@livekit/agents-plugin-phonic 1.0.46

package/dist/realtime/realtime_model.cjs

@@ -0,0 +1,431 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var realtime_model_exports = {};
+__export(realtime_model_exports, {
+  RealtimeModel: () => RealtimeModel,
+  RealtimeSession: () => RealtimeSession
+});
+module.exports = __toCommonJS(realtime_model_exports);
+var import_agents = require("@livekit/agents");
+var import_rtc_node = require("@livekit/rtc-node");
+var import_phonic = require("phonic");
+const PHONIC_INPUT_SAMPLE_RATE = 44100;
+const PHONIC_OUTPUT_SAMPLE_RATE = 44100;
+const PHONIC_NUM_CHANNELS = 1;
+const PHONIC_INPUT_FRAME_MS = 20;
+const DEFAULT_MODEL = "merritt";
+const WS_CLOSE_NORMAL = 1e3;
+class RealtimeModel extends import_agents.llm.RealtimeModel {
+  /** @internal */
+  _options;
+  get model() {
+    return this._options.model;
+  }
+  constructor(options = {}) {
+    super({
+      messageTruncation: false,
+      turnDetection: true,
+      userTranscription: true,
+      // TODO @Phonic-Co: Implement tool support
+      // Phonic has automatic tool reply generation, but tools are not supported with LiveKit Agents yet.
+      autoToolReplyGeneration: true,
+      audioOutput: true
+    });
+    const apiKey = options.apiKey || process.env.PHONIC_API_KEY;
+    if (!apiKey) {
+      throw new Error("Phonic API key is required. Provide apiKey or set PHONIC_API_KEY.");
+    }
+    this._options = {
+      apiKey,
+      voice: options.voice,
+      phonicAgent: options.phonicAgent,
+      project: options.project,
+      welcomeMessage: options.welcomeMessage,
+      generateWelcomeMessage: options.generateWelcomeMessage,
+      languages: options.languages,
+      audioSpeed: options.audioSpeed,
+      phonicTools: options.phonicTools,
+      boostedKeywords: options.boostedKeywords,
+      generateNoInputPokeText: options.generateNoInputPokeText,
+      noInputPokeSec: options.noInputPokeSec,
+      noInputPokeText: options.noInputPokeText,
+      noInputEndConversationSec: options.noInputEndConversationSec,
+      connOptions: options.connOptions ?? import_agents.DEFAULT_API_CONNECT_OPTIONS,
+      model: options.model ?? DEFAULT_MODEL,
+      baseUrl: options.baseUrl
+    };
+  }
+  /**
+   * Create a new realtime session
+   */
+  session() {
+    return new RealtimeSession(this);
+  }
+  async close() {
+  }
+}
+class RealtimeSession extends import_agents.llm.RealtimeSession {
+  _tools = {};
+  _chatCtx = import_agents.llm.ChatContext.empty();
+  options;
+  bstream;
+  inputResampler;
+  inputResamplerInputRate;
+  currentGeneration;
+  conversationId;
+  client;
+  socket;
+  logger = (0, import_agents.log)();
+  closed = false;
+  configSent = false;
+  instructionsReady;
+  resolveInstructionsReady;
+  connectTask;
+  constructor(realtimeModel) {
+    super(realtimeModel);
+    this.options = realtimeModel._options;
+    this.resolveInstructionsReady = () => {
+    };
+    this.instructionsReady = new Promise((resolve) => {
+      this.resolveInstructionsReady = resolve;
+    });
+    this.client = new import_phonic.PhonicClient({
+      apiKey: this.options.apiKey,
+      baseUrl: this.options.baseUrl
+    });
+    this.bstream = new import_agents.AudioByteStream(
+      PHONIC_INPUT_SAMPLE_RATE,
+      PHONIC_NUM_CHANNELS,
+      PHONIC_INPUT_SAMPLE_RATE * PHONIC_INPUT_FRAME_MS / 1e3
+    );
+    this.connectTask = this.connect().catch((error) => {
+      const normalizedError = error instanceof Error ? error : new Error(String(error));
+      this.emitError(normalizedError, false);
+    });
+  }
+  get chatCtx() {
+    return this._chatCtx.copy();
+  }
+  get tools() {
+    return { ...this._tools };
+  }
+  async updateInstructions(instructions) {
+    if (this.configSent) {
+      this.logger.warn(
+        "updateInstructions called after config was already sent. Phonic does not support updating instructions mid-session."
