Chroma Documentation

What Chroma is

Chroma describes itself as open source data infrastructure for AI: a search and vector database you can start with pip install chromadb or npm install chromadb. In its simplest mode it runs inside your own program, in memory or saved to a folder, so there is no server to set up. When you need one, chroma run starts it as a client-server database, and Chroma Cloud is a hosted serverless option.

What the docs cover

• Getting started: creating a client, a collection, adding documents and querying them.
• Embeddings: Chroma can embed your documents with a default model, or you can plug in an embedding function from a provider or an open model.
• Querying: similarity search with the number of results, plus metadata filters and document text filters.
• Running modes: in-memory, persistent on disk, client-server and the cloud service.
• Integrations: LangChain, LlamaIndex and other frameworks commonly used for RAG.

Is it worth learning?

As a first vector store, yes. You can see the whole idea of embeddings, similarity and metadata filters in about twenty lines of Python, which makes it ideal for a notebook, a hackathon or a final year project prototype. It is a good way to understand what pgvector, Qdrant or a managed vector service do, before choosing one for production.

Things to watch

An in-process database lives inside one program, so it is not shared between several app servers. Decide early where the real data lives: often the source documents stay in PostgreSQL or object storage and the vector store only holds embeddings plus an id. Keep a script that can rebuild the collection from scratch, because you will change embedding models.

Prerequisites and time

Basic Python or JavaScript. An evening to build a "chat with my notes" retrieval step; there is also an official Colab notebook to try without installing anything.

Free or paid?

The open source library and server are free under Apache 2.0. Chroma Cloud is paid, with trial credits for new accounts.

Best for: students and developers taking their first steps with embeddings, semantic search and RAG.