AI stack for phone calls.
Low latency, high quality speech APIs. Streaming and async jobs. Create a voice agent with one API, instead of stacking STT, TTS, turn detection, and an LLM.
Talk to Eva in the browser, same Omni agent as the demo line.
Talk to Eva, the same Omni agent that answers the phone. The call may be recorded.
already on the OpenAI SDK?
390 ms in-region · human pace
client = OpenAI(
base_url="https://api.pyai.com/v1",
api_key="pyai_test_…",
)change one line. keep your code.
or call +1 (252) 888-7924. same agent.
Speech APIs, the agent, and the rest of the call.
What is in the PyAI stack?
Speech-to-text. Streaming and async. Finals you can clean.
numerals · smart_format · dictation · drop_fillers · vocabulary
Text-to-speech. Streaming and async. Same catalog the agent uses.
stock · cloned · first byte in-region
Delivery control on the voice. Joy, urgency, empathy.
delivery, not just words
Who answered, including iPhone screening.
ten classes · a reason on every decision · one line of TwiML
A scorecard on the call you already ran.
PASS / WARN / FAIL · cited rule · audit hash
Notes on hang-up. No bot to invite.
summary · disposition · next step · your fields
Already on Twilio
One line of TwiML forks the call. Keep the rest of your stack.
<Stream url="wss://api.pyai.com/v1/amd/stream"> <Parameter name="api_key" value="YOUR_KEY"/> </Stream>AMD drop-in
Why one engine, not four vendors?
Vapi, Retell, and Bland stitch four vendors into every call, and the advertised platform fee is only one of the bills. PyAI runs speech + brain on one engine at $0.05/min. Managed telephony is $0.01/min separately.
Vapi / Retell / Bland: four vendors on every call
The advertised platform fee is just the orchestration layer. Add the model, the voice, and telephony and the all-in minute lands at $0.10-0.31/min on the packaged platforms, plus the latency of every hop.
public all-in ranges, Jun 2026 · verify on each provider’s site
PyAI: one engine, explicit line items
speech + brain, one speech-to-speech model
managed telephony $0.01/min · listed separately
One engine means no stacked model markups and no stacked latency. No platform fee to clear before the rate counts, no four-way bill to reconcile.
Bring your number: Twilio, JustCall, Telnyx, any SIP trunk.
Push outcomes to: HubSpot, Salesforce, Zapier, webhooks.
What happens after the phone is ringing?
From appointment booking to a scorecard on every scanned call. Numbers from teams running PyAI in production.
“Our AI Agents book appointments 24/7. That drove 57% incremental revenue for our customers. It’s our hottest-selling product.”
{
"event": "amd",
"call_id": "CA7d0…",
"answered_by": "screening",
"answered_by_twilio": "machine_start",
"confidence": 0.96,
"decision_ms": 720,
"reason": "screening phrase: 'the person you're calling is using \
a screening service' at 0.6s"
}Every AMD decision carries a reason you can read. See how AMD works.
How do I test without billing?
Sandbox keys never bill. Live keys consume prepaid credit. Speech + brain is $0.05/min. Managed telephony is $0.01/min and listed separately. The board lives on the rate card.
Using Cursor, Claude Code, Codex, or Lovable?
Start with one prompt. We hand your agent the API contract, the SDKs, and a free test key, so it writes correct PyAI code on the first try.
Questions about PyAI
What is PyAI?
PyAI is the AI stack for phone calls. Hear and Speak are low-latency, high-quality speech APIs for streaming and async jobs. Omni is one API for a voice agent, instead of stacking STT, TTS, turn detection, and an LLM. AMD, Trace, Recap, and Cast sit on the same stack.
How fast is PyAI on a phone call?
Omni is about 390 ms median voice-to-voice in-region, at human conversational pace. Hear's first partial is about 200 ms in-region and revisable. These are our measurements, not an SLA.
Is PyAI OpenAI-compatible?
Yes. Point your OpenAI SDK at https://api.pyai.com/v1 with your PyAI key. Transcription and speech keep the same request and response shapes, and errors come back in the OpenAI envelope.
How much does it cost?
Hear is $0.001/min. Omni API is $0.05/min for speech + brain. Agents is Live Beta at $0.08/min. Managed telephony is a separate $0.01/min. Sandbox keys never bill; live usage requires available credit.
How does PyAI compare with Vapi?
PyAI Omni is $0.05/min. Managed telephony is $0.01/min separately. Vapi advertises about $0.05/min as a platform fee; all-in typically lands around $0.10-$0.31/min once you add STT, an LLM, TTS, and telephony.
How do I start?
Get an API key. Sandbox keys never bill and need no card. Point the OpenAI SDK at https://api.pyai.com/v1, or talk to Eva on this page. A live key needs credit when you go to production.
Is PyAI built for telephony?
Yes. Hear is tuned for 8 kHz call audio with a ~200 ms first partial (in-region), and Omni runs phone agents at human conversational pace, about 390 ms median voice-to-voice in-region, with barge-in.
Get an API key. Make a phone agent.
Sandbox keys never bill. Fund a live key when you go to production.