Skip to main content
The Browser lives in the Work tab, right next to your chat. Open a web page beside the code that drives it, inspect an element, and hand exactly what you’re looking at to an agent — no copy-pasting URLs or describing the bug in prose.

​
Code and page, side by side

The browser opens on the Work right-edge sidebar, in the same surface as your agent chat — so the page you’re testing and the agent fixing it share one screen. Change code in a lane, reload the page, and see the result without leaving ADE. Unlike the lane-scoped tools, the browser belongs to the ADE window: each window owns its own tabs and active page, while all windows share one signed-in session, so a site you log into once stays logged in across them. A local dev server with a self-signed HTTPS certificate can be trusted once per profile; the trust covers that address and that certificate only.
The ADE browser pane next to an agent chat

The built-in browser beside the agent chat — inspect a page and hand it to the agent.

​
Inspect and hand off

Switch the pane into inspect mode and a DevTools-style outline tracks the element under your pointer. Click to commit it, and ADE captures the element — its markup, a stable selector, what’s rendered, and a screenshot — then attaches it to the active chat as a context chip. Instead of “the button looks wrong,” the agent gets the actual element it needs to fix.

​
What it’s for

Verify a change

Reload the running app and confirm an agent’s fix actually landed.

Capture context

Select an element or page state and pass it straight into a chat.

Capture proof

Grab what you see as an artifact attached to the work.
Point the browser at a dev server started from a lane-scoped shell or agent session. The page and the process that serves it stay in the same lane, so a reload reflects the lane’s latest code.

​
Drive it from anywhere

The same browser is scriptable. The ade browser CLI surface lets an agent — or you — navigate, screenshot, inspect, and pick context from the shared tabs, so a chat agent can drive the page it’s reasoning about.
  • Every agent sees every tab. ade browser status lists every tab of the project on every machine ADE can reach, with its machine, its owner (by chat title and lane), who opened it, and where the next open will go.
  • One task stays on one machine. ADE remembers which machine’s browser holds each tab. A command that names a tab goes to that machine, a command without one goes where the chat’s own tab is, and a plain open follows you. Errors name the machine that refused.
  • Isolated tabs open on your computer, so you can see them.
  • Any agent can close any tab, and close says whose tab it closed. Agents are told to check the owner first.
  • A project holds 20 tabs by default; change it with Tabs per project in Settings › Behaviour. Archiving or deleting a lane frees its tabs without closing them. After a restart, tab ids and openers are kept, but an agent’s tab comes back only if a person used it.

Agent chat

Hand browser context to an agent and let it work.

Computer use & proof

Drive apps and capture evidence of what happened.