One account, four ways to use it

Start working in a browser, with nothing to install

Everyone gets an isolated cloud container where the agent reads and writes files, runs commands and installs dependencies. It works on a phone too: add the page to your home screen and it behaves like an app. Pages the agent builds can be opened on the public internet through My outputs.

• 3 free machine hours each, and only the minutes the agent is while running are counted
• Files and sessions are kept, so everything is still there when you close the browser and come back
• After 15 minutes idle the container pauses automatically and stops costing anything

Runs on your own machine and uses no cloud machine hours

Native macOS and Windows apps. Sign in inside the app, confirm the device in your browser, and you are ready to work. Tasks run on your own machine, and only model calls go through our gateway and draw credits.

An OpenAI-compatible API, with keys that never reach the client

Upstream model keys exist only on the server. The client holds a device token you can revoke at any time, and changing your password signs out every device. Usage is billed in credits in real time, and every charge is visible in the Console.

• Both the OpenAI and the Anthropic protocol are supported
• Concurrency is capped by your plan; requests above the cap queue instead of failing
• Failed calls and empty results are not billed

Let the agent look up what happened today

Live web retrieval is built in. Results carry source links, so anything written into an answer can be traced back. You do not need a separate key for a search service.