+      );
+      return;
+    }
+    this.options.instructions = instructions;
+    this.resolveInstructionsReady();
+  }
+  async updateChatCtx(_chatCtx) {
+    this.logger.warn("updateChatCtx is not supported by the Phonic realtime model.");
+  }
+  async updateTools(tools) {
+    if (Object.keys(tools).length > 0) {
+      this.logger.warn("Tool use is not supported by the Phonic realtime model.");
+    }
+  }
+  updateOptions(_options) {
+    this.logger.warn("updateOptions is not supported by the Phonic realtime model.");
+  }
+  pushAudio(frame) {
+    if (this.closed) {
+      return;
+    }
+    for (const resampledFrame of this.resampleAudio(frame)) {
+      for (const chunk of this.bstream.write(resampledFrame.data.buffer)) {
+        const bytes = Buffer.from(chunk.data.buffer, chunk.data.byteOffset, chunk.data.byteLength);
+        const payload = {
+          type: "audio_chunk",
+          audio: bytes.toString("base64")
+        };
+        if (!this.socket) {
+          continue;
+        }
+        this.socket.sendAudioChunk(payload);
+      }
+    }
+  }
+  // TODO @Phonic-Co: Implement generateReply
+  async generateReply(_instructions) {
+    throw new Error(
+      "generateReply is not yet supported by the Phonic realtime model. Consider using `welcomeMessage` instead."
+    );
+  }
+  async commitAudio() {
+    this.logger.warn("commitAudio is not supported by the Phonic realtime model.");
+  }
+  async clearAudio() {
+    this.logger.warn("clearAudio is not supported by the Phonic realtime model.");
+  }
+  async interrupt() {
+    this.logger.warn("interrupt is not supported by the Phonic realtime model.");
+  }
+  async truncate(_options) {
+    this.logger.warn("truncate is not supported by the Phonic realtime model.");
+  }
+  async close() {
+    var _a;
+    this.closed = true;
+    this.resolveInstructionsReady();
+    this.closeCurrentGeneration({ interrupted: false });
+    (_a = this.socket) == null ? void 0 : _a.close();
+    await this.connectTask;
+    await super.close();
+  }
+  async connect() {
+    this.socket = await this.client.conversations.connect({
+      reconnectAttempts: this.options.connOptions.maxRetry
+    });
+    if (this.closed) {
+      this.socket.close();
+      return;
+    }
+    this.socket.on(
+      "message",
+      (message) => this.handleServerMessage(message)
+    );
+    this.socket.on("error", (error) => this.emitError(error, false));
+    this.socket.on("close", (event) => {
+      this.closeCurrentGeneration({ interrupted: false });
+      if (!this.closed && event.code !== WS_CLOSE_NORMAL) {
+        this.emitError(new Error(`Phonic STS socket closed with code ${event.code ?? -1}`), false);
+      }
+    });
+    await this.socket.waitForOpen();
+    await this.instructionsReady;
+    if (this.closed) return;
+    this.configSent = true;
+    this.socket.sendConfig({
+      type: "config",
+      model: this.options.model,
+      agent: this.options.phonicAgent,
+      project: this.options.project,
+      welcome_message: this.options.welcomeMessage,
+      generate_welcome_message: this.options.generateWelcomeMessage,
+      system_prompt: this.options.instructions,
+      voice_id: this.options.voice,
+      input_format: "pcm_44100",
+      output_format: "pcm_44100",
+      recognized_languages: this.options.languages,
+      audio_speed: this.options.audioSpeed,
+      tools: this.options.phonicTools,
+      boosted_keywords: this.options.boostedKeywords,
+      generate_no_input_poke_text: this.options.generateNoInputPokeText,
+      no_input_poke_sec: this.options.noInputPokeSec,
+      no_input_poke_text: this.options.noInputPokeText,
+      no_input_end_conversation_sec: this.options.noInputEndConversationSec
+    });
+  }
+  handleServerMessage(message) {
+    if (this.closed) {
+      return;
+    }
+    switch (message.type) {
+      case "assistant_started_speaking":
+        this.startNewAssistantTurn();
+        break;
+      case "assistant_finished_speaking":
+        this.finishAssistantTurn();
+        break;
+      case "audio_chunk":
+        this.handleAudioChunk(message);
+        break;
+      case "input_text":
+        this.handleInputText(message);
+        break;
+      case "user_started_speaking":
+        this.handleInputSpeechStarted();
+        break;
+      case "user_finished_speaking":
+        this.handleInputSpeechStopped();
+        break;
+      case "error":
+        this.emitError(new Error(message.error.message), false);
+        break;
+      case "tool_call":
+        this.emitError(
+          new Error(
+            `WebSocket tool calls are not yet supported by the Phonic realtime model with LiveKit Agents.`
+          ),
+          false
+        );
+        break;
+      case "assistant_ended_conversation":
+        this.emitError(
+          new Error(
+            "assistant_ended_conversation is not supported by the Phonic realtime model with LiveKit Agents."
