winstonkoh87/Athena-Public

Maintains persistent memory as local Markdown files synchronized across any LLM provider, enabling context-aware reasoning that adapts to individual user circumstances rather than generic statistical averages. Uses a 2K–20K token bootstrap system that scales with task complexity while preserving 80–98% of context window, paired with constitutional governance (6 laws, 4 capability levels) to bound autonomous decision-making. Integrates with ChatGPT, Claude, Gemini, and any model via a unified reasoning layer that treats the LLM as a swappable engine while the memory and decision protocols remain fixed.

429 stars.