Architecture

The two halves of MeghaOS and how they fit together.

MeghaOS has two halves that work together: a desktop experience that you see and interact with, and an AI assistant — the "brain" — that decides what that experience should be. Neither hard-codes screens; the assistant generates a description of the interface and the desktop renders it.

The Desktop

The desktop behaves like a window manager for AI-generated content. It:

  • renders A2UI descriptions into live widgets at runtime,
  • manages the windowed canvas, virtual workspaces, and built-in apps (Linux DE & OS),
  • sends your requests to the assistant and applies real-time updates it pushes back.

On the Linux editions this is your desktop environment. On macOS, the same experience runs as a preview app alongside your Mac desktop. See Editions & Platforms.

The Assistant

The assistant is the brain. It receives requests, plans, runs capabilities, and composes UI:

  • UI composition — turns a request into an A2UI description in a single AI call (optionally embedding just-in-time code).
  • Orchestration — decomposes multi-step requests into a graph and runs the steps.
  • Capabilities — hundreds of tools for the real world (email, files, finance, messaging, system control…).
  • Memory — a semantic store of facts that personalizes responses over time.

How they communicate

Desktop → Assistant

Your requests and the current on-screen state are sent to the assistant.

Assistant → Desktop

The assistant streams real-time updates back — stream ticks, notifications, and targeted widget refreshes — so the UI changes without re-asking.

Task execution

Requests that imply doing something (not just rendering) run through the Orchestrator, which decomposes them into a graph of steps executed in parallel by the Workflow Engine. A simpler sequential path is used as a fallback if needed.

Continue with the Data Flow to trace a single request end to end.