agentmemory: an LLM Wiki for Your Coding Agent

agentmemory is like a Karpathy-style LLM wiki but for your coding agent.

Local MCP server. Hooks into Claude Code, Cursor, Gemini CLI. Context is captured automatically through 12 hooks, nothing to call manually. Retrieval stacks BM25 + vector + graph with RRF fusion.

A few things worth noting:

• Graceful degradation, server down? Falls back to in-memory instead of breaking • 51 MCP tools available, 7 by default (unlock the rest with an env var) • Memory decays and reinforces over time following Ebbinghaus curve (not a flat key-value store) • Real-time viewer on localhost:3113 — you can actually watch what your agent is memorizing as it works

Honest pov: benchmark numbers (95% Retrieval@5) are self-reported. v0.9.0 dropped last month, so it's young. If you need battle-tested, Mem0 is still the safer bet.