A stream of thought becomes
a stream of product.

Developers. Ghost Studio is a developer tool: it turns your spoken instructions into tasks for the coding engine configured on your own laptop — features, fixes, refactors, tests, and reviews — so real development keeps moving while you are away from your desk. It is not a general assistant, a hosted IDE, or a no-code app builder: you still work in real projects, on your own machine, with the engine and models you configure.

Whatever you would type — plain sentences, no commands or syntax to memorize. “Fix the failing login test,” “make the background black,” “refactor checkout and run the tests” all work the same way: what you say is transcribed and becomes the task your configured coding engine builds from. Steering is spoken too — say which model should run a task, or send a transcript you already have. The stream keeps your earlier speech as context, so you can think out loud instead of composing perfect prompts.

You can give voice instructions away from your desk while your phone is connected to your running laptop. The local transcription path uses voice activity detection to filter non-speech segments; browser dictation depends on your browser’s speech service. Neither identifies you as the only speaker: we cannot guarantee only your voice or zero background noise. Nearby conversations can be transcribed, and accuracy depends on your microphone and surroundings. Use a close microphone and review the transcript when using it in noisy places.

Your own computer is the execution hub; your phone sends instructions to its configured coding engine. Use a supported laptop or desktop you own and authorize. This is coding-agent access, not general remote desktop control or a grant of administrator/root privileges. Agents inherit the local process’s OS permissions; choosing a project is not an OS sandbox. Compatibility depends on your operating system and coding engine, and pairing and license setup are required.

Yes — your phone is the remote; your computer runs the companion and local coding engine. You can be away from it, its display can be off, and it can sit on the lock screen — nobody has to be at the keyboard — but the computer must remain powered on, awake, and online. The companion holds the machine awake while it runs, so you are not babysitting the display. A sleeping or shut-down computer cannot receive or run local tasks. The HTTPS connection only carries instructions; it does not keep the computer running or grant administrator/root privileges. A hosted workspace we operate is still on the roadmap.

No — the lock screen does not stop the work. Lock the screen, let the display go dark, and walk away: your spoken tasks keep landing, building, and replying while nobody is at the keyboard. You never have to sit back down and unlock it for a task to arrive. The honest limit is power, not presence: a closed lid can still put the machine to sleep, and a sleeping or powered-off computer cannot take work — keep it on with the lid open and it keeps building around the clock.

Not yet. The current phone companion is an installable progressive web app (PWA): the pairing QR your laptop shows carries its authorized link — scan it, open the page, and add it to your home screen. A native iOS client is in development; it has not been submitted to or approved by Apple for the App Store. Its workflow requires explicit consent to trust and pair your phone with your laptop, and microphone permission for user-started voice capture. The phone sends your instructions; coding tasks run on your laptop, not on your iPhone.

By scanning one QR code — your laptop shows a personal pairing QR and your iPhone scans it. The code carries the authorized connection, so there is no link to type, and after the first scan your phone reconnects on its own. A published companion release and an eligible license are required — registering interest alone does not enable a connection.

The local workflow needs a supported coding engine configured and authenticated on your own machine, such as Devin Local or an IDE integration. That engine works within the project access you authorize. A hosted experience without local installs is still on the roadmap.

No — hosted coding is not generally available. The current companion connects to your own running laptop; this website does not provide hosted coding workers. Subscriptions are sold through Stripe checkout on the account page.

Yes — open the model dropdown and choose from the connected engine’s entire catalog: hundreds of models from every major lab can sit in that one list, so you can code with almost any model out there. You can also just say which model to use — the instruction carries your choice to the engine. Switch as often as you like — the next task can run on a different model than the last. Model choices still depend on that engine and your provider account; execution requires the exact model to be supported and available on your account, and choosing a name does not grant model access or bypass quotas.

As strong as the engine and model you connect. Ghost Studio is the steering, not the engine — point it at a capable coding engine and a frontier model on your own laptop, and spoken instructions become real development work: features, fixes, refactors, and tests inside your actual project. Your engine and model choices — not a plan tier — set the ceiling.

A CLI integration does not automatically grant redistribution or resale rights. Any commercial Devin integration needs appropriate provider terms, authentication, usage accounting, and quota handling. Ghost Studio does not bundle Devin or share a personal subscription with customers.

You don't press send. The stream listens continuously and treats a pause as the end of a thought — the chunk of speech you just finished is sent to your laptop as its own task while whatever you say next is still being captured. Start a new idea where the last one ended and each piece lands as a complete instruction; long one-take recordings work too, shipping as a single task.

No — the instruction comes from whatever the microphone hears: your voice, a teammate's, or a whole group in a meeting or design discussion. Hand the phone around or set it in the middle of the room; the transcript becomes the task your configured laptop builds from, on whatever coding engine and model you connected.

Instructions stay in page memory until you export a task brief, and disappear on reset or reload. Optional browser speech recognition requires consent and may send audio to the browser’s speech provider. Studio does not store audio, call coding models, upload code, or connect to your repository in the preview. Use sample instructions, not secrets.

$50 per month on the monthly plan, billed through Stripe — cancel anytime. An annual option is also offered; the account page shows the exact configured price and billing period before you pay. Provider model usage is separate unless a plan explicitly includes it.

That is the intended hosted design: explicitly selected repositories, isolated workspaces, visible model and usage information, and a review step before changes are published. These safeguards still need implementation and verification before live hosted execution launches.