Skip to content
QuickStatus.dev

Monitor the voice APIs your agent depends on

A voice agent is a chain: speech-to-text, a language model, text-to-speech, and usually a telephony or voice platform in front. When one link is down, the call fails. When one link is slow, the agent goes quiet and the caller hangs up, while every provider's status page still says operational.

QuickStatus.dev sends a small real request to each provider with your own key, from up to five regions, and alerts you when one fails or gets slower than your agent can wait. Free for 10 checks.

Start free, no card Read the check docs

For voice, slow is an outage too

A web page that takes three seconds is annoying. A voice agent that takes three seconds to answer sounds broken. That is why a voice stack needs an alert for slow, not only for down.

  • Down. Timeouts, 5xx, 429s under normal traffic, a 401 after a key rotation, or a 200 with an error in the body.
  • Slow. The API still answers 200, but later than your agent can wait. Callers hear silence.
  • Partly down. One region or one model fails while the rest work. A broad “partial outage” banner on the provider's page is easy to miss.

What to check in a voice stack

StageExamplesWhat to send
Speech-to-textDeepgram, AssemblyAI, OpenAIA transcription request for a one-second clip you host
Language modelOpenAI, Anthropic, GroqThe smallest chat request your agent sends. See Monitor the OpenAI API
Text-to-speechElevenLabs, Cartesia, OpenAIA synthesis request for one word, such as OK
Your own endpointsTool and webhook URLs your voice platform callsThe method and body the platform sends

The last row is the one teams forget. Platforms such as Vapi and Retell call your server for tools and call events. If that URL is down, the call breaks even though every provider is fine.

Set up a text-to-speech canary

  1. In the console, open Checks, click Add check, and pick Website or API.
  2. URL: the synthesis endpoint, for example https://api.elevenlabs.io/v1/text-to-speech/<voice id>. Set Up when to Status is 200 OK.
  3. Under HTTP request, set the method to POST. Under Request headers, add your key in the header the provider expects (xi-api-key for ElevenLabs) and Content-Type: application/json.
  4. In HTTP body, send one word:
    {"text": "OK", "model_id": "<the model you ship>"}
    The response is audio, so add no body assertion. The status code is enough.
  5. Under Slow responses, set Slow above (ms) to the latency your agent cannot live past, with Measured on set to Total time. With a one-word request, total time comes close to time to first audio.
  6. Under When to alert, set Mark down after to 2 and Request timeout to 5 or 10 seconds.
  7. Pick regions near your callers. Free checks from 1 region; Pro can use all five: San Francisco, Ashburn, Montreal, Frankfurt, and Singapore.
  8. Under Status page, link the check to a “Voice responses” component.

Save, then press Run now once to confirm the request returns 200 before you rely on it.

Speech-to-text works the same way

Point a POST check at the transcription endpoint and pass a short clip you host somewhere fast. For Deepgram that is https://api.deepgram.com/v1/listen with Authorization: Token <key> and the body {"url": "https://your-cdn.example/hello.wav"}. Add a Body contains assertion for transcript, so an empty result counts as a failure.

What it costs

You do not need to synthesize audio every minute to know the API is reachable. Point a frequent GET check at a free authenticated endpoint, such as the provider's voices or models list, and run the canary every 5 or 15 minutes. OK every 5 minutes is about 17,000 characters a month per region. Use a separate key with a spending cap, not a low rate limit, which can cause the same 429s you are watching for.

What happens when it gets slow

  • Slow, not down. A run that works but crosses Slow above marks the check Slow. You get an alert, the linked component shows Degraded performance, and uptime is not affected. You get another alert when it is fast again.
  • Runs in a row, not one bad run. Slow and down both use Mark down after, so one slow request is stored and nothing is sent.
  • Per region. Every run stores its duration and where it ran from, so you can tell a provider slowdown from routing trouble near one city.
  • Your customers hear it from you. When the canary is confirmed down, the linked component goes down and status page subscribers get the update. Alerts go by email, or on Pro also by Slack, Discord, PagerDuty, Pushover, or webhook.

What it does not do

  • It does not listen to calls or grade conversations. Use a voice QA tool for that. QuickStatus.dev watches the APIs underneath.
  • It checks over HTTPS. It does not open realtime WebSocket sessions, so for streaming speech the REST endpoint of the same provider is the closest signal.
  • It measures total time or time to first byte, not time to first audio on a stream. A one-word request keeps total time close to it.

The longer version, with what to alert on and why: TTS latency monitoring: slow or down?

Pricing

  • Free: 10 checks, 5-minute interval, 1 region, 14 days of results, 1 status page, email alerts
  • Pro, $15/month: 25 checks, 60-second interval, 5 regions, 90 days of results, and Slack, Discord, PagerDuty, Pushover, and webhook alerts

See pricing

Voice API monitoring FAQ

Does this listen to my calls?

No. QuickStatus.dev does not record or grade conversations. It sends its own small requests to the APIs your agent uses and tells you when they fail or slow down.

What should I set Slow above to?

Run the check for a week from the regions your callers are in and look at the durations. Then set Slow above a little past the slowest runs you can accept.

Will one slow request page me?

No. Slow uses the same Mark down after count as down, 3 runs in a row by default. You can set it to 1, 2, 3, or 5.

Can I monitor Vapi or Retell?

You can check their public API like any other HTTPS endpoint. The more useful check is usually your own server URL that the platform calls for tools and call events, because when it fails the call breaks while every provider looks fine.