Chroma

Chroma is an open-source vector database designed to be the easiest possible on-ramp: pip install chromadb and you have a working vector store in the same process as your Python code, no server required. It persists to local disk by default, can run as a standalone server when you outgrow embedded mode, and offers Chroma Cloud as a managed option. Its API is deliberately tiny — collections, add, query — and it can even run your embedding function for you, so you pass raw text and Chroma handles vectorization.

Its value to you is speed of iteration. When you’re prototyping a RAG pipeline, the vector database should be the least interesting part of your afternoon, and Chroma makes it exactly that: no Docker, no API keys, no schema design. That’s why it’s the default vector store in countless tutorials and a first-class integration in LangChain and LlamaIndex — the muscle memory you build here transfers directly to those frameworks.

The workflow: client = chromadb.PersistentClient(path="./db"), then collection = client.create_collection("docs"), collection.add(documents=[...], ids=[...], metadatas=[...]), and collection.query(query_texts=["..."], n_results=5). Metadata filters ride along via a where clause. Treat embedded Chroma as your development and small-production tool; when you need serious scale, high availability, or multi-region deployment, you’ll reach for the client-server mode, the cloud offering, or one of the heavier siblings covered next.