llumos is open source, local-first, and community-driven. Your apps, your data, your platform. No exceptions.
Three principles that guide every decision we make.
Your apps run in Electron on your desktop. Your data lives on your filesystem at ~/llumos/apps/. Nothing is uploaded. No telemetry. No analytics. No account required. No cloud dependency. If the internet disappears tomorrow, every app you've built still works perfectly. This isn't a feature — it's a philosophy.
This isn't a black box. The agent prompts are markdown files you can read and edit. The engine is TypeScript you can fork. The architecture is documented. When you modify llumos, you're modifying your own copy — not requesting a feature from a product team that may never ship it.
Most tools add AI as an afterthought — a chatbot sidebar, an autocomplete suggestion. llumos is built from the ground up around AI agents. Workspace agents build individual apps with full tool access. The Orchestrator coordinates across your entire ecosystem. Simulation subagents verify builds before you see them. This is what AI-native software looks like.
The codebase is designed to be readable. Here's how to jump in.
The llumos window opens. You're ready to develop.
Start with electron/agent/prompts/ — the agent behavior is defined in markdown files you can read in 5 minutes.
Then look at electron/main.ts and electron/ipc/handlers.ts to understand how apps are managed.
The React UI is in src/ — standard components, Zustand stores, Tailwind CSS.
Browse open issues on GitHub. Areas where contributions are especially welcome:
Standard GitHub flow: branch, commit, push, open a pull request.
We review PRs promptly and value clear commit messages.
Open source means everyone can shape the future of personal software.