muninn
Applied AI Engineer Core
Updated: 2026-02-18
Pluggable memory service for LLM agents with SQLite durability, audit logging, and staged confirmation workflows.
Impact: Turns ad-hoc memory writes into traceable, policy-gated memory operations.
What I built
- Built namespace-scoped storage for entities, facts, episodes, and preferences.
- Implemented FTS, vector, and hybrid retrieval with memory-card rendering.
- Added pending-candidate review and confirmation endpoints.
Proof: Run `scripts/smoke.sh` and call `/v0/memory/*` endpoints documented in README.
PythonFastAPISQLiteFTS5Pydantic