.rs

@@ -95,7 +95,7 @@ impl Connection {
 
         let mode = Self::parse_create_mode_str(mode)?;
 
-        let batches = Box::new(ArrowArrayStreamReader::from_pyarrow(data)?);
+        let batches = ArrowArrayStreamReader::from_pyarrow(data)?;
         future_into_py(self_.py(), async move {
             let table = inner
                 .create_table(name, batches)

python/src/index.rs

@@ -0,0 +1,109 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::sync::Mutex;
+
+use lancedb::{
+    index::{scalar::BTreeIndexBuilder, vector::IvfPqIndexBuilder, Index as LanceDbIndex},
+    DistanceType,
+};
+use pyo3::{
+    exceptions::{PyRuntimeError, PyValueError},
+    pyclass, pymethods, PyResult,
+};
+
+#[pyclass]
+pub struct Index {
+    inner: Mutex<Option<LanceDbIndex>>,
+}
+
+impl Index {
+    pub fn consume(&self) -> PyResult<LanceDbIndex> {
+        self.inner
+            .lock()
+            .unwrap()
+            .take()
+            .ok_or_else(|| PyRuntimeError::new_err("cannot use an Index more than once"))
+    }
+}
+
+#[pymethods]
+impl Index {
+    #[staticmethod]
+    pub fn ivf_pq(
+        distance_type: Option<String>,
+        num_partitions: Option<u32>,
+        num_sub_vectors: Option<u32>,
+        max_iterations: Option<u32>,
+        sample_rate: Option<u32>,
+    ) -> PyResult<Self> {
+        let mut ivf_pq_builder = IvfPqIndexBuilder::default();
+        if let Some(distance_type) = distance_type {
+            let distance_type = match distance_type.as_str() {
+                "l2" => Ok(DistanceType::L2),
+                "cosine" => Ok(DistanceType::Cosine),
+                "dot" => Ok(DistanceType::Dot),
+                _ => Err(PyValueError::new_err(format!(
+                    "Invalid distance type '{}'.  Must be one of l2, cosine, or dot",
+                    distance_type
+                ))),
+            }?;
+            ivf_pq_builder = ivf_pq_builder.distance_type(distance_type);
+        }
+        if let Some(num_partitions) = num_partitions {
+            ivf_pq_builder = ivf_pq_builder.num_partitions(num_partitions);
+        }
+        if let Some(num_sub_vectors) = num_sub_vectors {
+            ivf_pq_builder = ivf_pq_builder.num_sub_vectors(num_sub_vectors);
+        }
+        if let Some(max_iterations) = max_iterations {
+            ivf_pq_builder = ivf_pq_builder.max_iterations(max_iterations);
+        }
+        if let Some(sample_rate) = sample_rate {
+            ivf_pq_builder = ivf_pq_builder.sample_rate(sample_rate);
+        }
+        Ok(Self {
+            inner: Mutex::new(Some(LanceDbIndex::IvfPq(ivf_pq_builder))),
+        })
+    }
+
+    #[staticmethod]
+    pub fn btree() -> PyResult<Self> {
+        Ok(Self {
+            inner: Mutex::new(Some(LanceDbIndex::BTree(BTreeIndexBuilder::default()))),
+        })
+    }
+}
+
+#[pyclass(get_all)]
+/// A description of an index currently configured on a column
+pub struct IndexConfig {
+    /// The type of the index
+    pub index_type: String,
+    /// The columns in the index
+    ///
+    /// Currently this is always a list of size 1.  In the future there may
+    /// be more columns to represent composite indices.
+    pub columns: Vec<String>,
+}
+
+impl From<lancedb::index::IndexConfig> for IndexConfig {
+    fn from(value: lancedb::index::IndexConfig) -> Self {
+        let index_type = format!("{:?}", value.index_type);
+        Self {
+            index_type,
+            columns: value.columns,
+        }
+    }
+}

python/src/lib.rs

@@ -12,13 +12,21 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use arrow::RecordBatchStream;
 use connection::{connect, Connection};
 use env_logger::Env;
+use index::{Index, IndexConfig};
 use pyo3::{pymodule, types::PyModule, wrap_pyfunction, PyResult, Python};
+use query::{Query, VectorQuery};
+use table::Table;
 
+pub mod arrow;
 pub mod connection;
 pub mod error;
+pub mod index;
+pub mod query;
 pub mod table;
+pub mod util;
 
 #[pymodule]
 pub fn _lancedb(_py: Python, m: &PyModule) -> PyResult<()> {
@@ -27,6 +35,12 @@ pub fn _lancedb(_py: Python, m: &PyModule) -> PyResult<()> {
         .write_style("LANCEDB_LOG_STYLE");
     env_logger::init_from_env(env);
     m.add_class::<Connection>()?;
+    m.add_class::<Table>()?;
+    m.add_class::<Index>()?;
+    m.add_class::<IndexConfig>()?;
+    m.add_class::<Query>()?;
+    m.add_class::<VectorQuery>()?;
+    m.add_class::<RecordBatchStream>()?;
     m.add_function(wrap_pyfunction!(connect, m)?)?;
     m.add("__version__", env!("CARGO_PKG_VERSION"))?;
     Ok(())

python/src/query.rs

@@ -0,0 +1,125 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use arrow::array::make_array;
+use arrow::array::ArrayData;
+use arrow::pyarrow::FromPyArrow;
+use lancedb::query::{
+    ExecutableQuery, Query as LanceDbQuery, QueryBase, Select, VectorQuery as LanceDbVectorQuery,
+};
+use pyo3::pyclass;
+use pyo3::pymethods;
+use pyo3::PyAny;
+use pyo3::PyRef;
+use pyo3::PyResult;
+use pyo3_asyncio::tokio::future_into_py;
+
+use crate::arrow::RecordBatchStream;
+use crate::error::PythonErrorExt;
+use crate::util::parse_distance_type;
+
+#[pyclass]
+pub struct Query {
+    inner: LanceDbQuery,
+}
+
+impl Query {
+    pub fn new(query: LanceDbQuery) -> Self {
+        Self { inner: query }
+    }
+}
+
+#[pymethods]
+impl Query {
+    pub fn r#where(&mut self, predicate: String) {
+        self.inner = self.inner.clone().only_if(predicate);
+    }
+
+    pub fn select(&mut self, columns: Vec<(String, String)>) {
+        self.inner = self.inner.clone().select(Select::dynamic(&columns));
+    }
+
+    pub fn limit(&mut self, limit: u32) {
+        self.inner = self.inner.clone().limit(limit as usize);
+    }
+
+    pub fn nearest_to(&mut self, vector: &PyAny) -> PyResult<VectorQuery> {
+        let data: ArrayData = ArrayData::from_pyarrow(vector)?;
+        let array = make_array(data);
+        let inner = self.inner.clone().nearest_to(array).infer_error()?;
+        Ok(VectorQuery { inner })
+    }
+
+    pub fn execute(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            let inner_stream = inner.execute().await.infer_error()?;
+            Ok(RecordBatchStream::new(inner_stream))
+        })
+    }
+}
+
+#[pyclass]
+pub struct VectorQuery {
+    inner: LanceDbVectorQuery,
+}
+
+#[pymethods]
+impl VectorQuery {
+    pub fn r#where(&mut self, predicate: String) {
+        self.inner = self.inner.clone().only_if(predicate);
+    }
+
+    pub fn select(&mut self, columns: Vec<(String, String)>) {
+        self.inner = self.inner.clone().select(Select::dynamic(&columns));
+    }
+
+    pub fn limit(&mut self, limit: u32) {
+        self.inner = self.inner.clone().limit(limit as usize);
+    }
+
+    pub fn column(&mut self, column: String) {
+        self.inner = self.inner.clone().column(&column);
+    }
+
+    pub fn distance_type(&mut self, distance_type: String) -> PyResult<()> {
+        let distance_type = parse_distance_type(distance_type)?;
+        self.inner = self.inner.clone().distance_type(distance_type);
+        Ok(())
+    }
+
+    pub fn postfilter(&mut self) {
+        self.inner = self.inner.clone().postfilter();
+    }
+
+    pub fn refine_factor(&mut self, refine_factor: u32) {
+        self.inner = self.inner.clone().refine_factor(refine_factor);
+    }
+
+    pub fn nprobes(&mut self, nprobe: u32) {
+        self.inner = self.inner.clone().nprobes(nprobe as usize);
+    }
+
+    pub fn bypass_vector_index(&mut self) {
+        self.inner = self.inner.clone().bypass_vector_index()
+    }
+
+    pub fn execute(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner.clone();
+        future_into_py(self_.py(), async move {
+            let inner_stream = inner.execute().await.infer_error()?;
+            Ok(RecordBatchStream::new(inner_stream))
+        })
+    }
+}

python/src/table.rs

@@ -5,11 +5,17 @@ use arrow::{
 use lancedb::table::{AddDataMode, Table as LanceDbTable};
 use pyo3::{
     exceptions::{PyRuntimeError, PyValueError},
-    pyclass, pymethods, PyAny, PyRef, PyResult, Python,
+    pyclass, pymethods,
+    types::{PyDict, PyString},
+    PyAny, PyRef, PyResult, Python,
 };
 use pyo3_asyncio::tokio::future_into_py;
 
