Chroma

What it is for

Chroma is an Apache-2.0 embedding database: pip install chromadb, add documents, then search by vector, BM25, regex, or metadata — locally or on Chroma Cloud. Use it when the job is a retrieval index you own, not when you need a hosted chat seat.

What makes it unique: the cover is the official chroma-core/chroma GitHub file list — chromadb/, clients/, Apache-2.0 in the About pane — not a wordmark.

Key features

• Vector search — semantic kNN over embeddings you store (trychroma.com; docs). Similarity, not a keyword.
• Sparse / BM25 / SPLADE — lexical rank on the same collection (homepage “Sparse vector search”). A term score, not a cosine.
• Full-text / regex — trigram and regex operators (changelog “Regex Search”). A string match, not a vector.
• Metadata filters — faceted where on the same query (homepage Metadata search). A filter, not a rank.
• Collection forking — copy-on-write versions for A/B and rollbacks (homepage Forking; Aug 2025 notes). A snapshot, not a search.

CLI / Python / TypeScript / Rust clients are install / SDK. Write / storage / query dollars are the meter. Embedding model names are SKU. Self-host vs Cloud is where it runs.

Who it's for

Best fit: app teams who will run chromadb locally or pay Cloud usage after the Starter credits, and already have an embedding model.

Pricing

Verified 2026-10-22 (Asia/Shanghai) from trychroma.com/pricing. Software is Apache 2.0; Cloud is usage.

• Software — Apache 2.0. License fee $0. pip install chromadb.
• Starter (Cloud) — $0/month + usage. $5 free credits to start. 10 databases; 10 team members.
• Team (Cloud) — $250/month + usage. $100 credits included (that pool does not roll). 100 databases; 30 members; Slack; SOC 2; volume discounts.
• Enterprise — custom. Unlimited DBs/members; single-tenant / BYOC; SLAs.

Usage rates on the same page: write $2.50 / GiB written; storage $0.33 / GiB / month; query $0.0075 / TiB queried; network $0.09 / GiB returned. Do not invent a Chroma seat dollar the cards omitted.

Limitations / costs

• You bring the embedding bill on self-host. Cloud write/storage/query is extra.
• Starter is $0 + usage after $5 credits. Team is the first $250 support card.
• Index, not a prompt CMS. For versioned prompts + evals, see PromptLayer.

Alternatives

• Mem0 — memory layer when you do not want to own an index.
• Zep — agent memory, not a general vector DB.
• LlamaIndex — retrieval framework that can sit on Chroma.

Official tutorial

We searched for a first-party Chroma YouTube getting-started lesson we could oEmbed as a first-hour install. Feature talks (Lexical Search, Schema APIs) are not embedded. Use written Python getting started (pip install chromadb) plus Pricing.

Website

https://www.trychroma.com

Repository

https://github.com/chroma-core/chroma

Source

Catalog row curated for AI Tools Nav. Cover is the official chroma-core/chroma GitHub file list (Apache 2.0). Cloud dollars checked on trychroma.com/pricing (Starter $0 + usage; Team $250 / mo). No first-party getting-started embed.

Last verified

2026-10-22 (Asia/Shanghai) editorial check.