Credits and auto-pause for scheduled tasks

Scheduled AI Tool runs and scheduled Joule runs use the organization's shared credit pool. Before a run starts, Ampere checks whether the organization can spend the required credits. Scheduled Tasks can also pause after repeated failures so the same problem does not keep repeating unattended.

Use this article when a task is paused, a run failed or finished partially, or the dashboard says a task needs attention.

What does and does not use credits

Browsing the Scheduled Tasks dashboard, opening task details, filtering, and changing settings do not spend credits.

Credits are used when work executes:

• A scheduled or Run now AI Tool run uses credits for that tool.
• A scheduled or Run now Joule task uses credits for the Joule session and any paid work it performs.

The task and run detail pages show recorded usage after billing settles. A missing or settling value should not be read as a zero-credit run.

What happens before a run starts

Ampere checks subscription access and available credits before starting scheduled work. It can reserve an amount for the run and reconcile the final usage afterward.

If auto-recharge is configured and available, billing can try to add credits first. If the organization still cannot fund the run, the execution does not continue and the task can pause.

New AI work remains unavailable until credits are added, the subscription balance resets, or the billing issue is resolved. Existing history and saved tasks remain available to review.

Understand Joule run outcomes

Scheduled Joule work can end in several ways:

• Succeeded: the run completed and its available outputs can be reviewed.
• Partial: some useful work or output exists, but the run did not finish every expected step.
• Failed: the run could not produce a successful result after its allowed handling.
• Needs attention: the run is waiting for a decision or exact approval that cannot be made automatically.
• Cancelled: the run was stopped before normal completion.

A partial or needs-attention result is not the same as an empty run. Open the run detail and linked Joule session when your access allows it. Review any files or reports that were produced before retrying.

How retries and repeated-failure pause work

Temporary failures can be retried inside the run lifecycle. The task's repeated-failure threshold controls when the task itself pauses after consecutive failed runs.

The creation and edit flows label this setting Pause after repeated failures or Auto-pause after failures. The current default is 5 consecutive failures, and the allowed range is 1 to 20.

A successful start resets consecutive failure tracking. Use a lower threshold when repeated attempts could waste credits or create noise. Use a higher threshold only when temporary source or provider failures are common and safe to retry.

Why a run needs attention

A scheduled run can need attention when it reaches an action that requires review. For eligible connected-account actions, a task can use valid task-scoped standing permission. If that permission is absent, expired, or invalidated by a material task change, the run can preserve the exact pending action for approval.

Destructive actions always require fresh approval. Do not loosen a task's safeguards just to avoid an attention state.

After an exact action is approved, the run can resume from its captured state. If you deny the action, review the preserved draft or partial output and update the task instructions when necessary.

Email expectations

Scheduled tasks can send run-update emails for succeeded, partial, and failed results when recipients are configured. Delivery is not guaranteed: recipient eligibility, preferences, rate limits, suppression, and state rechecks can skip an email.

Email is a convenience, not the source of truth. Use the dashboard and run detail to confirm the actual outcome, auto-pause reason, attention state, and outputs.

Before you resume a paused task

Open the task details and check:

1. The latest run status and visible error or attention message.
2. The organization's usable credit balance and subscription state.
3. The saved AI Tool inputs or Joule prompt, project, attachments, and settings.
4. Any pending approval and whether the exact action is still appropriate.
5. The consecutive failure count, next-run timing, and run limit.
6. The saved email recipients if people rely on run updates.

Fix the cause before selecting Resume. Resuming without correcting missing credits, invalid inputs, unavailable sources, access loss, or an approval requirement can lead to another immediate pause.