Chroma

Chroma bills itself as "the open-source data infrastructure for AI", an embedding and hybrid search database, Apache 2.0 licensed with 29K+ GitHub stars, built around a deliberately small surface area: a handful of core functions handle adding documents, querying, and managing collections, with tokenization, embedding, and indexing all handled automatically underneath. That minimalism is the whole pitch, an in-memory client or a single chroma run --path command gets a persistent local instance running in seconds, no infrastructure decisions required before the first query.

The same simplicity extends to querying: dense vector search, metadata filtering, and full-text/document content filtering are all exposed through one consistent API, with client libraries for Python and JavaScript covering the two ecosystems most RAG tooling is built in. That low ceiling to a working prototype has made it the default starting point in countless LangChain and LlamaIndex tutorials, even though it competes with more infrastructure-heavy options once an application needs to scale.

Scaling up doesn't require switching databases: Chroma Cloud, now generally available (out of alpha as of the 1.4.1 release), offers the same API as a serverless, auto-scaling managed service with SOC 2 Type II compliance. Pricing is usage-based from a $0/month starting point with included credits, billed on data written, storage, and queries, plus a Team tier with a flat monthly platform fee and sales-led Enterprise pricing, so the free self-hosted path and the managed path share the same code without a rewrite.