agents 0.22.0 → 0.23.0

package/dist/voice/react.js

@@ -0,0 +1,234 @@
+import { VoiceClient, WebSocketVoiceTransport } from "./client.js";
+import { useCallback, useEffect, useMemo, useRef, useState } from "react";
+//#region src/voice/react.tsx
+/**
+* React hook for voice-to-text input. Captures microphone audio, streams it
+* to a server-side VoiceAgent for STT, and returns the transcript as a string.
+*
+* Unlike `useVoiceAgent`, this hook is optimised for dictation — it accumulates
+* user transcripts into a single string and ignores assistant responses / TTS.
+*
+* @example
+* ```tsx
+* const { transcript, interimTranscript, isListening, start, stop } = useVoiceInput({
+*   agent: "voice-input-agent"
+* });
+*
+* <textarea value={transcript + (interimTranscript ? " " + interimTranscript : "")} />
+* <button onClick={isListening ? stop : start}>
+*   {isListening ? "Stop" : "Dictate"}
+* </button>
+* ```
+*/
+function useVoiceInput(options) {
+	const connectionKey = useMemo(() => `${options.agent}:${options.name ?? "default"}:${options.host ?? ""}:${options.silenceThreshold ?? ""}:${options.silenceDurationMs ?? ""}`, [
+		options.agent,
+		options.name,
+		options.host,
+		options.silenceThreshold,
+		options.silenceDurationMs
+	]);
+	const clientRef = useRef(null);
+	const [transcript, setTranscript] = useState("");
+	const [interimTranscript, setInterimTranscript] = useState(null);
+	const [isListening, setIsListening] = useState(false);
+	const [turnMetrics, setTurnMetrics] = useState(null);
+	const [audioLevel, setAudioLevel] = useState(0);
+	const [isMuted, setIsMuted] = useState(false);
+	const [error, setError] = useState(null);
+	useEffect(() => {
+		setIsListening(false);
+		setInterimTranscript(null);
+		setTurnMetrics(null);
+		setAudioLevel(0);
+		setIsMuted(false);
+		setError(null);
+		const client = new VoiceClient({
+			agent: options.agent,
+			name: options.name,
+			host: options.host,
+			silenceThreshold: options.silenceThreshold,
+			silenceDurationMs: options.silenceDurationMs
+		});
+		clientRef.current = client;
+		client.connect();
+		const onTranscript = () => {
+			setTranscript(client.transcript.filter((m) => m.role === "user").map((m) => m.text).join(" "));
+		};
+		const onInterim = () => setInterimTranscript(client.interimTranscript);
+		const onTurnMetrics = (value) => setTurnMetrics(value);
+		const onAudioLevel = () => setAudioLevel(client.audioLevel);
+		const onMute = () => setIsMuted(client.isMuted);
+		const onError = () => setError(client.error);
+		const onStatus = () => {
+			const s = client.status;
+			setIsListening(s === "listening" || s === "thinking");
+		};
+		client.addEventListener("transcriptchange", onTranscript);
+		client.addEventListener("interimtranscript", onInterim);
+		client.addEventListener("turnmetrics", onTurnMetrics);
+		client.addEventListener("audiolevelchange", onAudioLevel);
+		client.addEventListener("mutechange", onMute);
+		client.addEventListener("error", onError);
+		client.addEventListener("statuschange", onStatus);
+		return () => {
+			client.removeEventListener("transcriptchange", onTranscript);
+			client.removeEventListener("interimtranscript", onInterim);
+			client.removeEventListener("turnmetrics", onTurnMetrics);
+			client.removeEventListener("audiolevelchange", onAudioLevel);
+			client.removeEventListener("mutechange", onMute);
+			client.removeEventListener("error", onError);
+			client.removeEventListener("statuschange", onStatus);
+			client.disconnect();
+		};
+	}, [connectionKey]);
+	return {
+		transcript,
+		interimTranscript,
+		turnMetrics,
+		isListening,
+		audioLevel,
+		isMuted,
+		error,
+		start: useCallback(() => clientRef.current.startCall(), []),
+		stop: useCallback(() => clientRef.current.endCall(), []),
+		toggleMute: useCallback(() => clientRef.current.toggleMute(), []),
+		clear: useCallback(() => setTranscript(""), [])
+	};
+}
+/**
+* React hook that wraps VoiceClient, syncing its state into React state.
+* All audio infrastructure (mic capture, playback, silence/interrupt detection,
+* voice protocol) is handled by VoiceClient — this hook just bridges to React.
+*
+* When the connection identity changes (agent, name, or host), the hook
+* automatically disconnects the old client, creates a new one, and reconnects.
+* The `onReconnect` callback fires when this happens.
+*/
+function useVoiceAgent(options) {
+	const queryString = useMemo(() => options.query ? JSON.stringify(options.query) : "", [options.query]);
+	const enabled = options.enabled ?? true;
+	const outputDeviceId = options.outputDeviceId;
+	const connectionKey = useMemo(() => `${options.agent}:${options.name ?? "default"}:${options.host ?? ""}:${options.silenceThreshold ?? ""}:${options.silenceDurationMs ?? ""}:${options.interruptThreshold ?? ""}:${options.interruptChunks ?? ""}:${queryString}`, [
+		options.agent,
+		options.name,
+		options.host,
+		options.silenceThreshold,
+		options.silenceDurationMs,
+		options.interruptThreshold,
+		options.interruptChunks,
+		queryString
+	]);
+	const effectKey = useMemo(() => `${enabled ? "enabled" : "disabled"}:${connectionKey}`, [enabled, connectionKey]);
+	const clientRef = useRef(null);
+	const activeKeyRef = useRef(null);
+	const onReconnectRef = useRef(options.onReconnect);
+	onReconnectRef.current = options.onReconnect;
+	const [status, setStatus] = useState("idle");
+	const [transcript, setTranscript] = useState([]);
+	const [metrics, setMetrics] = useState(null);
+	const [turnMetrics, setTurnMetrics] = useState(null);
+	const [audioLevel, setAudioLevel] = useState(0);
+	const [isMuted, setIsMuted] = useState(false);
+	const [connected, setConnected] = useState(false);
+	const [error, setError] = useState(null);
+	const [outputDeviceError, setOutputDeviceError] = useState(null);
+	const [interimTranscript, setInterimTranscript] = useState(null);
+	const [lastCustomMessage, setLastCustomMessage] = useState(null);
+	useEffect(() => {
+		setStatus("idle");
+		setTranscript([]);
+		setMetrics(null);
+		setTurnMetrics(null);
+		setAudioLevel(0);
+		setIsMuted(false);
+		setConnected(false);
+		setError(null);
+		setOutputDeviceError(null);
+		setInterimTranscript(null);
+		setLastCustomMessage(null);
+		if (!enabled) {
+			clientRef.current = null;
+			activeKeyRef.current = null;
+			return;
+		}
+		const isReconnect = activeKeyRef.current !== null && activeKeyRef.current !== connectionKey;
+		activeKeyRef.current = connectionKey;
+		if (isReconnect) onReconnectRef.current?.();
+		const { enabled: _enabled, onReconnect: _onReconnect, ...clientOptions } = options;
+		const client = new VoiceClient(clientOptions);
+		clientRef.current = client;
+		client.connect();
+		const onStatus = (s) => setStatus(s);
+		const onTranscript = (t) => setTranscript(t);
+		const onMetrics = (m) => setMetrics(m);
+		const onTurnMetrics = (value) => setTurnMetrics(value);
+		const onAudioLevel = (level) => setAudioLevel(level);
+		const onMute = (muted) => setIsMuted(muted);
+		const onConnection = (c) => setConnected(c);
+		const onError = (e) => setError(e);
+		const onOutputDeviceError = (e) => setOutputDeviceError(e);
+		const onInterim = (text) => setInterimTranscript(text);
+		client.addEventListener("statuschange", onStatus);
+		client.addEventListener("transcriptchange", onTranscript);
+		client.addEventListener("interimtranscript", onInterim);
+		client.addEventListener("metricschange", onMetrics);
+		client.addEventListener("turnmetrics", onTurnMetrics);
+		client.addEventListener("audiolevelchange", onAudioLevel);
+		client.addEventListener("mutechange", onMute);
+		client.addEventListener("connectionchange", onConnection);
+		client.addEventListener("error", onError);
+		client.addEventListener("outputdeviceerror", onOutputDeviceError);
+		return () => {
+			client.removeEventListener("statuschange", onStatus);
+			client.removeEventListener("transcriptchange", onTranscript);
+			client.removeEventListener("interimtranscript", onInterim);
+			client.removeEventListener("metricschange", onMetrics);
+			client.removeEventListener("turnmetrics", onTurnMetrics);
+			client.removeEventListener("audiolevelchange", onAudioLevel);
+			client.removeEventListener("mutechange", onMute);
+			client.removeEventListener("connectionchange", onConnection);
+			client.removeEventListener("error", onError);
+			client.removeEventListener("outputdeviceerror", onOutputDeviceError);
+			if (clientRef.current === client) clientRef.current = null;
+			client.disconnect();
+		};
+	}, [effectKey]);
+	const startCall = useCallback(() => clientRef.current?.startCall() ?? Promise.resolve(), []);
+	const endCall = useCallback(() => clientRef.current?.endCall(), []);
+	const toggleMute = useCallback(() => clientRef.current?.toggleMute(), []);
+	const sendText = useCallback((text) => clientRef.current?.sendText(text), []);
+	const sendJSON = useCallback((data) => clientRef.current?.sendJSON(data), []);
+	useEffect(() => {
+		const client = clientRef.current;
+		if (!client) return;
+		const onCustom = (msg) => setLastCustomMessage(msg);
+		client.addEventListener("custommessage", onCustom);
+		return () => client.removeEventListener("custommessage", onCustom);
+	}, [effectKey]);
+	useEffect(() => {
+		clientRef.current?.setOutputDevice(outputDeviceId);
+	}, [effectKey, outputDeviceId]);
+	return {
+		status,
+		transcript,
+		interimTranscript,
+		metrics,
+		turnMetrics,
+		audioLevel,
+		isMuted,
+		connected,
+		error,
+		outputDeviceError,
+		startCall,
+		endCall,
+		toggleMute,
+		sendText,
+		sendJSON,
+		lastCustomMessage
+	};
+}
+//#endregion
+export { WebSocketVoiceTransport, useVoiceAgent, useVoiceInput };
+
+//# sourceMappingURL=react.js.map

