Where can I monitor the health of integration actions?

What the page shows

Every task your flow steps and the agent run in your integrations lands here: how often it ran, how many failed and why. The store agent's tools (product search, order status…) are counted here too. If a connection breaks or a step stops working, your team is notified.

Open the “Action statistics” tab on the Integrations page. Every member can view it.

The same data also appears in the “What it does” cards on the provider's page and in the agent's Analytics tab.

Parts of the page

• Filters: the date range sits next to the tabs; the default is the last 7 days. The “Integration” filter sits below the alerts and doesn't filter them; “Clear” resets it.
• Summary tiles: “Succeeded”, “Failed”, “Success rate” and “Average duration”.
• Charts: “Runs per day”, “Failed runs per day” and “Per agent” (runs and success rate).
• The “Per action” table: runs and failures for each action, success rate and “Most common error”. Click a row to open the integration's page.
• “Calls”: every run one by one, newest first, filtered by “Status” and “Source” (“Flow step” or “AI Assistant”). Click a row to see the error, the provider's response and HTTP status, the number of attempts if it was retried, a link to the created record and the fields sent to the provider. Email addresses and long numbers are hidden; sent fields and the error text are kept for 7 days, while the run itself stays in the list. Runs stopped by the hourly limit are counted too.

A stored result returned on page reload isn't counted; only real runs are.

Alerts

An alert opens in three cases:

• Connection: opens on the first authorization error. Steps in your agents take their error branch meanwhile; you need to reconnect.
• Provider: opens based on the time window and threshold on the Tools page (by default, 5 failures in 15 minutes).
• Flow step: opens when a step in a flow hits a settings error 3 times in a row. Use “Open agent” to check the step's settings.

Errors like “Required field empty” and limit errors are counted, but they don't open an alert.

Alerts go to the channels linked to the “Integration step failure” notification on the “Alerts and notifications” page; in a new workspace, owners and admins get them by email. When things recover, a “working again” message goes out too.

In the catalog, an integration with an open alert shows “Needs attention” instead of “Connected”. Closed alerts are listed in “Alert history”, right below the open alerts, as “Recovered”, “Reconnected” or “Expired”. “Expired” means there was no new failure for 7 days.

Common errors

• “Authorization rejected” / “Reconnection needed”: remove the connection and connect again.
• “Hourly limit reached”: the workspace's hourly limit was reached. The limit is 20 for deletions and cancellations, and 100 for sends (such as email).
• “Required field empty”: a required field on the step, or the variable that fills it, was empty.
• “Target is a value the visitor typed”: the target of an irreversible action can't be the visitor's answer; choose the target on the step.
• “No channel selected” / “No list selected”: choose the channel or list on the step.