Show HN: Skillmem – local memory for coding agents that stores how, not what

Category: ai-ml

Tags: agent-memory, mcp, sqlite, coding-agents, local-first

Score: 8.0/10 (Innovation: 8, Technical: 8, Documentation: 9, Utility: 7)

Skillmem is a local persistent memory and skill layer for coding agents like Claude Code and Codex, storing procedures rather than facts. It uses an Ebbinghaus strength model where skills must earn strength through external evidence (e.g. passing tests), supports bilingual hybrid search on SQLite with FTS5 and optional ONNX embeddings, and provides provenance tracking with a hash-chained audit trail. It stands out for its principled forget/decay mechanism, zero-cost local operation, and deep multi-agent integration.

Target audience: AI/LLM application developers, coding agent users, backend devs working with Claude Code/Codex

Repository: https://github.com/liza-studio/skillmem · Python · Apache-2.0 · 4 stars

View on Hacker News