Browser automations
Hand the agent a browser. It opens pages, clicks, fills forms, and screenshots — and you watch the live session.
What powers it
Browser automation in Froots is wrapped around the agent-browser CLI — a Playwright-driven binary that exposes browser primitives as agent tools. The browser-automation skill is one of the three hardcoded system skills; it’s gated on the agent-browser binary being present and toggleable from the Skills tab.
The agent gets six tool calls: browser_open, browser_snapshot, browser_click, browser_fill, browser_screenshot, browser_scroll. Each call streams into the UI so you can see what the agent saw at every step.
What it’s good at
- One-off scrapes — go to a site, pull the table, drop it in a note.
- Multi-page workflows — search results → click result → extract → repeat.
- Form filling — fill the application, capture a screenshot, submit on your approval.
- Visual verification — “does the staging site actually render correctly?”
Watch and intervene
What it isn’t
Not a headless scraper for millions of pages. The browser runs locally with one tab; this is one agent, one session, watching what it’s doing.