-use crate::error::PythonErrorExt;
+use crate::{
+    error::PythonErrorExt,
+    index::{Index, IndexConfig},
+    query::Query,
+};
 
 #[pyclass]
 pub struct Table {
@@ -58,7 +64,7 @@ impl Table {
     }
 
     pub fn add<'a>(self_: PyRef<'a, Self>, data: &PyAny, mode: String) -> PyResult<&'a PyAny> {
-        let batches = Box::new(ArrowArrayStreamReader::from_pyarrow(data)?);
+        let batches = ArrowArrayStreamReader::from_pyarrow(data)?;
         let mut op = self_.inner_ref()?.add(batches);
         if mode == "append" {
             op = op.mode(AddDataMode::Append);
@@ -74,6 +80,28 @@ impl Table {
         })
     }
 
+    pub fn update<'a>(
+        self_: PyRef<'a, Self>,
+        updates: &PyDict,
+        r#where: Option<String>,
+    ) -> PyResult<&'a PyAny> {
+        let mut op = self_.inner_ref()?.update();
+        if let Some(only_if) = r#where {
+            op = op.only_if(only_if);
+        }
+        for (column_name, value) in updates.into_iter() {
+            let column_name: &PyString = column_name.downcast()?;
+            let column_name = column_name.to_str()?.to_string();
+            let value: &PyString = value.downcast()?;
+            let value = value.to_str()?.to_string();
+            op = op.column(column_name, value);
+        }
+        future_into_py(self_.py(), async move {
+            op.execute().await.infer_error()?;
+            Ok(())
+        })
+    }
+
     pub fn count_rows(self_: PyRef<'_, Self>, filter: Option<String>) -> PyResult<&PyAny> {
         let inner = self_.inner_ref()?.clone();
         future_into_py(self_.py(), async move {
@@ -81,10 +109,79 @@ impl Table {
         })
     }
 
+    pub fn create_index<'a>(
+        self_: PyRef<'a, Self>,
+        column: String,
+        index: Option<&Index>,
+        replace: Option<bool>,
+    ) -> PyResult<&'a PyAny> {
+        let index = if let Some(index) = index {
+            index.consume()?
+        } else {
+            lancedb::index::Index::Auto
+        };
+        let mut op = self_.inner_ref()?.create_index(&[column], index);
+        if let Some(replace) = replace {
+            op = op.replace(replace);
+        }
+
+        future_into_py(self_.py(), async move {
+            op.execute().await.infer_error()?;
+            Ok(())
+        })
+    }
+
+    pub fn list_indices(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            Ok(inner
+                .list_indices()
+                .await
+                .infer_error()?
+                .into_iter()
+                .map(IndexConfig::from)
+                .collect::<Vec<_>>())
+        })
+    }
+
     pub fn __repr__(&self) -> String {
         match &self.inner {
             None => format!("ClosedTable({})", self.name),
             Some(inner) => inner.to_string(),
         }
     }
+
+    pub fn version(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(
+            self_.py(),
+            async move { inner.version().await.infer_error() },
+        )
+    }
+
+    pub fn checkout(self_: PyRef<'_, Self>, version: u64) -> PyResult<&PyAny> {
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            inner.checkout(version).await.infer_error()
+        })
+    }
+
+    pub fn checkout_latest(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(self_.py(), async move {
+            inner.checkout_latest().await.infer_error()
+        })
+    }
+
+    pub fn restore(self_: PyRef<'_, Self>) -> PyResult<&PyAny> {
+        let inner = self_.inner_ref()?.clone();
+        future_into_py(
+            self_.py(),
+            async move { inner.restore().await.infer_error() },
+        )
+    }
+
+    pub fn query(&self) -> Query {
+        Query::new(self.inner_ref().unwrap().query())
+    }
 }

python/src/util.rs

@@ -0,0 +1,51 @@
+use std::sync::Mutex;
+
+use lancedb::DistanceType;
+use pyo3::{
+    exceptions::{PyRuntimeError, PyValueError},
+    PyResult,
+};
+
+/// A wrapper around a rust builder
+///
+/// Rust builders are often implemented so that the builder methods
+/// consume the builder and return a new one. This is not compatible
+/// with the pyo3, which, being garbage collected, cannot easily obtain
+/// ownership of an object.
+///
+/// This wrapper converts the compile-time safety of rust into runtime
+/// errors if any attempt to use the builder happens after it is consumed.
+pub struct BuilderWrapper<T> {
+    name: String,
+    inner: Mutex<Option<T>>,
+}
+
+impl<T> BuilderWrapper<T> {
+    pub fn new(name: impl AsRef<str>, inner: T) -> Self {
+        Self {
+            name: name.as_ref().to_string(),
+            inner: Mutex::new(Some(inner)),
+        }
+    }
+
+    pub fn consume<O>(&self, mod_fn: impl FnOnce(T) -> O) -> PyResult<O> {
+        let mut inner = self.inner.lock().unwrap();
+        let inner_builder = inner.take().ok_or_else(|| {
+            PyRuntimeError::new_err(format!("{} has already been consumed", self.name))
+        })?;
+        let result = mod_fn(inner_builder);
+        Ok(result)
+    }
+}
+
+pub fn parse_distance_type(distance_type: impl AsRef<str>) -> PyResult<DistanceType> {
+    match distance_type.as_ref().to_lowercase().as_str() {
+        "l2" => Ok(DistanceType::L2),
+        "cosine" => Ok(DistanceType::Cosine),
+        "dot" => Ok(DistanceType::Dot),
+        _ => Err(PyValueError::new_err(format!(
+            "Invalid distance type '{}'.  Must be one of l2, cosine, or dot",
+            distance_type.as_ref()
+        ))),
+    }
+}

rust/ffi/node/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb-node"
-version = "0.4.12"
+version = "0.4.13"
 description = "Serverless, low-latency vector database for AI applications"
 license.workspace = true
 edition.workspace = true

rust/ffi/node/src/index/scalar.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use lancedb::index::{scalar::BTreeIndexBuilder, Index};
 use neon::{
     context::{Context, FunctionContext},
     result::JsResult,
@@ -33,9 +34,9 @@ pub fn table_create_scalar_index(mut cx: FunctionContext) -> JsResult<JsPromise>
 
     rt.spawn(async move {
         let idx_result = table
-            .create_index(&[&column])
+            .create_index(&[column], Index::BTree(BTreeIndexBuilder::default()))
             .replace(replace)
-            .build()
+            .execute()
             .await;
 
         deferred.settle_with(&channel, move |mut cx| {

rust/ffi/node/src/index/vector.rs

@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use lance_linalg::distance::MetricType;
-use lancedb::index::IndexBuilder;
+use lancedb::index::vector::IvfPqIndexBuilder;
+use lancedb::index::Index;
+use lancedb::DistanceType;
 use neon::context::FunctionContext;
 use neon::prelude::*;
 use std::convert::TryFrom;
 
-use crate::error::Error::InvalidIndexType;
 use crate::error::ResultExt;
 use crate::neon_ext::js_object_ext::JsObjectExt;
 use crate::runtime;
@@ -39,13 +39,20 @@ pub fn table_create_vector_index(mut cx: FunctionContext) -> JsResult<JsPromise>
         .map(|s| s.value(&mut cx))
         .unwrap_or("vector".to_string()); // Backward compatibility
 
+    let replace = index_params
+        .get_opt::<JsBoolean, _, _>(&mut cx, "replace")?
+        .map(|r| r.value(&mut cx));
+
     let tbl = table.clone();
-    let index_builder = tbl.create_index(&[&column_name]);
-    let index_builder =
-        get_index_params_builder(&mut cx, index_params, index_builder).or_throw(&mut cx)?;
+    let ivf_pq_builder = get_index_params_builder(&mut cx, index_params).or_throw(&mut cx)?;
+
+    let mut index_builder = tbl.create_index(&[column_name], Index::IvfPq(ivf_pq_builder));
+    if let Some(replace) = replace {
+        index_builder = index_builder.replace(replace);
+    }
 
