What it does
Shared memory infrastructure for a multi-agent ecosystem. Any agent on any CLI can recall, build on, and assess confidence in prior work across session and project boundaries — without memory decaying into noise as the system scales.
Architecture / Key capabilities
- Three-layer memory stack — OpenMemory (session-scoped, fast recall), Basic Memory (file-backed, durable conventions and architecture), and project
_docs/(ground truth per-project state) - Drift detection — Monitors stored memories for semantic drift over time, flagging when recalled context no longer matches current project reality
- Extraction normalization — Standardizes how memories are written and retrieved across heterogeneous agent runtimes so recall is consistent regardless of which agent stored the memory
- Metacognitive reliability scoring — Each memory carries confidence metadata so consuming agents can weigh recall quality rather than treating all memories as equally trustworthy
- Cross-project, cross-entity access — Serves 11+ entities across 44+ projects, with namespace isolation preventing bleed between unrelated contexts
Key numbers
- 11+ entities served
- 44+ projects connected
- 3 memory layers (OpenMemory, Basic Memory, project
_docs/)
Current phase
Metacognitive Phase 1 COMPLETE. Q1 enrichments + Enhancement B shipped. Six-factor decay live. Phase 2 in progress — ia review --stale shipped by interagent-claude. Establishment log pilot spec written.
Status
Active — Phase 2 measurement underway. ia review stale --brief wired into session-start.
Links
MISSING — Repository URL