Skip to main content

Why this is a common failure point

Voice agents fail silently. A rotated API key, an expired phone number, a provider outage or a bad deploy can leave the agent answering calls with silence, or not answering at all, and nobody finds out until a customer complains. Dashboards of past calls won’t warn you: if calls stop connecting, there is nothing new to look at.

How Roark tests it

A health check calls your agent on a schedule with Roark’s fixed probe flow and checks the basics: it answered, it spoke, and it responded quickly. Because the probe never changes, every check is comparable with the last, and a failure alerts you before your callers notice.

Setting it up

The probe call is fixed, so there are no flows to pick. Choose the agents and the cadence (hourly by default). Health checks have their own home under Health checks, which links straight to this template.

What it measures

The template seeds these, and you can add or remove metrics in Advanced before running.

What to look for

  • Alert on failure. Set an alert on the plan so a failed probe reaches you, not just the dashboard.
  • Watch time to first word drift. A slow climb over days often comes before an outage.

Over the API

Run it with POST /v1/simulation/run and "template": "health-check". GET /v1/simulation/template lists what each template accepts.