Inbox
The Inbox is one place for everything running, everything that needs your decision and everything that just finished, across every chat and workflow run you have; on a governed deployment it also shows approvals raised by API-started runs, for everyone assigned. There is nothing to set up.
Steps
- Click Inbox in the dock. The badge counts items running or waiting for you (1 to 99, then "99+"); finished items are not counted.
- Choose All, Needs you, Running or Done; each carries a count ("N+" on Done while more history exists). Finished runs are Done; approvals and publish reviews are Needs you; everything else is Running; All shows running and needs-you cards together. A run that raised several cards shows only its latest under Running and one card per held step under Needs you.
- Act on a card: approve or send back, stop the run, or Open the conversation or run.
What you should see
The header reads "Everything that needs you, is running, or just finished."; on a governed deployment it adds "Approvals from API-triggered runs appear here for everyone assigned."
| Card | Shows | Actions |
|---|---|---|
| Running | The conversation title (or "Untitled chat"), a status chip, a progress line ("n/N nodes done" with the current step's label, or "Step: n/N" for older workflow runs) and a segmented bar | Stop asks "Stop this run?" (chat) or "Remove this activity?" (workflow) with Keep and Confirm, then reads "Stopping…" until the card moves to Done; Open |
| Approval | The step's handle and title; a badge Awaiting review, Awaiting go-ahead or Wants to run a command ("Needs you" on older runs); the same ask as on the board (the objective, the files with Open in Preview, or the command); "You sent back" with your note if you returned work; "What each action does" beneath the buttons | The approve action, Send back (the note is required) and, here only, Stop; Open jumps to the step's card. On a governed deployment an API run badge (with the key id when known): "Raised by an API-triggered run. Everyone assigned to this deployment sees it; whoever acts is recorded on the run." |
| Publish | "AMPG" (Augmented Multi Resolution Property Graph), the graph's name and "vN is ready"; "A workflow run built version N of" the graph "from vM (rewritten with the run's learnings)."; badge Publish; how many agents "will be re-pinned on adopt" | The run's buttons; what adopting does is on Knowledge graphs: use, move agents, delete |
| Done | Badge Completed, Failed or Cancelled; "Finished" with the local date and time (or "Workflow run" / "Chat run" when none was recorded) | The whole card opens the run (click, Enter or Space); chat and workflow history merge newest first, 20 per page, with Show more (N loaded) |
Empty filters read "Nothing here yet" ("Running agents and workflows will show up here the moment they start."), "Nothing needs you" ("Approvals and publish reviews will show up here the moment a run needs your decision.") or "Nothing finished yet" ("Completed, failed, and cancelled chat and workflow runs will show up here.").
Notes
- The Inbox, the dock badge and the banners update over a live connection from the deployment; if it drops, Studio re-reads the Inbox every 30 seconds until it returns.
- Cards have no read or unread state; a card leaves when its run moves on or you stop it. The Inbox shows your own runs plus, on a governed deployment, approvals from API-started runs; never other users' chats.
- "Completed workflow-run history failed to load:" with a reason: Done could not fetch workflow history; chat history still shows. "Action failed" under a card's buttons: the decision did not reach the run; if it was decided elsewhere the card shows its current state, otherwise retry. A card you acted on is still there: it refreshes within seconds.