Common problems
Start with the usual question: at which moment did it stop? Building, publishing, triggering, and running fail for different reasons.
While building
| Symptom | Cause | What to do |
|---|---|---|
| I cannot save: conflict | Someone else changed the draft | Reload and reapply your change |
| A value does not appear in the picker | It comes from a step that runs later | Reorder the flow |
| "The item" does not appear | You are outside the "For each item" step | Those sources only exist inside the loop |
| A tool appears as not allowed | The space has not allowed it | See Allowed resources |
| A tool appears as "not confirmed" | It could not be checked right now | Use the check-again button |
While publishing
Publishing validates everything before creating the version. The most common refusals:
| Message | Meaning | What to do |
|---|---|---|
| Unreachable step | It is not reached from the start | Connect it to the flow |
| Step with no exit | Only the Result may have no exit | Connect it to the next step |
| Cycle between steps | The flow loops back on itself | Redo that section |
| No Result | Every agent needs at least one | Add the Result step |
| Reference to a later step | A value is used before it exists | Reorder, or use a different value |
| Address not allowed | The host is not on the space's list | Ask for it to be allowed |
| A value that looks like a credential | A key written straight into a header | Use a secret reference |
| Limit above the ceiling | The agent declares more than the space allows | Reduce it, or ask for the ceiling to be reviewed |
| No service account | AI steps require a service account | See Service account |
| Version already used | Each number is unique within the agent | Choose the next one |
While triggering
| Symptom | Cause | What to do |
|---|---|---|
| The schedule does not fire | The version it points to is no longer active | Check the environment's active version |
| The schedule stopped after a change | The input data no longer fits | Fix the schedule's data |
| The HTTP call returns an invalid credential | A wrong credential, or a nonexistent identifier | Check both |
| The HTTP call refuses the input | The data does not match what the agent declares | Adjust the body |
| The HTTP call returns limit reached | Rate, concurrency, or size | See HTTP addresses |
| The address expired | development addresses are temporary | Create another, or make it permanent in production |
While running
| Symptom | Likely cause | What to do |
|---|---|---|
| Immediate failure | No active version in the environment, or an inactive service account | See Publishing |
| An AI step fails because the output does not match the format | A declared field did not come back | Reduce the fields or make the objective more explicit |
| An AI step fails because it asked for information | The objective depends on something not handed over | Complete the input data |
| An external call is blocked | The destination resolved to an internal address | Confirm the system's public address |
| The run expires | The total deadline ran out | Raise the timeout or reduce the work |
| The run looks stuck | It is waiting for an approval, a wait, or a retry | Check the current step in the timeline |
| A child run failed | The problem is in the called agent | Open the child run |
When the result "worked" but does not look right
Two causes, in this order:
- An error was handled by a diversion. The result declares the steps with handled failures. Look for that list before anything else.
- The AI step had nothing to verify. With no declared fields and no required tool, nothing allows stating that it met its objective. See AI task.
When to ask for help
Have these at hand:
- the agent and the version;
- the run's identifier;
- which step it stopped at and the error code.
With that, whoever helps sees the same timeline you do.
Next steps
Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.