package/dist/voice/react.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"react.js","names":[],"sources":["../../src/voice/react.tsx"],"sourcesContent":["import { useEffect, useRef, useState, useCallback, useMemo } from \"react\";\nimport {\n  VoiceClient,\n  type VoiceClientOptions,\n  type VoiceStatus,\n  type TranscriptMessage,\n  type VoicePipelineMetrics,\n  type VoiceTurnMetrics\n} from \"./client\";\n\n// Re-export types so consumers can import everything from agents/voice/react\nexport type {\n  VoiceStatus,\n  VoiceRole,\n  VoiceAudioFormat,\n  VoiceAudioInput,\n  VoiceTransport,\n  TranscriptMessage,\n  VoicePipelineMetrics,\n  VoiceTurnMetrics,\n  VoiceTurnOutcome,\n  VoiceTurnSource,\n  VoiceClientOptions,\n  VoiceClientEvent,\n  VoiceClientEventMap,\n  VoiceConnectionDiagnostic,\n  VoiceTransportCloseInfo,\n  VoiceError,\n  VoiceErrorCode,\n  VoiceErrorStage\n} from \"./client\";\nexport { WebSocketVoiceTransport } from \"./client\";\n\n/** Options accepted by useVoiceAgent. */\nexport interface UseVoiceAgentOptions extends VoiceClientOptions {\n  /**\n   * Whether the hook should create and connect a VoiceClient.\n   *\n   * @default true\n   */\n  enabled?: boolean;\n  /**\n   * Called when the hook reconnects due to option changes (e.g., agent name\n   * or instance name changed). Use this to show a toast or notification.\n   */\n  onReconnect?: () => void;\n}\n\nexport interface UseVoiceAgentReturn {\n  status: VoiceStatus;\n  transcript: TranscriptMessage[];\n  /**\n   * The current interim (partial) transcript from streaming STT.\n   * Updates in real time as the user speaks. null when not available.\n   */\n  interimTranscript: string | null;\n  metrics: VoicePipelineMetrics | null;\n  /** Last stable terminal summary for a speech or text turn. */\n  turnMetrics: VoiceTurnMetrics | null;\n  audioLevel: number;\n  isMuted: boolean;\n  connected: boolean;\n  error: string | null;\n  outputDeviceError: string | null;\n  startCall: () => Promise<void>;\n  endCall: () => void;\n  toggleMute: () => void;\n  sendText: (text: string) => void;\n  /** Send arbitrary JSON to the agent (app-level messages). */\n  sendJSON: (data: Record<string, unknown>) => void;\n  /** The last non-voice-protocol message received from the server. */\n  lastCustomMessage: unknown;\n}\n\n// ---------------------------------------------------------------------------\n// useVoiceInput — lightweight hook for voice-to-text dictation\n// ---------------------------------------------------------------------------\n\n/** Options accepted by useVoiceInput. */\nexport interface UseVoiceInputOptions {\n  /** Agent name (matches the server-side Durable Object class). */\n  agent: string;\n  /** Instance name for the agent. @default \"default\" */\n  name?: string;\n  /** Host to connect to. @default window.location.host */\n  host?: string;\n\n  /** RMS threshold below which audio is considered silence. @default 0.04 */\n  silenceThreshold?: number;\n  /** How long silence must last before sending end_of_speech (ms). @default 500 */\n  silenceDurationMs?: number;\n}\n\nexport interface UseVoiceInputReturn {\n  /** Accumulated final transcript text from all utterances. */\n  transcript: string;\n  /**\n   * Current interim (partial) transcript from streaming STT.\n   * Updates in real time as the user speaks. null when not available.\n   */\n  interimTranscript: string | null;\n  /** Last stable terminal summary for a speech turn. */\n  turnMetrics: VoiceTurnMetrics | null;\n  /** Whether the mic is actively listening. */\n  isListening: boolean;\n  /** Current audio level (0–1) for visual feedback (e.g. waveform). */\n  audioLevel: number;\n  /** Whether the mic is muted. */\n  isMuted: boolean;\n  /** Any error message. */\n  error: string | null;\n  /** Start listening — requests mic permission and begins streaming audio. */\n  start: () => Promise<void>;\n  /** Stop listening — releases the mic. */\n  stop: () => void;\n  /** Toggle mute (mic stays open but audio is not sent). */\n  toggleMute: () => void;\n  /** Clear the accumulated transcript. */\n  clear: () => void;\n}\n\n/**\n * React hook for voice-to-text input. Captures microphone audio, streams it\n * to a server-side VoiceAgent for STT, and returns the transcript as a string.\n *\n * Unlike `useVoiceAgent`, this hook is optimised for dictation — it accumulates\n * user transcripts into a single string and ignores assistant responses / TTS.\n *\n * @example\n * ```tsx\n * const { transcript, interimTranscript, isListening, start, stop } = useVoiceInput({\n *   agent: \"voice-input-agent\"\n * });\n *\n * <textarea value={transcript + (interimTranscript ? \" \" + interimTranscript : \"\")} />\n * <button onClick={isListening ? stop : start}>\n *   {isListening ? \"Stop\" : \"Dictate\"}\n * </button>\n * ```\n */\nexport function useVoiceInput(\n  options: UseVoiceInputOptions\n): UseVoiceInputReturn {\n  const connectionKey = useMemo(\n    () =>\n      `${options.agent}:${options.name ?? \"default\"}:${options.host ?? \"\"}:${options.silenceThreshold ?? \"\"}:${options.silenceDurationMs ?? \"\"}`,\n    [\n      options.agent,\n      options.name,\n      options.host,\n      options.silenceThreshold,\n      options.silenceDurationMs\n    ]\n  );\n\n  const clientRef = useRef<VoiceClient | null>(null);\n\n  const [transcript, setTranscript] = useState(\"\");\n  const [interimTranscript, setInterimTranscript] = useState<string | null>(\n    null\n  );\n  const [isListening, setIsListening] = useState(false);\n  const [turnMetrics, setTurnMetrics] = useState<VoiceTurnMetrics | null>(null);\n  const [audioLevel, setAudioLevel] = useState(0);\n  const [isMuted, setIsMuted] = useState(false);\n  const [error, setError] = useState<string | null>(null);\n\n  // Connect on mount or when connection identity changes\n  useEffect(() => {\n    setIsListening(false);\n    setInterimTranscript(null);\n    setTurnMetrics(null);\n    setAudioLevel(0);\n    setIsMuted(false);\n    setError(null);\n\n    const client = new VoiceClient({\n      agent: options.agent,\n      name: options.name,\n      host: options.host,\n      silenceThreshold: options.silenceThreshold,\n      silenceDurationMs: options.silenceDurationMs\n    });\n    clientRef.current = client;\n    client.connect();\n\n    // Sync user transcripts into a single accumulated string\n    const onTranscript = () => {\n      const msgs = client.transcript;\n      const userTexts = msgs\n        .filter((m) => m.role === \"user\")\n        .map((m) => m.text);\n      setTranscript(userTexts.join(\" \"));\n    };\n\n    const onInterim = () => setInterimTranscript(client.interimTranscript);\n    const onTurnMetrics = (value: VoiceTurnMetrics) => setTurnMetrics(value);\n    const onAudioLevel = () => setAudioLevel(client.audioLevel);\n    const onMute = () => setIsMuted(client.isMuted);\n    const onError = () => setError(client.error);\n\n    const onStatus = () => {\n      const s = client.status;\n      setIsListening(s === \"listening\" || s === \"thinking\");\n    };\n\n    client.addEventListener(\"transcriptchange\", onTranscript);\n    client.addEventListener(\"interimtranscript\", onInterim);\n    client.addEventListener(\"turnmetrics\", onTurnMetrics);\n    client.addEventListener(\"audiolevelchange\", onAudioLevel);\n    client.addEventListener(\"mutechange\", onMute);\n    client.addEventListener(\"error\", onError);\n    client.addEventListener(\"statuschange\", onStatus);\n\n    return () => {\n      client.removeEventListener(\"transcriptchange\", onTranscript);\n      client.removeEventListener(\"interimtranscript\", onInterim);\n      client.removeEventListener(\"turnmetrics\", onTurnMetrics);\n      client.removeEventListener(\"audiolevelchange\", onAudioLevel);\n      client.removeEventListener(\"mutechange\", onMute);\n      client.removeEventListener(\"error\", onError);\n      client.removeEventListener(\"statuschange\", onStatus);\n      client.disconnect();\n    };\n    // eslint-disable-next-line react-hooks/exhaustive-deps -- reconnect when connection identity changes\n  }, [connectionKey]);\n\n  const start = useCallback(() => clientRef.current!.startCall(), []);\n  const stop = useCallback(() => clientRef.current!.endCall(), []);\n  const toggleMute = useCallback(() => clientRef.current!.toggleMute(), []);\n  const clear = useCallback(() => setTranscript(\"\"), []);\n\n  return {\n    transcript,\n    interimTranscript,\n    turnMetrics,\n    isListening,\n    audioLevel,\n    isMuted,\n    error,\n    start,\n    stop,\n    toggleMute,\n    clear\n  };\n}\n\n// ---------------------------------------------------------------------------\n// useVoiceAgent — full-featured hook for conversational voice agents\n// ---------------------------------------------------------------------------\n\n/**\n * React hook that wraps VoiceClient, syncing its state into React state.\n * All audio infrastructure (mic capture, playback, silence/interrupt detection,\n * voice protocol) is handled by VoiceClient — this hook just bridges to React.\n *\n * When the connection identity changes (agent, name, or host), the hook\n * automatically disconnects the old client, creates a new one, and reconnects.\n * The `onReconnect` callback fires when this happens.\n */\nexport function useVoiceAgent(\n  options: UseVoiceAgentOptions\n): UseVoiceAgentReturn {\n  // Derive a stable key from the connection-identity fields.\n  // When this changes, we tear down the old client and create a new one.