Skip to content
ESLog inStart free
AI voice agents · powered by Pulsar 1

AI voice agents at the speed of conversation.

Pulsar 1, our real-time voice model, starts speaking in ~100 ms with the same voices and the same quality as Voyager 1. Build it into your own agent, or get a ready-made one that answers your business calls.

  • ~100 ms to first audio
  • 55+ languages
  • The voices of Voyager 1
Sample call
Pulsar 1 · es-ESReal Pulsar 1 audio, in Spanish
Hola, Lucía. Soy Clara, de la clínica dental Alba. ¿En qué te puedo ayudar?Hi, Lucía. This is Clara, from Alba Dental Clinic. How can I help you?
144 ms to first audio
Hola, Clara. Tenía cita el jueves, pero me ha surgido un lío en el trabajo y no voy a poder ir.Hi, Clara. I had an appointment on Thursday, but something’s come up at work and I can’t make it.
[serene] Vale, entiendo lo que me cuentas. [comforting] Lo miramos ahora mismo y lo dejamos solucionado.Okay, I understand what you’re telling me. We’ll look at it right now and get it sorted.
151 ms to first audio
¿Podría ser el martes por la mañana?Could we do Tuesday morning?
[upbeat] Perfecto, ya tienes la cita confirmada para el martes a las diez.Perfect, your appointment is confirmed for Tuesday at ten.
152 ms to first audio

Real Pulsar 1 audio, in Spanish, with the translation underneath. Each badge is that reply’s time to first audio, measured on the server; stop it mid-sentence to see the cancel.

Latency under load

~100 msto first audio · median

With one conversation, or with 48.

Median time to first audio stays between 93 and 113 ms from 1 to 48 simultaneous conversations. That is the number that matters when your agent is talking to a lot of people at once.

Time to first audio by simultaneous conversations

How we measured it: server-side, on 16 October 2026, from the moment the text arrives to the first audio out, excluding network and LLM. Median and 95th percentile by number of simultaneous conversations.

95–125 msA typical agent sentence (~3.5 s of audio), delivered in full.
113 msMedian with 48 conversations at once.
3×More than three times faster to first audio than Voyager 1, with the same voice.

In conversation, the typical gap between two people is around 200 ms. Pulsar 1 has the voice out in half that.

Same model, built to respond

Fast, not flat.

Pulsar 1 speaks with the voices, languages and emotions of Voyager 1. Your agent never trades its voice for speed.

800+

voices: the same as the Voicefinder library.

55+

languages, including native Spanish, Catalan, Galician and Basque.

[upbeat][hushed][unhappy]

Emotion tags right in the text, just like Voyager 1.

Two ways in

Build it yourself, or get it done.

For developersPulsar 1 API

Your agent, our voice.

You bring the speech recognition, the LLM and the telephony. Pulsar 1 brings the voice, in real time, over WebSocket.

  • speak, flush and cancel: three messages, full control
  • Per-turn interruptions with context_id
  • μ-law and A-law at 8 kHz for telephony; 16-bit PCM and float
  • Audio by sentence: the first chunk can arrive at the first comma
For businessesVoicefinder Agents

A receptionist that never gets tired.

It answers your calls, books appointments in your calendar and hands over to a person when needed. No coding.

  • Answers 24/7 on your number
  • Bookings, changes and cancellations in your calendar
  • Reminders by SMS, WhatsApp or email
  • A summary of every call in Slack, Teams or your inbox
For developers

Three messages. Full control.

Open a WebSocket, send text as your LLM writes it and play the audio as it arrives. If the caller cuts in, cancel that turn and carry on.

