Browser
What it does
Navigates web pages, interacts with elements, extracts content, fills forms, and takes screenshots using a headless browser. The assistant's eyes and hands on the internet. Activate with /browser or let the assistant load it. Operations run as assistant browser commands.
Setup
None. Works immediately. For sites that need a login, store credentials in the vault first (“Store my GitHub login”). Credentials are scoped to that domain.
Permissions
- Runs in the sandbox. No host-desktop grant required.
- Credential fill uses stored vault credentials for auto-login.
- Navigating and extracting are lower risk. Filling credentials and submitting forms still go through approvals unless you wrote a trust rule.
Common prompts
| You say | What happens |
|---|---|
| “Go to example.com and tell me what is on the page” | Navigates, extracts, summarizes |
| “Search for flights from JFK to Lisbon in June” | Navigates a travel site, extracts results |
| “Fill out this form with my info” | Navigates and fills fields from USER.md / vault |
| “Log into my Jira and check my open tickets” | Uses stored credentials, extracts data |
| “Read this article and summarize it” | Fetches and summarizes |
Configuration
- No config for basic browsing.
- Commands include navigate, snapshot, screenshot, click, type, press-key, wait-for, extract, fill-credential, close.
- This is not your desktop Firefox/Chrome. No cookies, no logged-in sessions, no bookmarks. Fresh every time.
Tips
- Prefer APIs when they exist. GitHub, Jira, and similar are more reliable through a connected skill than through the browser.
- JavaScript-heavy sites usually work. Occasional SPAs render poorly. Firecrawl BYOK can help for fetch; see Firecrawl.
- Screenshots go to the model you configured. Do not leave secrets on the page if you asked for a screenshot.
Related: Features, Web search.