We review your voice agent and tell you what it overspends on, where it loses latency and what breaks in telephony.
For agencies, multi-tenant SaaS and small teams on Vapi, Retell, LiveKit or Pipecat. Read-only exports. Mutual NDA.
agentaudit demo generates. Not a client. Reference values come from our own cost model (model reference, verify), not a market price list.These come from teams and builders describing their own production agents in public posts and forums.
Speech-to-text, LLM, text-to-speech, telephony and platform fees arrive on separate invoices. One developer was surprised that a 10-minute call costs $1.50–2.00 once they are added up.
Calls under 10 seconds and calls that end on silence are billed too.
One team's SIP trunk disappeared after 7–8 hours in production, and outbound calls failed until it was recreated. Another got 404 "No trunk found" on every inbound call.
A leading "+" in the number is enough for some carriers to return 404.
One team's dashboards said the agent was fast while the customer heard ~950 ms. Debugging means opening each provider's dashboard and matching timestamps.
One team called it "archaeology".
You keep control of your systems. We work from exports and recordings you send.
We agree the scope (Quick or Full), the dates and sign a mutual NDA.
Whatever you have of:
Findings ranked by severity, estimated monthly savings, a prioritized action plan and a readout call. In 3 or 5 business days.
The numbers in every audit come from agentaudit, our open-source CLI. Run it on your Vapi or Retell export and get cost per minute by component vs a reference, wasted minutes and SIP errors explained in plain words. It runs on your machine; nothing is uploaded.
Leave your email and we send you the install guide and an example report.
We only use your email to contact you about this. Privacy.
$ pip install agentaudit $ agentaudit run --vapi calls.json --sip-log sip.log --out report.md # no data yet? this writes sample inputs + a full example report $ agentaudit demo
## 1. Executive summary - **All-in cost per minute:** $0.147 vs $0.063 reference (model reference, verify). - **Estimated savings:** up to $1,492/month from 6 cost findings (estimates; they can overlap). - **Top issues:** - [CRITICAL] 1 in 20 turns takes over 3.5 s (LAT-02) - [HIGH] Text-to-speech (TTS) costs 2.6x the model reference (COST-02) ## 5. Telephony and SIP - Call attempts seen: 40; failed at SIP level: 4 (10.0%). - Log patterns: stun media x1, no trunk found x2. - 404 responses on '+'-prefixed numbers: 2.
Excerpt from the example report generated by agentaudit demo on made-up data (lines wrapped to fit). Not a client.
The tool gives the numbers. The audit adds a person who reads your configs, listens to your calls and tells you what to fix first.
3 business days · paid upfront
5 business days · 50% upfront, 50% on delivery
Billed monthly in advance · cancel any month
Founding clients: the first 3 clients pay $1,200 (Quick) or $2,800 (Full) in exchange for a written, anonymized case study and a 30-minute feedback call.
Invoiced in USD (EUR possible).
Read-only exports only: call exports, invoices, SIP logs, a sample of recordings and your agent configs. We never need write access to production, your provider accounts or your phone numbers.
We sign a mutual NDA before you send anything. The analysis runs locally with our open-source tool, so your call data is not uploaded to any third-party service. We delete your data within 30 days of delivery. We only publish an anonymized case study if you agree in writing.
No. We check your calls and configs against a practical checklist (AI disclosure, recording consent, outbound calling rules such as TCPA, data retention) and flag what looks risky. Confirm anything that matters with your lawyer.
Agents built on Vapi, Retell, LiveKit and Pipecat, with carriers such as Twilio or Telnyx and the usual speech-to-text, LLM and text-to-speech providers. If your setup is different, ask on the call.
Eval and observability platforms (Hamming, Coval, Cekura, Roark and others) are products your team sets up and runs over time. This is a one-off, fixed-price review by a person, across your invoices, telephony, latency and configs, ending in a written list of what to fix. The two can work together.
Yes. It covers the cost numbers, waste and SIP error explanations. The audit adds the parts a script cannot do: listening to recordings, reading configs, checking compliance and putting the fixes in order.
Book a short call. We look at your setup and tell you whether a Quick or a Full audit makes sense.