ChromaDB Vector Store AI Skill Guide (Claude)

Overview & Engine Architecture

Chroma is an embedding-first vector store: documents (or precomputed vectors) land in collections, each locked to a fixed embedding dimension and distance space after first write. Clients run embedded (PersistentClient / ephemeral Client) or client–server (HttpClient / AsyncHttpClient). Agents treat the embedding function + distance metric as a retrieval contract, persist paths intentionally, and never mix models across upsert and query without a new collection rebuild.