A regular agent reply is text — a plan, a draft, an answer. Orvoq Act goes further: it lets an agent actually operate a browser or a compute terminal to get something done, not just describe how.
Each of these runs in its own isolated sandbox, destroyed once the execution finishes — never on your own machine, and never shared between executions.
While it's running, the transcript shows a live card with a concrete description of what's happening right now — which page it's on, how many files it's processed — not a generic “working…” spinner. Click into it for the full browser live view or compute terminal log.
Was this article helpful?