yoheinakajima/babyagi

BabyAGI is an experimental Python framework for building a self-building autonomous agent, that is, a program meant to write and manage its own pieces of code over time. The README is upfront about its status: the original BabyAGI from March 2023 introduced task planning as a method for developing autonomous agents and has been archived in a separate snapshot repo; this newer BabyAGI is a different attempt by the same author, who explicitly warns it is not meant for production use and is intended as a place to share ideas and let experienced developers experiment. The way it works centres on something the README calls "functionz," a function framework for storing, managing, and executing Python functions out of a database. Instead of writing your logic in plain modules, you decorate functions with @babyagi.register_function() and they are saved into the system along with metadata describing what they import, which other functions they depend on, and which secret keys they need (for example an openai_api_key). The framework then resolves these dependencies automatically when a function is called, so a function like hello_world can transparently use another function world that it depends on. There is also a load_function mechanism for loading entire packs of related functions from a file path, and built-in function packs ship with the project. On top of this, BabyAGI runs a web dashboard you mount in a small application and visit at a URL like /dashboard. From the dashboard you can register, update, or remove functions, visualise the dependency graph between them, add and manage secret keys, set up triggers that run one function in response to another, and read a comprehensive log of every execution, including arguments, outputs, execution time, and any errors. You would use this if you want to experiment with the "agent that grows its own toolbox" idea, a system where new functions can be added at runtime, can call each other, and can be triggered by events, all observable from a single dashboard. It is installed via pip install babyagi and is written in Python. The full README is longer than what was provided.