Tell us what you want to build.

You don't need to know how to code to build. Describe what you want, follow the preview, and click what you'd like to change. Terminals, files, and Git are still there whenever you want to go deeper.

Cockpit charges only its subscription, never per token. AI usage stays on your own account, just as it does today, with no middleman or markup. If you already subscribe to Claude or ChatGPT, Cockpit works with that subscription.

Before every request in the Create screen, Cockpit saves a project version. Compare before and after, see changed files, and restore whenever you like. Your current state is saved before restoring, so you can go back again.

No. Cockpit runs on your computer with your files, Git, and AI accounts. It only contacts this website for licensing: your key, a random computer identifier, its name, and usage counts such as open projects and agents, unless you disable them in Settings › License. Never project names, paths, conversations, or code. In Live, voice and screen sharing travel directly between participants; only room information and chat pass through the Live service.

Anytime, under Manage subscription in the dashboard. No need to contact anyone, and no cancellation fee.

Claude Code, Codex, and Gemini (through Antigravity) use your account sign-in and existing subscription. DeepSeek, Kimi, Grok, and OpenAI use API keys; OpenCode and Grok CLI use their own CLI authentication. OpenRouter, OpenCode Go, FreeLLMAPI, OmniRoute, and any OpenAI-compatible API give you access to hundreds of models, including Qwen, GLM, MiniMax, Llama, Mistral, and local models. Each team member can use a different AI and model.

Both include all of Cockpit and a full team: up to 16 AI agents per mission, live previews, verification, versions, and mobile access. The difference is capacity. On Pro, up to 3 computers and 2 concurrent Cockpit instances. On Max, up to 10 computers, 5 concurrent Cockpit instances, more offline days, and priority support.

For cloud AI, yes, just as with any other way of using it. A local model running on your computer doesn't need a connection. The license is checked online periodically. If your connection drops, work doesn't stop immediately: Cockpit stays unlocked for up to 7 days on Pro and 30 on Max, until it can check again, always within your paid period. Work already running in a mission finishes; new tasks wait until the license is restored. Your projects always stay on your computer.

It's a way to see your team as characters in an animated office. See who's working, who's waiting for your reply, and open each agent's terminal. It's built with the open-source Habblaud project, included with Cockpit, and also offers a list view if you prefer less animation.

Yes, with Live: voice calls, screen sharing, optional camera, and chat, inside Cockpit with the same account. Guests can only enter the room, without access to your projects. Live is a pilot with two-person rooms.

Yes. Switch the interface between Portuguese and English, and choose the agents' response language separately. Project names, requests, and code stay exactly as you wrote them.

When an agent delivers, Cockpit runs your project's existing type checks and tests. If a check fails, the task goes back to the agent with the error log, up to twice. Only you can approve a delivery with failing tests. Projects without tests are marked as unverified, never approved.

Yes. It runs directly on Linux and macOS, and through WSL2 on Windows. The download page lists the requirements for each system.

Yes. Cockpit runs on your computer and you access it in your phone's browser. Invite the device by link or QR code and choose view-only or full control. All four work modes and Progress fit on a phone, and delivery, review, and blocker alerts can also reach you on Telegram.

Cockpit asks for a subscription to keep building. Nothing is deleted: projects, versions, and history stay on your computer, just as you left them.