Skip to main content

Agent runs

Agent Runs are background AI work started by TOW when an operation needs more processing time than a normal page action should take.

There is no separate end-user Agent Runs page in the main navigation. Instead, TOW shows Agent run status in the sidebar and on the items being processed.

Maintainers can read Background AI system for the architecture, model boundaries, and operational controls behind this status.

Where to see status

LocationWhat it shows
Sidebar footerOverall active Agent runs for your user: idle, running, or queued.
Item badgeAgent running on a specific snapshot, daily session, roadmap generation, or related target.
Page refreshSome pages refresh or invalidate data when active runs complete.

The sidebar text can show:

  • Agents idle when no active runs are visible.
  • Agents: N running, M queued when work is active.

What starts Agent Runs

Agent Runs can be started by:

  • Submitting a project Check-In.
  • Generating or regenerating a roadmap.
  • Generating a snapshot.
  • Reviewing chat turns for memory diffs.
  • Reviewing docs for memory changes.
  • Running manual or nightly ticket conflict scans.
  • Running background operating reviews and pulses.

The exact output depends on the source action. A run may produce a summary, a memory diff, an Inbox item, ticket automation, a refreshed pulse, or a generated snapshot.

Disabled Agents

Agent Runs are not created when their effective Agent is disabled. Direct AI-only actions show a clear disabled state or return an error naming the disabling scope. Mixed actions can still save non-AI data, then skip enrichment.

Admins can review or change policy in Agent configuration.

Status meanings

StatusMeaning
QueuedTOW accepted the work and is waiting for a worker to process it.
RunningA worker is currently processing the run.
CompletedThe run finished and remains in the audit table.
FailedThe run failed. Recent failures may appear in status polling.

While a run is active

  1. Keep working if the next task does not depend on the result.
  2. Avoid submitting the same expensive action repeatedly.
  3. Wait for Agent status to return to idle.
  4. Refresh the page or revisit the affected item.
  5. Open Inbox or Memory Diffs if you expected review items.

Review output

Agent Runs often produce proposals rather than direct changes. Look for output in:

  • Inbox.
  • Memory Diffs.
  • Today.
  • The related Check-In session.
  • Snapshots.
  • Tickets or Roadmap.
caution

AI review: A completed run does not mean its recommendations are approved. Review memory diffs, proposals, snapshots, and pending actions before treating them as accepted workspace state.