     rt.spawn(async move {
-        let idx_result = index_builder.build().await;
+        let idx_result = index_builder.execute().await;
         deferred.settle_with(&channel, move |mut cx| {
             idx_result.or_throw(&mut cx)?;
             Ok(cx.boxed(JsTable::from(table)))
@@ -57,26 +64,17 @@ pub fn table_create_vector_index(mut cx: FunctionContext) -> JsResult<JsPromise>
 fn get_index_params_builder(
     cx: &mut FunctionContext,
     obj: Handle<JsObject>,
-    builder: IndexBuilder,
-) -> crate::error::Result<IndexBuilder> {
-    let mut builder = match obj.get::<JsString, _, _>(cx, "type")?.value(cx).as_str() {
-        "ivf_pq" => builder.ivf_pq(),
-        _ => {
-            return Err(InvalidIndexType {
-                index_type: "".into(),
-            })
-        }
-    };
-
-    if let Some(index_name) = obj.get_opt::<JsString, _, _>(cx, "index_name")? {
-        builder = builder.name(index_name.value(cx).as_str());
+) -> crate::error::Result<IvfPqIndexBuilder> {
+    if obj.get_opt::<JsString, _, _>(cx, "index_name")?.is_some() {
+        return Err(crate::error::Error::LanceDB {
+            message: "Setting the index_name is no longer supported".to_string(),
+        });
     }
-
+    let mut builder = IvfPqIndexBuilder::default();
     if let Some(metric_type) = obj.get_opt::<JsString, _, _>(cx, "metric_type")? {
-        let metric_type = MetricType::try_from(metric_type.value(cx).as_str())?;
-        builder = builder.metric_type(metric_type);
+        let distance_type = DistanceType::try_from(metric_type.value(cx).as_str())?;
+        builder = builder.distance_type(distance_type);
     }
-
     if let Some(np) = obj.get_opt_u32(cx, "num_partitions")? {
         builder = builder.num_partitions(np);
     }
@@ -86,11 +84,5 @@ fn get_index_params_builder(
     if let Some(max_iters) = obj.get_opt_u32(cx, "max_iters")? {
         builder = builder.max_iterations(max_iters);
     }
-    if let Some(num_bits) = obj.get_opt_u32(cx, "num_bits")? {
-        builder = builder.num_bits(num_bits);
-    }
-    if let Some(replace) = obj.get_opt::<JsBoolean, _, _>(cx, "replace")? {
-        builder = builder.replace(replace.value(cx));
-    }
     Ok(builder)
 }

rust/ffi/node/src/query.rs

@@ -2,7 +2,8 @@ use std::convert::TryFrom;
 use std::ops::Deref;
 
 use futures::{TryFutureExt, TryStreamExt};
-use lance_linalg::distance::MetricType;
+use lancedb::query::{ExecutableQuery, QueryBase, Select};
+use lancedb::DistanceType;
 use neon::context::FunctionContext;
 use neon::handle::Handle;
 use neon::prelude::*;
@@ -56,53 +57,72 @@ impl JsQuery {
         let channel = cx.channel();
         let table = js_table.table.clone();
 
-        let query_vector = query_obj.get_opt::<JsArray, _, _>(&mut cx, "_queryVector")?;
         let mut builder = table.query();
-        if let Some(query) = query_vector.map(|q| convert::js_array_to_vec(q.deref(), &mut cx)) {
-            builder = builder.nearest_to(&query);
-            if let Some(metric_type) = query_obj
-                .get_opt::<JsString, _, _>(&mut cx, "_metricType")?
-                .map(|s| s.value(&mut cx))
-                .map(|s| MetricType::try_from(s.as_str()).unwrap())
-            {
-                builder = builder.metric_type(metric_type);
-            }
-
-            let nprobes = query_obj.get_usize(&mut cx, "_nprobes").or_throw(&mut cx)?;
-            builder = builder.nprobes(nprobes);
-        };
-
         if let Some(filter) = query_obj
             .get_opt::<JsString, _, _>(&mut cx, "_filter")?
             .map(|s| s.value(&mut cx))
         {
-            builder = builder.filter(filter);
+            builder = builder.only_if(filter);
         }
         if let Some(select) = select {
-            builder = builder.select(select.as_slice());
+            builder = builder.select(Select::columns(select.as_slice()));
         }
         if let Some(limit) = limit {
             builder = builder.limit(limit as usize);
         };
 
-        builder = builder.prefilter(prefilter);
+        let query_vector = query_obj.get_opt::<JsArray, _, _>(&mut cx, "_queryVector")?;
+        if let Some(query) = query_vector.map(|q| convert::js_array_to_vec(q.deref(), &mut cx)) {
+            let mut vector_builder = builder.nearest_to(query).unwrap();
+            if let Some(distance_type) = query_obj
+                .get_opt::<JsString, _, _>(&mut cx, "_metricType")?
+                .map(|s| s.value(&mut cx))
+                .map(|s| DistanceType::try_from(s.as_str()).unwrap())
+            {
+                vector_builder = vector_builder.distance_type(distance_type);
+            }
 
-        rt.spawn(async move {
-            let record_batch_stream = builder.execute_stream();
-            let results = record_batch_stream
-                .and_then(|stream| {
-                    stream
-                        .try_collect::<Vec<_>>()
-                        .map_err(lancedb::error::Error::from)
-                })
-                .await;
+            let nprobes = query_obj.get_usize(&mut cx, "_nprobes").or_throw(&mut cx)?;
+            vector_builder = vector_builder.nprobes(nprobes);
 
-            deferred.settle_with(&channel, move |mut cx| {
-                let results = results.or_throw(&mut cx)?;
-                let buffer = record_batch_to_buffer(results).or_throw(&mut cx)?;
-                convert::new_js_buffer(buffer, &mut cx, is_electron)
+            if !prefilter {
+                vector_builder = vector_builder.postfilter();
+            }
+            rt.spawn(async move {
+                let results = vector_builder
+                    .execute()
+                    .and_then(|stream| {
+                        stream
+                            .try_collect::<Vec<_>>()
+                            .map_err(lancedb::error::Error::from)
+                    })
+                    .await;
+
+                deferred.settle_with(&channel, move |mut cx| {
+                    let results = results.or_throw(&mut cx)?;
+                    let buffer = record_batch_to_buffer(results).or_throw(&mut cx)?;
+                    convert::new_js_buffer(buffer, &mut cx, is_electron)
+                });
             });
-        });
+        } else {
+            rt.spawn(async move {
+                let results = builder
+                    .execute()
+                    .and_then(|stream| {
+                        stream
+                            .try_collect::<Vec<_>>()
+                            .map_err(lancedb::error::Error::from)
+                    })
+                    .await;
+
+                deferred.settle_with(&channel, move |mut cx| {
+                    let results = results.or_throw(&mut cx)?;
+                    let buffer = record_batch_to_buffer(results).or_throw(&mut cx)?;
+                    convert::new_js_buffer(buffer, &mut cx, is_electron)
+                });
+            });
+        };
+
         Ok(promise)
     }
 }

rust/ffi/node/src/table.rs

@@ -80,7 +80,7 @@ impl JsTable {
         rt.spawn(async move {
             let batch_reader = RecordBatchIterator::new(batches.into_iter().map(Ok), schema);
             let table_rst = database
-                .create_table(&table_name, Box::new(batch_reader))
+                .create_table(&table_name, batch_reader)
                 .write_options(WriteOptions {
                     lance_write_params: Some(params),
                 })
@@ -126,7 +126,7 @@ impl JsTable {
         rt.spawn(async move {
             let batch_reader = RecordBatchIterator::new(batches.into_iter().map(Ok), schema);
             let add_result = table
-                .add(Box::new(batch_reader))
+                .add(batch_reader)
                 .write_options(WriteOptions {
                     lance_write_params: Some(params),
                 })
@@ -297,11 +297,14 @@ impl JsTable {
 
             let predicate = predicate.as_deref();
 
-            let update_result = table
-                .as_native()
-                .unwrap()
-                .update(predicate, updates_arg)
-                .await;
+            let mut update_op = table.update();
+            if let Some(predicate) = predicate {
+                update_op = update_op.only_if(predicate);
+            }
+            for (column, value) in updates_arg {
+                update_op = update_op.column(column, value);
+            }
+            let update_result = update_op.execute().await;
             deferred.settle_with(&channel, move |mut cx| {
                 update_result.or_throw(&mut cx)?;
                 Ok(cx.boxed(Self::from(table)))

rust/lancedb/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "lancedb"
-version = "0.4.12"
+version = "0.4.13"
 edition.workspace = true
 description = "LanceDB: A serverless, low-latency vector database for AI applications"
 license.workspace = true
@@ -26,6 +26,7 @@ lance = { workspace = true }
 lance-index = { workspace = true }
 lance-linalg = { workspace = true }
 lance-testing = { workspace = true }
+pin-project = { workspace = true }
 tokio = { version = "1.23", features = ["rt-multi-thread"] }
 log.workspace = true
 async-trait = "0"

