Chroma

Chroma (ChromaDB) is the "pip install and go" vector database — open-source, Python-native, and built around developer experience. It's the fastest way to get a RAG prototype running, which is why it's the default vector store in so many LangChain and LlamaIndex tutorials. You can go from zero to a working semantic search in a handful of lines.

💡 In one line: Chroma is a lightweight, open-source vector database designed for quick, local, developer-friendly RAG — install, add documents, query.

What is Chroma?

Chroma is an open-source (Apache 2.0) vector database from Chroma Inc. It has Python and JavaScript/TypeScript SDKs, runs in-memory by default (or persistent, or client-server), and stores vectors + the original documents + metadata together. Under the hood it uses HNSW for ANN search. There's also Chroma Cloud, a managed serverless option.

Key Features

  • Minimalist API — pip install chromadb, create a collection, add documents, query.
  • Auto-embedding — a built-in default model (all-MiniLM-L6-v2), or pluggable functions for OpenAI, Cohere, Hugging Face — or pass raw vectors yourself.
  • Metadata filtering and document storage built in.
  • Multimodal support (CLIP + image loaders).
  • Chroma Cloud (2026) — serverless, object-storage-backed, with hybrid search (dense + sparse + full-text + regex + metadata).

How It Works

Chroma embeds your documents automatically on add, then embeds the query and searches. 

Whiteboard
Whiteboard diagram


Deployment Modes

Chroma scales from a notebook to the cloud without changing the API. 

Whiteboard
Whiteboard diagram

Code Example


 
Chroma auto-embeds both the documents and the query — no separate embedding step needed.
 
Strengths & Trade-offs
 
Strengths
 
     
  • Dead-simple and fast to start — great for prototyping and RAG.
    •  
  • Auto-embedding and local/private by default.
    •  
  • Open-source, and the default in popular frameworks.
    •  
 
Trade-offs
 
     
  • The classic open-source version is single-node, so raw scale to many millions of vectors is limited.
    •  
  • Historically no built-in hybrid/keyword search or multi-tenancy (Chroma Cloud now adds hybrid search).
    •  
  • For very large, multi-tenant production, teams often choose Pinecone, Weaviate, or Milvus.
    •  
 
When to Use It
 
     
  • Prototyping, local development, notebooks, and small-to-mid RAG.
    •  
  • Agent memory and knowledge bases that live on your machine.
    •  
  • Chroma Cloud when you want managed, serverless scale with hybrid search.
    •  
  • Reach for others at very large, multi-tenant scale.
    •  
 
A Note on Currency
 
Chroma Cloud went GA in 2026 with hybrid search and an object-storage architecture, while the open-source version remains free (Apache 2.0) and popular for local dev. Check trychroma.com for the latest.
 
Summary
 
     
  • Chroma is a lightweight, open-source vector database built for developer experience.
    •  
  • It auto-embeds documents and queries, stores vectors + docs + metadata, and uses HNSW.
    •  
  • It runs in-memory, persistent, client-server, or on Chroma Cloud.
    •  
  • It's ideal for prototyping and local RAG; huge multi-tenant scale suits Pinecone/Weaviate/Milvus.
    •  
  • Chroma Cloud (2026) adds serverless scale and hybrid search on the same open codebase. EOF echo created