This project provides a simple, self-building autonomous agent designed to iteratively create and register tools to complete user-defined tasks. It's targeted at developers and researchers exploring agent capabilities, offering dynamic tool creation and automatic package management for complex problem-solving without constant human intervention.

How It Works

BabyAGI 2o dynamically generates and registers Python functions as tools based on LLM outputs to accomplish tasks. It automatically installs necessary packages for these tools and handles errors through iteration. The agent's core advantage lies in its self-modification capability, allowing it to adapt and expand its functionality as needed, with the goal of integrating with the BabyAGI 2 framework for tool persistence.

Quick Start & Requirements

• Install dependencies: pip install litellm
• Set environment variables: LITELLM_MODEL (e.g., gpt-4) and relevant API keys (e.g., OPENAI_API_KEY).
• Run: python main.py
• Prerequisites: Python 3.7+, pip, LLM with tool-calling capabilities.
• Recommended: Use a virtual environment.
• Setup: Minimal, primarily API key configuration.
• More info: Replit fork

Highlighted Details

• Dynamically creates and registers tools.
• Automatically installs required packages for tools.
• Supports multiple LLMs via litellm (e.g., OpenAI, Anthropic).
• Iterative error handling and learning.

Maintenance & Community

This is an experimental project. Contributions are welcome, particularly for integration with the BabyAGI framework. Direct contact via X/Twitter is suggested for PRs.

Licensing & Compatibility

MIT License. Permissive for commercial use and closed-source linking.

Limitations & Caveats

Executes LLM-generated code and installs dependencies, requiring a secure execution environment. The project is experimental and may have unpredictable behavior.