Your AI agent took 400 calls today. Did it stick to the brief?

AI voice agent QA means checking what your agent said against what you told it: its prompt, your price list and your rules. ListenFox reads every call your agent handles, points to the line where it quoted the wrong price or broke one of your rules, and marks the calls where a booking slipped away.

Start free trial

First 100 calls free. No card needed.

See an example call

A few minutes

Connect the platform your agent runs on.

Retell AI, Vapi and Bland AI connect with your API key, and each call is sent over by webhook as it ends. Built on ElevenLabs, Synthflow or your own stack? Connect Twilio or Plivo if your calls are recorded there, upload the recordings, or pick a Google Drive folder.

An inbound webhook, so any platform can send calls straight in, is coming soon.

All integrations

Connect a source

  • Retell AI
  • Vapi
  • Bland AI
  • Twilio
  • Plivo
  • Upload
  • Google Drive

Once

Give it the brief you gave your agent.

Upload your agent’s prompt, price list and policies as PDF, Word, text or Markdown files, or give it your website. ListenFox drafts a one-page facts sheet. You edit it and confirm it, and nothing is checked against facts you haven’t approved.

Mark the prompt as a playbook too. Then anything the agent says that breaks one of its rules, such as offering a discount you never give, is flagged as well.

Facts sheetConfirmed

Prices
Diagnostic visit $89
Hours
Mon to Sat, 8 am to 6 pm. Closed Sundays
We don’t offer
Duct cleaning

Playbook: agent prompt v4

Never waive the diagnostic fee. Transfer to a person when a caller smells gas.

Example for a sample heating and cooling company

On every call

Every call gets checked.

Each recording is transcribed again with the speakers separated, so the check doesn’t rely on your agent’s own record of the call. Then it’s read for the outcome, what the caller wanted and what they asked for that the agent couldn’t offer.

Anything the agent said that contradicts your facts or your playbook is flagged with the quote, the moment it was said and the fact it broke.

Your teamCaller

Contradicts your facts sheet 0:11

Said “$69 to come out”; facts say the diagnostic visit is $89.

Example flag

What it catches.

Each finding comes from the call itself, so you can play the moment and hear it for yourself.

Wrong prices

A price outside your list for the same item. “Quoted $69; price list says $89.”

Broken rules

Refunds, deposits, cancellations, warranties, or any rule in your playbook, stated differently on the call.

Hours and availability

A Sunday booking when you’re closed on Sundays, or a same-day visit you don’t offer.

Services you don’t sell

Saying yes to a job that’s listed as not offered, or turning down one you do offer.

Bookings that slipped

Real prospects who hung up with no next step are flagged as a missed booking. What they asked for and couldn’t get is listed too.

Your own questions

Add yes-or-no questions in plain words, such as “Did the agent confirm the address?” Every call gets an answer.

One call, checked against the brief.

An AI receptionist for a sample heating and cooling company, checked against the facts sheet and playbook above. Highlighted words are what got flagged.

Example call

Caller 0:05

Hi, my AC stopped cooling overnight. What do you charge to come out?

Ava, AI agent 0:11

It’s $69 to come out, and we waive that if you go ahead with the repair.

Caller 0:24

Okay. Could someone come this Sunday?

Ava, AI agent 0:29

Sunday works. I’ve booked a technician for 10 am.

Caller 0:41

Great, thanks.

Your people and your agents, held to the same facts.

Most businesses run both: the AI picks up first and people take the harder calls. Both are checked against the same facts sheet and playbook, so you can see which side drifts.

Example flags
KindWhat was saidAgentWhen
Price“It’s $69 to come out.”Ava (AI)Mon, Oct 5, 0:11
Hours“Sunday works.”Ava (AI)Mon, Oct 5, 0:29
Policy“You get the deposit back if you cancel on the day.”MarcusSun, Oct 4, 2:47
Service“Yes, we clean ducts too.”Ava (AI)Sat, Oct 3, 1:05
  • The policy check, outcomes, missed bookings and your own questions run on every call, whoever answered it.
  • Retell sends each call with the agent’s name, so its calls sit next to your team’s and you can filter by agent. Vapi and Bland calls arrive without a name for now.
  • A trigger can post each flag to Slack, Telegram, email or your own webhook, with the quote and the moment.

Coming soon for AI agents.

Not available yet: a brief score for each agent tracked week by week, calls before and after a prompt change side by side, and handling scores and transfer checks for AI agents. Each will be marked live on the changelog.

How does AI voice agent QA work?

Check what it said against what you told it. ListenFox transcribes every call your Retell AI, Vapi or Bland AI agent takes, compares it with the prices, hours and rules you gave the agent, and flags each slip with the quote and the moment, plus calls where a booking slipped away.

Questions about AI agents

Does it use the transcript from my voice platform?

No. ListenFox transcribes the recording itself, with the speakers separated, so the check doesn’t depend on your agent’s own account of the call.

Will it catch every made-up answer?

No. It flags what contradicts your confirmed facts sheet and playbook. If the sheet says nothing about a topic, nothing is flagged on it, so the fuller the sheet, the more it catches.

Is this voice agent evaluation or pre-launch testing?

Evaluation of real calls, after launch. Testing tools ring an agent with scripted callers before release; ListenFox does not simulate calls. It checks every call your agent really took, so it works as voice agent monitoring once the agent is live.

Does it change how my agent behaves on the call?

No. It reads calls after they end and sits nowhere in your agent’s call path. Changing the prompt is up to you.

We build voice agents for clients. Can we keep them apart?

Yes. Each client can have its own workspace, with its own facts sheet, playbook and team.

What does it cost?

The same plans as for calls your team takes. See pricing.

Bring a week of your agent’s calls.

Plans from US$9 a month.

Start a free trial, upload your agent’s prompt and price list, and see its calls checked. Or book a demo and we’ll check them with you.

Three steps from here: create your account, connect your calls and read your first findings.