Hi HN, I spent the last 14 months pairing with my own AI agent for everything from coding to life admin (1,000+ logged sessions). I got tired of the "Subscription Cycle" (paying $20/mo for a wrapper that forgets context next week), so I built a sovereign framework: "Project Athena". It is a local-first system that lets you build an AI agent with persistent memory, protocols, and knowledge graphs that live on your machine, not in a siloed cloud account. *Key Features:* - *Local-First Memory*: Markdown files + local vector store. You own the data. - *Protocol Library*: 69+ decision frameworks (e.g. "Law of Ruin") that the agent follows. - *GraphRAG Integration*: Uses GraphRAG to detect communities/patterns in your notes (for about $0.50/run via Gemini). - *Trilateral Feedback*: Can cross-check decisions against multiple models (Claude, Gemini, GPT) automatically. *Why I built it:* I wanted an "Exocortex" – a second brain that grows with me. Most agents are "Stateless Servants" (they do a task and reset). Athena is stateful. It remembers that I prefer TDD, that I hate Tailwind, and that I'm optimizing for low-latency. It’s open source (MIT). The repo is a "Reference Implementation" – a starter pack you can fork to build your own personal OS. Repo and Docs in the link!