\n  const queryString = useMemo(\n    () => (options.query ? JSON.stringify(options.query) : \"\"),\n    [options.query]\n  );\n\n  const enabled = options.enabled ?? true;\n  const outputDeviceId = options.outputDeviceId;\n\n  const connectionKey = useMemo(\n    () =>\n      `${options.agent}:${options.name ?? \"default\"}:${options.host ?? \"\"}:${options.silenceThreshold ?? \"\"}:${options.silenceDurationMs ?? \"\"}:${options.interruptThreshold ?? \"\"}:${options.interruptChunks ?? \"\"}:${queryString}`,\n    [\n      options.agent,\n      options.name,\n      options.host,\n      options.silenceThreshold,\n      options.silenceDurationMs,\n      options.interruptThreshold,\n      options.interruptChunks,\n      queryString\n    ]\n  );\n\n  const effectKey = useMemo(\n    () => `${enabled ? \"enabled\" : \"disabled\"}:${connectionKey}`,\n    [enabled, connectionKey]\n  );\n\n  const clientRef = useRef<VoiceClient | null>(null);\n  const activeKeyRef = useRef<string | null>(null);\n  const onReconnectRef = useRef(options.onReconnect);\n  onReconnectRef.current = options.onReconnect;\n\n  // React state mirrors VoiceClient state\n  const [status, setStatus] = useState<VoiceStatus>(\"idle\");\n  const [transcript, setTranscript] = useState<TranscriptMessage[]>([]);\n  const [metrics, setMetrics] = useState<VoicePipelineMetrics | null>(null);\n  const [turnMetrics, setTurnMetrics] = useState<VoiceTurnMetrics | null>(null);\n  const [audioLevel, setAudioLevel] = useState(0);\n  const [isMuted, setIsMuted] = useState(false);\n  const [connected, setConnected] = useState(false);\n  const [error, setError] = useState<string | null>(null);\n  const [outputDeviceError, setOutputDeviceError] = useState<string | null>(\n    null\n  );\n  const [interimTranscript, setInterimTranscript] = useState<string | null>(\n    null\n  );\n  const [lastCustomMessage, setLastCustomMessage] = useState<unknown>(null);\n\n  // Connect on mount or when connection identity changes\n  useEffect(() => {\n    // Reset state for a fresh or disabled connection\n    setStatus(\"idle\");\n    setTranscript([]);\n    setMetrics(null);\n    setTurnMetrics(null);\n    setAudioLevel(0);\n    setIsMuted(false);\n    setConnected(false);\n    setError(null);\n    setOutputDeviceError(null);\n    setInterimTranscript(null);\n    setLastCustomMessage(null);\n\n    if (!enabled) {\n      clientRef.current = null;\n      activeKeyRef.current = null;\n      return;\n    }\n\n    const isReconnect =\n      activeKeyRef.current !== null && activeKeyRef.current !== connectionKey;\n    activeKeyRef.current = connectionKey;\n\n    // Fire reconnect callback (e.g., to show a toast)\n    if (isReconnect) {\n      onReconnectRef.current?.();\n    }\n\n    const {\n      enabled: _enabled,\n      onReconnect: _onReconnect,\n      ...clientOptions\n    } = options;\n    const client = new VoiceClient(clientOptions);\n    clientRef.current = client;\n    client.connect();\n\n    // Sync handlers — read state from client and push to React\n    const onStatus = (s: VoiceStatus) => setStatus(s);\n    const onTranscript = (t: TranscriptMessage[]) => setTranscript(t);\n    const onMetrics = (m: VoicePipelineMetrics | null) => setMetrics(m);\n    const onTurnMetrics = (value: VoiceTurnMetrics) => setTurnMetrics(value);\n    const onAudioLevel = (level: number) => setAudioLevel(level);\n    const onMute = (muted: boolean) => setIsMuted(muted);\n    const onConnection = (c: boolean) => setConnected(c);\n    const onError = (e: string | null) => setError(e);\n    const onOutputDeviceError = (e: string | null) => setOutputDeviceError(e);\n    const onInterim = (text: string | null) => setInterimTranscript(text);\n\n    client.addEventListener(\"statuschange\", onStatus);\n    client.addEventListener(\"transcriptchange\", onTranscript);\n    client.addEventListener(\"interimtranscript\", onInterim);\n    client.addEventListener(\"metricschange\", onMetrics);\n    client.addEventListener(\"turnmetrics\", onTurnMetrics);\n    client.addEventListener(\"audiolevelchange\", onAudioLevel);\n    client.addEventListener(\"mutechange\", onMute);\n    client.addEventListener(\"connectionchange\", onConnection);\n    client.addEventListener(\"error\", onError);\n    client.addEventListener(\"outputdeviceerror\", onOutputDeviceError);\n\n    return () => {\n      client.removeEventListener(\"statuschange\", onStatus);\n      client.removeEventListener(\"transcriptchange\", onTranscript);\n      client.removeEventListener(\"interimtranscript\", onInterim);\n      client.removeEventListener(\"metricschange\", onMetrics);\n      client.removeEventListener(\"turnmetrics\", onTurnMetrics);\n      client.removeEventListener(\"audiolevelchange\", onAudioLevel);\n      client.removeEventListener(\"mutechange\", onMute);\n      client.removeEventListener(\"connectionchange\", onConnection);\n      client.removeEventListener(\"error\", onError);\n      client.removeEventListener(\"outputdeviceerror\", onOutputDeviceError);\n      if (clientRef.current === client) {\n        clientRef.current = null;\n      }\n      client.disconnect();\n    };\n    // eslint-disable-next-line react-hooks/exhaustive-deps -- reconnect when connection identity changes\n  }, [effectKey]);\n\n  // Stable action callbacks — always use the latest client\n  const startCall = useCallback(\n    () => clientRef.current?.startCall() ?? Promise.resolve(),\n    []\n  );\n  const endCall = useCallback(() => clientRef.current?.endCall(), []);\n  const toggleMute = useCallback(() => clientRef.current?.toggleMute(), []);\n  const sendText = useCallback(\n    (text: string) => clientRef.current?.sendText(text),\n    []\n  );\n  const sendJSON = useCallback(\n    (data: Record<string, unknown>) => clientRef.current?.sendJSON(data),\n    []\n  );\n\n  // Listen for custom messages — needs a separate effect since it must\n  // attach to the latest client.\n  useEffect(() => {\n    const client = clientRef.current;\n    if (!client) return;\n    const onCustom = (msg: unknown) => setLastCustomMessage(msg);\n    client.addEventListener(\"custommessage\", onCustom);\n    return () => client.removeEventListener(\"custommessage\", onCustom);\n  }, [effectKey]);\n\n  useEffect(() => {\n    void clientRef.current?.setOutputDevice(outputDeviceId);\n  }, [effectKey, outputDeviceId]);\n\n  return {\n    status,\n    transcript,\n    interimTranscript,\n    metrics,\n    turnMetrics,\n    audioLevel,\n    isMuted,\n    connected,\n    error,\n    outputDeviceError,\n    startCall,\n    endCall,\n    toggleMute,\n    sendText,\n    sendJSON,\n    lastCustomMessage\n  };\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;;;AA4IA,SAAgB,cACd,SACqB;CACrB,MAAM,gBAAgB,cAElB,GAAG,QAAQ,MAAM,GAAG,QAAQ,QAAQ,UAAU,GAAG,QAAQ,QAAQ,GAAG,GAAG,QAAQ,oBAAoB,GAAG,GAAG,QAAQ,qBAAqB,MACxI;EACE,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR,QAAQ;CACV,CACF;CAEA,MAAM,YAAY,OAA2B,IAAI;CAEjD,MAAM,CAAC,YAAY,iBAAiB,SAAS,EAAE;CAC/C,MAAM,CAAC,mBAAmB,wBAAwB,SAChD,IACF;CACA,MAAM,CAAC,aAAa,kBAAkB,SAAS,KAAK;CACpD,MAAM,CAAC,aAAa,kBAAkB,SAAkC,IAAI;CAC5E,MAAM,CAAC,YAAY,iBAAiB,SAAS,CAAC;CAC9C,MAAM,CAAC,SAAS,cAAc,SAAS,KAAK;CAC5C,MAAM,CAAC,OAAO,YAAY,SAAwB,IAAI;CAGtD,gBAAgB;EACd,eAAe,KAAK;EACpB,qBAAqB,IAAI;EACzB,eAAe,IAAI;EACnB,cAAc,CAAC;EACf,WAAW,KAAK;EAChB,SAAS,IAAI;EAEb,MAAM,SAAS,IAAI,YAAY;GAC7B,OAAO,QAAQ;GACf,MAAM,QAAQ;GACd,MAAM,QAAQ;GACd,kBAAkB,QAAQ;GAC1B,mBAAmB,QAAQ;EAC7B,CAAC;EACD,UAAU,UAAU;EACpB,OAAO,QAAQ;EAGf,MAAM,qBAAqB;GAKzB,cAJa,OAAO,WAEjB,QAAQ,MAAM,EAAE,SAAS,MAAM,CAAC,CAChC,KAAK,MAAM,EAAE,IACM,CAAC,CAAC,KAAK,GAAG,CAAC;EACnC;EAEA,MAAM,kBAAkB,qBAAqB,OAAO,iBAAiB;EACrE,MAAM,iBAAiB,UAA4B,eAAe,KAAK;EACvE,MAAM,qBAAqB,cAAc,OAAO,UAAU;EAC1D,MAAM,eAAe,WAAW,OAAO,OAAO;EAC9C,MAAM,gBAAgB,SAAS,OAAO,KAAK;EAE3C,MAAM,iBAAiB;GACrB,MAAM,IAAI,OAAO;GACjB,eAAe,MAAM,eAAe,MAAM,UAAU;EACtD;EAEA,OAAO,iBAAiB,oBAAoB,YAAY;EACxD,OAAO,iBAAiB,qBAAqB,SAAS;EACtD,OAAO,iBAAiB,eAAe,aAAa;EACpD,OAAO,iBAAiB,oBAAoB,YAAY;EACxD,OAAO,iBAAiB,cAAc,MAAM;EAC5C,OAAO,iBAAiB,SAAS,OAAO;EACxC,OAAO,iBAAiB,gBAAgB,QAAQ;EAEhD,aAAa;GACX,OAAO,oBAAoB,oBAAoB,YAAY;GAC3D,OAAO,oBAAoB,qBAAqB,SAAS;GACzD,OAAO,oBAAoB,eAAe,aAAa;GACvD,OAAO,oBAAoB,oBAAoB,YAAY;GAC3D,OAAO,oBAAoB,cAAc,MAAM;GAC/C,OAAO,oBAAoB,SAAS,OAAO;GAC3C,OAAO,oBAAoB,gBAAgB,QAAQ;GACnD,OAAO,WAAW;EACpB;CAEF,GAAG,CAAC,aAAa,CAAC;CAOlB,OAAO;EACL;EACA;EACA;EACA;EACA;EACA;EACA;EACA,OAbY,kBAAkB,UAAU,QAAS,UAAU,GAAG,CAAC,CAa3D;EACJ,MAbW,kBAAkB,UAAU,QAAS,QAAQ,GAAG,CAAC,CAazD;EACH,YAbiB,kBAAkB,UAAU,QAAS,WAAW,GAAG,CAAC,CAa5D;EACT,OAbY,kBAAkB,cAAc,EAAE,GAAG,CAAC,CAa9C;CACN;AACF;;;;;;;;;;AAeA,SAAgB,cACd,SACqB;CAGrB,MAAM,cAAc,cACX,QAAQ,QAAQ,KAAK,UAAU,QAAQ,KAAK,IAAI,IACvD,CAAC,QAAQ,KAAK,CAChB;CAEA,MAAM,UAAU,QAAQ,WAAW;CACnC,MAAM,iBAAiB,QAAQ;CAE/B,MAAM,gBAAgB,cAElB,GAAG,QAAQ,MAAM,GAAG,QAAQ,QAAQ,UAAU,GAAG,QAAQ,QAAQ,GAAG,GAAG,QAAQ,oBAAoB,GAAG,GAAG,QAAQ,qBAAqB,GAAG,GAAG,QAAQ,sBAAsB,GAAG,GAAG,QAAQ,mBAAmB,GAAG,GAAG,eACnN;EACE,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR,QAAQ;EACR;CACF,CACF;CAEA,MAAM,YAAY,cACV,GAAG,UAAU,YAAY,WAAW,GAAG,iBAC7C,CAAC,SAAS,aAAa,CACzB;CAEA,MAAM,YAAY,OAA2B,IAAI;CACjD,MAAM,eAAe,OAAsB,IAAI;CAC/C,MAAM,iBAAiB,OAAO,QAAQ,WAAW;CACjD,eAAe,UAAU,QAAQ;CAGjC,MAAM,CAAC,QAAQ,aAAa,SAAsB,MAAM;CACxD,MAAM,CAAC,YAAY,iBAAiB,SAA8B,CAAC,CAAC;CACpE,MAAM,CAAC,SAAS,cAAc,SAAsC,IAAI;CACxE,MAAM,CAAC,aAAa,kBAAkB,SAAkC,IAAI;CAC5E,MAAM,CAAC,YAAY,iBAAiB,SAAS,CAAC;CAC9C,MAAM,CAAC,SAAS,cAAc,SAAS,KAAK;CAC5C,MAAM,CAAC,WAAW,gBAAgB,SAAS,KAAK;CAChD,MAAM,CAAC,OAAO,YAAY,SAAwB,IAAI;CACtD,MAAM,CAAC,mBAAmB,wBAAwB,SAChD,IACF;CACA,MAAM,CAAC,mBAAmB,wBAAwB,SAChD,IACF;CACA,MAAM,CAAC,mBAAmB,wBAAwB,SAAkB,IAAI;CAGxE,gBAAgB;EAEd,UAAU,MAAM;EAChB,cAAc,CAAC,CAAC;EAChB,WAAW,IAAI;EACf,eAAe,IAAI;EACnB,cAAc,CAAC;EACf,WAAW,KAAK;EAChB,aAAa,KAAK;EAClB,SAAS,IAAI;EACb,qBAAqB,IAAI;EACzB,qBAAqB,IAAI;EACzB,qBAAqB,IAAI;EAEzB,IAAI,CAAC,SAAS;GACZ,UAAU,UAAU;GACpB,aAAa,UAAU;GACvB;EACF;EAEA,MAAM,cACJ,aAAa,YAAY,QAAQ,aAAa,YAAY;EAC5D,aAAa,UAAU;EAGvB,IAAI,aACF,eAAe,UAAU;EAG3B,MAAM,EACJ,SAAS,UACT,aAAa,cACb,GAAG,kBACD;EACJ,MAAM,SAAS,IAAI,YAAY,aAAa;EAC5C,UAAU,UAAU;EACpB,OAAO,QAAQ;EAGf,MAAM,YAAY,MAAmB,UAAU,CAAC;EAChD,MAAM,gBAAgB,MAA2B,cAAc,CAAC;EAChE,MAAM,aAAa,MAAmC,WAAW,CAAC;EAClE,MAAM,iBAAiB,UAA4B,eAAe,KAAK;EACvE,MAAM,gBAAgB,UAAkB,cAAc,KAAK;EAC3D,MAAM,UAAU,UAAmB,WAAW,KAAK;EACnD,MAAM,gBAAgB,MAAe,aAAa,CAAC;EACnD,MAAM,WAAW,MAAqB,SAAS,CAAC;EAChD,MAAM,uBAAuB,MAAqB,qBAAqB,CAAC;EACxE,MAAM,aAAa,SAAwB,qBAAqB,IAAI;EAEpE,OAAO,iBAAiB,gBAAgB,QAAQ;EAChD,OAAO,iBAAiB,oBAAoB,YAAY;EACxD,OAAO,iBAAiB,qBAAqB,SAAS;EACtD,OAAO,iBAAiB,iBAAiB,SAAS;EAClD,OAAO,iBAAiB,eAAe,aAAa;EACpD,OAAO,iBAAiB,oBAAoB,YAAY;EACxD,OAAO,iBAAiB,cAAc,MAAM;EAC5C,OAAO,iBAAiB,oBAAoB,YAAY;EACxD,OAAO,iBAAiB,SAAS,OAAO;EACxC,OAAO,iBAAiB,qBAAqB,mBAAmB;EAEhE,aAAa;GACX,OAAO,oBAAoB,gBAAgB,QAAQ;GACnD,OAAO,oBAAoB,oBAAoB,YAAY;GAC3D,OAAO,oBAAoB,qBAAqB,SAAS;GACzD,OAAO,oBAAoB,iBAAiB,SAAS;GACrD,OAAO,oBAAoB,eAAe,aAAa;GACvD,OAAO,oBAAoB,oBAAoB,YAAY;GAC3D,OAAO,oBAAoB,cAAc,MAAM;GAC/C,OAAO,oBAAoB,oBAAoB,YAAY;GAC3D,OAAO,oBAAoB,SAAS,OAAO;GAC3C,OAAO,oBAAoB,qBAAqB,mBAAmB;GACnE,IAAI,UAAU,YAAY,QACxB,UAAU,UAAU;GAEtB,OAAO,WAAW;EACpB;CAEF,GAAG,CAAC,SAAS,CAAC;CAGd,MAAM,YAAY,kBACV,UAAU,SAAS,UAAU,KAAK,QAAQ,QAAQ,GACxD,CAAC,CACH;CACA,MAAM,UAAU,kBAAkB,UAAU,SAAS,QAAQ,GAAG,CAAC,CAAC;CAClE,MAAM,aAAa,kBAAkB,UAAU,SAAS,WAAW,GAAG,CAAC,CAAC;CACxE,MAAM,WAAW,aACd,SAAiB,UAAU,SAAS,SAAS,IAAI,GAClD,CAAC,CACH;CACA,MAAM,WAAW,aACd,SAAkC,UAAU,SAAS,SAAS,IAAI,GACnE,CAAC,CACH;CAIA,gBAAgB;EACd,MAAM,SAAS,UAAU;EACzB,IAAI,CAAC,QAAQ;EACb,MAAM,YAAY,QAAiB,qBAAqB,GAAG;EAC3D,OAAO,iBAAiB,iBAAiB,QAAQ;EACjD,aAAa,OAAO,oBAAoB,iBAAiB,QAAQ;CACnE,GAAG,CAAC,SAAS,CAAC;CAEd,gBAAgB;EACd,UAAe,SAAS,gBAAgB,cAAc;CACxD,GAAG,CAAC,WAAW,cAAc,CAAC;CAE9B,OAAO;EACL;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;CACF;AACF"}