+          ),
+          false
+        );
+        break;
+      case "conversation_created":
+        this.conversationId = message.conversation_id;
+        this.logger.info(`Phonic Conversation began with ID: ${this.conversationId}`);
+        break;
+      case "assistant_chose_not_to_respond":
+      case "ready_to_start_conversation":
+      case "input_cancelled":
+      case "tool_call_output_processed":
+      case "tool_call_interrupted":
+      case "dtmf":
+      default:
+        break;
+    }
+  }
+  handleAudioChunk(message) {
+    const gen = this.currentGeneration;
+    if (!gen) return;
+    if (message.text) {
+      gen.outputText += message.text;
+      gen.textChannel.write(message.text);
+    }
+    if (message.audio) {
+      const bytes = Buffer.from(message.audio, "base64");
+      const sampleCount = Math.floor(bytes.byteLength / Int16Array.BYTES_PER_ELEMENT);
+      if (sampleCount > 0) {
+        const pcm = new Int16Array(
+          bytes.buffer.slice(
+            bytes.byteOffset,
+            bytes.byteOffset + sampleCount * Int16Array.BYTES_PER_ELEMENT
+          )
+        );
+        const frame = new import_rtc_node.AudioFrame(
+          pcm,
+          PHONIC_OUTPUT_SAMPLE_RATE,
+          PHONIC_NUM_CHANNELS,
+          sampleCount / PHONIC_NUM_CHANNELS
+        );
+        gen.audioChannel.write(frame);
+      }
+    }
+  }
+  handleInputText(message) {
+    const itemId = (0, import_agents.shortuuid)("PI_");
+    this.emit("input_audio_transcription_completed", {
+      itemId,
+      transcript: message.text,
+      isFinal: true
+    });
+    this._chatCtx.addMessage({
+      role: "user",
+      content: message.text,
+      id: itemId
+    });
+  }
+  handleInputSpeechStarted() {
+    this.emit("input_speech_started", {});
+    this.closeCurrentGeneration({ interrupted: true });
+  }
+  handleInputSpeechStopped() {
+    this.emit("input_speech_stopped", {
+      userTranscriptionEnabled: true
+    });
+  }
+  startNewAssistantTurn() {
+    if (this.currentGeneration) {
+      this.closeCurrentGeneration({ interrupted: true });
+    }
+    const responseId = (0, import_agents.shortuuid)("PS_");
+    const textChannel = import_agents.stream.createStreamChannel();
+    const audioChannel = import_agents.stream.createStreamChannel();
+    const functionChannel = import_agents.stream.createStreamChannel();
+    const messageChannel = import_agents.stream.createStreamChannel();
+    messageChannel.write({
+      messageId: responseId,
+      textStream: textChannel.stream(),
+      audioStream: audioChannel.stream(),
+      modalities: Promise.resolve(["audio", "text"])
+    });
+    this.currentGeneration = {
+      responseId,
+      messageChannel,
+      functionChannel,
+      textChannel,
+      audioChannel,
+      outputText: ""
+    };
+    this.emit("generation_created", {
+      messageStream: messageChannel.stream(),
+      functionStream: functionChannel.stream(),
+      userInitiated: false,
+      responseId
+    });
+  }
+  finishAssistantTurn() {
+    this.closeCurrentGeneration({ interrupted: false });
+  }
+  closeCurrentGeneration({ interrupted }) {
+    const gen = this.currentGeneration;
+    if (!gen) return;
+    if (gen.outputText) {
+      this._chatCtx.addMessage({
+        role: "assistant",
+        content: gen.outputText,
+        id: gen.responseId,
+        interrupted
+      });
+    }
+    gen.textChannel.close();
+    gen.audioChannel.close();
+    gen.functionChannel.close();
+    gen.messageChannel.close();
+    this.currentGeneration = void 0;
+  }
+  emitError(error, recoverable) {
+    this.emit("error", {
+      timestamp: Date.now(),
+      label: "phonic_realtime",
+      type: "realtime_model_error",
