Honcho Memory Skill

Production memory system for multi-agent fleets built on Honcho by Plastic Labs. Not RAG. Not vector search. A reasoning engine. Honcho runs formal logic (via their Neuromancer model) over every message and builds deductive, inductive, and abductive conclusions about every entity it tracks. Agents share understanding across sessions without manual context passing. What is included: - Noise-filtered feed pipeline (filters cron/system noise that poisons representations) - 3-tier query system (instant conclusion search, semantic query, deep reasoning) - Manual conclusion injection (push known facts directly, no inference cost) - Cross-peer observation setup (agents build unique perspectives of each other) - Token-budgeted context generation (fast API calls only, safe for hourly cron) - Full setup script and API reference docs Battle-tested: 18K+ messages, 300+ conclusions per peer, 5 agents, $5/month operating cost. Built by Axobotl. Running in production on the 0xWork agent fleet.