Skip to content

minecraft-agent

Minecraft agent where an LLM plans and Jev selects each player action, beating the game's dragon in under nine minutes in one run.

Projectprototype

Overview

This Minecraft agent uses GPT-6 Astra or GPT-5.6 Sol to plan and Jev to select player actions in Minecraft Java 1.16.5, using the official vanilla server and Mineflayer. A read-only Java sensor can report the dragon's head position without changing game rules or entities, and each selected action is sent through the normal player protocol.

The latest verified run, starting with an empty inventory in a fresh Survival world on Peaceful, completed in 8 minutes 43 seconds, 40% shorter than the previous recording: it travelled via the Nether, killed the Ender Dragon with six bed explosions on the first landing and reached the exit with full health and no deaths, with no restarts or operator repairs. The run used 131 Jev decisions and 35 planner calls, and all run checks and 29 local tests passed. Camera control uses continuous turns with rate limits, and a hidden native client renders a recorded video with an overlay of models, time, coordinates, objective and action. Recordings and credentials stay local; the OpenRouter key is fetched from Google Secret Manager.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich player action should be taken next to advance the planner's objective?—

How it works

  1. LLM planner sets the current objective
  2. Jev selects the next player action
  3. Mineflayer executes it through the normal protocol
  4. Sensors report state; planner updates when needed
  5. Run is recorded and verified against checks

Strengths

  • Completed the game in a verified single run
  • Few planner calls relative to Jev decisions
  • No game-rule changes

Caveats

  • Peaceful difficulty
  • Credential setup tied to Google Secret Manager
  • Recordings not included in the repo

At a glance

Type
Project
Status
prototype
Language
JavaScript
Runtime
Node with Mineflayer and vanilla server
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Which player action to take next, given the planner's objective.

How fast was the best run?

8 minutes 43 seconds, per the README.

Does it cheat?

The sensor is read-only and actions go through the normal player protocol.

Related projects

  • Real-time bot battles where you write a fighter's strategy in plain English and a System One model pilots it several times a second.

    ProjectbetaTypeScript
  • Bot that plays Red Alert 2 web skirmishes, with code handling mechanics and a decision model making stance, threat and timing calls.

    ServicebetaPython
  • Multiplayer 3D science-fantasy version of Houston's River Oaks District with Jev-powered NPC companion, chauffeur and smart drive.

    ProjectbetaTypeScript
  • Browser 3D planetary rover sandbox where a TypeSafe controller chooses the next action from complete candidates, versus a baseline.

    ProjectbetaTypeScript
  • Doom runs as a spacecraft payload over real flight software; Jev pilots from the ground every half second while Claude reviews.

    ProjectprototypePython
  • Minecraft agent where Hermes plans, Jev picks one bounded action at a time and a Mineflayer harness validates and executes.

    ProjectprototypeJavaScript