Operations agent: runs, retries and reports

Every time a trigger starts the agent, a run is recorded. Runs survive restarts and updates: every step is saved as it completes, and a run picked up by another server continues from where it stopped.

Open a run#

The agent's Runs page lists every run with its status, the trigger, how long it worked and what it cost. Open a run to see each step on a timeline:

• how long it ran and how many bytes it received and produced;
• the JSON it received and produced, exactly what the next step got;
• when a remote endpoint failed, what the other side answered (the HTTP status and the start of its answer) and every failed attempt;
• the report's answers and the Excel download, when a report step ran;
• the run's Billing block: what it handled and what it cost.

On the board, click a step after a run to see the same data in the inspector.

Step statuses#

• Completed, Running, Queued, Waiting (for the steps before it).
• Waiting for retry: a step failed and will try again at the time shown; the steps after it wait too. The run continues on its own.
• Failed: the step used up its attempts. The steps after it are skipped, unless you drew a failure branch.
• Skipped: the step was on a branch that was not taken, or it had nothing to do, for example no recording on a missed call. A run whose steps had nothing to do still completes.
• Partly failed on the run: some branches completed and some did not.

Retry policy#

Every step has a retry policy (the Retry policy panel in the inspector): how many attempts, from 1 to 24, and how many minutes between them, from 1 to 1440. The defaults:

• steps that call a remote endpoint (webhook, dispatcher, recording download, storage): 24 attempts, one hour apart, until the endpoint answers 2xx;
• transcription, summary and email: 3 attempts, ten minutes apart; report from questions: 2 attempts, ten minutes apart.

A remote endpoint has received the call when it answers 2xx; any other answer, or none, is a failed attempt. A setting only you can fix, such as a missing recipient or an address that is not allowed, fails at once instead of retrying all day. The hours a run waits for a retry are not billed.

In a run, press Retry now on a failed or waiting step: it runs again at once, then everything after it. A webhook may in rare cases be delivered twice; it carries an idempotency key your endpoint can use to ignore the duplicate.

Logs#

The agent's Logs page shows every step of the last 100 runs on one line, newest first, with its attempt and message. Filter it to Failed or Waiting for retry to see at a glance what needs attention; click a line to open its run.

All runs of the project#

Runs in the left menu of the console lists the runs of every operations agent of the project in a period you choose (this month by default), filterable by agent. Above the list: the number of runs, automation minutes, plan minutes used, the list value and what was charged. Export to Excel downloads the period, up to 5,000 runs per file; narrow the dates for more.

The dashboard can show calls, runs or both per day, week or month: choose Operations or All agents in its agent type filter.

When something goes wrong#

• A run did not start: check the agent is enabled and has the right trigger. For a PBX trigger, open Integrations and check that the PBX shows Connected and the call appears in its recent calls; the All requests from the PBX tab shows every request the PBX sent, including the refused ones. For a webhook, use Listen to see whether the request arrives.
• A step keeps retrying: open the run and read What the other side answered. Fix the endpoint, then press Retry now.
• A value arrives empty: open the run, click the step that should have produced it and check the field name in its output; use Listen to see the real trigger data.