Vector search for developers who'd rather not think about vectors

What it does

Chroma is an open-source database for AI search. It stores documents, automatically converts them into vector embeddings, and indexes them for similarity search. You can also supply your own embeddings if you prefer. It offers metadata filtering and runs either in-memory for prototyping or as a persistent client-server deployment.

The interesting bit

The interesting bit is how aggressively it hides the vector plumbing. You feed it raw text and metadata; Chroma handles the tokenization, embedding, and indexing itself, exposing only a handful of collection and query methods. It is essentially a document database that happens to speak embeddings.

Key highlights

• Core API is just four functions: create a collection, add documents, query, and fetch by ID.
• Handles tokenization, embedding, and indexing automatically, or accepts custom embeddings.
• Supports metadata and document-content filters during similarity search.
• Runs in-memory for quick scripts or as a persistent server; clients exist for Python and JavaScript.
• Apache 2.0, with a managed “Chroma Cloud” tier if you want someone else to host it.

Caveats

• A row-based API is promised but not yet available; the current interface appears to be batch-oriented.
• The project describes itself as “rapidly developing,” which is a polite warning that APIs may still shift.

Verdict

Pick this up if you need a low-friction vector store for prototypes or AI agents. Look elsewhere if your workload demands deep operational visibility or strict enterprise SLAs; the README is essentially a quickstart guide.