+      error,
+      recoverable
+    });
+  }
+  *resampleAudio(frame) {
+    if (this.inputResampler) {
+      if (frame.sampleRate !== this.inputResamplerInputRate) {
+        this.inputResampler = void 0;
+        this.inputResamplerInputRate = void 0;
+      }
+    }
+    if (this.inputResampler === void 0 && (frame.sampleRate !== PHONIC_INPUT_SAMPLE_RATE || frame.channels !== PHONIC_NUM_CHANNELS)) {
+      this.inputResampler = new import_rtc_node.AudioResampler(
+        frame.sampleRate,
+        PHONIC_INPUT_SAMPLE_RATE,
+        PHONIC_NUM_CHANNELS
+      );
+      this.inputResamplerInputRate = frame.sampleRate;
+    }
+    if (this.inputResampler) {
+      for (const resampledFrame of this.inputResampler.push(frame)) {
+        yield resampledFrame;
+      }
+    } else {
+      yield frame;
+    }
+  }
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  RealtimeModel,
+  RealtimeSession
+});
+//# sourceMappingURL=realtime_model.cjs.map

package/dist/realtime/realtime_model.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/realtime/realtime_model.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2026 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport type { APIConnectOptions } from '@livekit/agents';\nimport {\n  AudioByteStream,\n  DEFAULT_API_CONNECT_OPTIONS,\n  llm,\n  log,\n  shortuuid,\n  stream,\n} from '@livekit/agents';\nimport { AudioFrame, AudioResampler } from '@livekit/rtc-node';\nimport type { Phonic } from 'phonic';\nimport { PhonicClient } from 'phonic';\nimport type { ServerEvent, Voice } from './api_proto.js';\n\nconst PHONIC_INPUT_SAMPLE_RATE = 44100;\nconst PHONIC_OUTPUT_SAMPLE_RATE = 44100;\nconst PHONIC_NUM_CHANNELS = 1;\nconst PHONIC_INPUT_FRAME_MS = 20;\nconst DEFAULT_MODEL = 'merritt';\nconst WS_CLOSE_NORMAL = 1000;\n\nexport interface RealtimeModelOptions {\n  apiKey: string;\n  model: string;\n  phonicAgent?: string;\n  voice?: Voice | string;\n  welcomeMessage?: string;\n  generateWelcomeMessage?: boolean;\n  project?: string;\n  connOptions: APIConnectOptions;\n  baseUrl?: string;\n  languages?: string[];\n  audioSpeed?: number;\n  phonicTools?: string[];\n  boostedKeywords?: string[];\n  generateNoInputPokeText?: boolean;\n  noInputPokeSec?: number;\n  noInputPokeText?: string;\n  noInputEndConversationSec?: number;\n  /** Set by `updateInstructions` via `voice.Agent` rather than the RealtimeModel constructor */\n  instructions?: string;\n}\n\nexport class RealtimeModel extends llm.RealtimeModel {\n  /** @internal */\n  _options: RealtimeModelOptions;\n\n  get model(): string {\n    return this._options.model;\n  }\n\n  constructor(\n    options: {\n      /**\n       * Phonic API key. If not provided, will attempt to read from PHONIC_API_KEY environment variable\n       */\n      apiKey?: string;\n      /**\n       * The name of the model to use. Defaults to 'merritt'\n       */\n      model?: Phonic.ConfigPayload['model'] | string;\n      /**\n       * Phonic agent to use for the conversation. Options explicitly set here will override the agent settings.\n       */\n      phonicAgent?: string;\n      /**\n       * Voice ID for agent outputs\n       */\n      voice?: Voice;\n      /**\n       * Welcome message for the agent to say when the conversation starts. Ignored when generateWelcomeMessage is true\n       */\n      welcomeMessage?: string;\n      /**\n       * When true, the welcome message will be automatically generated and welcomeMessage will be ignored\n       */\n      generateWelcomeMessage?: boolean;\n      /**\n       * Project name to use for the conversation. Defaults to `main`\n       */\n      project?: string;\n      /**\n       * ISO 639-1 language codes the agent should recognize and