Chroma

Chroma is “the open-source data infrastructure for AI”: a database for embeddings with metadata, supporting dense, sparse and hybrid search. A different note, chroma-context-1, covers a Chroma model rather than the database.

Repository snapshot (GitHub API, 2026-11-19)

ItemDetail
Repochroma-core/chroma, Rust, Apache-2.0
Created2022-10-05
Activitylast push 2026-11-13; GitHub’s latest-release tag is 1.5.9 (2026-05-05), though more recent tags exist
Size29,715 stars, 2,609 forks

Capabilities (docs and README)

  • Index: HNSW by default for local collections (tunable ef_construction, ef_search, max_neighbors); SPANN in distributed and Cloud deployments.
  • Metrics: l2 (default), ip and cosine, set per collection and limited by the embedding function.
  • Dense, sparse and hybrid search; keyword and regex search without embeddings; full-text search on Chroma Cloud.
  • Metadata filters (where) and document-content filters (where_document).
  • Built-in embedding and tokenisation, or bring your own embeddings.
  • No dimension or collection-size limit was stated on the pages read.

Deployment and billing

In-memory client for prototyping, persistent local storage, client-server mode (chroma run), Docker per the README, and Chroma Cloud (managed, serverless). Cloud plans are Starter (with free credits), Team and Enterprise; usage is billed on write, storage, query and network. Rates are on the pricing page and are not recorded here. A deployment-modes docs page returned 404, so modes come from the README.

Where it fits

Chroma is the easiest on-ramp for prototypes and RAG tutorials, with a managed path later. Compare lancedb, pgvector, qdrant and pinecone in vector-databases-compared.

Sources