rust/lancedb/examples/ivf_pq.rs

@@ -0,0 +1,165 @@
+// Copyright 2024 Lance Developers.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! This example demonstrates setting advanced parameters when building an IVF PQ index
+//!
+//! Snippets from this example are used in the documentation on ANN indices.
+
+use std::sync::Arc;
+
+use arrow_array::types::Float32Type;
+use arrow_array::{
+    FixedSizeListArray, Int32Array, RecordBatch, RecordBatchIterator, RecordBatchReader,
+};
+use arrow_schema::{DataType, Field, Schema};
+
+use futures::TryStreamExt;
+use lancedb::connection::Connection;
+use lancedb::index::vector::IvfPqIndexBuilder;
+use lancedb::index::Index;
+use lancedb::query::{ExecutableQuery, QueryBase};
+use lancedb::{connect, DistanceType, Result, Table};
+
+#[tokio::main]
+async fn main() -> Result<()> {
+    if std::path::Path::new("data").exists() {
+        std::fs::remove_dir_all("data").unwrap();
+    }
+    let uri = "data/sample-lancedb";
+    let db = connect(uri).execute().await?;
+    let tbl = create_table(&db).await?;
+
+    create_index(&tbl).await?;
+    search_index(&tbl).await?;
+    Ok(())
+}
+
+fn create_some_records() -> Result<Box<dyn RecordBatchReader + Send>> {
+    const TOTAL: usize = 1000;
+    const DIM: usize = 128;
+
+    let schema = Arc::new(Schema::new(vec![
+        Field::new("id", DataType::Int32, false),
+        Field::new(
+            "vector",
+            DataType::FixedSizeList(
+                Arc::new(Field::new("item", DataType::Float32, true)),
+                DIM as i32,
+            ),
+            true,
+        ),
+    ]));
+
+    // Create a RecordBatch stream.
+    let batches = RecordBatchIterator::new(
+        vec![RecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(Int32Array::from_iter_values(0..TOTAL as i32)),
+                Arc::new(
+                    FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
+                        (0..TOTAL).map(|_| Some(vec![Some(1.0); DIM])),
+                        DIM as i32,
+                    ),
+                ),
+            ],
+        )
+        .unwrap()]
+        .into_iter()
+        .map(Ok),
+        schema.clone(),
+    );
+    Ok(Box::new(batches))
+}
+
+async fn create_table(db: &Connection) -> Result<Table> {
+    let initial_data: Box<dyn RecordBatchReader + Send> = create_some_records()?;
+    let tbl = db
+        .create_table("my_table", Box::new(initial_data))
+        .execute()
+        .await
+        .unwrap();
+    Ok(tbl)
+}
+
+async fn create_index(table: &Table) -> Result<()> {
+    // --8<-- [start:create_index]
+    // For this example, `table` is a lancedb::Table with a column named
+    // "vector" that is a vector column with dimension 128.
+
+    // By default, if the column "vector" appears to be a vector column,
+    // then an IVF_PQ index with reasonable defaults is created.
+    table
+        .create_index(&["vector"], Index::Auto)
+        .execute()
+        .await?;
+    // For advanced cases, it is also possible to specifically request an
+    // IVF_PQ index and provide custom parameters.
+    table
+        .create_index(
+            &["vector"],
+            Index::IvfPq(
+                // Here we specify advanced indexing parameters.  In this case
+                // we are creating an index that my have better recall than the
+                // default but is also larger and slower.
+                IvfPqIndexBuilder::default()
+                    // This overrides the default distance type of L2
+                    .distance_type(DistanceType::Cosine)
+                    // With 1000 rows this have been ~31 by default
+                    .num_partitions(50)
+                    // With dimension 128 this would have been 8 by default
+                    .num_sub_vectors(16),
+            ),
+        )
+        .execute()
+        .await?;
+    // --8<-- [end:create_index]
+    Ok(())
+}
+
+async fn search_index(table: &Table) -> Result<()> {
+    // --8<-- [start:search1]
+    let query_vector = [1.0; 128];
+    // By default the index will find the 10 closest results using default
+    // search parameters that give a reasonable tradeoff between accuracy
+    // and search latency
+    let mut results = table
+        .vector_search(&query_vector)?
+        // Note: you should always set the distance_type to match the value used
+        // to train the index
+        .distance_type(DistanceType::Cosine)
+        .execute()
+        .await?;
+    while let Some(batch) = results.try_next().await? {
+        println!("{:?}", batch);
+    }
+    // We can also provide custom search parameters.  Here we perform a
+    // slower but more accurate search
+    let mut results = table
+        .vector_search(&query_vector)?
+        .distance_type(DistanceType::Cosine)
+        // Override the default of 10 to get more rows
+        .limit(15)
+        // Override the default of 20 to search more partitions
+        .nprobes(30)
+        // Override the default of None to apply a refine step
+        .refine_factor(1)
+        .execute()
+        .await?;
+    while let Some(batch) = results.try_next().await? {
+        println!("{:?}", batch);
+    }
+    Ok(())
+    // --8<-- [end:search1]
+}

rust/lancedb/examples/simple.rs

@@ -12,6 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+//! This example demonstrates basic usage of LanceDb.
+//!
+//! Snippets from this example are used in the quickstart documentation.
+
 use std::sync::Arc;
 
 use arrow_array::types::Float32Type;
@@ -19,7 +23,10 @@ use arrow_array::{FixedSizeListArray, Int32Array, RecordBatch, RecordBatchIterat
 use arrow_schema::{DataType, Field, Schema};
 use futures::TryStreamExt;
 
+use lancedb::arrow::IntoArrow;
 use lancedb::connection::Connection;
+use lancedb::index::Index;
+use lancedb::query::{ExecutableQuery, QueryBase};
 use lancedb::{connect, Result, Table as LanceDbTable};
 
 #[tokio::main]
@@ -56,14 +63,14 @@ async fn main() -> Result<()> {
 async fn open_with_existing_tbl() -> Result<()> {
     let uri = "data/sample-lancedb";
     let db = connect(uri).execute().await?;
-    // --8<-- [start:open_with_existing_file]
-    let _ = db.open_table("my_table").execute().await.unwrap();
-    // --8<-- [end:open_with_existing_file]
+    #[allow(unused_variables)]
+    // --8<-- [start:open_existing_tbl]
+    let table = db.open_table("my_table").execute().await.unwrap();
+    // --8<-- [end:open_existing_tbl]
     Ok(())
 }
 
-async fn create_table(db: &Connection) -> Result<LanceDbTable> {
-    // --8<-- [start:create_table]
+fn create_some_records() -> Result<impl IntoArrow> {
     const TOTAL: usize = 1000;
     const DIM: usize = 128;
 
@@ -98,33 +105,22 @@ async fn create_table(db: &Connection) -> Result<LanceDbTable> {
         .map(Ok),
         schema.clone(),
     );
+    Ok(Box::new(batches))
+}
+
+async fn create_table(db: &Connection) -> Result<LanceDbTable> {
+    // --8<-- [start:create_table]
+    let initial_data = create_some_records()?;
     let tbl = db
-        .create_table("my_table", Box::new(batches))
+        .create_table("my_table", initial_data)
         .execute()
         .await
         .unwrap();
     // --8<-- [end:create_table]
 
-    let new_batches = RecordBatchIterator::new(
-        vec![RecordBatch::try_new(
-            schema.clone(),
-            vec![
-                Arc::new(Int32Array::from_iter_values(0..TOTAL as i32)),
-                Arc::new(
-                    FixedSizeListArray::from_iter_primitive::<Float32Type, _, _>(
-                        (0..TOTAL).map(|_| Some(vec![Some(1.0); DIM])),
-                        DIM as i32,
-                    ),
-                ),
-            ],
-        )
-        .unwrap()]
-        .into_iter()
-        .map(Ok),
-        schema.clone(),
-    );
     // --8<-- [start:add]
-    tbl.add(Box::new(new_batches)).execute().await.unwrap();
+    let new_data = create_some_records()?;
+    tbl.add(new_data).execute().await.unwrap();
     // --8<-- [end:add]
 
     Ok(tbl)
@@ -142,23 +138,19 @@ async fn create_empty_table(db: &Connection) -> Result<LanceDbTable> {
 
 async fn create_index(table: &LanceDbTable) -> Result<()> {
     // --8<-- [start:create_index]
-    table
-        .create_index(&["vector"])
-        .ivf_pq()
-        .num_partitions(8)
-        .build()
-        .await
+    table.create_index(&["vector"], Index::Auto).execute().await
     // --8<-- [end:create_index]
 }
 
 async fn search(table: &LanceDbTable) -> Result<Vec<RecordBatch>> {
     // --8<-- [start:search]
-    Ok(table
-        .search(&[1.0; 128])
+    table
+        .query()
         .limit(2)
-        .execute_stream()
+        .nearest_to(&[1.0; 128])?
+        .execute()
         .await?
         .try_collect::<Vec<_>>()
-        .await?)
+        .await
     // --8<-- [end:search]
 }