package/dist/voice/sfu.d.ts

@@ -0,0 +1,71 @@
+//#region src/voice/sfu.d.ts
+/**
+ * Pure utility functions for the Cloudflare Realtime SFU integration.
+ *
+ * Extracted from sfu.ts for testability. These handle:
+ * - Protobuf varint encoding/decoding
+ * - SFU WebSocket adapter protobuf packet encoding/decoding
+ * - Audio format conversion (48kHz stereo ↔ 16kHz mono)
+ */
+declare function decodeVarint(
+  buf: Uint8Array,
+  offset: number
+): {
+  value: number;
+  bytesRead: number;
+};
+declare function encodeVarint(value: number): Uint8Array;
+/** Extract the PCM payload from a protobuf Packet message. */
+declare function extractPayloadFromProtobuf(
+  data: ArrayBuffer
+): Uint8Array | null;
+/** Encode PCM payload into a protobuf Packet message (for ingest/buffer mode — just payload). */
+declare function encodePayloadToProtobuf(payload: Uint8Array): ArrayBuffer;
+/** Downsample 48kHz stereo interleaved PCM to 16kHz mono PCM (both 16-bit LE). */
+declare function downsample48kStereoTo16kMono(
+  stereo48k: Uint8Array
+): ArrayBuffer;
+/** Upsample 16kHz mono PCM to 48kHz stereo interleaved PCM (both 16-bit LE). */
+declare function upsample16kMonoTo48kStereo(mono16k: ArrayBuffer): Uint8Array;
+interface SFUConfig {
+  appId: string;
+  apiToken: string;
+}
+declare function sfuFetch(
+  config: SFUConfig,
+  path: string,
+  body: unknown
+): Promise<unknown>;
+declare function createSFUSession(config: SFUConfig): Promise<{
+  sessionId: string;
+}>;
+declare function addSFUTracks(
+  config: SFUConfig,
+  sessionId: string,
+  body: unknown
+): Promise<unknown>;
+declare function renegotiateSFUSession(
+  config: SFUConfig,
+  sessionId: string,
+  sdp: string
+): Promise<unknown>;
+declare function createSFUWebSocketAdapter(
+  config: SFUConfig,
+  tracks: unknown[]
+): Promise<unknown>;
+//#endregion
+export {
+  SFUConfig,
+  addSFUTracks,
+  createSFUSession,
+  createSFUWebSocketAdapter,
+  decodeVarint,
+  downsample48kStereoTo16kMono,
+  encodePayloadToProtobuf,
+  encodeVarint,
+  extractPayloadFromProtobuf,
+  renegotiateSFUSession,
+  sfuFetch,
+  upsample16kMonoTo48kStereo
+};
+//# sourceMappingURL=sfu.d.ts.map

