[peteroyce]

Closes #477

Adds a useDeepgramAgent React hook that wraps the Voice Agent WebSocket API, giving React developers a declarative interface without manual WebSocket management.

Usage

import { useDeepgramAgent } from "@deepgram/sdk/react";

function VoiceAgent() {
  const { connect, disconnect, isConnected, transcript, agentText, error, sendFunctionResult } =
    useDeepgramAgent({
      apiKey: process.env.DEEPGRAM_API_KEY,
      agent: {
        think: { provider: { type: "open_ai" }, model: "gpt-4o-mini" },
        speak: { model: "aura-asteria-en" },
      },
      onFunctionCall: async (name, params) => ({ result: "done" }),
      onTranscript: (text, isFinal) => console.log("User:", text),
      onAgentResponse: (text) => console.log("Agent:", text),
    });

  return (
    <button onClick={isConnected ? disconnect : connect}>
      {isConnected ? "End" : "Start"} Conversation
    </button>
  );
}

What's included

• src/react/useDeepgramAgent.ts — hook implementation
• src/react/index.ts — barrel export
• package.json — ./react export path + React as optional peer dependency (≥16.8)

State exposed

Value	Type	Description
connect	() => void	Opens the WebSocket session
disconnect	() => void	Closes the session
isConnected	boolean	WebSocket open and ready
isListening	boolean	User is speaking
isSpeaking	boolean	Agent audio is playing
transcript	string	Latest user transcript
agentText	string	Latest agent response text
error	Error | null	Last error, if any
sendFunctionResult	(id, result) => void	Manual function call response

Notes

• React is an optional peer dependency — the ./react sub-path is tree-shaken; non-React consumers are unaffected
• Callback refs are used to avoid stale closure issues with onFunctionCall, onTranscript, and onAgentResponse
• The hook sends the initial Settings message automatically on open
• Cleanup on unmount closes the WebSocket automatically

[GregHolmes]

Hi @peteroyce,

Thank you for your contribution, but I don’t think we should merge this into the SDK. We intentionally keep the core SDK framework-agnostic, and React integrations belong in @deepgram/react (https://github.com/deepgram/react), which already provides this hook and the supporting microphone/playback lifecycle.

@deepgram/react does need some maintenance because it currently resolves through an older SDK version. I've got this on my to do list over the next couple weeks though.