rust/lancedb/src/arrow.rs

@@ -12,4 +12,110 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-pub use lance::arrow::*;
+use std::{pin::Pin, sync::Arc};
+
+pub use arrow_array;
+pub use arrow_schema;
+use futures::{Stream, StreamExt};
+
+use crate::error::Result;
+
+/// An iterator of batches that also has a schema
+pub trait RecordBatchReader: Iterator<Item = Result<arrow_array::RecordBatch>> {
+    /// Returns the schema of this `RecordBatchReader`.
+    ///
+    /// Implementation of this trait should guarantee that all `RecordBatch`'s returned by this
+    /// reader should have the same schema as returned from this method.
+    fn schema(&self) -> Arc<arrow_schema::Schema>;
+}
+
+/// A simple RecordBatchReader formed from the two parts (iterator + schema)
+pub struct SimpleRecordBatchReader<I: Iterator<Item = Result<arrow_array::RecordBatch>>> {
+    pub schema: Arc<arrow_schema::Schema>,
+    pub batches: I,
+}
+
+impl<I: Iterator<Item = Result<arrow_array::RecordBatch>>> Iterator for SimpleRecordBatchReader<I> {
+    type Item = Result<arrow_array::RecordBatch>;
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.batches.next()
+    }
+}
+
+impl<I: Iterator<Item = Result<arrow_array::RecordBatch>>> RecordBatchReader
+    for SimpleRecordBatchReader<I>
+{
+    fn schema(&self) -> Arc<arrow_schema::Schema> {
+        self.schema.clone()
+    }
+}
+
+/// A stream of batches that also has a schema
+pub trait RecordBatchStream: Stream<Item = Result<arrow_array::RecordBatch>> {
+    /// Returns the schema of this `RecordBatchStream`.
+    ///
+    /// Implementation of this trait should guarantee that all `RecordBatch`'s returned by this
+    /// stream should have the same schema as returned from this method.
+    fn schema(&self) -> Arc<arrow_schema::Schema>;
+}
+
+/// A boxed RecordBatchStream that is also Send
+pub type SendableRecordBatchStream = Pin<Box<dyn RecordBatchStream + Send>>;
+
+impl<I: lance::io::RecordBatchStream + 'static> From<I> for SendableRecordBatchStream {
+    fn from(stream: I) -> Self {
+        let schema = stream.schema();
+        let mapped_stream = Box::pin(stream.map(|r| r.map_err(Into::into)));
+        Box::pin(SimpleRecordBatchStream {
+            schema,
+            stream: mapped_stream,
+        })
+    }
+}
+
+/// A simple RecordBatchStream formed from the two parts (stream + schema)
+#[pin_project::pin_project]
+pub struct SimpleRecordBatchStream<S: Stream<Item = Result<arrow_array::RecordBatch>>> {
+    pub schema: Arc<arrow_schema::Schema>,
+    #[pin]
+    pub stream: S,
+}
+
+impl<S: Stream<Item = Result<arrow_array::RecordBatch>>> Stream for SimpleRecordBatchStream<S> {
+    type Item = Result<arrow_array::RecordBatch>;
+
+    fn poll_next(
+        self: Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+    ) -> std::task::Poll<Option<Self::Item>> {
+        let this = self.project();
+        this.stream.poll_next(cx)
+    }
+}
+
+impl<S: Stream<Item = Result<arrow_array::RecordBatch>>> RecordBatchStream
+    for SimpleRecordBatchStream<S>
+{
+    fn schema(&self) -> Arc<arrow_schema::Schema> {
+        self.schema.clone()
+    }
+}
+
+/// A trait for converting incoming data to Arrow
+///
+/// Integrations should implement this trait to allow data to be
+/// imported directly from the integration.  For example, implementing
+/// this trait for `Vec<Vec<...>>` would allow the `Vec` to be directly
+/// used in methods like [`crate::connection::Connection::create_table`]
+/// or [`crate::table::Table::add`]
+pub trait IntoArrow {
+    /// Convert the data into an Arrow array
+    fn into_arrow(self) -> Result<Box<dyn arrow_array::RecordBatchReader + Send>>;
+}
+
+impl<T: arrow_array::RecordBatchReader + Send + 'static> IntoArrow for T {
+    fn into_arrow(self) -> Result<Box<dyn arrow_array::RecordBatchReader + Send>> {
+        Ok(Box::new(self))
+    }
+}

rust/lancedb/src/connection.rs

@@ -27,6 +27,7 @@ use object_store::{
 };
 use snafu::prelude::*;
 
+use crate::arrow::IntoArrow;
 use crate::error::{CreateDirSnafu, Error, InvalidTableNameSnafu, Result};
 use crate::io::object_store::MirroringObjectStoreWrapper;
 use crate::table::{NativeTable, WriteOptions};
@@ -116,23 +117,27 @@ impl TableNamesBuilder {
     }
 }
 
+pub struct NoData {}
+
+impl IntoArrow for NoData {
+    fn into_arrow(self) -> Result<Box<dyn arrow_array::RecordBatchReader + Send>> {
+        unreachable!("NoData should never be converted to Arrow")
+    }
+}
+
 /// A builder for configuring a [`Connection::create_table`] operation
-pub struct CreateTableBuilder<const HAS_DATA: bool> {
+pub struct CreateTableBuilder<const HAS_DATA: bool, T: IntoArrow> {
     parent: Arc<dyn ConnectionInternal>,
     pub(crate) name: String,
-    pub(crate) data: Option<Box<dyn RecordBatchReader + Send>>,
+    pub(crate) data: Option<T>,
     pub(crate) schema: Option<SchemaRef>,
     pub(crate) mode: CreateTableMode,
     pub(crate) write_options: WriteOptions,
 }
 
 // Builder methods that only apply when we have initial data
-impl CreateTableBuilder<true> {
-    fn new(
-        parent: Arc<dyn ConnectionInternal>,
-        name: String,
-        data: Box<dyn RecordBatchReader + Send>,
-    ) -> Self {
+impl<T: IntoArrow> CreateTableBuilder<true, T> {
+    fn new(parent: Arc<dyn ConnectionInternal>, name: String, data: T) -> Self {
         Self {
             parent,
             name,
@@ -151,12 +156,32 @@ impl CreateTableBuilder<true> {
 
     /// Execute the create table operation
     pub async fn execute(self) -> Result<Table> {
-        self.parent.clone().do_create_table(self).await
+        let parent = self.parent.clone();
+        let (data, builder) = self.extract_data()?;
+        parent.do_create_table(builder, data).await
+    }
+
+    fn extract_data(
+        mut self,
+    ) -> Result<(
+        Box<dyn RecordBatchReader + Send>,
+        CreateTableBuilder<false, NoData>,
+    )> {
+        let data = self.data.take().unwrap().into_arrow()?;
+        let builder = CreateTableBuilder::<false, NoData> {
+            parent: self.parent,
+            name: self.name,
+            data: None,
+            schema: self.schema,
+            mode: self.mode,
+            write_options: self.write_options,
+        };
+        Ok((data, builder))
     }
 }
 
 // Builder methods that only apply when we do not have initial data
-impl CreateTableBuilder<false> {
+impl CreateTableBuilder<false, NoData> {
     fn new(parent: Arc<dyn ConnectionInternal>, name: String, schema: SchemaRef) -> Self {
         Self {
             parent,
@@ -174,7 +199,7 @@ impl CreateTableBuilder<false> {
     }
 }
 
-impl<const HAS_DATA: bool> CreateTableBuilder<HAS_DATA> {
+impl<const HAS_DATA: bool, T: IntoArrow> CreateTableBuilder<HAS_DATA, T> {
     /// Set the mode for creating the table
     ///
     /// This controls what happens if a table with the given name already exists
@@ -237,17 +262,24 @@ pub(crate) trait ConnectionInternal:
     Send + Sync + std::fmt::Debug + std::fmt::Display + 'static
 {
     async fn table_names(&self, options: TableNamesBuilder) -> Result<Vec<String>>;
-    async fn do_create_table(&self, options: CreateTableBuilder<true>) -> Result<Table>;
+    async fn do_create_table(
+        &self,
+        options: CreateTableBuilder<false, NoData>,
+        data: Box<dyn RecordBatchReader + Send>,
+    ) -> Result<Table>;
     async fn do_open_table(&self, options: OpenTableBuilder) -> Result<Table>;
     async fn drop_table(&self, name: &str) -> Result<()>;
     async fn drop_db(&self) -> Result<()>;
 
-    async fn do_create_empty_table(&self, options: CreateTableBuilder<false>) -> Result<Table> {
-        let batches = RecordBatchIterator::new(vec![], options.schema.unwrap());
-        let opts = CreateTableBuilder::<true>::new(options.parent, options.name, Box::new(batches))
-            .mode(options.mode)
-            .write_options(options.write_options);
-        self.do_create_table(opts).await
+    async fn do_create_empty_table(
+        &self,
+        options: CreateTableBuilder<false, NoData>,
+    ) -> Result<Table> {
+        let batches = Box::new(RecordBatchIterator::new(
+            vec![],
+            options.schema.as_ref().unwrap().clone(),
+        ));
+        self.do_create_table(options, batches).await
     }
 }
 
@@ -285,12 +317,12 @@ impl Connection {
     ///
     /// * `name` - The name of the table
     /// * `initial_data` - The initial data to write to the table
-    pub fn create_table(
+    pub fn create_table<T: IntoArrow>(
         &self,
         name: impl Into<String>,
-        initial_data: Box<dyn RecordBatchReader + Send>,
-    ) -> CreateTableBuilder<true> {
-        CreateTableBuilder::<true>::new(self.internal.clone(), name.into(), initial_data)
+        initial_data: T,
+    ) -> CreateTableBuilder<true, T> {
+        CreateTableBuilder::<true, T>::new(self.internal.clone(), name.into(), initial_data)
     }
 