speak
Sends text to a turn (context_id). Send it in pieces, as your LLM produces it.
flush
Marks the turn as complete, so the rest of it is spoken.
cancel
Stops that turn without closing the connection. Use it when the caller interrupts.
Output formats
μ-law 8 kHzA-law 8 kHz16-bit PCMPCM float24 kHz native · resampling
import WebSocket from "ws";
// Illustrative example: Pulsar 1’s endpoint, authentication and model_id
// will be published at launch.
const ws = new WebSocket(PULSAR_WSS_URL, {
headers: { "xi-api-key": process.env.VOICEFINDER_API_KEY }, // provisional
});
const turn = "t1";
ws.on("open", () => {
ws.send(JSON.stringify({
type: "speak",
context_id: turn,
model_id: "pulsar", // provisional
voice_id: VOICE_ID,
output_format: "ulaw_8000", // ready for telephony
text: "[upbeat] Hi! I’m here to help. What can I do for you?",
}));
ws.send(JSON.stringify({ type: "flush", context_id: turn }));
});
ws.on("message", (audio) => call.write(audio)); // first audio in ~100 ms
// If the caller cuts in, stop just that turn
call.on("barge-in", () =>
ws.send(JSON.stringify({ type: "cancel", context_id: turn })));
Illustrative example. Pulsar 1’s public endpoint, authentication and model_id will be published at launch.
Use cases

Where a voice agent does the talking.

Both

Customer service

Answers questions, checks an order or opens a ticket, with no hold music and no phone menus.

Both

Appointments and bookings

Books, moves and cancels appointments against the real calendar, and confirms by SMS or WhatsApp.

Both

Outbound sales, with consent

Calls only people who asked for it: quote follow-ups, confirmations and renewals.

Voicefinder Agents only calls with consent, between 9 am and 9 pm.

Pulsar 1 API

In-app support

Give your product’s assistant a voice that answers while users keep going.

Pulsar 1 API

Video games

Characters that answer in real time, with emotion, in each player’s language.

Pulsar 1 API

Telephony and IVR

Swap “press 1” for a conversation. μ-law and A-law, the formats of telephony, out of the box.

Integrations

Works with the tools you already use.

These are the Voicefinder Agents integrations. With Pulsar 1 you connect whatever you like: it is an API.

Calls

Twilio

Your number: answers calls and sends SMS.

Messaging

WhatsApp Business

Confirmations and reminders on WhatsApp.

Calendar

iCal calendar

Google, Apple or Outlook by subscription; busy time blocks the schedule.

Alerts

Slack

A summary of every conversation, message and booking.

Alerts

Microsoft Teams

The same alerts, in a Teams channel.

Automation

n8n

Every booking, message or conversation reaches your flow, signed.

Automation

Zapier

A zap for every event.

Automation

Make

Scenarios triggered by every booking, message or conversation.

Tool

Webhook

The agent calls your API mid-call.

CRM

Attio

Every customer in your CRM, with a note for each booking or message.

Agents

MCP

Your calendar and customers as tools for Claude and other agents.

Payments

Stripe

Payment links on your own account.

FAQ

What people ask about AI voice agents.

What is an AI voice agent?

Software that talks on the phone or in your app the way a person would: it listens, understands with a language model and replies with a voice. Voicefinder gives you the voice (Pulsar 1) or the whole agent (Voicefinder Agents).

What’s the difference between Pulsar 1 and Voicefinder Agents?

Pulsar 1 is the real-time voice model, through an API, for teams building their own agent. Voicefinder Agents is the ready-made app for businesses: phone number, calendar, alerts and integrations. It runs on Pulsar 1.

How fast does it answer?

Pulsar 1 starts speaking with a median of 93 to 113 ms, from 1 to 48 simultaneous conversations (measured server-side on 16 October 2026, excluding network). The full response time of an agent also includes the speech recognition and the LLM you choose.

Does it sound worse because it is fast?

No. Pulsar 1 is the same model as Voyager 1, built to respond: the same voices, the same 55+ languages, the same emotions and the same voice quality.

Does it work over the phone?

Yes. Pulsar 1 outputs μ-law and A-law at 8 kHz, the telephony formats, as well as PCM. Voicefinder Agents answers on your number through Twilio and sends confirmations and reminders by SMS, WhatsApp or email.

Can callers interrupt the agent?

Yes. Each turn has its own context_id: cancel stops that turn without closing the connection, and the next one starts clean.

When can I use it, and how much does it cost?

Pulsar 1 and Voicefinder Agents are in early access. Pricing will be published at launch: request access and we will tell you more.

Pick your way in.

Pulsar 1 and Voicefinder Agents are in early access. Tell us what you want to build.