ChromaDB-rs

A Rust based client library for the Chroma vector database.

⚙️ Running ChromaDB

ℹ Chroma can be run in-memory in Python (without Docker), but this feature is not yet available in other languages. To use this library you either need a hosted or local version of ChromaDB running.

If you can run docker-compose up -d --build you can run Chroma.

git clone https://github.com/chroma-core/chroma.git
cd chroma
# Run a ChromaDB instance at localhost:8000
docker-compose up -d --build

More information about deploying Chroma to production can be found here.

🚀 Installing the library

cargo add chromadb

The library crate can be found at crates.io.

📖 Documentation

The library reference can be found here.

🔍 Overview

The library provides 2 modules to interact with the ChromaDB server via API V1:

• client - To interface with the ChromaDB server.
• collection - To interface with an associated ChromaDB collection.

You can connect to ChromaDB by instantiating a ChromaClient

use chromadb::v2::ChromaClient;
use chromadb::v2::collection::{ChromaCollection, GetQuery, GetResult, CollectionEntries};
use serde_json::json;

// With default ChromaClientOptions
// Defaults to http://localhost:8000
let client: ChromaClient = ChromaClient::new(Default::default());

// With custom ChromaClientOptions
let client: ChromaClient = ChromaClient::new(ChromaClientOptions { url: "<CHROMADB_URL>".into() });

Now that a client is instantiated, we can interface with the ChromaDB server.

// Get or create a collection with the given name and no metadata.
let collection: ChromaCollection = client.get_or_create_collection("my_collection", None).await?;

// Get the UUID of the collection
let collection_uuid = collection.id();
println!("Collection UUID: {}", collection_uuid);

With a collection instance, we can perform queries on the database

// Upsert some embeddings with documents and no metadata.
let collection_entries = CollectionEntries {
    ids: vec!["demo-id-1".into(), "demo-id-2".into()],
    embeddings: Some(vec![vec![0.0_f32; 768], vec![0.0_f32; 768]]),
    metadatas: None,
    documents: Some(vec![
        "Some document about 9 octopus recipies".into(),
        "Some other document about DCEU Superman Vs CW Superman".into()
    ])
 };
 
let result: bool = collection.upsert(collection_entries, None).await?;

// Create a filter object to filter by document content.
let where_document = json!({
    "$contains": "Superman"
     });
 
// Get embeddings from a collection with filters and limit set to 1. 
// An empty IDs vec will return all embeddings.
let get_query = GetQuery {
     ids: vec![],
     where_metadata: None,
     limit: Some(1),
     offset: None,
     where_document: Some(where_document),
     include: Some(vec!["documents".into(),"embeddings".into()])
 };
let get_result: GetResult = collection.get(get_query).await?;
println!("Get result: {:?}", get_result);

Find more information about the available filters and options in the get() documentation.

Performing a similarity search

//Instantiate QueryOptions to perform a similarity search on the collection
//Alternatively, an embedding_function can also be provided with query_texts to perform the search
let query = QueryOptions {
    query_texts: None,
    query_embeddings: Some(vec![vec![0.0_f32; 768], vec![0.0_f32; 768]]),
    where_metadata: None,
    where_document: None,
    n_results: Some(5),
    include: None,
 };
 
let query_result: QueryResult = collection.query(query, None).await?;
println!("Query result: {:?}", query_result);

Support for Embedding providers

This crate has built-in support for OpenAI and SBERT embeddings.

To use OpenAI embeddings, enable the openai feature in your Cargo.toml.

let collection: ChromaCollection = client.get_or_create_collection("openai_collection", None).await?;

let collection_entries = CollectionEntries {
  ids: vec!["demo-id-1", "demo-id-2"],
  embeddings: None,
  metadatas: None,
  documents: Some(vec![
           "Some document about 9 octopus recipies",
           "Some other document about DCEU Superman Vs CW Superman"])
};

// Use OpenAI embeddings
let openai_embeddings = OpenAIEmbeddings::new(Default::default());

collection.upsert(collection_entries, Some(Box::new(openai_embeddings))).await?;

Sponsors

OpenSauced provides insights into open source projects by using data science in git commits.

⚖️ LICENSE

MIT © 2023