Skip to main content
Inbox is the user action queue. Where Home tells you what is happening, Inbox is where you work through what needs a response.

​
What Inbox collects

Inbox aggregates the action categories that need attention:
  • Approvals — governed actions waiting on you
  • Failed runs — runs that ended in a failure or timeout state
  • Alerts — system or agent-generated alerts
  • Join requests — requests from agents or users to access Studio
Without Inbox, you would need to check each underlying page separately. Inbox surfaces everything that needs attention in one queue.

​
Tabs

Start with Pending when checking in after agents have been running. It shows only what still needs action.

​
Category filters

Category filtering is only available on the All tab. The dropdown lets you narrow to a specific category:
  • All categories — all events
  • Join requests — incoming access requests
  • Approvals — approval requests from agents
  • Failed runs — execution failures
  • Alerts — system and agent alerts

​
Working with failed runs

Inbox gives you the most context for failed runs outside of the full run detail page. For each failed run, Inbox can show:
  • the related issue, if one exists
  • the agent that ran it
  • the failure state
  • the failure message
  • a Retry action to requeue the run
  • an Open run link to the full run detail
Use Retry for transient failures. Open the run detail when the failure message is unclear or the issue may need to be changed before retrying.

​
Quick navigation

From any item in Inbox, you can jump directly to:
  • an approval → opens the approval detail page
  • an issue → opens the issue detail page
  • a run → opens the run detail page
This makes Inbox a fast triage surface: scan the queue, open items that need more context, act or dismiss. For full approval context, open the approval detail from Inbox. When checking in after agents have been running, work through Inbox in this order:
  1. Clear pending approvals — agents cannot continue until these are resolved
  2. Review failed runs — retry or investigate
  3. Review alerts — system-level signals that may indicate deeper issues
  4. Review join requests — approve only agents or users that should have access