speak\n       */\n      languages?: string[];\n      /**\n       * Audio playback speed\n       */\n      audioSpeed?: number;\n      /**\n       * Phonic tool names available to the assistant\n       */\n      phonicTools?: string[];\n      /**\n       * Keywords to boost in speech recognition\n       */\n      boostedKeywords?: string[];\n      /**\n       * Auto-generate poke text when user is silent\n       */\n      generateNoInputPokeText?: boolean;\n      /**\n       * Seconds of silence before sending poke message\n       */\n      noInputPokeSec?: number;\n      /**\n       * Poke message text (ignored when generateNoInputPokeText is true)\n       */\n      noInputPokeText?: string;\n      /**\n       * Seconds of silence before ending conversation\n       */\n      noInputEndConversationSec?: number;\n      /**\n       * Connection options for the API connection\n       */\n      connOptions?: APIConnectOptions;\n      baseUrl?: string;\n    } = {},\n  ) {\n    super({\n      messageTruncation: false,\n      turnDetection: true,\n      userTranscription: true,\n      // TODO @Phonic-Co: Implement tool support\n      // Phonic has automatic tool reply generation, but tools are not supported with LiveKit Agents yet.\n      autoToolReplyGeneration: true,\n      audioOutput: true,\n    });\n\n    const apiKey = options.apiKey || process.env.PHONIC_API_KEY;\n    if (!apiKey) {\n      throw new Error('Phonic API key is required. Provide apiKey or set PHONIC_API_KEY.');\n    }\n\n    this._options = {\n      apiKey,\n      voice: options.voice,\n      phonicAgent: options.phonicAgent,\n      project: options.project,\n      welcomeMessage: options.welcomeMessage,\n      generateWelcomeMessage: options.generateWelcomeMessage,\n      languages: options.languages,\n      audioSpeed: options.audioSpeed,\n      phonicTools: options.phonicTools,\n      boostedKeywords: options.boostedKeywords,\n      generateNoInputPokeText: options.generateNoInputPokeText,\n      noInputPokeSec: options.noInputPokeSec,\n      noInputPokeText: options.noInputPokeText,\n      noInputEndConversationSec: options.noInputEndConversationSec,\n      connOptions: options.connOptions ?? DEFAULT_API_CONNECT_OPTIONS,\n      model: options.model ?? DEFAULT_MODEL,\n      baseUrl: options.baseUrl,\n    };\n  }\n\n  /**\n   * Create a new realtime session\n   */\n  session(): RealtimeSession {\n    return new RealtimeSession(this);\n  }\n\n  async close(): Promise<void> {}\n}\n\ninterface GenerationState {\n  responseId: string;\n  messageChannel: stream.StreamChannel<llm.MessageGeneration>;\n  functionChannel: stream.StreamChannel<llm.FunctionCall>;\n  textChannel: stream.StreamChannel<string>;\n  audioChannel: stream.StreamChannel<AudioFrame>;\n  outputText: string;\n}\n\n/**\n * Realtime session for Phonic (https://docs.phonic.co/)\n */\nexport class RealtimeSession extends llm.RealtimeSession {\n  private _tools: llm.ToolContext = {};\n  private _chatCtx = llm.ChatContext.empty();\n\n  private options: RealtimeModelOptions;\n  private bstream: AudioByteStream;\n  private inputResampler?: AudioResampler;\n  private inputResamplerInputRate?: number;\n\n  private currentGeneration?: GenerationState;\n  private conversationId?: string;\n\n  private client: PhonicClient;\n  private socket?: Awaited<ReturnType<PhonicClient['conversations']['connect']>>;\n  private logger = log();\n  private closed = false;\n  private configSent = false;\n  private instructionsReady: Promise<void>;\n  private resolveInstructionsReady: () => void;\n  private connectTask: Promise<void>;\n\n  constructor(realtimeModel: RealtimeModel) {\n    super(realtimeModel);\n    this.options = realtimeModel._options;\n\n    this.resolveInstructionsReady = () => {};\n    this.instructionsReady = new Promise<void>((resolve) => {\n      this.resolveInstructionsReady = resolve;\n    });\n\n    this.client = new PhonicClient({\n      apiKey: this.options.apiKey,\n      baseUrl: this.options.baseUrl,\n    });\n    this.bstream = new AudioByteStream(\n      PHONIC_INPUT_SAMPLE_RATE,\n      PHONIC_NUM_CHANNELS,\n      (PHONIC_INPUT_SAMPLE_RATE * PHONIC_INPUT_FRAME_MS) / 1000,\n    );\n    this.connectTask = this.connect().catch((error: unknown) => {\n      const normalizedError = error instanceof Error ? error : new Error(String(error));\n      this.emitError(normalizedError, false);\n    });\n  }\n\n  get chatCtx(): llm.ChatContext {\n    return this._chatCtx.copy();\n  }\n\n  get tools(): llm.ToolContext {\n    return { ...this._tools };\n  }\n\n  async updateInstructions(instructions: string): Promise<void> {\n    if (this.configSent) {\n      this.logger.warn(\n        'updateInstructions called after config was already sent. Phonic does not support updating instructions mid-session.',\n      );\n      return;\n    }\n    this.options.instructions = instructions;\n    this.resolveInstructionsReady();\n  }\n\n  async updateChatCtx(_chatCtx: llm.ChatContext): Promise<void> {\n    this.logger.warn('updateChatCtx is not supported by the Phonic realtime model.');\n  }\n\n  async updateTools(tools: llm.ToolContext): Promise<void> {\n    if (Object.keys(tools).length > 0) {\n      this.logger.warn('Tool use is not supported by the Phonic realtime model.');\n    }\n  }\n\n  updateOptions(_options: { toolChoice?: llm.ToolChoice | null }): void {\n    this.logger.warn('updateOptions is not supported by the Phonic realtime model.');\n  }\n\n  pushAudio(frame: AudioFrame): void {\n    if (this.closed) {\n      return;\n    }\n\n    for (const resampledFrame of this.resampleAudio(frame)) {\n      for (const chunk of this.bstream.write(resampledFrame.data.buffer as ArrayBuffer)) {\n        const bytes = Buffer.from(chunk.data.buffer, chunk.data.byteOffset, chunk.data.byteLength);\n        const payload: Phonic.AudioChunkPayload = {\n          type: 'audio_chunk',\n          audio: bytes.toString('base64'),\n        };\n\n        if (!this.socket) {\n          continue;\n        }\n        this.socket.sendAudioChunk(payload);\n      }\n    }\n  }\n\n  // TODO @Phonic-Co: Implement generateReply\n  async generateReply(_instructions?: string): Promise<llm.GenerationCreatedEvent> {\n    throw new Error(\n      'generateReply is not yet supported by the Phonic realtime model. Consider using `welcomeMessage` instead.',\n    );\n  }\n\n  async commitAudio(): Promise<void> {\n    this.logger.warn('commitAudio is not supported by the Phonic realtime model.');\n  }\n  async clearAudio(): Promise<void> {\n    this.logger.warn('clearAudio is not supported by the Phonic realtime model.');\n  }\n\n  async interrupt(): Promise<void> {\n    this.logger.warn('interrupt is not supported by the Phonic realtime model.');\n  }\n\n  async truncate(_options: { messageId: string; audioEndMs: number; audioTranscript?: string }) {\n    this.logger.warn('truncate is not supported by the Phonic realtime model.');\n  }\n\n  async close(): Promise<void> {\n    this.closed = true;\n    this.resolveInstructionsReady();\n    this.closeCurrentGeneration({ interrupted: false });\n    this.socket?.close();\n    await this.connectTask;\n    await super.close();\n  }\n\n  private async connect(): Promise<void> {\n    this.socket = await this.client.conversations.connect({\n      reconnectAttempts: this.options.connOptions.maxRetry,\n    });\n\n    if (this.closed) {\n      this.socket.close();\n      return;\n    }\n\n    this.socket.on('message', (message: unknown) =>\n      this.handleServerMessage(message as ServerEvent),\n    );\n    this.socket.on('error', (error: Error) => this.emitError(error, false));\n    this.socket.on('close', (event: { code?: number }) => {\n      this.closeCurrentGeneration({ interrupted: false });\n      if (!this.closed && event.code !