A calm operator cockpit for running a portfolio of businesses with AI agents, built from scratch, owned outright, and engineered around flat-rate subscriptions.
Personas · the agent roster and its routingFig. 01
OkenOS is a personal control center built to manage multi-brand workflows, agent fleets, and execution logs from a single dashboard.
Instead of relying on fragmented web GUIs or expensive metered API usage, OkenOS routes tasks locally through flat-rate AI subscriptions, serving as the command bridge for autonomous development and operations.
Built specifically to run on flat-rate plans (Claude Code on Max, ChatGPT Pro via Codex CLI) rather than metered API keys. Ideation and speccing are handled by lightweight models, reserving premium model compute exclusively for execution.
Features automated systems like Dream Review that analyze weekly logs and error repetition. Deterministic evidence packs are generated directly in code so models interpret verified metrics without hallucinating data.
Reads directly off disk using plain JSON (oken-data/*.json) and connects straight to your Obsidian markdown vault ([[wikilinks]]) with 3D graph visualization. Zero external telemetry or third-party database dependencies.
Houses specialized operational views, including an agent-driven Kanban board, structured goal decomposition, and multi-model roundtables (Claude, GPT, Kimi) working off shared workspace context.
The cockpit runs at 127.0.0.1:4400 and the layout follows the viewport, so the same roster, board and logs are legible on a phone on the same network as they are on the desk.
Personas on a phoneFig. 02
Next.js 15
React 19 · TypeScript 5.7
Tailwind CSS
Custom node server.mjs
Dual WebSockets
/api/pty & /api/agent
node-pty
@xterm/xterm
Integrated terminal
@anthropic-ai/claude-agent-sdk
Codex CLI
OpenRouter
Local JSON
Markdown filesystem
No database