Back to Projects

muninn

Applied AI Engineer Core

Updated: 2026-03-31

Memory substrate with scoped cards, evidence links, policy-state learning, and deterministic rehydration across MCP and HTTP compatibility surfaces.

Impact: Separates durable memory, provenance, and behavioral policy into auditable contracts that multiple runtimes can share.

What I built

  • Runs a human-memory runtime for MCP tools with canonical space resolution and staged rehydration bundles.
  • Persists cards, evidence, lineage, and policy-state signals in scoped SQLite stores with structured telemetry.
  • Keeps legacy `/v0/memory/*` compatibility via an explicit v0 runtime adapter instead of hidden side paths.

Proof: Run `muninn status`, call `muninn.rehydrate.bundle`, and verify `muninn.cards.upsert` writes against `human_memory.db` as documented.

PythonFastAPISQLiteFTS5MCP