The OpenClaw browser tool

The OpenClaw browser tool runs a dedicated Chrome, Brave, Edge or Chromium profile that the agent controls. It goes through a small control service inside the Gateway, loopback only, and it is isolated from your personal browser: the profile is named openclaw, carries an orange accent so you can tell it apart, and never touches your own profile. Here is what the tool gives an agent, the three profiles the docs define, the two ways it ends up disabled, and where the browser can live when it is not on the Gateway host.

What the agent gets

• Deterministic tab control: list, open, focus, close.
• Actions: click, type, drag, select; snapshots, screenshots and PDFs; question answering over the readable text of a page without returning a whole snapshot.
• Downloads land under a managed directory and come back as metadata once the final URL has passed policy.
• A bundled browser-automation skill that teaches the loop: check status and tabs, label task tabs, snapshot before acting, snapshot again after a change, recover a stale reference once, and report a login, a captcha or a camera prompt as manual action instead of guessing.

Three profiles

openclaw is the managed, isolated browser and the default. user attaches to your real signed-in Chrome through Chrome DevTools MCP, and Chrome shows a blocking allow-remote-debugging prompt the first time, so someone has to be at the computer. chrome is the extension profile for the same signed-in Chrome, driven through the OpenClaw extension, which is why it works from a phone with nobody at the desk. The docs’ rule: managed by default, the extension when signed-in sessions matter and you are away, the DevTools profile when they matter and you are there.

This browser is not your daily driver. It is a safe, isolated surface for agent automation and verification.

The OpenClaw Chrome extension covers the extension route in detail, including what it can and cannot see.

Two ways it is disabled

• Browser disabled means the plugin or the browser enabled flag is off; the default needs both on.
• A missing browser command, or an agent that says the tool is unavailable, is usually a plugins allowlist that omits browser with no root browser block in the config; add it to the list, or add the block.
• The coding tool profile includes web search and web fetch but not the browser; add it with the also-allow list, and a sub-agent needs it at the profile stage, not only in the sub-agent policy.

Remote and hosted browsers

The browser can run on another machine or at a hosted CDP provider: a node browser proxy with no configuration, Browserless on the same host or hosted, Browserbase, Notte, or any direct WebSocket CDP endpoint. Navigation goes through an SSRF policy either way, and the control API has its own auth. OpenClaw sandboxing explained explains how the browser fits under a sandbox, and OpenClaw security best practices is the checklist.

On Diali

On Diali the managed browser is part of every plan and runs beside the assistant on its isolated instance, so there is no Chromium to install and no port to reach; your own signed-in browser stays yours unless you pair it through the extension. Hosted OpenClaw on Diali is the assistant.

• A separate profile, loopback control, never your own browser.
• openclaw by default; chrome from a phone; user when you are at the desk.
• Disabled means the plugin, the flag, or the allowlist.