Genesis
An open cognitive architecture for a self-directed personal AI agent, combining persistent memory, self-directed learning and a graduated autonomy model with over 60 tools, meant to be cloned and run as one's own instance.
Genesis is built around a continuous cognitive cycle rather than a request-response loop: it remembers past interactions, learns from them between sessions, and decides on its own what to work on next instead of waiting to be told, earning wider autonomy over time as it demonstrates reliable behavior on smaller tasks. The author is explicit that this is not true AGI in a from-first-principles sense, since the underlying reasoning still runs on an existing LLM rather than a purpose-built architecture, but frames it as an open-source approximation of the behaviors people associate with a persistent personal agent.
The project is a full codebase meant to be cloned, run as a personal instance, and extended, with its memory system, learning process, autonomy rules and tool integrations documented as separate modules a developer can build on.
It is written in Python and licensed under MIT.