Mnemosyne

Mnemosyne

Evidence-backed memory for AI agents

Description

An AI agent's memory holds stale conclusions and keeps acting on requirements that changed long ago. Mnemosyne is persistent agent memory tied to evidence: it recovers original sources and stops following a lesson when its source changes.

It is local-first with temporal history, tested skills and gradual migration, on SQLite and vectors with MCP support.

Features



Evidence-linked:Traceable sources.

Auto-retire:Stale lessons dropped.

Temporal history:Knowledge over time.

Local-first:SQLite.