Quick Start
Voice Mirror is a voice-native IDE: you build real apps by voice, watch them render live in the App Preview, and the AI sees and drives the same running surface you do. The loop is simple — voice → build → see → fix.
This guide gets you there in about 5 minutes.
1. Install
Section titled “1. Install”Download the Windows installer from the latest release and run it. Installed builds auto-update to new stable releases.
The alpha installer is unsigned — Windows SmartScreen will warn on first run. Click More info → Run anyway. Prefer the bleeding edge? There’s a nightly channel too.
Building from source (or on macOS/Linux)? See the Installation guide for prerequisites and full setup.
2. Launch
Section titled “2. Launch”Launch Voice Mirror from the Start menu (or npm run dev from a source
checkout). The floating orb appears on your desktop. There are three ways to
interact:
| Mode | How | Best For |
|---|---|---|
| Wake Word | Say “Hey Claude” (default) | Hands-free |
| Push-to-Talk | Hold the bound key/button (Windows) | Quick commands |
| Call Mode | Always-on listening | Continuous conversation |
3. The wow moment: build it, watch it, let the AI drive it
Section titled “3. The wow moment: build it, watch it, let the AI drive it”This is the loop Voice Mirror is built around. Ask it to build and launch a small app, then watch it come to life in the App Preview while the AI clicks through it.
“Hey Claude, build me a small to-do web app and launch it so I can see it.”
What happens:
- Voice → build — the AI scaffolds the app, writes the code, and starts the dev server for you.
- See — the running app streams live, true-size into the App Preview. You and the AI are looking at the same surface.
- Drive — ask the AI to test it: “Add a couple of tasks and check the box on the first one.” It reads the app’s element tree, then clicks and types in the live preview while you watch.
- Fix — spot something off? “The checkbox doesn’t strike through the text — fix that.” The AI edits the code, the preview reloads, and you verify it together.
The App Preview drives web, Tauri, WebView2, and Electron apps, and even native Windows apps (Notepad, Calculator, Settings) through the same interface — the AI can’t tell which engine is underneath. Focus stays in sync: the preview auto-follows whichever window you or the AI last touched.
App Preview and native-app driving are Windows-only for now. macOS and Linux support is planned.
4. A few more things to try
Section titled “4. A few more things to try”- “Hey Claude, what’s on my screen?” — captures and analyzes your screen.
- “Search for the latest Svelte 5 docs.” — web search via built-in browser automation.
- “Remember that the API key lives in .env.” — stores to persistent memory across sessions.
5. Explore the workspace
Section titled “5. Explore the workspace”Beyond the orb, Voice Mirror opens into a full VS Code-style IDE workspace:
- Chat panel — talk to your AI provider with full markdown rendering and inline tool-activity cards.
- App Preview / live preview — the see-and-drive surface, with dev-server detection and auto-start (Node and Python).
- File tree + code editor — CodeMirror 6 editor with syntax highlighting, go-to-definition, references, rename, and format via LSP.
- Integrated terminals — connected to AI providers, with dev-server start/stop/restart from the status bar.
- Command palette — quick file, line, and symbol navigation.
What’s Different?
Section titled “What’s Different?”Without Voice Mirror
Section titled “Without Voice Mirror”Your AI assistant:
- Forgets everything between conversations
- Can only respond to text — no voice, no screen awareness
- Can’t see or interact with the apps you’re building
- Lives in a separate window from your code
With Voice Mirror
Section titled “With Voice Mirror”Your AI assistant:
- Lives as a floating orb, expandable to a full IDE workspace
- Responds to voice commands hands-free
- Builds apps by voice and watches them render live in the App Preview
- Sees and drives the running app — the same surface you watch
- Remembers project conventions and decisions across sessions
- Controls the browser, runs terminal commands, and edits files
- Provides 45 MCP tools across 5 groups for deep integration
- Works with 75+ models via OpenCode, plus Claude Code, Codex, Gemini CLI, and more
Next Steps
Section titled “Next Steps”- Installation — full setup guide with prerequisites and configuration
- Voice Mirror Architecture — how the App Preview and see-and-drive loop work
- MCP Tools Reference — full tool documentation