yoheinakajima/babyagi

BabyAGI is an experimental Python framework for what its author calls a "self-building autonomous agent", a program meant to grow its own capabilities by adding and managing functions over time. The README notes that the original BabyAGI from March 2023, which popularised task-planning as a method for building autonomous agents, has been archived, this newer repository is a redesign around a different core idea, and the author explicitly says it is meant to share ideas and spark discussion, not for production use. The heart of the project is a small framework called functionz, which stores, manages and executes Python functions from a database rather than only from imported source files. Functions are registered with decorators and can declare metadata: which external libraries they import, which other registered functions they depend on, which secret keys they need, and a human-readable description. functionz uses a graph-based structure to track imports and dependencies, automatically resolves and loads them when a function is called, and writes logs of every execution including inputs, outputs, execution time and errors. It also supports "triggers" that fire one function in response to events in another, and you can group related functions into reusable "function packs." Around all of this sits a web dashboard for registering and updating functions, visualising dependencies, managing secret keys and reviewing logs. You would use BabyAGI when you want to experiment with agent-like systems that build up a library of capabilities at runtime, or to prototype workflows where small Python functions are composed dynamically. Installation is via pip install babyagi.