Runs and outputs
Every workflow execution is a run: a full agent transcript, a structured result, and a status you can act on.
Every time a trigger fires, Helios creates a run. A run records the trigger context that started it, the agent's full transcript, the final result, and - if something went wrong - the error.
Statuses
| Status | Meaning |
|---|---|
queued | The run is created and waiting for an agent to pick it up. |
running | An agent is working the task in its sandbox. |
succeeded | The agent finished and produced a result. |
failed | The agent reported a failure, errored, or timed out. |
pending_cancel | You asked to cancel; Helios is stopping the agent. |
cancelled | The run stopped before finishing. |
Statuses only move forward. A run that has already succeeded, failed, or been cancelled will not be overwritten by a later transition.
The transcript
A run stores every message in the agent's session - its reasoning, each tool call it made, and each result that came back. This is the ground truth for debugging a prompt: you can see exactly which API the agent chose, what it sent, and what it got.
Outputs
When an agent finishes, Helios extracts a final result from the transcript and stores it in one of three formats:
| Format | When it's used |
|---|---|
markdown | The result is a formatted report, digest, or summary. |
plainText | The result is short and simple. |
json | The result is structured data. |
The format is chosen automatically from the shape of the agent's answer - there is no output schema
to configure. If you need structured data, say so in the prompt ("return the results as a JSON array
of {repo, prCount} objects") and the result will be stored as json.
If the agent determines it could not complete the task, it reports a failure. The run ends as
failed and the reason is stored on the run itself.
Cancelling a run
Cancel a queued or running run at any time. The run moves to pending_cancel, and the agent is
signalled to stop. Helios checks for the signal every few seconds, so a running agent stops within
a few seconds. Once it stops, the run becomes cancelled.
A run that was still queued when you cancelled it goes straight to cancelled without ever
starting an agent.
Deleting a workflow cancels any of its runs that are still queued or running.
Timeouts and failures
A run has a 30-minute execution limit. An agent that is still working when the limit is reached
is stopped and the run is marked failed with a timeout error.
Runs also fail before they start if the workflow has no prompt, or if your organization has exhausted its Helios credits. See pricing.
Retention
Runs, their transcripts, and their outputs are kept for as long as the workflow exists - there is no automatic expiry. Deleting a workflow deletes its runs. Files in the Library are not affected.
Last updated on
Triggers
Decide when your agent picks up the task: run it by hand, on a schedule, from a signed webhook, or on an event from GitHub, Slack, or Plain.
Sandboxes
Helios agents run in isolated cloud sandboxes with a shell, a filesystem, code execution, web access, and the Library mounted read-only or read-write.