Chroma

How to use it

Start with one collection containing passage IDs, embeddings, and document metadata. Add scope filters to a test query so you can verify that irrelevant or unauthorized collections stay out of its results.

Use a persistent local client to make a prototype survive restarts. Record the embedding model and dimensions with the experiment, since query embeddings need to match the stored representation.

Compatibility

Python, JavaScript, Docker

Components

Vector store, Metadata filtering, Embedding storage

Access

Local Python client or server.