Docs
Run your first task.
A guided walkthrough of one full task, from a well-written prompt to reading the result, with everything you'll see the agent do along the way.
New here? Start with Getting started to install and sign in first. This page assumes the app is open and you're signed in.
Write a good prompt
The agent plans its own steps, so you don't describe clicks, you describe the goal. Good prompts are specific about three things: the starting point, the outcome, and the format.
Effective
Search Wikipedia for the 10 most populous countries and give me a table of country and population.
Too vague
Find some population stuff.
Prefer several small, focused tasks over one sprawling one. Tight prompts are cheaper in credits, easier to verify, and let you catch a wrong turn early.
The five-step flow
Open the command bar
Press Cmd + K on macOS or Ctrl + K on Windows and Linux to focus the task input. This is where every task begins.
Describe the task in plain language
State the goal, not the clicks. Name the site or starting point, the outcome you want, and the shape of the result, for example "list the top 5 posts as a table with titles and links." The agent figures out the individual steps.
Review the plan
Before touching anything, the agent lays out a step-by-step plan. Read it. If a step looks wrong or too broad, refine your prompt and let it re-plan before it runs.
Watch it execute, and approve when asked
The agent drives the live browser: navigating, clicking, typing, and reading pages across tabs. Anything sensitive pauses for your approval. You can pause, take over manually, or undo at any time.
Read the result
When it finishes, the agent reports what it did and delivers the output. Copy or export it, or send a follow-up in the same thread to refine or extend the task.
Next steps
- Permissions: understand every approval prompt and set per-site rules.
- Memory: how the agent carries context between tasks.
- Troubleshooting: what to do if a task stalls or a result looks off.