chroma

Install chromadb, create a client and collection, then call collection.add with documents, metadatas, and ids. Use collection.query with query_texts and n_results to retrieve the most similar documents along with their distances and metadata.

Chroma is open-source and self-hosted with metadata filtering, suited for local development and RAG prototypes. Pinecone is a managed cloud service with auto-scaling, while FAISS provides pure similarity search without metadata storage.

Yes, use chromadb.PersistentClient with a path argument to persist data to disk automatically. Reload the same path later and call get_collection to access your stored documents and embeddings.

Yes, Chroma integrates with both frameworks. Use langchain_chroma.Chroma to create a vector store from documents, or ChromaVectorStore in LlamaIndex backed by a Chroma collection for indexing and querying.

Pass a where clause to collection.query with exact matches, comparison operators like $gt and $lte, or logical operators $and and $or. This narrows the search space before similarity ranking is applied.

Avoid Chroma when you need managed auto-scaling cloud infrastructure, where Pinecone fits better, or when you need maximum query performance at very large scale, where Rust-based Qdrant may be preferable.