AI · Multi-Agent · In progress
Neocortex Jarvis Life OS
A local-first system of agents for the parts of my day that used to live in separate apps.
Solo engineer · product, architecture, backend, and UI
The product
Neocortex is a personal operating system I am still building. Fitness, career, learning, and planning each have an agent, with a web command center and a phone HUD on one backend.
One job per agent
Fifteen agents. Jarvis plans the day, Stark scores career work, Hercules tracks biometrics, Turing runs review, Cicero coaches interviews.
The backend
FastAPI, a SQLite write queue, and a six-provider model chain so a failed provider is not a failed day.
Two surfaces
A Next.js command center and an Expo HUD for the phone, including Health Connect, meal photos, and a morning brief.
How I worked
- Split the day into domains before writing agents, so each one had a job and a way to be turned off.
- Built the backend, the web HUD, and the phone HUD against the same local data.
- Added a safety layer with a kill switch per agent, and backend tests, before widening the scope.
Challenges
Scope is the risk
Fifteen agents only stay understandable if each one owns one thing. A shared junk drawer would have made the system impossible to debug.
Providers fail
The chain has to answer from the next model without the plan for the day disappearing.
Finishing the loop
A long side project dies when the next idea is more interesting than the last one working. The work is keeping the current loop finished.
What shipped
15 agents orchestrated
Jarvis, Stark, Hercules, Turing, Cicero — each agent owns one domain of my life.
- FastAPI
- Python
- Next.js
- Expo
- SQLite
- n8n
- Groq
- Gemini