     /// Create an empty table with a given schema
@@ -303,8 +335,8 @@ impl Connection {
         &self,
         name: impl Into<String>,
         schema: SchemaRef,
-    ) -> CreateTableBuilder<false> {
-        CreateTableBuilder::<false>::new(self.internal.clone(), name.into(), schema)
+    ) -> CreateTableBuilder<false, NoData> {
+        CreateTableBuilder::<false, NoData>::new(self.internal.clone(), name.into(), schema)
     }
 
     /// Open an existing table in the database
@@ -356,6 +388,15 @@ pub struct ConnectBuilder {
     aws_creds: Option<AwsCredential>,
 
     /// The interval at which to check for updates from other processes.
+    ///
+    /// If None, then consistency is not checked. For performance
+    /// reasons, this is the default. For strong consistency, set this to
+    /// zero seconds. Then every read will check for updates from other
+    /// processes. As a compromise, you can set this to a non-zero timedelta
+    /// for eventual consistency. If more than that interval has passed since
+    /// the last check, then the table will be checked for updates. Note: this
+    /// consistency only applies to read operations. Write operations are
+    /// always consistent.
     read_consistency_interval: Option<std::time::Duration>,
 }
 
@@ -685,7 +726,11 @@ impl ConnectionInternal for Database {
         Ok(f)
     }
 
-    async fn do_create_table(&self, options: CreateTableBuilder<true>) -> Result<Table> {
+    async fn do_create_table(
+        &self,
+        options: CreateTableBuilder<false, NoData>,
+        data: Box<dyn RecordBatchReader + Send>,
+    ) -> Result<Table> {
         let table_uri = self.table_uri(&options.name)?;
 
         let mut write_params = options.write_options.lance_write_params.unwrap_or_default();
@@ -696,7 +741,7 @@ impl ConnectionInternal for Database {
         match NativeTable::create(
             &table_uri,
             &options.name,
-            options.data.unwrap(),
+            data,
             self.store_wrapper.clone(),
             Some(write_params),
             self.read_consistency_interval,

rust/lancedb/src/index.rs

@@ -12,181 +12,69 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::{cmp::max, sync::Arc};
-
-use lance_index::IndexType;
-pub use lance_linalg::distance::MetricType;
-
-pub mod vector;
+use std::sync::Arc;
 
 use crate::{table::TableInternal, Result};
 
-/// Index Parameters.
-pub enum IndexParams {
-    Scalar {
-        replace: bool,
-    },
-    IvfPq {
-        replace: bool,
-        metric_type: MetricType,
-        num_partitions: u64,
-        num_sub_vectors: u32,
-        num_bits: u32,
-        sample_rate: u32,
-        max_iterations: u32,
-    },
+use self::{scalar::BTreeIndexBuilder, vector::IvfPqIndexBuilder};
+
+pub mod scalar;
+pub mod vector;
+
+pub enum Index {
+    Auto,
+    BTree(BTreeIndexBuilder),
+    IvfPq(IvfPqIndexBuilder),
 }
 
-/// Builder for Index Parameters.
-
+/// Builder for the create_index operation
+///
+/// The methods on this builder are used to specify options common to all indices.
 pub struct IndexBuilder {
     parent: Arc<dyn TableInternal>,
+    pub(crate) index: Index,
     pub(crate) columns: Vec<String>,
-    // General parameters
-    /// Index name.
-    pub(crate) name: Option<String>,
-    /// Replace the existing index.
     pub(crate) replace: bool,
-
-    pub(crate) index_type: IndexType,
-
-    // Scalar index parameters
-    // Nothing to set here.
-
-    // IVF_PQ parameters
-    pub(crate) metric_type: MetricType,
-    pub(crate) num_partitions: Option<u32>,
-    // PQ related
-    pub(crate) num_sub_vectors: Option<u32>,
-    pub(crate) num_bits: u32,
-
-    /// The rate to find samples to train kmeans.
-    pub(crate) sample_rate: u32,
-    /// Max iteration to train kmeans.
-    pub(crate) max_iterations: u32,
 }
 
 impl IndexBuilder {
-    pub(crate) fn new(parent: Arc<dyn TableInternal>, columns: &[&str]) -> Self {
+    pub(crate) fn new(parent: Arc<dyn TableInternal>, columns: Vec<String>, index: Index) -> Self {
         Self {
             parent,
-            columns: columns.iter().map(|c| c.to_string()).collect(),
-            name: None,
+            index,
+            columns,
             replace: true,
-            index_type: IndexType::Scalar,
-            metric_type: MetricType::L2,
-            num_partitions: None,
-            num_sub_vectors: None,
-            num_bits: 8,
-            sample_rate: 256,
-            max_iterations: 50,
         }
     }
 
-    /// Build a Scalar Index.
+    /// Whether to replace the existing index, the default is `true`.
     ///
-    /// Accepted parameters:
-    ///  - `replace`: Replace the existing index.
-    ///  - `name`: Index name. Default: `None`
-    pub fn scalar(mut self) -> Self {
-        self.index_type = IndexType::Scalar;
-        self
-    }
-
-    /// Build an IVF PQ index.
-    ///
-    /// Accepted parameters:
-    /// - `replace`: Replace the existing index.
-    /// - `name`: Index name. Default: `None`
-    /// - `metric_type`: [MetricType] to use to build Vector Index.
-    /// - `num_partitions`: Number of IVF partitions.
-    /// - `num_sub_vectors`: Number of sub-vectors of PQ.
-    /// - `num_bits`: Number of bits used for PQ centroids.
-    /// - `sample_rate`: The rate to find samples to train kmeans.
-    /// - `max_iterations`: Max iteration to train kmeans.
-    pub fn ivf_pq(mut self) -> Self {
-        self.index_type = IndexType::Vector;
-        self
-    }
-
-    /// The columns to build index on.
-    pub fn columns(mut self, cols: &[&str]) -> Self {
-        self.columns = cols.iter().map(|s| s.to_string()).collect();
-        self
-    }
-
-    /// Whether to replace the existing index, default is `true`.
+    /// If this is false, and another index already exists on the same columns
+    /// and the same name, then an error will be returned.  This is true even if
+    /// that index is out of date.
     pub fn replace(mut self, v: bool) -> Self {
         self.replace = v;
         self
     }
 
-    /// Set the index name.
-    pub fn name(mut self, name: &str) -> Self {
-        self.name = Some(name.to_string());
-        self
+    pub async fn execute(self) -> Result<()> {
+        self.parent.clone().create_index(self).await
     }
+}
 
-    /// [MetricType] to use to build Vector Index.
+#[derive(Debug, Clone, PartialEq)]
+pub enum IndexType {
+    IvfPq,
+    BTree,
+}
+
+/// A description of an index currently configured on a column
+pub struct IndexConfig {
+    /// The type of the index
+    pub index_type: IndexType,
+    /// The columns in the index
     ///
-    /// Default value is [MetricType::L2].
-    pub fn metric_type(mut self, metric_type: MetricType) -> Self {
-        self.metric_type = metric_type;
-        self
-    }
-
-    /// Number of IVF partitions.
-    pub fn num_partitions(mut self, num_partitions: u32) -> Self {
-        self.num_partitions = Some(num_partitions);
-        self
-    }
-
-    /// Number of sub-vectors of PQ.
-    pub fn num_sub_vectors(mut self, num_sub_vectors: u32) -> Self {
-        self.num_sub_vectors = Some(num_sub_vectors);
-        self
-    }
-
-    /// Number of bits used for PQ centroids.
-    pub fn num_bits(mut self, num_bits: u32) -> Self {
-        self.num_bits = num_bits;
-        self
-    }
-
-    /// The rate to find samples to train kmeans.
-    pub fn sample_rate(mut self, sample_rate: u32) -> Self {
-        self.sample_rate = sample_rate;
-        self
-    }
-
-    /// Max iteration to train kmeans.
-    pub fn max_iterations(mut self, max_iterations: u32) -> Self {
-        self.max_iterations = max_iterations;
-        self
-    }
-
-    /// Build the parameters.
-    pub async fn build(self) -> Result<()> {
-        self.parent.clone().do_create_index(self).await
-    }
-}
-
-pub(crate) fn suggested_num_partitions(rows: usize) -> u32 {
-    let num_partitions = (rows as f64).sqrt() as u32;
-    max(1, num_partitions)
-}
-
-pub(crate) fn suggested_num_sub_vectors(dim: u32) -> u32 {
-    if dim % 16 == 0 {
-        // Should be more aggressive than this default.
-        dim / 16
-    } else if dim % 8 == 0 {
-        dim / 8
-    } else {
-        log::warn!(
-            "The dimension of the vector is not divisible by 8 or 16, \
-                which may cause performance degradation in PQ"
-        );
-        1
-    }
+    /// Currently this is always a Vec of size 1.  In the future there may
+    /// be more columns to represent composite indices.
+    pub columns: Vec<String>,
 }

