Scope‑isolated AI retrieval library that keeps data from leaking across matters.
Chancel enforces strict scope isolation for AI retrieval, ensuring that a client’s data never crosses a matter boundary. It implements three backends—isolated, filtered, and shared—each mapping logical collections to physical storage with different security guarantees. The library is written in Python, runs offline for demos, and integrates with any vector‑store provider, making it easy to add provable isolation to existing RAG pipelines. Developers can use the CLI to run safety demos and verify that their retrieval logic respects boundaries.
View on GitHub →Nobel-Co/chancel