package/dist/voice/sfu.js

@@ -0,0 +1,157 @@
+//#region src/voice/sfu.ts
+/**
+* Pure utility functions for the Cloudflare Realtime SFU integration.
+*
+* Extracted from sfu.ts for testability. These handle:
+* - Protobuf varint encoding/decoding
+* - SFU WebSocket adapter protobuf packet encoding/decoding
+* - Audio format conversion (48kHz stereo ↔ 16kHz mono)
+*/
+function decodeVarint(buf, offset) {
+	let value = 0;
+	let shift = 0;
+	let bytesRead = 0;
+	while (offset + bytesRead < buf.length) {
+		const byte = buf[offset + bytesRead];
+		value |= (byte & 127) << shift;
+		bytesRead++;
+		if ((byte & 128) === 0) break;
+		shift += 7;
+	}
+	return {
+		value,
+		bytesRead
+	};
+}
+function encodeVarint(value) {
+	const bytes = [];
+	while (value > 127) {
+		bytes.push(value & 127 | 128);
+		value >>>= 7;
+	}
+	bytes.push(value & 127);
+	return new Uint8Array(bytes);
+}
+/** Extract the PCM payload from a protobuf Packet message. */
+function extractPayloadFromProtobuf(data) {
+	const buf = new Uint8Array(data);
+	let offset = 0;
+	while (offset < buf.length) {
+		const { value: tag, bytesRead: tagBytes } = decodeVarint(buf, offset);
+		offset += tagBytes;
+		const fieldNumber = tag >>> 3;
+		const wireType = tag & 7;
+		if (wireType === 0) {
+			const { bytesRead } = decodeVarint(buf, offset);
+			offset += bytesRead;
+		} else if (wireType === 2) {
+			const { value: length, bytesRead: lenBytes } = decodeVarint(buf, offset);
+			offset += lenBytes;
+			if (fieldNumber === 5) return buf.slice(offset, offset + length);
+			offset += length;
+		} else break;
+	}
+	return null;
+}
+/** Encode PCM payload into a protobuf Packet message (for ingest/buffer mode — just payload). */
+function encodePayloadToProtobuf(payload) {
+	const tagBytes = encodeVarint(42);
+	const lengthBytes = encodeVarint(payload.length);
+	const result = new Uint8Array(tagBytes.length + lengthBytes.length + payload.length);
+	result.set(tagBytes, 0);
+	result.set(lengthBytes, tagBytes.length);
+	result.set(payload, tagBytes.length + lengthBytes.length);
+	return result.buffer;
+}
+/** Downsample 48kHz stereo interleaved PCM to 16kHz mono PCM (both 16-bit LE). */
+function downsample48kStereoTo16kMono(stereo48k) {
+	const inputView = new DataView(stereo48k.buffer, stereo48k.byteOffset, stereo48k.byteLength);
+	const inputSamples = stereo48k.byteLength / 4;
+	const outputSamples = Math.floor(inputSamples / 3);
+	const output = /* @__PURE__ */ new ArrayBuffer(outputSamples * 2);
+	const outputView = new DataView(output);
+	for (let i = 0; i < outputSamples; i++) {
+		const srcOffset = i * 3 * 4;
+		if (srcOffset + 3 >= stereo48k.byteLength) break;
+		const left = inputView.getInt16(srcOffset, true);
+		const right = inputView.getInt16(srcOffset + 2, true);
+		const mono = Math.round((left + right) / 2);
+		outputView.setInt16(i * 2, mono, true);
+	}
+	return output;
+}
+/** Upsample 16kHz mono PCM to 48kHz stereo interleaved PCM (both 16-bit LE). */
+function upsample16kMonoTo48kStereo(mono16k) {
+	const inputView = new DataView(mono16k);
+	const inputSamples = mono16k.byteLength / 2;
+	const outputSamples = inputSamples * 3;
+	const output = /* @__PURE__ */ new ArrayBuffer(outputSamples * 4);
+	const outputView = new DataView(output);
+	for (let i = 0; i < inputSamples; i++) {
+		const sample = inputView.getInt16(i * 2, true);
+		for (let j = 0; j < 3; j++) {
+			const outOffset = (i * 3 + j) * 4;
+			outputView.setInt16(outOffset, sample, true);
+			outputView.setInt16(outOffset + 2, sample, true);
+		}
+	}
+	return new Uint8Array(output);
+}
+const SFU_API_BASE = "https://rtc.live.cloudflare.com/v1";
+async function sfuFetch(config, path, body) {
+	const url = `${SFU_API_BASE}/apps/${config.appId}${path}`;
+	const response = await fetch(url, {
+		method: "POST",
+		headers: {
+			Authorization: `Bearer ${config.apiToken}`,
+			"Content-Type": "application/json"
+		},
+		body: JSON.stringify(body)
+	});
+	if (!response.ok) {
+		const text = await response.text();
+		throw new Error(`SFU API error ${response.status}: ${text}`);
+	}
+	return response.json();
+}
+async function createSFUSession(config) {
+	const url = `${SFU_API_BASE}/apps/${config.appId}/sessions/new`;
+	const response = await fetch(url, {
+		method: "POST",
+		headers: { Authorization: `Bearer ${config.apiToken}` }
+	});
+	if (!response.ok) {
+		const text = await response.text();
+		throw new Error(`SFU API error ${response.status}: ${text}`);
+	}
+	return response.json();
+}
+async function addSFUTracks(config, sessionId, body) {
+	return sfuFetch(config, `/sessions/${sessionId}/tracks/new`, body);
+}
+async function renegotiateSFUSession(config, sessionId, sdp) {
+	const url = `${SFU_API_BASE}/apps/${config.appId}/sessions/${sessionId}/renegotiate`;
+	const response = await fetch(url, {
+		method: "PUT",
+		headers: {
+			Authorization: `Bearer ${config.apiToken}`,
+			"Content-Type": "application/json"
+		},
+		body: JSON.stringify({ sessionDescription: {
+			type: "answer",
+			sdp
+		} })
+	});
+	if (!response.ok) {
+		const text = await response.text();
+		throw new Error(`SFU renegotiate error ${response.status}: ${text}`);
+	}
+	return response.json();
+}
+async function createSFUWebSocketAdapter(config, tracks) {
+	return sfuFetch(config, "/adapters/websocket/new", { tracks });
+}
+//#endregion
+export { addSFUTracks, createSFUSession, createSFUWebSocketAdapter, decodeVarint, downsample48kStereoTo16kMono, encodePayloadToProtobuf, encodeVarint, extractPayloadFromProtobuf, renegotiateSFUSession, sfuFetch, upsample16kMonoTo48kStereo };
+
+//# sourceMappingURL=sfu.js.map

