Troubleshooting

I cannot access a venture or setting

1. Confirm the active organization in the organization switcher.
2. Confirm that the venture belongs to that organization.
3. Refresh the session by signing out and back in.
4. Ask an Owner or Admin to verify membership and role.

A “not found” response can also mean the page is outside your role or venture scope.

A workflow will not start

Check the launch form, active venture, agent state, selected model, provider credential, repository or project-management reference, and plan limits. Developer workflows additionally require runner infrastructure.

If Organ reports that the spending threshold was reached, review the venture Billing tab before starting more work. Raising a threshold does not change the cost of work already recorded.

For a model error, confirm both halves of the selection. Changing provider clears the previous model. “Inherit” may resolve through the workflow default, venture global setting, system default, and compatible agent model; inspect the run's recorded configuration instead of assuming the current settings are what it used.

A workflow is stuck or failed

Open the workflow detail and identify the last phase and error. Resolve pending input or approval. Rotate an expired credential, correct an invalid external reference, or select an available model before retrying. Use cancellation for work that should no longer proceed.

If a Developer run was interrupted, wait for its recovery state to settle or contact support. Do not repeatedly start duplicate runs for the same task.

Costs are missing or surprising

Usage appears only when the execution harness reports it and the workflow persists it. Compare the date range, venture, workflow, phase, resolved model, retry history, and prompt-cache counters.

Use this sequence:

1. Open the run and locate the highest-cost or failed phase.
2. Open the token popover and compare actual tokens with total tokens; cache reads/writes explain the difference.
3. Check the per-model rows when the phase used multiple models.
4. Look for earlier failed attempts—retrying does not refund consumed model tokens.
5. Confirm whether the run snapshotted an older venture setting or a per-run override.
6. For compute, wait until the run is complete and the hourly attribution view refreshes.
7. Compare the same UTC calendar window and account for trial/plan credits.

A zero compute value can mean the run is still active, does not have attributable runtime detail, or has not reached the next hourly refresh.

Provider and billing invoices remain the commercial source of truth. See Compute, tokens, and model selection for exact formulas and limitations.

A configured spending threshold does not cancel active runs. Use provider-side limits for strict financial enforcement.

Content is not on the public blog

Confirm that the draft is approved and eligible for publication. For an external channel, check its publication record and destination error. Multi-destination publishing may require operator configuration.

An integration stopped working

Check whether the credential expired, lost scope, or was revoked. Replace it in Credentials, reconnect the integration, and retry the smallest safe operation. Never paste secrets into task text, logs, content, or resource metadata.

Search returns no support result

Try a product noun such as “agent,” “workflow,” “credential,” or “publishing.” Search indexes this knowledge base, not venture records. If search itself fails, refresh the page and contact support with the article URL and time.