Please do not open a public issue for security problems. Use GitHub's private vulnerability reporting for this repository. You should get an acknowledgement within three working days.

This project is a reference implementation. The threat model lists what it is designed to defend against and what is deliberately out of scope.

• Put the service behind a gateway that authenticates users and forwards their groups, and set ARAG_SERVICE_TOKEN so nothing else can assert an identity.
• Sync document ACLs from the source systems at ingest. A document with a stale ACL is a leak this service cannot detect.
• Review quarantined chunks (arag.query logs and the answer trace): they are usually a poisoned or mis-sourced document that should be removed from the index.
• Keep model and embedding API keys in a secret manager, never in .env files that are committed.