mindcraft-bots/mindcraft

What it solves

Mindcraft allows users to create intelligent Minecraft agents that can interact with the game world, perform tasks, and communicate using Large Language Models (LLMs). It bridges the gap between high-level LLM reasoning and the low-level game actions required to navigate and manipulate the Minecraft environment.

How it works

The project uses the Mineflayer API to interface with Minecraft Java Edition. It connects an LLM (via various API providers like OpenAI, Google, Anthropic, etc.) to the game, allowing the bot to process information and execute actions. The system supports:

• Bot Profiles: JSON configurations that define the LLM backend, behavior prompts, and examples.
• Dynamic Coding: The ability for the bot to write and execute code to perform new actions (sandboxed, but requires explicit enabling).
• Task Management: A system to launch bots with specific goals, such as gathering specific items or constructing blueprints.
• Multimodal Capabilities: Support for vision models for image interpretation and voice synthesis for narrating responses.

Who it’s for

• Minecraft players who want to add AI-driven companions to their worlds.
• Researchers studying embodied reasoning and multi-agent collaboration in virtual environments.
• Developers interested in integrating LLMs with game engines.

Highlights

• Broad Model Support: Compatible with a wide range of APIs including OpenAI, Gemini, Claude, DeepSeek, and local models via Ollama.
• Customizable Profiles: Users can define unique personalities and capabilities for different bots through JSON profiles.
• ** لقب Embodied Reasoning**: Based on research into multi-agent LLM frameworks for embodied reasoning.
• Task-Driven Automation: Includes built-in support for executing specific goal-oriented tasks like gathering resources.