Local UI testing, console inspection, screenshots, documentation research, and approved deployment workflows.
Features
Browser agent
The browser is a real Chromium view embedded in the Vector window, not a screenshot service, and it is attached to the current repository and session. Open a local preview or URL, interact manually, or ask Vector to inspect and test the page as part of the engineering task.
Observe, repair, retest
The agent opens the running product, clicks and types, reads console and network failures, captures what it saw, repairs the code, and retests.
Sign-in and sensitive steps
The agent refuses to type into password, one-time-code and card fields. It pauses and hands the live browser to you, then resumes with what it can see.
Take over for sign-in, secrets, payments, destructive changes, and other sensitive actions.
Reading a page, and handing it to you
From desktop 1.99.111: the agent reads a page's main text with its links, lists its links, finds text on it, and sees a short outline of what it can click or type into. When a page needs your sign-in, or a paywall you can get past, the agent hands it to you in Preview: Sign in or get past the paywall yourself in Preview, then press Continue. It then reads the page you reached.
What a page may do
From desktop 1.99.111: the agent browser grants a page no permission but fullscreen: no camera, microphone, location, notifications, clipboard or devices, and it never asks you on a page's behalf. Downloads are cancelled: Downloads are off in the agent browser. Approving a site with Always allow covers that site, never every site. Everything the agent reads from a page reaches the model fenced as untrusted, with hidden text removed and lines addressed to AI assistants flagged.
Deep research gives each research agent a browser of its own, apart from this one and from each other.