unknown

A developer describes building and abandoning two AI memory systems (a Laravel/Neo4j package and a Go rewrite) before creating MuninnDB, a fundamentally different approach to AI agent memory. Unlike conventional retrieval systems that passively store and fetch embeddings, MuninnDB implements cognitive psychology concepts from ACT-R theory: temporal decay (activation strength fades over time without access), Hebbian association learning (co-occurring concepts auto-link), Bayesian confidence scoring (certainty updates with new/contradictory evidence), and push triggers (proactive surfacing of relevant memories before being asked). The system runs entirely locally with a bundled embedding model, no LLM required, and exposes an MCP interface plus web UI. Currently in alpha at v0.2.4.