# Athena

Athena is a local-first personal knowledge management agent that keeps a person's memory, reasoning structure and governance rules as plain files on their own machine, so that context can move between ChatGPT, Claude, Gemini or any other model instead of resetting with each one.

Athena stores what a user has taught it as plain Markdown files on the user's own disk rather than inside one provider's cloud account, so the files can be version-controlled, edited directly, and pointed at whatever model the user switches to next, rather than being lost to a model update or locked to one platform.

It is described as a memory, reasoning and governance layer that sits in front of any LLM: it recalls relevant past context rather than replaying everything, applies structured reasoning that questions the framing of a request rather than only optimizing within it, and enforces bounded autonomy through a stated set of constitutional rules and capability levels. Boot cost for loading its context scales with the task, from a light few thousand tokens for ordinary chat up to a larger load for its most thorough startup mode.

It also supports scheduled tasks and a self-improvement loop, described in its own documentation, and is written in Python.

Website: https://winstonkoh87.com/athena/