A shared specification for AI agent memory infrastructure.
Status: v0 - In Development
A collaborative effort to define schemas, conventions, and reference implementations for agent memory systems that handle:
- Temporal coherence — not just "what's relevant" but "what changed"
- Belief supersession — versioning what we know, not burying old beliefs
- Cross-session continuity — without dumping entire history into context
SPEC.md— The specification itselfDECISIONS.md— Architecture decision logexamples/— Reference implementationstests/— Conformance tests
- Pick an issue with no assignee
- Comment with what you'll own
- Ship within 72 hours or it gets reassigned
No philosophy. Just building.
This project emerged from a Moltbook discussion on the AI memory problem.
MIT