Tool call statistics and failure alerts
Monitor how well your tools work, inspect failed calls and set alert thresholds.
Where is it?
At the top of the “Tools” page there are three tabs: “Tools”, “Call statistics” and “Logging and alert settings”. All members can see the statistics; changing the settings requires the owner or admin role.
Call statistics
Choose a date range at the top (today, yesterday, the last 7, 30 or 90 days, all time, or a custom range). The default is the last 7 days.
A call counts as succeeded when your tool returns a 2xx response, even if the response itself says there was an error. If retry on failure is on, all attempts count as one call.
- Summary: number of succeeded and failed calls, success rate and average duration.
- Charts: calls per day, failed calls per day and a per-tool breakdown. Test calls aren't counted in the charts, but they do appear in the call log.
- Open alerts: tools that keep failing are listed in a single box at the top of the page, one per line; the link on each line takes you to that tool's failed calls. The history of closed alerts is at the bottom of the page.
- Call log: filter by tool, status and source (agent, choice step, tool call step, test). Clicking a row shows the error details, the HTTP status, duration, response size, the tool's error text, the fields sent (email addresses and long numbers are hidden), and the agent and session. The body of a successful response isn't stored.
The store agent's tools (product search, order status…) aren't counted here; you'll find them under Integrations › “Action statistics”.
Logging and alert settings
- “Retention period”: how many days detailed call records are kept (1–30 days, default 7). Counts are permanent; only the details are deleted when the period ends.
- “Time window” and “Threshold”: an alert opens when at least the threshold number of calls (1–1000, default 5) fail within the time window (5–1440 minutes, default 15).
When a call succeeds and the failure count drops below the threshold, the alert closes by itself and a “working again” notification goes out. A tool with an open alert shows the “Failing” badge in the Tools list.
The time window and threshold are also used for the provider failure alert in integrations; integration alerts go out with the “Integration step failure” notification. Choose which channel tool alerts go to (Slack, Teams, Telegram, email) with the “Tool failure” notification on the “Alerts and notifications” page.
Related articles
- Tools: connecting the agent to your own systemsAdd your own HTTPS endpoints that the agent calls during a conversation to fetch live data or take an action.
- Alerts and notifications: which event goes to which channel?Routing new request, credit and error alerts to Slack, Teams, Telegram or email, and tracking deliveries.
- Where can I monitor the health of integration actions?The “Action statistics” page shows every task your flow steps and the agent run in integrations, the failures and why.