Steve  by YuvDwi

AI agents for embodied gameplay and complex task execution

Created 3 weeks ago

New!

1,008 stars

Top 37.0% on SourcePulse

GitHubView on GitHub
Project Summary

This project provides AI agents, named "Steve," that can play Minecraft, automating tasks through natural language commands. It targets Minecraft players and AI researchers interested in embodied agents, offering benefits like autonomous gameplay, collaborative task execution, and a novel interaction paradigm.

How It Works

Each Steve agent operates on a ReAct-style loop: Reason, Act, and Observe. Natural language instructions are processed by an LLM (Groq, OpenAI, or Gemini) which breaks them down into executable code. This code interacts directly with Minecraft's game mechanics. For multi-agent coordination, a server-side manager divides tasks like building into spatial sections, assigns agents, prevents conflicts, and rebalances workloads deterministically.

Quick Start & Requirements

  • Prerequisites: Minecraft 1.20.1 with Forge, Java 17, and an API key for a supported LLM (OpenAI, Groq, or Gemini).
  • Installation: Download the JAR from the releases page, place it in the Minecraft mods folder, launch the game, and configure your API key in config/steve-common.toml.
  • Usage: Spawn an agent with the command /steve spawn <name> and press 'K' to open the command panel.
  • Links: Project repository: https://github.com/YuvDwi/Steve

Highlighted Details

  • Resource Extraction: Agents autonomously determine optimal mining locations and strategies.
  • Autonomous Building: Agents plan structures, manage materials, and construct them block by block.
  • Collaborative Execution: Multiple agents can automatically partition and parallelize complex tasks like building a castle.
  • Natural Language Interface: Agents interpret and execute commands such as "mine 20 iron ore" or "build a house near me."

Maintenance & Community

The project is inspired by LangChain/AutoGPT and utilizes Minecraft Forge. Planned features include crafting, voice commands, persistent memory via vector databases, and asynchronous action execution. Community interaction details like Discord/Slack are not specified.

Licensing & Compatibility

The project is licensed under the MIT license, permitting commercial use and closed-source linking. It requires Minecraft 1.20.1 with Forge.

Limitations & Caveats

Agent intelligence is dependent on the LLM used; GPT-3.5 may make suboptimal decisions compared to GPT-4. Currently, agents cannot craft tools. Actions are synchronous, preventing multitasking. Agent memory and world context reset upon game restart, though persistent memory is planned.

Health Check
Last Commit

3 weeks ago

Responsiveness

Inactive

Pull Requests (30d)
7
Issues (30d)
25
Star History
1,013 stars in the last 24 days

Explore Similar Projects

Starred by Andrew Ng Andrew Ng(Founder of DeepLearning.AI; Cofounder of Coursera; Professor at Stanford), Thomas Wolf Thomas Wolf(Cofounder of Hugging Face), and
4 more.

ag2 by ag2ai

0.7%
4k
AgentOS for building AI agents and facilitating multi-agent cooperation
Created 1 year ago
Updated 3 days ago
Feedback? Help us improve.