package/dist/voice/sfu.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sfu.js","names":[],"sources":["../../src/voice/sfu.ts"],"sourcesContent":["/**\n * Pure utility functions for the Cloudflare Realtime SFU integration.\n *\n * Extracted from sfu.ts for testability. These handle:\n * - Protobuf varint encoding/decoding\n * - SFU WebSocket adapter protobuf packet encoding/decoding\n * - Audio format conversion (48kHz stereo ↔ 16kHz mono)\n */\n\n// --- Protobuf helpers ---\n// The SFU WebSocket adapter uses a simple protobuf message:\n//   message Packet {\n//     uint32 sequenceNumber = 1;\n//     uint32 timestamp = 2;\n//     bytes payload = 5;\n//   }\n\nexport function decodeVarint(\n  buf: Uint8Array,\n  offset: number\n): { value: number; bytesRead: number } {\n  let value = 0;\n  let shift = 0;\n  let bytesRead = 0;\n  while (offset + bytesRead < buf.length) {\n    const byte = buf[offset + bytesRead];\n    value |= (byte & 0x7f) << shift;\n    bytesRead++;\n    if ((byte & 0x80) === 0) break;\n    shift += 7;\n  }\n  return { value, bytesRead };\n}\n\nexport function encodeVarint(value: number): Uint8Array {\n  const bytes: number[] = [];\n  while (value > 0x7f) {\n    bytes.push((value & 0x7f) | 0x80);\n    value >>>= 7;\n  }\n  bytes.push(value & 0x7f);\n  return new Uint8Array(bytes);\n}\n\n/** Extract the PCM payload from a protobuf Packet message. */\nexport function extractPayloadFromProtobuf(\n  data: ArrayBuffer\n): Uint8Array | null {\n  const buf = new Uint8Array(data);\n  let offset = 0;\n\n  while (offset < buf.length) {\n    const { value: tag, bytesRead: tagBytes } = decodeVarint(buf, offset);\n    offset += tagBytes;\n\n    const fieldNumber = tag >>> 3;\n    const wireType = tag & 0x07;\n\n    if (wireType === 0) {\n      // Varint\n      const { bytesRead } = decodeVarint(buf, offset);\n      offset += bytesRead;\n    } else if (wireType === 2) {\n      // Length-delimited (bytes)\n      const { value: length, bytesRead: lenBytes } = decodeVarint(buf, offset);\n      offset += lenBytes;\n\n      if (fieldNumber === 5) {\n        // payload field\n        return buf.slice(offset, offset + length);\n      }\n      offset += length;\n    } else {\n      // Unknown wire type — skip\n      break;\n    }\n  }\n\n  return null;\n}\n\n/** Encode PCM payload into a protobuf Packet message (for ingest/buffer mode — just payload). */\nexport function encodePayloadToProtobuf(payload: Uint8Array): ArrayBuffer {\n  // Field 5, wire type 2 (length-delimited): tag = (5 << 3) | 2 = 42\n  const tagBytes = encodeVarint(42);\n  const lengthBytes = encodeVarint(payload.length);\n\n  const result = new Uint8Array(\n    tagBytes.length + lengthBytes.length + payload.length\n  );\n  result.set(tagBytes, 0);\n  result.set(lengthBytes, tagBytes.length);\n  result.set(payload, tagBytes.length + lengthBytes.length);\n\n  return result.buffer;\n}\n\n// --- Audio conversion ---\n\n/** Downsample 48kHz stereo interleaved PCM to 16kHz mono PCM (both 16-bit LE). */\nexport function downsample48kStereoTo16kMono(\n  stereo48k: Uint8Array\n): ArrayBuffer {\n  // Input: 48kHz stereo 16-bit LE → 2 channels × 2 bytes = 4 bytes per sample pair\n  // Output: 16kHz mono 16-bit LE → 2 bytes per sample\n  // Ratio: 48000/16000 = 3, plus stereo→mono = average of L+R\n\n  const inputView = new DataView(\n    stereo48k.buffer,\n    stereo48k.byteOffset,\n    stereo48k.byteLength\n  );\n  const inputSamples = stereo48k.byteLength / 4; // stereo sample pairs\n  const outputSamples = Math.floor(inputSamples / 3);\n  const output = new ArrayBuffer(outputSamples * 2);\n  const outputView = new DataView(output);\n\n  for (let i = 0; i < outputSamples; i++) {\n    const srcOffset = i * 3 * 4; // 3x downsample, 4 bytes per stereo pair\n    if (srcOffset + 3 >= stereo48k.byteLength) break;\n    const left = inputView.getInt16(srcOffset, true);\n    const right = inputView.getInt16(srcOffset + 2, true);\n    const mono = Math.round((left + right) / 2);\n    outputView.setInt16(i * 2, mono, true);\n  }\n\n  return output;\n}\n\n/** Upsample 16kHz mono PCM to 48kHz stereo interleaved PCM (both 16-bit LE). */\nexport function upsample16kMonoTo48kStereo(mono16k: ArrayBuffer): Uint8Array {\n  const inputView = new DataView(mono16k);\n  const inputSamples = mono16k.byteLength / 2;\n  const outputSamples = inputSamples * 3; // 3x upsample\n  const output = new ArrayBuffer(outputSamples * 4); // stereo = 4 bytes per pair\n  const outputView = new DataView(output);\n\n  for (let i = 0; i < inputSamples; i++) {\n    const sample = inputView.getInt16(i * 2, true);\n    // Write 3 stereo samples (simple sample duplication)\n    for (let j = 0; j < 3; j++) {\n      const outOffset = (i * 3 + j) * 4;\n      outputView.setInt16(outOffset, sample, true); // left\n      outputView.setInt16(outOffset + 2, sample, true); // right\n    }\n  }\n\n  return new Uint8Array(output);\n}\n\n// --- SFU API helpers ---\n\nexport interface SFUConfig {\n  appId: string;\n  apiToken: string;\n}\n\nconst SFU_API_BASE = \"https://rtc.live.cloudflare.com/v1\";\n\nexport async function sfuFetch(\n  config: SFUConfig,\n  path: string,\n  body: unknown\n): Promise<unknown> {\n  const url = `${SFU_API_BASE}/apps/${config.appId}${path}`;\n  const response = await fetch(url, {\n    method: \"POST\",\n    headers: {\n      Authorization: `Bearer ${config.apiToken}`,\n      \"Content-Type\": \"application/json\"\n    },\n    body: JSON.stringify(body)\n  });\n  if (!response.ok) {\n    const text = await response.text();\n    throw new Error(`SFU API error ${response.status}: ${text}`);\n  }\n  return response.json();\n}\n\nexport async function createSFUSession(\n  config: SFUConfig\n): Promise<{ sessionId: string }> {\n  const url = `${SFU_API_BASE}/apps/${config.appId}/sessions/new`;\n  const response = await fetch(url, {\n    method: \"POST\",\n    headers: {\n      Authorization: `Bearer ${config.apiToken}`\n    }\n  });\n  if (!response.ok) {\n    const text = await response.text();\n    throw new Error(`SFU API error ${response.status}: ${text}`);\n  }\n  return response.json() as Promise<{ sessionId: string }>;\n}\n\nexport async function addSFUTracks(\n  config: SFUConfig,\n  sessionId: string,\n  body: unknown\n): Promise<unknown> {\n  return sfuFetch(config, `/sessions/${sessionId}/tracks/new`, body);\n}\n\nexport async function renegotiateSFUSession(\n  config: SFUConfig,\n  sessionId: string,\n  sdp: string\n): Promise<unknown> {\n  const url = `${SFU_API_BASE}/apps/${config.appId}/sessions/${sessionId}/renegotiate`;\n  const response = await fetch(url, {\n    method: \"PUT\",\n    headers: {\n      Authorization: `Bearer ${config.apiToken}`,\n      \"Content-Type\": \"application/json\"\n    },\n    body: JSON.stringify({\n      sessionDescription: { type: \"answer\", sdp }\n    })\n  });\n  if (!response.ok) {\n    const text = await response.text();\n    throw new Error(`SFU renegotiate error ${response.status}: ${text}`);\n  }\n  return response.json();\n}\n\nexport async function createSFUWebSocketAdapter(\n  config: SFUConfig,\n  tracks: unknown[]\n): Promise<unknown> {\n  return sfuFetch(config, \"/adapters/websocket/new\", { tracks });\n}\n"],"mappings":";;;;;;;;;AAiBA,SAAgB,aACd,KACA,QACsC;CACtC,IAAI,QAAQ;CACZ,IAAI,QAAQ;CACZ,IAAI,YAAY;CAChB,OAAO,SAAS,YAAY,IAAI,QAAQ;EACtC,MAAM,OAAO,IAAI,SAAS;EAC1B,UAAU,OAAO,QAAS;EAC1B;EACA,KAAK,OAAO,SAAU,GAAG;EACzB,SAAS;CACX;CACA,OAAO;EAAE;EAAO;CAAU;AAC5B;AAEA,SAAgB,aAAa,OAA2B;CACtD,MAAM,QAAkB,CAAC;CACzB,OAAO,QAAQ,KAAM;EACnB,MAAM,KAAM,QAAQ,MAAQ,GAAI;EAChC,WAAW;CACb;CACA,MAAM,KAAK,QAAQ,GAAI;CACvB,OAAO,IAAI,WAAW,KAAK;AAC7B;;AAGA,SAAgB,2BACd,MACmB;CACnB,MAAM,MAAM,IAAI,WAAW,IAAI;CAC/B,IAAI,SAAS;CAEb,OAAO,SAAS,IAAI,QAAQ;EAC1B,MAAM,EAAE,OAAO,KAAK,WAAW,aAAa,aAAa,KAAK,MAAM;EACpE,UAAU;EAEV,MAAM,cAAc,QAAQ;EAC5B,MAAM,WAAW,MAAM;EAEvB,IAAI,aAAa,GAAG;GAElB,MAAM,EAAE,cAAc,aAAa,KAAK,MAAM;GAC9C,UAAU;EACZ,OAAO,IAAI,aAAa,GAAG;GAEzB,MAAM,EAAE,OAAO,QAAQ,WAAW,aAAa,aAAa,KAAK,MAAM;GACvE,UAAU;GAEV,IAAI,gBAAgB,GAElB,OAAO,IAAI,MAAM,QAAQ,SAAS,MAAM;GAE1C,UAAU;EACZ,OAEE;CAEJ;CAEA,OAAO;AACT;;AAGA,SAAgB,wBAAwB,SAAkC;CAExE,MAAM,WAAW,aAAa,EAAE;CAChC,MAAM,cAAc,aAAa,QAAQ,MAAM;CAE/C,MAAM,SAAS,IAAI,WACjB,SAAS,SAAS,YAAY,SAAS,QAAQ,MACjD;CACA,OAAO,IAAI,UAAU,CAAC;CACtB,OAAO,IAAI,aAAa,SAAS,MAAM;CACvC,OAAO,IAAI,SAAS,SAAS,SAAS,YAAY,MAAM;CAExD,OAAO,OAAO;AAChB;;AAKA,SAAgB,6BACd,WACa;CAKb,MAAM,YAAY,IAAI,SACpB,UAAU,QACV,UAAU,YACV,UAAU,UACZ;CACA,MAAM,eAAe,UAAU,aAAa;CAC5C,MAAM,gBAAgB,KAAK,MAAM,eAAe,CAAC;CACjD,MAAM,yBAAS,IAAI,YAAY,gBAAgB,CAAC;CAChD,MAAM,aAAa,IAAI,SAAS,MAAM;CAEtC,KAAK,IAAI,IAAI,GAAG,IAAI,eAAe,KAAK;EACtC,MAAM,YAAY,IAAI,IAAI;EAC1B,IAAI,YAAY,KAAK,UAAU,YAAY;EAC3C,MAAM,OAAO,UAAU,SAAS,WAAW,IAAI;EAC/C,MAAM,QAAQ,UAAU,SAAS,YAAY,GAAG,IAAI;EACpD,MAAM,OAAO,KAAK,OAAO,OAAO,SAAS,CAAC;EAC1C,WAAW,SAAS,IAAI,GAAG,MAAM,IAAI;CACvC;CAEA,OAAO;AACT;;AAGA,SAAgB,2BAA2B,SAAkC;CAC3E,MAAM,YAAY,IAAI,SAAS,OAAO;CACtC,MAAM,eAAe,QAAQ,aAAa;CAC1C,MAAM,gBAAgB,eAAe;CACrC,MAAM,yBAAS,IAAI,YAAY,gBAAgB,CAAC;CAChD,MAAM,aAAa,IAAI,SAAS,MAAM;CAEtC,KAAK,IAAI,IAAI,GAAG,IAAI,cAAc,KAAK;EACrC,MAAM,SAAS,UAAU,SAAS,IAAI,GAAG,IAAI;EAE7C,KAAK,IAAI,IAAI,GAAG,IAAI,GAAG,KAAK;GAC1B,MAAM,aAAa,IAAI,IAAI,KAAK;GAChC,WAAW,SAAS,WAAW,QAAQ,IAAI;GAC3C,WAAW,SAAS,YAAY,GAAG,QAAQ,IAAI;EACjD;CACF;CAEA,OAAO,IAAI,WAAW,MAAM;AAC9B;AASA,MAAM,eAAe;AAErB,eAAsB,SACpB,QACA,MACA,MACkB;CAClB,MAAM,MAAM,GAAG,aAAa,QAAQ,OAAO,QAAQ;CACnD,MAAM,WAAW,MAAM,MAAM,KAAK;EAChC,QAAQ;EACR,SAAS;GACP,eAAe,UAAU,OAAO;GAChC,gBAAgB;EAClB;EACA,MAAM,KAAK,UAAU,IAAI;CAC3B,CAAC;CACD,IAAI,CAAC,SAAS,IAAI;EAChB,MAAM,OAAO,MAAM,SAAS,KAAK;EACjC,MAAM,IAAI,MAAM,iBAAiB,SAAS,OAAO,IAAI,MAAM;CAC7D;CACA,OAAO,SAAS,KAAK;AACvB;AAEA,eAAsB,iBACpB,QACgC;CAChC,MAAM,MAAM,GAAG,aAAa,QAAQ,OAAO,MAAM;CACjD,MAAM,WAAW,MAAM,MAAM,KAAK;EAChC,QAAQ;EACR,SAAS,EACP,eAAe,UAAU,OAAO,WAClC;CACF,CAAC;CACD,IAAI,CAAC,SAAS,IAAI;EAChB,MAAM,OAAO,MAAM,SAAS,KAAK;EACjC,MAAM,IAAI,MAAM,iBAAiB,SAAS,OAAO,IAAI,MAAM;CAC7D;CACA,OAAO,SAAS,KAAK;AACvB;AAEA,eAAsB,aACpB,QACA,WACA,MACkB;CAClB,OAAO,SAAS,QAAQ,aAAa,UAAU,cAAc,IAAI;AACnE;AAEA,eAAsB,sBACpB,QACA,WACA,KACkB;CAClB,MAAM,MAAM,GAAG,aAAa,QAAQ,OAAO,MAAM,YAAY,UAAU;CACvE,MAAM,WAAW,MAAM,MAAM,KAAK;EAChC,QAAQ;EACR,SAAS;GACP,eAAe,UAAU,OAAO;GAChC,gBAAgB;EAClB;EACA,MAAM,KAAK,UAAU,EACnB,oBAAoB;GAAE,MAAM;GAAU;EAAI,EAC5C,CAAC;CACH,CAAC;CACD,IAAI,CAAC,SAAS,IAAI;EAChB,MAAM,OAAO,MAAM,SAAS,KAAK;EACjC,MAAM,IAAI,MAAM,yBAAyB,SAAS,OAAO,IAAI,MAAM;CACrE;CACA,OAAO,SAAS,KAAK;AACvB;AAEA,eAAsB,0BACpB,QACA,QACkB;CAClB,OAAO,SAAS,QAAQ,2BAA2B,EAAE,OAAO,CAAC;AAC/D"}

