Understanding agent approval requests

Agents in Orvoq can do real work on their own — but anything with a consequence outside the conversation waits for a person first. When that happens, you'll see an approval request like this:

Why this happens

Every agent has a list of tools it's allowed to use — searching the web, reading a document, sending a message, and so on. Tools with an effect outside the session (sending an email, publishing something, spending money) are marked as needing approval by default. You can see exactly which tools an agent has, and which ones need your sign-off, on its profile under Workspace → Agents.

Who can respond

An approval request is visible to everyone watching that session or Run — not just whoever happened to trigger it. Anyone with sufficient permission can approve, reject, or act on it. Whichever response comes first is the one that counts; if two people respond at nearly the same time, the second response is simply discarded rather than causing an error.

Your options

• Approve — the agent proceeds exactly as proposed
• Reject — the agent stops; it won't silently retry the same action
• Modify instructions — tell the agent to do something slightly different before continuing, without losing its progress so far
• Override output — edit what the agent produced directly, rather than sending it back to try again

Where to see everything waiting on you

Rather than checking each session individually, open Runtime Consolefrom the left nav — it shows every active agent across every session in the workspace, with a filter for “Needs your approval” so nothing gets missed.