Your cloud computer
Your assistant has its own private computing environment — here's what that means and what it can do.
Your assistant has a computer
Most AI assistants exist only as conversations. They can generate text, but when it comes to actually doing things — running code, browsing a real website, calling an API, executing a script — there's nothing there.
Your assistant is different. It has access to a private cloud computing environment: a sandboxed workspace that spins up when your assistant needs to do real work. Not a simulation of a computer. An actual one.
What runs there
When your assistant is working on a task, it can use its workspace to:
Browse the web — not by fetching a URL and parsing the HTML, but by opening a real browser, navigating to pages, clicking through interfaces, and reading what's actually on screen. This is how your assistant can check a booking link for available times, interact with a web form, or navigate a site that requires JavaScript to load.
Run skills that need compute — some skills are packages of code that need to actually run somewhere. Those run in your workspace, isolated from everyone else's. When a skill needs a runtime or a local process, it gets one.
Execute multi-step work — complex tasks involve more than a single API call. Your assistant can chain operations, process results, loop over data, and build up context within a session — all within the workspace.
It's private and per-user
Your workspace is yours. No one else's skills or tasks run in it. When you're not actively using it, it's idle. When a task needs it, it starts up in the background before your assistant's first tool call — so you don't wait for a cold start mid-task.
After 30 minutes of inactivity, the workspace shuts down automatically. Your data isn't persisted there — it's an execution environment, not storage. Files your assistant creates during tasks are saved to your Drive, not left in the workspace.
The local extension — your own files
The cloud workspace handles tasks that need compute. But some of the most valuable work involves files that already exist — on your laptop, in your documents folder, in a project directory.
The desktop app bridges the two. By opening the desktop app and granting it access to a folder, you extend your assistant's reach to your local machine. Your assistant can read files from those folders, search across them, and (with your approval) make edits.
Every approved write is backed up before it's applied. You can undo any change from the workspace history panel. Your assistant never sees files outside the folders you've explicitly granted — and certain paths (credentials, keys, .env files) are always off-limits regardless.
The desktop app is required for local file access. If you're in the browser, your assistant can tell you what it would do, but can't touch your local filesystem until you open the app.
What your assistant will never do
Regardless of what's available in the workspace:
- It will not access other users' environments or data
- It will not read credential files, private keys, or
.envfiles from your local machine - It will not make changes to your local files without showing you exactly what it plans to change and getting your approval
- It will not keep running after you close the task — no background processes linger once work is complete