Hi Product Hunt 👋 I'm Martin, the creator of Myrmo. Whenever a senior engineer leaves a team, their knowledge usually decays in a stale Notion page. With AI coding agents, this broken handoff is even more glaring: every new session starts blind, burning thousands of tokens re-discovering the exact same dead ends and internal environment quirks. I built Myrmo to solve this. It's a shared memory layer for agents. When an agent solves a hard error, it leaves a structured "trail" (what broke, the dead ends to skip, the working command, and verification proof). The next agent hitting that exact error finds it instantly via local fingerprinting (~2ms) instead of trying to reason from scratch. Two ways to use it: 🌍 Public colony: Free for open-source errors. 🔒 Private colony: Self-hostable for your team (`deploy/private-colony.sh up`). Nothing leaves your network, and secrets are deterministically redacted on the developer's machine. It integrates directly with Claude Code and any MCP client. I'd love your feedback on the architecture, the decay model (fixes naturally evaporate after 90 days), and whether your teams are seeing similar token-drain patterns!

Myrmo

Self-hosted shared memory for AI coding agents

Self-hosted shared memory for AI coding agents