package/dist/voice/text.d.ts

@@ -0,0 +1,6 @@
+import {
+  n as TextSource,
+  r as iterateText,
+  t as SentenceChunker
+} from "../sentence-chunker-BAidJ4DA.js";
+export { SentenceChunker, type TextSource, iterateText };

package/dist/voice/text.js

@@ -0,0 +1,2 @@
+import { r as SentenceChunker, t as iterateText } from "../text-stream-CpdiKrJB.js";
+export { SentenceChunker, iterateText };

package/dist/voice/types.d.ts

@@ -0,0 +1,58 @@
+import {
+  C as VoiceTransport,
+  D as VoiceTurnSource,
+  E as VoiceTurnOutcome,
+  S as VoiceStatus,
+  T as VoiceTurnMetrics,
+  _ as VoiceErrorStage,
+  a as TranscriberSessionOptions,
+  b as VoiceRole,
+  c as VoiceAudioFormat,
+  d as VoiceCompletionOutcome,
+  f as VoiceCompletionOutcomeCode,
+  g as VoiceErrorCode,
+  h as VoiceError,
+  i as TranscriberSession,
+  l as VoiceAudioInput,
+  m as VoiceDiagnosticsOptions,
+  n as TTSProvider,
+  o as TranscriptMessage,
+  p as VoiceDiagnosticEvent,
+  r as Transcriber,
+  s as VOICE_PROTOCOL_VERSION,
+  t as StreamingTTSProvider,
+  u as VoiceClientMessage,
+  v as VoiceModelFinishReason,
+  w as VoiceTransportCloseInfo,
+  x as VoiceServerMessage,
+  y as VoicePipelineMetrics
+} from "../types-_Faxb570.js";
+export {
+  StreamingTTSProvider,
+  TTSProvider,
+  Transcriber,
+  TranscriberSession,
+  TranscriberSessionOptions,
+  TranscriptMessage,
+  VOICE_PROTOCOL_VERSION,
+  VoiceAudioFormat,
+  VoiceAudioInput,
+  VoiceClientMessage,
+  VoiceCompletionOutcome,
+  VoiceCompletionOutcomeCode,
+  VoiceDiagnosticEvent,
+  VoiceDiagnosticsOptions,
+  VoiceError,
+  VoiceErrorCode,
+  VoiceErrorStage,
+  VoiceModelFinishReason,
+  VoicePipelineMetrics,
+  VoiceRole,
+  VoiceServerMessage,
+  VoiceStatus,
+  VoiceTransport,
+  VoiceTransportCloseInfo,
+  VoiceTurnMetrics,
+  VoiceTurnOutcome,
+  VoiceTurnSource
+};

package/dist/voice/types.js

@@ -0,0 +1,18 @@
+//#region src/voice/types.ts
+/**
+* Shared types for the voice pipeline.
+*
+* Used by both the server (index.ts) and client (client.ts)
+* to ensure protocol consistency.
+*/
+/**
+* Current voice protocol version.
+* Bump this when making backwards-incompatible wire protocol changes.
+* The server sends this in the initial `welcome` message so clients
+* can detect version mismatches.
+*/
+const VOICE_PROTOCOL_VERSION = 1;
+//#endregion
+export { VOICE_PROTOCOL_VERSION };
+
+//# sourceMappingURL=types.js.map

