Where Hotato fits next to your QA platform.
Hotato is not your full QA platform. For most teams it sits beside a broad QA tool and a runtime interruption layer. If another tool is the better fit for what you need, use it, by name. This page says which is which.
Two jobs. Route by name.
Hotato is not your full QA platform. Here is exactly what each side covers.
Use a broad QA platform
Hamming, Cekura, Coval, Bluejay, Roark, Vapi, Retell, or a similar platform.
Broad session QA, synthetic simulation, task success, transcript rubrics, compliance workflows, production dashboards, load testing.
Use Hotato
Labelled fixtures from real calls: local, private timing evidence, CI-enforced regression tests via hotato verify, turn-taking proof, and refusing unsafe threshold bandaids.
Hotato answers: "Did this exact production timing failure come back?"
Hotato does not answer: "Was the whole call successful?"
Three questions, three tools.
The cards above cover two. Here is the third, and the question that routes to each.
| What you need | Best fit | Why |
|---|---|---|
| Prevent an interruption problem in the moment, at runtime. Predict endpointing, suppress barge-in on noise, tune the live turn detector. | Krisp, Pipecat, LiveKit and similar runtime layers |
They act during the live call. Hotato never runs at runtime and never touches a live call. |
If your open question is "is my agent good?", start with a QA platform. If it is "is my agent interrupting well right now?", start with a runtime layer. If it is "did the talk-over bug we fixed last month come back in this release?", that is Hotato.
Narrow on purpose.
Private
Scoring, scanning, reports, fixtures, and verification run offline. Audio stays on your machine unless you explicitly pull it from your own stack. Nothing is uploaded to Attention Labs.
Deterministic
No learned score, no sampling. The same recording produces the same timing numbers every run, so a red build means the audio changed.
Regression-shaped
A confirmed bug becomes a fixture that gates CI. Verify proves a fix across the whole battery and reports coincidence, not causation.
Three signals only
Talking over the caller, false-stopping on a backchannel, yielding too slowly. It does not grade content, intent, or outcomes.
When an example shows a call failing on a named stack, it is a provider-default run: one assistant, one config, one date, one scripted caller, on that provider's out-of-the-box interruption settings. It demonstrates the threshold funnel, not a vendor benchmark and not a ranking. Any stack, tuned, can pass the same fixtures. Hotato never publishes a scoreboard.
Two different jobs, often confused.
A runtime layer improves the median live call. A regression layer stops a good fix from quietly rotting three releases later. You want both.
Hotato takes the moment a runtime layer got wrong, freezes it as a fixture with hotato fixture promote, and fails CI if it regresses. hotato verify proves a fix across the whole battery and reports coincidence, not causation.