Assembling...
Voice-capable variant of the Phi small language model family, integrating speech understanding and generation into a compact multimodal architecture. Designed for on-device and edge scenarios with Azure cloud fallback.
Talk naturally with Phi Voice
Start a conversation and speak freely. The AI will listen and respond naturally — no buttons between messages.
Demo Mode · Voice: Browser
Compact multimodal model runs voice AI directly on-device
MIT license enables flexible deployment across edge and cloud
Seamless fallback from on-device to Azure cloud for complex queries
Deploy AI voice agents that handle customer inquiries with natural conversation flow and real-time responses.
Build always-on voice assistants for enterprise applications with full-duplex capabilities.
Enable voice-first healthcare consultations with HIPAA-compliant conversational AI.
Replace traditional IVR menus with natural language voice agents that understand intent.
// Phi Voice — Conversational Voice Session
import { VoiceSession } from "@arkitekton/voice";
const session = await VoiceSession.create({
model: "vm-ms-004",
vendor: "microsoft",
config: {
fullDuplex: true,
language: "en-US",
turnDetection: "server_vad",
},
});
session.on("speech_started", () => {
console.log("Agent is speaking...");
});
session.on("transcript", (text) => {
console.log("User said:", text);
});
// Connect to audio stream
const mic = await navigator.mediaDevices.getUserMedia({ audio: true });
session.connect(mic);Enterprise speech services across Azure and research labs