An LLM-driven Minecraft bot with a strategic/tactical planning loop, built with mineflayer and TypeScript.
A mineflayer-based Minecraft bot whose behavior is
driven by an LLM (via an OpenAI-compatible API). The architecture splits decision-making into a
StrategicPlanner (longer-horizon goals) and a TacticalPlanner (moment-to-moment execution),
with working memory persisted to SQLite. Development is tracked through a structured phase/plan
system in .planning/STATE.md.
The goal is a bot that "feels like a competent, persistent player — not a command executor."
Note: This is a separate, earlier experiment from
minecraft-server, which uses a different architecture (a decoupled runtime/brain split over WebSocket) for the same general idea of an LLM-controlled Minecraft agent. The two don't share code.
- TypeScript, Bun/tsx
- mineflayer, mineflayer-pathfinder, mineflayer-collectblock, mineflayer-auto-eat
- OpenAI SDK (pointed at an OpenAI-compatible provider)
- better-sqlite3, Zod
Experimental / in development — see .planning/STATE.md for the current milestone and phase.