Examples

The repo ships runnable examples under examples/. Each one is deployed live to Cloudflare Workers on every push — staging tracks dev, production tracks main releases.

playground — full local pipeline explorer

The flagship example: a full local pipeline with a live mic meter, VAD state, latency metrics, and a model picker — built to show every stage of the on-device voice loop (mic → VAD → STT → LLM → TTS) at once, not just the end result.

cd examples/playground
npm install
npm run serve   # builds + serves; open the printed URL (WebGPU required)

teams-react — Teams-style local call

A complete Teams-style call screen built entirely from the @valora-ai/react room/participant surface, running a fully local voice assistant — no server, no mocks. Silero VAD → Moonshine STT → LFM2.5 WebGPU LLM → Kokoro TTS, built with createVoiceAgent and passed directly to @valora-ai/react's VoiceRoom.

cd examples/teams-react
npm install
npm run serve   # builds + serves; open the printed URL (WebGPU required)

First load downloads the models (progress shown), then speak or type and Valora replies with real on-device inference + speech.

shadcn-react — shadcn/ui-flavored components

The same headless primitives as teams-react, skinned with shadcn/ui components instead.

cd examples/shadcn-react
npm install
npm run serve

Snippet examples

Smaller single-file React examples live in examples/:

• react-chat.tsx — useLocalModel with a model picker
• react-local-realtime.tsx — useLocalRealtime over an assembled local VoiceAgent
• react-tts.tsx — useLocalTTS
• react-vad.tsx — useVAD
• react-voice-loop.tsx — mic → VAD → STT loop
• ai-sdk.ts — the valora AI SDK provider, no server