winstonkoh87/Athena-Public

What it solves

Athena addresses the problem of unreliable, opaque, and vendor-locked AI memory. Most AI assistants rely on platform-specific memory that is lost when switching models (e.g., from GPT to Claude) or reset during model updates. Athena moves the memory layer to the user's local machine, ensuring that personal context, decision frameworks, and session history are owned and portable across any LLM provider.

How it works

Athena acts as a memory and governance layer between an AI-enabled IDE (like Cursor or VS Code) and a reasoning engine. Instead of relying on cloud memory, it uses a local workspace of hundreds of small, addressable Markdown files and Python scripts. These files serve as "memory nodes" that an agent can query surgically via semantic search or filename, rather than loading a massive monolith of text. This modular architecture allows for a lightweight boot (2K–20K tokens) while maintaining a massive, compounding history of thousands of sessions.

Who it’s for

Users who want a sovereign, long-term memory system for their AI assistants that persists across different LLM providers and the same time, avoids vendor lock-in, and uses their personal context to improve AI reasoning and decision-making.

Highlights

• Local-First Ownership: All memory is stored as plain Markdown files on the user's disk, allowing for version control via Git.
• Model Agnostic: The same memory substrate can be used with ChatGPT, Claude, Gemini, or any other model.
• Agentic Governance: Includes a set of constitutional laws and capability levels to prevent "irreversible ruin" and ensure the AI pushes back against the user when grounded in documented patterns.
• Surgical Retrieval: Uses a modular file system optimized for AI agents to query specific nodes of information without wasting context window space.
• Human Augmentation: Focuses on augmenting human cognition by using personal context to provide calibrated, situation-specific answers rather than generic statistical averages.