Managing Your Pipeline
Use the Agent to inspect connected apps, sources, integrations, and jobs, then make permission-scoped operational changes safely.
Last reviewed July 13, 2026
The Agent can inspect pipeline resources and, when your role and access mode allow it, trigger supported operational actions. Start with current state and the first error before asking it to retry or change anything.
Inspect current state
Ask for the resource and time window you need:
List sources that have failed in the last 24 hours.Show the latest job for the Shopify source and its first error.Which integrations are paused?Does the Mixpanel app need new credentials?Show currently running jobs and how long they have been running.
The Agent can correlate connected apps, sources, destinations, integrations, and job records when the relationships are available to your account.
Make an operational change
Write actions can include supported operations such as triggering a sync, pausing or resuming a resource, updating a configuration, or cancelling a job. Availability depends on your permission and the tool surface.
Before approving or relying on a change, confirm:
- The active team workspace.
- The exact resource name and ID.
- Whether the action affects a source, destination, integration, or job.
- The time range and data volume for a sync or backfill.
- Whether a retry can duplicate partially delivered data.
Safe retry workflow
- Read the first failure and identify the root cause.
- Fix credentials, permissions, schema, or configuration first.
- Retry a small range when the connector supports it.
- Watch the new job in Activity Logs.
- Compare accepted and rejected row counts.
- Expand the range only after the small run succeeds.
Examples
Find the latest failed Google Ads source job, explain the first error, and tell me what must be fixed. Do not trigger a retry.
Show the source ID and selected time window, then trigger a seven-day sync only after I confirm.
List paused integrations and distinguish manual pauses from repeated-failure pauses.
Troubleshooting
| Symptom | Check |
|---|---|
| Write tool is unavailable | Role, feature status, and read-only access mode |
| Resource name is ambiguous | Ask for IDs and connected app names before acting |
| Retry fails immediately | Credentials, permissions, required configuration, and provider status |
| Job runs but sends no rows | Source filters, date window, model output, and destination identity requirements |
| Counts change after retry | Partial delivery and destination deduplication behavior |