Describe what you want. An AI agent writes the code, builds the app, and launches it in your workspace, whether that workspace is hosted in the cloud or running locally on your desktop.
Tell the AI what you want in plain English. Be as detailed or as vague as you like -- the agent asks clarifying questions, makes smart defaults, and starts building.
The agent writes React components, installs dependencies, runs the build, and self-verifies with a simulation subagent. No templates -- every line of code is generated for your specific request.
Your app opens in its own native window. Don't like something? Say "make the cells bigger" and it rebuilds instantly. Use undo to roll back any change. Every iteration is a conversation, not a deployment.
The Orchestrator sees all your apps. One prompt connects them: "When a meeting ends, send the notes to my email app and draft a follow-up." It modifies both apps, wires the IPC, and rebuilds everything.
Workspace agents build individual apps. The Orchestrator connects them all.
Every app gets window.llumos injected at runtime. No configuration required.
Send targeted messages between specific apps. The Orchestrator wires the connections; apps just call send().
Persistent key-value storage. Apps share reference data through well-known key conventions like shared:contacts.
Fire-and-forget messages to all running apps. Useful for system-wide events like theme changes or status updates.
Every app moves through a clear lifecycle. Iterate as many times as you want.
Apps run in their own BrowserWindow. Full OS access, native menus, system notifications. Not sandboxed web tabs.
Every generated app uses React 18 and Vite. Hot module replacement during development. Instant builds.
Apps live at ~/llumos/apps/<app-id>/. Metadata in apps.json. Your filesystem, your data.
Push any app to GitHub with one click. Version control built in. Share your source code.
One prompt, one app, your workspace. Start in the browser now or run it locally when you want full desktop control.