Chroma — Open-Source Vector Database for AI

What it is

Chroma is an open-source vector database designed as the data infrastructure layer for AI applications. It provides a minimal API -- create a collection, add documents, query by similarity, and retrieve by ID -- that handles tokenization, embedding, and indexing automatically. Chroma supports Python, JavaScript/TypeScript, Go, and Rust clients.

Chroma is built for developers creating RAG pipelines, semantic search engines, recommendation systems, or any AI application that needs vector similarity search. It works well with Claude, OpenAI, and other LLM providers as the retrieval backend.

How it saves time or tokens

Chroma eliminates the complexity of managing embeddings manually. You pass raw text documents to collection.add() and Chroma handles embedding generation, indexing, and storage. When querying, you pass a natural-language query and get back the most semantically similar documents. This reduces the amount of context you need to pass to an LLM, directly cutting token consumption in RAG workflows. The simple API means less code to write and fewer tokens spent generating database logic.

How to use

1. Install the client: pip install chromadb (Python) or npm install chromadb (JavaScript).
2. Create a client and collection: client = chromadb.Client() then collection = client.create_collection('docs').
3. Add documents with collection.add() and query with collection.query() using natural language.

Example

import chromadb

client = chromadb.Client()
collection = client.create_collection('my_docs')

# Add documents (Chroma embeds them automatically)
collection.add(
    documents=[
        'AI agents can browse the web autonomously',
        'Vector databases store embeddings for similarity search',
        'RAG retrieves relevant context before generating answers'
    ],
    ids=['doc1', 'doc2', 'doc3']
)

# Query by natural language
results = collection.query(
    query_texts=['How do retrieval systems work?'],
    n_results=2
)
print(results['documents'])

Related on TokRepo

• RAG tools -- retrieval-augmented generation tools and frameworks
• AI tools for coding -- developer tools enhanced with AI capabilities

Common pitfalls

• Using the default in-memory client for production. For persistent storage, use chromadb.PersistentClient(path='./chroma_data') or connect to a Chroma server.
• Not specifying an embedding function when your documents need a specific model. Chroma uses a default model but you can pass a custom one via embedding_function parameter.
• Adding documents without meaningful IDs. Chroma uses IDs for deduplication and updates, so random UUIDs make it hard to update or delete specific documents later.