rust/lancedb/src/index/scalar.rs

@@ -0,0 +1,30 @@
+//! Scalar indices are exact indices that are used to quickly satisfy a variety of filters
+//! against a column of scalar values.
+//!
+//! Scalar indices are currently supported on numeric, string, boolean, and temporal columns.
+//!
+//! A scalar index will help with queries with filters like `x > 10`, `x < 10`, `x = 10`,
+//! etc.  Scalar indices can also speed up prefiltering for vector searches.  A single
+//! vector search with prefiltering can use both a scalar index and a vector index.
+
+/// Builder for a btree index
+///
+/// A btree index is an index on scalar columns.  The index stores a copy of the column
+/// in sorted order.  A header entry is created for each block of rows (currently the
+/// block size is fixed at 4096).  These header entries are stored in a separate
+/// cacheable structure (a btree).  To search for data the header is used to determine
+/// which blocks need to be read from disk.
+///
+/// For example, a btree index in a table with 1Bi rows requires sizeof(Scalar) * 256Ki
+/// bytes of memory and will generally need to read sizeof(Scalar) * 4096 bytes to find
+/// the correct row ids.
+///
+/// This index is good for scalar columns with mostly distinct values and does best when
+/// the query is highly selective.
+///
+/// The btree index does not currently have any parameters though parameters such as the
+/// block size may be added in the future.
+#[derive(Default, Debug, Clone)]
+pub struct BTreeIndexBuilder {}
+
+impl BTreeIndexBuilder {}

rust/lancedb/src/index/vector.rs

@@ -12,10 +12,19 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+//! Vector indices are approximate indices that are used to find rows similar to
+//! a query vector.  Vector indices speed up vector searches.
+//!
+//! Vector indices are only supported on fixed-size-list (tensor) columns of floating point
+//! values
+use std::cmp::max;
+
 use serde::Deserialize;
 
 use lance::table::format::{Index, Manifest};
 
+use crate::DistanceType;
+
 pub struct VectorIndex {
     pub columns: Vec<String>,
     pub index_name: String,
@@ -42,3 +51,145 @@ pub struct VectorIndexStatistics {
     pub num_indexed_rows: usize,
     pub num_unindexed_rows: usize,
 }
+
+/// Builder for an IVF PQ index.
+///
+/// This index stores a compressed (quantized) copy of every vector.  These vectors
+/// are grouped into partitions of similar vectors.  Each partition keeps track of
+/// a centroid which is the average value of all vectors in the group.
+///
+/// During a query the centroids are compared with the query vector to find the closest
+/// partitions.  The compressed vectors in these partitions are then searched to find
+/// the closest vectors.
+///
+/// The compression scheme is called product quantization.  Each vector is divided into
+/// subvectors and then each subvector is quantized into a small number of bits.  the
+/// parameters `num_bits` and `num_subvectors` control this process, providing a tradeoff
+/// between index size (and thus search speed) and index accuracy.
+///
+/// The partitioning process is called IVF and the `num_partitions` parameter controls how
+/// many groups to create.
+///
+/// Note that training an IVF PQ index on a large dataset is a slow operation and
+/// currently is also a memory intensive operation.
+#[derive(Debug, Clone)]
+pub struct IvfPqIndexBuilder {
+    pub(crate) distance_type: DistanceType,
+    pub(crate) num_partitions: Option<u32>,
+    pub(crate) num_sub_vectors: Option<u32>,
+    pub(crate) sample_rate: u32,
+    pub(crate) max_iterations: u32,
+}
+
+impl Default for IvfPqIndexBuilder {
+    fn default() -> Self {
+        Self {
+            distance_type: DistanceType::L2,
+            num_partitions: None,
+            num_sub_vectors: None,
+            sample_rate: 256,
+            max_iterations: 50,
+        }
+    }
+}
+
+impl IvfPqIndexBuilder {
+    /// [DistanceType] to use to build the index.
+    ///
+    /// Default value is [DistanceType::L2].
+    ///
+    /// This is used when training the index to calculate the IVF partitions (vectors are
+    /// grouped in partitions with similar vectors according to this distance type) and to
+    /// calculate a subvector's code during quantization.
+    ///
+    /// The metric type used to train an index MUST match the metric type used to search the
+    /// index.  Failure to do so will yield inaccurate results.
+    pub fn distance_type(mut self, distance_type: DistanceType) -> Self {
+        self.distance_type = distance_type;
+        self
+    }
+
+    /// The number of IVF partitions to create.
+    ///
+    /// This value should generally scale with the number of rows in the dataset.  By default
+    /// the number of partitions is the square root of the number of rows.
+    ///
+    /// If this value is too large then the first part of the search (picking the right partition)
+    /// will be slow.  If this value is too small then the second part of the search (searching
+    /// within a partition) will be slow.
+    pub fn num_partitions(mut self, num_partitions: u32) -> Self {
+        self.num_partitions = Some(num_partitions);
+        self
+    }
+
+    /// Number of sub-vectors of PQ.
+    ///
+    /// This value controls how much the vector is compressed during the quantization step.
+    /// The more sub vectors there are the less the vector is compressed.  The default is
+    /// the dimension of the vector divided by 16.  If the dimension is not evenly divisible
+    /// by 16 we use the dimension divded by 8.
+    ///
+    /// The above two cases are highly preferred.  Having 8 or 16 values per subvector allows
+    /// us to use efficient SIMD instructions.
+    ///
+    /// If the dimension is not visible by 8 then we use 1 subvector.  This is not ideal and
+    /// will likely result in poor performance.
+    pub fn num_sub_vectors(mut self, num_sub_vectors: u32) -> Self {
+        self.num_sub_vectors = Some(num_sub_vectors);
+        self
+    }
+
+    /// The rate used to calculate the number of training vectors for kmeans.
+    ///
+    /// When an IVF PQ index is trained, we need to calculate partitions.  These are groups
+    /// of vectors that are similar to each other.  To do this we use an algorithm called kmeans.
+    ///
+    /// Running kmeans on a large dataset can be slow.  To speed this up we run kmeans on a
+    /// random sample of the data.  This parameter controls the size of the sample.  The total
+    /// number of vectors used to train the index is `sample_rate * num_partitions`.
+    ///
+    /// Increasing this value might improve the quality of the index but in most cases the
+    /// default should be sufficient.
+    ///
+    /// The default value is 256.
+    pub fn sample_rate(mut self, sample_rate: u32) -> Self {
+        self.sample_rate = sample_rate;
+        self
+    }
+
+    /// Max iterations to train kmeans.
+    ///
+    /// When training an IVF PQ index we use kmeans to calculate the partitions.  This parameter
+    /// controls how many iterations of kmeans to run.
+    ///
+    /// Increasing this might improve the quality of the index but in most cases the parameter
+    /// is unused because kmeans will converge with fewer iterations.  The parameter is only
+    /// used in cases where kmeans does not appear to converge.  In those cases it is unlikely
+    /// that setting this larger will lead to the index converging anyways.
+    ///
+    /// The default value is 50.
+    pub fn max_iterations(mut self, max_iterations: u32) -> Self {
+        self.max_iterations = max_iterations;
+        self
+    }
+}
+
+pub(crate) fn suggested_num_partitions(rows: usize) -> u32 {
+    let num_partitions = (rows as f64).sqrt() as u32;
+    max(1, num_partitions)
+}
+
+pub(crate) fn suggested_num_sub_vectors(dim: u32) -> u32 {
+    if dim % 16 == 0 {
+        // Should be more aggressive than this default.
+        dim / 16
+    } else if dim % 8 == 0 {
+        dim / 8
+    } else {
+        log::warn!(
+            "The dimension of the vector is not divisible by 8 or 16, \
+                which may cause performance degradation in PQ"
+        );
+        1
+    }
+}

rust/lancedb/src/io/object_store.rs

@@ -342,7 +342,11 @@ mod test {
     use object_store::local::LocalFileSystem;
     use tempfile;
 
-    use crate::{connect, table::WriteOptions};
+    use crate::{
+        connect,
+        query::{ExecutableQuery, QueryBase},
+        table::WriteOptions,
+    };
 
