ChromaDB
ChromaDB (Chroma) is an open-source embedding/vector database designed to be simple to run locally or embed directly in an application, commonly used for storing and querying vector embeddings in retrieval-augmented generation (RAG) and…
Definition
ChromaDB (Chroma) is an open-source embedding/vector database designed to be simple to run locally or embed directly in an application, commonly used for storing and querying vector embeddings in retrieval-augmented generation (RAG) and LLM application prototypes. It offers a lightweight Python/JS API and can run in-memory, persisted to disk, or as a standalone server.
Overview
Chroma was built with developer experience as a primary goal: where many vector databases require deploying and managing a separate service, Chroma can be installed as a Python package and run entirely in-process (in-memory or backed by local disk persistence) with just a few lines of code, making it popular for prototyping, notebooks, and small-to-medium production use cases. It also supports a client-server mode for larger deployments, where a Chroma server process handles requests from multiple application instances. Chroma's core API centers on 'collections' — named sets of embeddings with associated documents and metadata. It integrates directly with popular embedding functions (OpenAI, Sentence Transformers, Cohere, and others) so users can pass raw text and let Chroma handle embedding generation, or supply pre-computed vectors themselves. Queries support nearest-neighbor similarity search combined with metadata filtering (`where` clauses), similar in spirit to other vector databases. Being open source (Apache 2.0 licensed core), Chroma is frequently used as the default or reference vector store in LLM application frameworks like LangChain and LlamaIndex, making it a common first choice for developers building their first RAG prototype before evaluating whether they need a more heavily managed or horizontally scalable solution like Pinecone or Weaviate for production scale. Chroma also offers a hosted cloud version for teams that want the open-source developer experience without operating the infrastructure themselves, positioning it between a lightweight local library and a full managed service.
Key Features
- Open-source, Apache 2.0 licensed core embedding database
- Runs in-process (embedded) or as a standalone client-server deployment
- Simple Python/JavaScript API centered on 'collections'
- Built-in integrations with common embedding providers
- Metadata filtering alongside vector similarity search
- Local persistence to disk without needing external infrastructure
- Widely used as the default vector store in LangChain/LlamaIndex tutorials
Use Cases
Alternatives
History
Chroma (often called ChromaDB) is an open-source embedding database — search infrastructure for AI applications — that stores vector embeddings together with documents and metadata for fast similarity search. It was co-founded in 2022 in San Francisco by Jeff Huber and Anton Troynikov, both with machine-learning engineering backgrounds. Released as a fully open-source Apache-2.0 project with an emphasis on developer ergonomics (a few lines of Python to get started), Chroma resonated immediately with the LangChain and LlamaIndex communities that were rapidly building retrieval-augmented generation (RAG) applications, making it one of the most widely adopted vector stores for prototyping and production LLM apps.
Sources
- Chroma — official website · as of 2026-07-17
- Chroma on GitHub — chroma-core/chroma · as of 2026-07-17