BabyAGI is an experimental framework for a self-building autonomous agent. The current project replaces the March 2023 task-planning agent (now archived as babyagi_archive) with a database-backed function framework called functionz, on the premise that the way to a general autonomous agent is to build the simplest thing that can build itself. The author states that it is not meant for production use.
functionz stores, manages, and executes Python functions from a database. You register functions with @babyagi.register_function, declaring imports, dependencies on other registered functions, required secret keys, and a description; the framework tracks these as a graph, loads dependencies automatically, and logs every execution with inputs, outputs, timing, and errors. Triggers run functions in response to events such as a function being added or updated, and function packs group related functions for loading from a file path.
A dashboard served at /dashboard lets you register and update functions, visualize dependencies, manage secret keys, review logs, and configure triggers. Two draft self-building agents show the intended direction: processuserinput decides whether to reuse existing functions or generate new ones from smaller components, and self_build generates a set of functions from a description.
Features
- Function registry: database-backed storage of functions with imports, dependencies, key dependencies, and descriptions
- Dependency graph: automatic resolution and loading of dependent functions and libraries before execution
- Execution logging: every call recorded with arguments, outputs, execution time, errors, and trigger relationships
- Triggers: functions that run automatically when other functions are added, updated, or executed
- Function packs: built-in default and AI packs, plus custom packs loaded from a file path
- Secret management: keys added from code or through the dashboard and injected into functions that declare them
- Dashboard: a web UI for managing functions, dependencies, keys, logs, and triggers
- AI functions: auto-generated descriptions and embeddings, and prompt-based selection of similar functions
- Self-building agents: draft packs that write new functions from existing ones