---
title: Browser
description: A built-in browser that agents can drive, with tabs per chat and your own sign-ins, plus a picker that hands any element to the agent.
sidebar:
  icon: globe-02
---

The desktop app has a built-in browser that agents can use for web tasks: reading docs, testing your
local dev server, filling in a form. It lives in the right sidebar, so you see every page the agent
opens.

## Open it

Press `mod+shift+b`, or the globe icon in the activity bar on the right edge of the window. Each chat
has its own tabs. They share your sign-ins and stay open until Glade quits.

The view has tabs, an address bar with back, forward and reload, and a content blocker per site.
While an agent uses the browser, a line under the address bar says what it is doing (**Agent is
clicking**, **Agent is reading page**) with a **Stop** button, and the globe icon pulses.

## Point at an element

The picker (**Pick an element**) lets you click anything on the page. The element is added to the
composer as a reference the agent can act on, with a screenshot attached. **Add a screenshot to chat**
attaches the whole page.

## Safety

- Page dialogs are dismissed and reported to the agent instead of interrupting you.
- Agents can run JavaScript only on your own local pages (`localhost`, `127.0.0.1` and similar), to
  inspect a dev app. Every other site is read through its content.
- The browser exists only in the desktop app, not on [SSH hosts](/machines/ssh-hosts) or paired
  devices.

## Under the hood

Agents drive the browser through `browser_*` tools on Glade's `glade` MCP server:

| Group    | Tools                                                                                                                                                   |
| -------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Look     | `browser_snapshot`, `browser_find`, `browser_get_text`, `browser_screenshot`                                                                            |
| Act      | `browser_click`, `browser_type`, `browser_fill`, `browser_select`, `browser_press`, `browser_hover`, `browser_drag`, `browser_scroll`, `browser_upload` |
| Navigate | `browser_navigate`, `browser_tabs`, `browser_wait`, `browser_zoom`                                                                                      |
| Inspect  | `browser_console`, `browser_network`, `browser_evaluate` (local pages only)                                                                             |
| Control  | `browser_dialog`, `browser_content_blocker`, `browser_batch`                                                                                            |

Agents work from the page's accessibility snapshot rather than pixels where they can, and
`browser_batch` runs a predictable sequence of steps in one call.