Quickstart

From a fresh clone to a running app, and your first AI-generated screen, in under five minutes.

Run the dev app

bun run tauri:dev

On Wayland (CachyOS, etc.) always use bun run tauri:dev. The script auto-sets WEBKIT_DISABLE_DMABUF_RENDERER=1 which avoids WebKitGTK protocol error 71.

What this does:

  1. Starts the Vite dev server on port 1420.
  2. Compiles and launches the Tauri shell.
  3. Wires the two together — frontend talks to the Rust backend over IPC.

The first compile takes a couple of minutes; subsequent runs are seconds.

First-run walkthrough

When the app opens, you’ll see a 9-tab header (Wizard, Screens, Components, Design, Workflows, APIs, Runner, Library, Assets). A productive first run looks like this:

  1. Open the Wizard — click the Wizard tab. This is the only panel that uses ask_user, so it’s the friendliest way to drive the model step by step.
  2. Answer the first question — the model will start with a short text or choice question (e.g. “What kind of app are you building?”). Type your answer and submit.
  3. Watch the live preview — as the model generates code, the preview iframe updates via postMessage({type:"reload"}) HMR.
  4. Annotate, if you want — click points or drag regions on the preview to send spatial feedback back to the model.
  5. Switch to Runner — the Runner tab shows the file tree, terminal (xterm.js), and live dev server. If the Wizard produced a scaffold, the dev server auto-starts.

Try other panels

  • Screens — chat + AI generation + device preview, with an embedded flow canvas
  • Components — prompt → component code, live preview
  • Design (Themes) — prompt → CSS theme, with the same preview
  • Workflows — node-based graph execution (React Flow)
  • APIs — HTTP request/response testing
  • Library — searchable library of everything you’ve generated

Build a production binary

bun tauri build

Outputs to src-tauri/target/release/bundle/:

  • Linux: .deb, .AppImage
  • macOS: .dmg, .app
  • Windows: .msi, .exe

What next