     #[tokio::test]
     async fn test_e2e() {
@@ -381,9 +385,11 @@ mod test {
         assert_eq!(t.count_rows(None).await.unwrap(), 100);
 
         let q = t
-            .search(&[0.1, 0.1, 0.1, 0.1])
+            .query()
             .limit(10)
-            .execute_stream()
+            .nearest_to(&[0.1, 0.1, 0.1, 0.1])
+            .unwrap()
+            .execute()
             .await
             .unwrap();
 

rust/lancedb/src/lib.rs

@@ -36,8 +36,6 @@
 //!
 //! ### Quick Start
 //!
-//! <div class="warning">Rust API is not stable yet, please expect breaking changes.</div>
-//!
 //! #### Connect to a database.
 //!
 //! ```rust
@@ -130,16 +128,15 @@
 //! # use arrow_array::{FixedSizeListArray, types::Float32Type, RecordBatch,
 //! #   RecordBatchIterator, Int32Array};
 //! # use arrow_schema::{Schema, Field, DataType};
+//! use lancedb::index::Index;
 //! # tokio::runtime::Runtime::new().unwrap().block_on(async {
 //! # let tmpdir = tempfile::tempdir().unwrap();
 //! # let db = lancedb::connect(tmpdir.path().to_str().unwrap()).execute().await.unwrap();
 //! # let tbl = db.open_table("idx_test").execute().await.unwrap();
-//! tbl.create_index(&["vector"])
-//!     .ivf_pq()
-//!     .num_partitions(256)
-//!     .build()
-//!     .await
-//!     .unwrap();
+//! tbl.create_index(&["vector"], Index::Auto)
+//!    .execute()
+//!    .await
+//!    .unwrap();
 //! # });
 //! ```
 //!
@@ -151,6 +148,7 @@
 //! # use arrow_schema::{DataType, Schema, Field};
 //! # use arrow_array::{RecordBatch, RecordBatchIterator};
 //! # use arrow_array::{FixedSizeListArray, Float32Array, Int32Array, types::Float32Type};
+//! # use lancedb::query::{ExecutableQuery, QueryBase};
 //! # tokio::runtime::Runtime::new().unwrap().block_on(async {
 //! # let tmpdir = tempfile::tempdir().unwrap();
 //! # let db = lancedb::connect(tmpdir.path().to_str().unwrap()).execute().await.unwrap();
@@ -171,8 +169,10 @@
 //! # db.create_table("my_table", Box::new(batches)).execute().await.unwrap();
 //! # let table = db.open_table("my_table").execute().await.unwrap();
 //! let results = table
-//!     .search(&[1.0; 128])
-//!     .execute_stream()
+//!     .query()
+//!     .nearest_to(&[1.0; 128])
+//!     .unwrap()
+//!     .execute()
 //!     .await
 //!     .unwrap()
 //!     .try_collect::<Vec<_>>()
@@ -181,6 +181,7 @@
 //! # });
 //! ```
 
+pub mod arrow;
 pub mod connection;
 pub mod data;
 pub mod error;
@@ -193,8 +194,72 @@ pub(crate) mod remote;
 pub mod table;
 pub mod utils;
 
+use std::fmt::Display;
+
+use serde::{Deserialize, Serialize};
+
+pub use connection::Connection;
 pub use error::{Error, Result};
+use lance_linalg::distance::DistanceType as LanceDistanceType;
 pub use table::Table;
 
+#[derive(Debug, Copy, Clone, PartialEq, Serialize, Deserialize)]
+#[non_exhaustive]
+pub enum DistanceType {
+    /// Euclidean distance. This is a very common distance metric that
+    /// accounts for both magnitude and direction when determining the distance
+    /// between vectors. L2 distance has a range of [0, ∞).
+    L2,
+    /// Cosine distance.  Cosine distance is a distance metric
+    /// calculated from the cosine similarity between two vectors. Cosine
+    /// similarity is a measure of similarity between two non-zero vectors of an
+    /// inner product space. It is defined to equal the cosine of the angle
+    /// between them.  Unlike L2, the cosine distance is not affected by the
+    /// magnitude of the vectors.  Cosine distance has a range of [0, 2].
+    ///
+    /// Note: the cosine distance is undefined when one (or both) of the vectors
+    /// are all zeros (there is no direction).  These vectors are invalid and may
+    /// never be returned from a vector search.
+    Cosine,
+    /// Dot product. Dot distance is the dot product of two vectors. Dot
+    /// 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.
+    Dot,
+}
+
+impl From<DistanceType> for LanceDistanceType {
+    fn from(value: DistanceType) -> Self {
+        match value {
+            DistanceType::L2 => LanceDistanceType::L2,
+            DistanceType::Cosine => LanceDistanceType::Cosine,
+            DistanceType::Dot => LanceDistanceType::Dot,
+        }
+    }
+}
+
+impl From<LanceDistanceType> for DistanceType {
+    fn from(value: LanceDistanceType) -> Self {
+        match value {
+            LanceDistanceType::L2 => DistanceType::L2,
+            LanceDistanceType::Cosine => DistanceType::Cosine,
+            LanceDistanceType::Dot => DistanceType::Dot,
+        }
+    }
+}
+
+impl<'a> TryFrom<&'a str> for DistanceType {
+    type Error = <LanceDistanceType as TryFrom<&'a str>>::Error;
+
+    fn try_from(value: &str) -> std::prelude::v1::Result<Self, Self::Error> {
+        LanceDistanceType::try_from(value).map(DistanceType::from)
+    }
+}
+
+impl Display for DistanceType {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        LanceDistanceType::from(*self).fmt(f)
+    }
+}
+
 /// Connect to a database
 pub use connection::connect;

rust/lancedb/src/remote/db.rs

@@ -14,13 +14,14 @@
 
 use std::sync::Arc;
 
+use arrow_array::RecordBatchReader;
 use async_trait::async_trait;
 use reqwest::header::CONTENT_TYPE;
 use serde::Deserialize;
 use tokio::task::spawn_blocking;
 
 use crate::connection::{
-    ConnectionInternal, CreateTableBuilder, OpenTableBuilder, TableNamesBuilder,
+    ConnectionInternal, CreateTableBuilder, NoData, OpenTableBuilder, TableNamesBuilder,
 };
 use crate::error::Result;
 use crate::Table;
@@ -74,8 +75,11 @@ impl ConnectionInternal for RemoteDatabase {
         Ok(rsp.json::<ListTablesResponse>().await?.tables)
     }
 
-    async fn do_create_table(&self, options: CreateTableBuilder<true>) -> Result<Table> {
-        let data = options.data.unwrap();
+    async fn do_create_table(
+        &self,
+        options: CreateTableBuilder<false, NoData>,
+        data: Box<dyn RecordBatchReader + Send>,
+    ) -> Result<Table> {
         // TODO: https://github.com/lancedb/lancedb/issues/1026
         // We should accept data from an async source.  In the meantime, spawn this as blocking
         // to make sure we don't block the tokio runtime if the source is slow.

rust/lancedb/src/remote/table.rs

@@ -4,12 +4,13 @@ use async_trait::async_trait;
 use lance::dataset::{scanner::DatasetRecordBatchStream, ColumnAlteration, NewColumnTransform};
 
 use crate::{
+    connection::NoData,
     error::Result,
-    index::IndexBuilder,
-    query::Query,
+    index::{IndexBuilder, IndexConfig},
+    query::{Query, QueryExecutionOptions, VectorQuery},
     table::{
         merge::MergeInsertBuilder, AddDataBuilder, NativeTable, OptimizeAction, OptimizeStats,
-        TableInternal,
+        TableInternal, UpdateBuilder,
     },
 };
 
@@ -45,25 +46,55 @@ impl TableInternal for RemoteTable {
     fn name(&self) -> &str {
         &self.name
     }
+    async fn version(&self) -> Result<u64> {
+        todo!()
+    }
+    async fn checkout(&self, _version: u64) -> Result<()> {
+        todo!()
+    }
+    async fn checkout_latest(&self) -> Result<()> {
+        todo!()
+    }
+    async fn restore(&self) -> Result<()> {
+        todo!()
+    }
     async fn schema(&self) -> Result<SchemaRef> {
         todo!()
     }
     async fn count_rows(&self, _filter: Option<String>) -> Result<usize> {
         todo!()
     }
-    async fn do_add(&self, _add: AddDataBuilder) -> Result<()> {
+    async fn add(
+        &self,
+        _add: AddDataBuilder<NoData>,
+        _data: Box<dyn RecordBatchReader + Send>,
+    ) -> Result<()> {
         todo!()
     }
-    async fn do_query(&self, _query: &Query) -> Result<DatasetRecordBatchStream> {
+    async fn plain_query(
+        &self,
+        _query: &Query,
+        _options: QueryExecutionOptions,
+    ) -> Result<DatasetRecordBatchStream> {
+        todo!()
+    }
+    async fn vector_query(
+        &self,
+        _query: &VectorQuery,
+        _options: QueryExecutionOptions,
+    ) -> Result<DatasetRecordBatchStream> {
+        todo!()
+    }
+    async fn update(&self, _update: UpdateBuilder) -> Result<()> {
         todo!()
     }
     async fn delete(&self, _predicate: &str) -> Result<()> {
         todo!()
     }
-    async fn do_create_index(&self, _index: IndexBuilder) -> Result<()> {
+    async fn create_index(&self, _index: IndexBuilder) -> Result<()> {
         todo!()
     }
-    async fn do_merge_insert(
+    async fn merge_insert(
         &self,
         _params: MergeInsertBuilder,
         _new_data: Box<dyn RecordBatchReader + Send>,
@@ -86,4 +117,7 @@ impl TableInternal for RemoteTable {
     async fn drop_columns(&self, _columns: &[&str]) -> Result<()> {
         todo!()
     }
+    async fn list_indices(&self) -> Result<Vec<IndexConfig>> {
+        todo!()
+    }
 }