package/dist/voice/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","names":[],"sources":["../../src/voice/types.ts"],"sourcesContent":["/**\n * Shared types for the voice pipeline.\n *\n * Used by both the server (index.ts) and client (client.ts)\n * to ensure protocol consistency.\n */\n\n// --- Protocol version ---\n\n/**\n * Current voice protocol version.\n * Bump this when making backwards-incompatible wire protocol changes.\n * The server sends this in the initial `welcome` message so clients\n * can detect version mismatches.\n */\nexport const VOICE_PROTOCOL_VERSION = 1;\n\n// --- Voice status ---\n\nexport type VoiceStatus = \"idle\" | \"listening\" | \"thinking\" | \"speaking\";\n\n// --- Diagnostics ---\n\n/** Server-side diagnostic configuration shared by both voice mixins. */\nexport interface VoiceDiagnosticsOptions {\n  /** Forward safe server diagnostics and enable browser console logging. */\n  browserConsole?: boolean;\n}\n\n/**\n * A bounded diagnostic record forwarded by the voice protocol.\n * Event names and metadata are intentionally open and are not stable API.\n */\nexport interface VoiceDiagnosticEvent {\n  event: string;\n  timestamp: number;\n  data?: Record<string, unknown>;\n}\n\n// --- Stable per-turn metrics ---\n\nexport type VoiceTurnSource = \"speech\" | \"text\";\n\nexport type VoiceTurnOutcome =\n  | \"completed\"\n  | \"no_output\"\n  | \"output_limit\"\n  | \"content_filtered\"\n  | \"model_error\"\n  | \"tts_error\"\n  | \"aborted\"\n  | \"skipped\"\n  | \"error\";\n\n/**\n * Stable, content-free summary of one allocated voice turn.\n *\n * `turnId`, `source`, and `outcome` are dimensions used to correlate and\n * interpret the measurements; every other field is a duration. Optional\n * timings are omitted when their lifecycle landmark was not reached. Timings\n * use one server clock, overlap, and are not additive. Browser playback timing\n * is excluded because browser and Worker clocks are independent.\n */\nexport interface VoiceTurnMetrics {\n  /** SDK-assigned correlation ID for this turn. */\n  turnId: string;\n  /** Whether the turn originated from finalized speech or a text message. */\n  source: VoiceTurnSource;\n  /** Terminal result, emitted exactly once for the allocated turn. */\n  outcome: VoiceTurnOutcome;\n  /** Turn allocation to terminal summary, in milliseconds. */\n  turnTotalMs: number;\n  /** Provider speech start to the first interim transcript. */\n  speechStartToFirstInterimMs?: number;\n  /** Provider speech start to the finalized transcript. */\n  speechStartToFinalMs?: number;\n  /** Time spent in the server's `afterTranscribe` hook. */\n  afterTranscribeMs?: number;\n  /** Model invocation to the first non-whitespace text delta. */\n  modelToFirstTextMs?: number;\n  /** Cumulative duration of reasoning blocks exposed by the model stream. */\n  exposedReasoningMs?: number;\n  /** Model invocation through normalized stream consumption. */\n  modelStreamConsumptionMs?: number;\n  /** Finalized input through the first server audio send. */\n  finalInputToFirstAudioMs?: number;\n\n  /** First TTS provider invocation through the first server audio send. */\n  ttsToFirstAudioMs?: number;\n  /** First TTS provider invocation through completion of all sentence work. */\n  ttsWallMs?: number;\n  /** Cumulative overlapping TTS sentence hook and provider work. */\n  ttsWorkMs?: number;\n}\n\n// --- Audio format ---\n\n/** Audio format the server uses for binary audio payloads. */\nexport type VoiceAudioFormat = \"mp3\" | \"pcm16\" | \"wav\" | \"opus\";\n\n// --- Conversation message role ---\n\nexport type VoiceRole = \"user\" | \"assistant\";\n\n// --- Structured voice errors ---\n\n/** Stable machine-readable error codes emitted by the voice protocol. */\nexport type VoiceErrorCode = \"stt_startup_failed\" | \"stt_connection_lost\";\n\n/** Stable pipeline stage associated with a structured voice error. */\nexport type VoiceErrorStage = \"stt\";\n\n/** Client-safe error detail. `message` remains for string-event compatibility. */\nexport interface VoiceError {\n  message: string;\n  code?: VoiceErrorCode;\n  stage?: VoiceErrorStage;\n  retryable?: boolean;\n}\n\n// --- LLM completion outcomes ---\n\n/** Stable machine-readable outcomes for non-ordinary LLM completions. */\nexport type VoiceCompletionOutcomeCode =\n  | \"no_output\"\n  | \"output_limit\"\n  | \"content_filtered\"\n  | \"model_error\";\n\n/** Normalized finish reasons emitted by the supported AI SDK stream shape. */\nexport type VoiceModelFinishReason =\n  | \"stop\"\n  | \"length\"\n  | \"content-filter\"\n  | \"tool-calls\"\n  | \"error\"\n  | \"other\";\n\n/** Bounded completion metadata exposed by the voice protocol. */\nexport interface VoiceCompletionOutcome {\n  code: VoiceCompletionOutcomeCode;\n  stage: \"llm\";\n  finishReason?: VoiceModelFinishReason;\n  partialOutput: boolean;\n}\n\n// --- Wire protocol: Client → Server ---\n\nexport type VoiceClientMessage =\n  | { type: \"hello\"; protocol_version?: number }\n  | { type: \"start_call\"; preferred_format?: VoiceAudioFormat }\n  | { type: \"end_call\" }\n  | { type: \"start_of_speech\" }\n  | { type: \"end_of_speech\" }\n  | { type: \"interrupt\" }\n  | { type: \"text_message\"; text: string };\n\n// --- Wire protocol: Server → Client ---\n\nexport type VoiceServerMessage =\n  | {\n      type: \"welcome\";\n      protocol_version: number;\n      diagnostics?: { browser_console: true };\n    }\n  | ({ type: \"diagnostic\" } & VoiceDiagnosticEvent)\n  | { type: \"status\"; status: VoiceStatus }\n  | { type: \"audio_config\"; format: VoiceAudioFormat; sampleRate?: number }\n  | { type: \"transcript\"; role: VoiceRole; text: string }\n  | { type: \"transcript_start\"; role: VoiceRole }\n  | { type: \"transcript_delta\"; text: string }\n  | { type: \"transcript_end\"; text: string }\n  | { type: \"transcript_interim\"; text: string }\n  | { type: \"playback_interrupt\" }\n  | {\n      type: \"metrics\";\n      llm_ms: number;\n      tts_ms: number;\n      first_audio_ms: number;\n      total_ms: number;\n    }\n  | ({ type: \"turn_metrics\" } & VoiceTurnMetrics)\n  | ({ type: \"completion_outcome\" } & VoiceCompletionOutcome)\n  | ({ type: \"error\" } & VoiceError);\n\n// --- Pipeline metrics (structured form for consumers) ---\n\n/**\n * Compact compatibility summary for successful, non-empty speech turns.\n * These overlapping latency landmarks and work totals are not additive.\n * Use `VoiceTurnMetrics` for stable detailed timing and terminal summaries of\n * unsuccessful, aborted, skipped, or text turns.\n */\nexport interface VoicePipelineMetrics {\n  /**\n   * Time from immediately before `onTurn()` until normalized model-stream\n   * consumption completes. This can include tool work and consumer waits while\n   * consuming the stream. It is not time to first text.\n   */\n  llm_ms: number;\n  /**\n   * Cumulative per-sentence work from immediately before `beforeSynthesize`\n   * until that sentence's synthesis and hook work settles. Sentence work can\n   * overlap both other sentences and model consumption, so this value can\n   * exceed wall time and overlaps the other metrics.\n   */\n  tts_ms: number;\n  /**\n   * Time from turn-pipeline start, before `afterTranscribe`, to the first server\n   * audio send. Includes post-STT hooks, model work, and TTS, but excludes STT\n   * and browser playback. `0` means the server sent no audio.\n   */\n  first_audio_ms: number;\n  /**\n   * Time from the same turn-pipeline start until model consumption and TTS\n   * draining complete. Measured before final context, persistence, and status\n   * work. Excludes STT and browser playback.\n   */\n  total_ms: number;\n}\n\n// --- Transcript message (client-side enriched form) ---\n\nexport interface TranscriptMessage {\n  role: VoiceRole;\n  text: string;\n  timestamp: number;\n}\n\n// --- Provider interfaces ---\n\nexport interface TTSProvider {\n  synthesize(text: string, signal?: AbortSignal): Promise<ArrayBuffer | null>;\n}\n\nexport interface StreamingTTSProvider {\n  synthesizeStream(\n    text: string,\n    signal?: AbortSignal\n  ): AsyncGenerator<ArrayBuffer>;\n}\n\n// --- Transcriber (continuous per-call STT) ---\n\n/**\n * Continuous speech-to-text provider.\n *\n * Creates a per-call session that receives audio continuously from\n * `start_call` to `end_call`. The model handles turn detection\n * internally — there is no client-side speech boundary signaling\n * required for STT.\n *\n * The session fires `onUtterance` when the model detects a complete\n * utterance (e.g. Flux `EndOfTurn`, Nova 3 `speech_final` +\n * endpointing). The voice pipeline maps this to `onTurn` (withVoice)\n * or `onTranscript` (withVoiceInput).\n */\nexport interface Transcriber {\n  /** Create a new transcription session for one call. */\n  createSession(options?: TranscriberSessionOptions): TranscriberSession;\n}\n\nexport interface TranscriberSessionOptions {\n  /** Language code (e.g. \"en\"). */\n  language?: string;\n  /**\n   * Called when the provider produces an interim (unstable) transcript.\n   * This text may change as more audio arrives.\n   */\n  onInterim?: (text: string) => void;\n  /**\n   * Called when the model detects the start of user speech.\n   *\n   * Providers can use this for low-latency barge-in before a final\n   * utterance is available. The transcript may be omitted or unstable.\n   */\n  onSpeechStart?: (text?: string) => void;\n  /**\n   * Called when the model detects a complete utterance.\n   * The transcript is the stable text for this turn.\n   *\n   * For Flux: fires on `EndOfTurn`.\n   * For Nova 3: fires on `Results` with `speech_final: true`.\n   */\n  onUtterance?: (transcript: string) => void;\n  /**\n   * Called when the session can no longer transcribe because its provider\n   * connection failed or closed unexpectedly. Providers must not call this\n   * for teardown initiated by {@link TranscriberSession.close}.\n   */\n  onFatalError?: (error: Error) => void;\n}\n\n/**\n * A per-call transcription session. Lives for the entire call duration.\n *\n * Unlike per-utterance sessions, this session is never finished or\n * aborted mid-call. It receives all audio continuously and the model\n * handles speech boundary detection. On interrupt, the LLM+TTS\n * pipeline is aborted but the transcriber session stays alive.\n */\nexport interface TranscriberSession {\n  /**\n   * Feed raw PCM audio (16kHz mono 16-bit LE).\n   * Fire-and-forget — the session buffers internally as needed.\n   */\n  feed(chunk: ArrayBuffer): void;\n\n  /**\n   * Resolves when the session is ready to accept audio and emit transcripts.\n   * Optional so existing custom transcribers can start synchronously.\n   */\n  waitUntilReady?(): Promise<void>;\n\n  /**\n   * Optional. Provide the agent's most recent spoken reply (the text sent to\n   * TTS) as conversational context for the next user turn.\n   *\n   * The pipeline calls this after the agent finishes speaking each reply and\n   * greeting. Providers that support context carryover (e.g. AssemblyAI's\n   * `agent_context`) use it to better recognize short or contextual answers\n   * (\"yes\", \"7pm\", an email spelled aloud). Providers that don't support it\n   * simply omit this method — it is a no-op for them.\n   */\n  updateAgentContext?(text: string): void;\n\n  /**\n   * Close the session and release resources.\n   * Called at end_call or disconnect — not on interrupt.\n   */\n  close(): void;\n}\n\n// --- Audio input ---\n\n/**\n * Pluggable audio input source for VoiceClient.\n *\n * When provided via `VoiceClientOptions.audioInput`, VoiceClient delegates\n * mic capture to this object instead of using its built-in AudioWorklet.\n * The audio input is responsible for capturing audio and routing it to the\n * server (however it chooses — WebRTC, SFU, direct binary, etc.).\n *\n * It must call `onAudioLevel` with RMS values so VoiceClient can run\n * silence detection, interrupt detection, and update the audio level UI.\n *\n * @example\n * ```typescript\n * class SFUAudioInput implements VoiceAudioInput {\n *   onAudioLevel: ((rms: number) => void) | null = null;\n *   async start() {\n *     // Set up WebRTC peer connection, SFU session, etc.\n *     // In a monitoring loop, call this.onAudioLevel?.(rms)\n *   }\n *   stop() {\n *     // Tear down WebRTC\n *   }\n * }\n * ```\n */\nexport interface VoiceAudioInput {\n  /** Start capturing audio. Called by VoiceClient on startCall(). */\n  start(): Promise<void>;\n  /** Stop capturing audio. Called by VoiceClient on endCall() or disconnect(). */\n  stop(): void;\n  /**\n   * Set by VoiceClient before start(). The audio input must call this\n   * with RMS audio level values on each frame so VoiceClient can run\n   * silence detection, interrupt detection, and update the UI.\n   */\n  onAudioLevel: ((rms: number) => void) | null;\n\n  /**\n   * Set by VoiceClient before start(). If the audio input provides\n   * raw PCM audio (16kHz mono 16-bit LE), call this callback and\n   * VoiceClient will forward the data to the server via its transport.\n   *\n   * This is needed when audio reaches the server through the same\n   * WebSocket as protocol messages (e.g. SFU in local dev where the\n   * SFU adapter can't connect back to localhost).\n   *\n   * If the audio input routes audio to the server through an external\n   * path (e.g. SFU WebSocket adapter in production), this can be left\n   * unused — the audio will arrive on a separate connection.\n   */\n  onAudioData?: ((pcm: ArrayBuffer) => void) | null;\n}\n\n// --- Voice transport ---\n\n/** Details a transport can provide when its connection closes. */\nexport interface VoiceTransportCloseInfo {\n  code?: number;\n  reason?: string;\n  wasClean?: boolean;\n}\n\n/**\n * Abstraction over the data channel between client and server.\n * The default implementation wraps PartySocket (WebSocket).\n * Implement this interface to use WebRTC, SFU, or other transports.\n */\nexport interface VoiceTransport {\n  /** Send a JSON-serializable message to the server. */\n  sendJSON(data: Record<string, unknown>): void;\n  /** Send raw binary audio to the server. */\n  sendBinary(data: ArrayBuffer): void;\n\n  /** Open the connection. */\n  connect(): void;\n  /** Close the connection and release resources. */\n  disconnect(): void;\n\n  /** Whether the transport is currently connected and ready to send. */\n  readonly connected: boolean;\n\n  // --- Event callbacks (set by VoiceClient) ---\n  onopen: (() => void) | null;\n  onclose: ((info?: VoiceTransportCloseInfo) => void) | null;\n  onerror: ((error?: unknown) => void) | null;\n  /** Called when a JSON string message arrives from the server. */\n  onmessage: ((data: string | ArrayBuffer | Blob) => void) | null;\n}\n"],"mappings":";;;;;;;;;;;;;AAeA,MAAa,yBAAyB"}