Getting Started

Valora is a set of focused packages. Pick the one that matches what you're building, or stack them for a fully local in-browser voice assistant.

@valora-ai/voice

npm install @valora-ai/voice — the voice-agent runtime: VAD → STT → LLM → TTS, one state machine, pluggable engines.

@valora-ai/ai-sdk

bun add @valora-ai/ai-sdk — an in-process AI SDK provider backed by local WebGPU engines.

@valora-ai/react

npm install @valora-ai/react — the call UI: orb, transcript, controls, on a re-skinnable headless layer.

Which do I want?

I want to...	Reach for
Run a model in the terminal or behind an OpenAI-compatible endpoint	`apps/cli` repo CLI
Use a local model with the Vercel AI SDK, in-process	`@valora-ai/ai-sdk`
Use local model/TTS/VAD/transcription React hooks	`@valora-ai/react/local`
Build a browser voice agent (VAD/STT/LLM/TTS loop)	`@valora-ai/voice`
Render a call screen (orb, transcript, controls) bound to that agent	`@valora-ai/react`
All of the above, wired together	All three — see Putting it together below

Requirements

WebGPU — Chrome/Edge, or Safari with WebGPU enabled.
Bun ≥ 1.3.14 for the repo CLI and package scripts.
React 18+ for @valora-ai/react.

Putting it together

Stack all three for a fully local, in-browser voice assistant (STT → LLM → TTS → call UI). The examples/teams-react app is a complete, runnable reference — no mocks, no server.

import '@valora-ai/react/styles.css';
import { VoiceRoom, BarVisualizer, Transcript } from '@valora-ai/react';
import { createVoiceAgent } from '@valora-ai/voice';
// local hooks live at @valora-ai/react/local — see the local runtime page

function Call({ agent }) {
  return (
    <VoiceRoom agent={agent}>
      <BarVisualizer />
      <Transcript />
    </VoiceRoom>
  );
}