Heimdall – Trust-verified knowledge layer for AI coding agents
Source
github.com
Author
arihantdeva
Date
Context
Heimdall adds checks against actual files to an AI coding AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →'s memory search. Its documented result labels distinguish a plausible semantic match from one supported by an existing path and stronger content checks. Moved files can be reattached to their records, while missing paths are marked for removal.
To refresh that memory, notifications tell a reconciler which path to inspect; the reconciler reads the file from disk rather than trusting the notification's description. This is intended to keep remembered code aligned with edited code. The README explicitly allows a delay between an edit and reconciliation, and its trust labels remain retrieval signals rather than proof that a result is the right solution.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.