Give your product a voice
your users can just talk to

Natural turn-taking

Vocra knows when to listen, when to pause, and when to jump in. It handles interruptions and backchannels the way a real conversation flows.

Sub-second latency

Speech in, speech out in around 320ms. The streaming pipeline means replies start before the sentence even finishes — no awkward dead air.

40+ languages, 100+ voices

Pick a lifelike voice or clone your own. Vocra speaks fluently across languages and switches mid-call without missing a beat.

A voice agent in a few lines

Drop Vocra into any stack. Bring your own model and tools, point it at a phone number or the web, and ship. We handle the audio, you handle the logic.

// start a live voice agent
import { Vocra } from "vocra";

const agent = new Vocra({
  voice: "nova",
  model: "your-llm",
  greeting: "Hey, how can I help?",
});

await agent.start();