M3M

# M3M — Memory x Method x Map > Tried RAG, vector DBs, memory servers — and your agent still forgets? Try this. --- ## The problem You carefully brief your agent. It nods along. Next session: ``` Agent: "What's the file path?" ``` Or you repeat the same rules. Over and over. It never sticks. **Here's what everyone misses:** you strapped a RAG backpack onto your agent. But the agent has no idea what's inside it. --- ## What M3M does Four files. | File | Job | |------|-----| | Scene router | Agent knows which tool to reach for, when | | Rules | You say it once. It stays. | | Error log | Fixed a bug? Never again. | | Session bridge | Picks up where you left off | No database. No vector embeddings. No API keys. Just markdown files. Your agent finally knows what it knows. --- ## One command ```bash python3 init-agent-memory.py . ``` - What's left from last session - Experience stacks up, agent gets better - Same mistake? Not twice. Want auto-reminders after every edit? ```bash python3 init-agent-memory.py . --with-hooks ``` Agent gets nudged to update its memory. No discipline required. --- ## Health check (v1.2) Skeletons rot. Check yours: ```bash python3 scripts/check-memory-health.py . ``` ``` 🔍 M3M memory health check 📋 rules 🟢 still fresh 📋 errors 🟡 worth a review 📋 routes 🟡 1 gap found 🏥 Overall: 🟡 not bad ``` Three checks. Red / yellow / green. Run it weekly. --- ## Before and after Without M3M: ``` Agent: "This won't work." ``` With M3M: ``` Agent: "We hit this last time. I know what to do." ``` --- MIT. Stars help more agents know what they know.