Hybrid recall — vector + BM25 + graph (Personalized PageRank), fused and
scored exactly as the MCP engraphis_recall tool sees it. Click a result for
its full record and supersession chain.
The live answer plus what it superseded — bi-temporal history a flat vector store can't answer. Superseded facts are closed, never deleted.
Every version of a fact in world-time order, including invalidated ones, each with its validity window.
Query-less recall: importance + recency + retention, plus the last session's summary and open threads for the selected repo.
Memory health at a glance: growth, retention distribution, what the decay model will archive next, resolver activity, and the busiest entities.
Distills recurring episodic memories into durable semantic digests and
archives fully-decayed transients (bi-temporal close — recoverable, audited). Same code
path as python -m scripts.consolidate.
Every governance and resolver action — invalidations, corrections, forgets, pins, evolution links — with actor and reason. Nothing is ever silently deleted. Export produces the full bi-temporal dump (memories, history, audit) as JSON.
Entities and relations extracted from your memories — the graph arm of recall is Personalized PageRank over this network (HippoRAG-style). Colors match entity type; click a node or a Top Connected row to focus it.
Legend
Top connected
Multi-user access to this Inspector: admins manage users and licenses, members govern memories, viewers read. Roles are enforced server-side on every request.