🤖 Browser Agent

Tell the agent what to do in your browser — it opens a tab and works it for you: navigate, click, type, and read, with every action visible in a labeled tab group. Runs on its own; the MCP Bridge is not required.

Provider

Model

Effort

No conversation yet. Try one of these, or write your own below.

Estimated conversation context: loading…

to run

Keep automation tabs in background (recommended) The agent works without taking your keyboard focus. Its tabs stay visible in labeled tab groups. Uncheck only when the agent must take focus; your next keystrokes can land in its tab.

🗂️ Session journal

The last 10 sessions are kept — every message, every reply, and every captured debug frame — so a session can be replayed here, audited, or handed to a coding agent. Exports include page content the agent read; review before sharing.

No sessions recorded yet — your first run will appear here.

🔬 Diagnostics — evidence bundle for improving the agent

Copy the compact diagnostics bundle for the tab this agent last drove. It combines the get-debug-pack evidence with build identity and worker activity. Sensitive values and fresh quota requests are omitted.