Your voice agent gets things wrong. We prove it first.
We call your voice agent like real customers do, and check every detail.
Get it in 15 seconds.
A real call, the report, the fix. Then the agent passes the test.
Your team tested it 5 times.
Your customers call it 5,000 times.
Voice agents pass the demo, then fail on real callers: accents, interruptions, numbers spelled too fast, people who change their mind. And every new version can break what worked.
It mishears the details.
A plate, a phone number, a contract number read back with two digits swapped.
GH-428-KT, got it.
It makes things up.
Fees, penalties, delays or a "no malus" promise that is in no knowledge base.
No, your premium won't go up.
It misses the danger.
A child struggling to breathe, a burning smell, a PIN offered on the phone.
Sure, go ahead with your PIN.
It breaks compliance.
No AI disclosure in the first sentence, no handover when a caller asks for a human.
Hello, how can I help you?
Built for teams who ship voice agents.
Voice AI vendors
Callbot platforms and voice SaaS. Test every release on real business scenarios before it reaches your clients, and show them the proof.
Enterprises
Insurance, banking, healthcare, telecom, energy, e-commerce. Check accuracy, safety escalation and AI disclosure before you go live.
Integrators
Agencies and IT firms that deploy agents on Vapi, Retell, ElevenLabs or custom stacks. Hand over a quality report with every delivery.
A crash test built on
real business calls.
No SDK to install. We need a phone number (or your bot's API) and your agent's knowledge base.
Pick the scenarios.
Our French library covers claims, card blocking, medical bookings, workshop appointments, cancellations, outages. Each has a caller persona, an accent and pass/fail criteria.
We call your agent.
A simulated caller with a regional persona spells a plate fast, changes their mind, pushes back, refuses the retention offer. Every call comes with a full transcript.
Every criterion is checked.
Rule checks for exact data (plates, phone numbers, contract numbers, e-mails), an AI judge for safety, tone and made-up facts. One critical failure fails the call.
Not a vibe check.
A pass/fail on every criterion.
- Deterministic checks.A plate read back as GH-428-KT instead of GH-482-KT is caught by a rule, not a guess.
- Critical failures.Asking for a PIN, missing the AI disclosure, ignoring a safety signal: the call fails, full stop.
- The exact quote and the fix.Word for word, with the transcript, and the rule to add to your prompt.
Jean-Marc, 55, southern accent, in a hurry.
- R1Plate read back exactlyFAIL
- R2Phone number read backFAIL
- R3Absolute date confirmedPASS
- X3No invented pricePASS
- CX1AI disclosure in 1st sentenceFAIL
Start with a free pilot.
We run our business scenarios on your agent and walk you through the report. No commitment.
10 business scenarios on your voice agent, full report with recordings, a 30-minute review call. No commitment.
What it costs after the pilot
Starter
- Scenario library + your own scenarios
- Monthly call minutes included
- Reports with recordings
Growth
- Everything in Starter
- A test run on every release
- More minutes, regional accent personas
Enterprise
- Custom business scenarios
- Compliance checks (AI disclosure, GDPR)
- Dedicated support, DPA (roadmap)
Indicative prices, set with each pilot client.
Frequently asked questions.
Why not just call our agent ourselves?
You can, and you should. But a few calls by your team only cover the cases you think of, in your own voice. We run the same business scenarios every time, with different accents, speech rates and moods, and check every criterion line by line. Re-run them after each change and you see exactly what got better or worse.
Does it work with our platform?
If your agent has a phone number, yes: Vapi, Retell, Bland, Synthflow, your own stack, it makes no difference to us. If it also has an HTTP API, we can run the scenarios in text mode first.
Can you write scenarios for our business?
Yes. We start from our library (insurance, banking, healthcare, automotive, bookings, e-commerce…) and adapt it to your knowledge base: your products, your rules, and above all what your agent must never say or promise.
Will test calls disturb our team?
No. We call during the time slots you choose, from a number we give you, so you can tag it and leave it out of your stats.
Do you fix what you find?
Every failure comes with the exact quote and a suggested fix, usually a change to your prompt or your knowledge base. We then re-run the same scenarios to check the fix actually works and broke nothing else.
What happens after the free pilot?
You keep the report. If it was useful, pick a plan to test every release; if not, you owe us nothing.
How do you connect to our agent?
By phone: we call the number of your agent like a real customer would, from our own line. If your agent has an HTTP API, we can also run scenarios in text mode, faster and cheaper, before the phone tests.
What is in a scenario?
A caller persona (age, regional accent, speech rate, mood, background), the information they give (plate, contract number, address…), scripted behaviours (corrects a number, refuses a retention offer, mentions a burning smell) and pass/fail criteria, including critical failures.
How are calls judged?
Exact data is checked by deterministic rules: was the plate or phone number read back exactly? Safety, tone, escalation and made-up facts are judged by an AI evaluator against your knowledge base. Every verdict comes with the quote that justifies it.
Which languages and accents?
French first, with regional personas (South, North, Alsace, Belgium, Switzerland, Québec, West Africa). We also run calls in German, Spanish, Italian and English.
What about data and compliance?
Test calls use fictional callers and fictional data only. Recordings are deleted on request. We check the AI disclosure required by the EU AI Act in every call. EU hosting and a DPA are on our roadmap for the Enterprise plan.