== WS_CLOSE_NORMAL) {\n        this.emitError(new Error(`Phonic STS socket closed with code ${event.code ?? -1}`), false);\n      }\n    });\n\n    await this.socket.waitForOpen();\n    await this.instructionsReady;\n    if (this.closed) return;\n    this.configSent = true;\n    this.socket.sendConfig({\n      type: 'config',\n      model: this.options.model as Phonic.ConfigPayload['model'],\n      agent: this.options.phonicAgent,\n      project: this.options.project,\n      welcome_message: this.options.welcomeMessage,\n      generate_welcome_message: this.options.generateWelcomeMessage,\n      system_prompt: this.options.instructions,\n      voice_id: this.options.voice,\n      input_format: 'pcm_44100',\n      output_format: 'pcm_44100',\n      recognized_languages: this.options.languages,\n      audio_speed: this.options.audioSpeed,\n      tools: this.options.phonicTools,\n      boosted_keywords: this.options.boostedKeywords,\n      generate_no_input_poke_text: this.options.generateNoInputPokeText,\n      no_input_poke_sec: this.options.noInputPokeSec,\n      no_input_poke_text: this.options.noInputPokeText,\n      no_input_end_conversation_sec: this.options.noInputEndConversationSec,\n    });\n  }\n\n  private handleServerMessage(message: ServerEvent): void {\n    if (this.closed) {\n      return;\n    }\n\n    switch (message.type) {\n      case 'assistant_started_speaking':\n        this.startNewAssistantTurn();\n        break;\n      case 'assistant_finished_speaking':\n        this.finishAssistantTurn();\n        break;\n      case 'audio_chunk':\n        this.handleAudioChunk(message);\n        break;\n      case 'input_text':\n        this.handleInputText(message);\n        break;\n      case 'user_started_speaking':\n        this.handleInputSpeechStarted();\n        break;\n      case 'user_finished_speaking':\n        this.handleInputSpeechStopped();\n        break;\n      case 'error':\n        this.emitError(new Error(message.error.message), false);\n        break;\n      case 'tool_call':\n        this.emitError(\n          new Error(\n            `WebSocket tool calls are not yet supported by the Phonic realtime model with LiveKit Agents.`,\n          ),\n          false,\n        );\n        break;\n      case 'assistant_ended_conversation':\n        this.emitError(\n          new Error(\n            'assistant_ended_conversation is not supported by the Phonic realtime model with LiveKit Agents.',\n          ),\n          false,\n        );\n        break;\n      case 'conversation_created':\n        this.conversationId = message.conversation_id;\n        this.logger.info(`Phonic Conversation began with ID: ${this.conversationId}`);\n        break;\n      case 'assistant_chose_not_to_respond':\n      case 'ready_to_start_conversation':\n      case 'input_cancelled':\n      case 'tool_call_output_processed':\n      case 'tool_call_interrupted':\n      case 'dtmf':\n      default:\n        break;\n    }\n  }\n\n  private handleAudioChunk(message: Phonic.AudioChunkResponsePayload): void {\n    /**\n     * Although Phonic sends audio chunks when the assistant is not speaking (i.e. containing silence or background noise),\n     * we only process the chunks when the assistant is speaking to align with the generations model, whereby new streams are created for each turn.\n     */\n    const gen = this.currentGeneration;\n    if (!gen) return;\n\n    if (message.text) {\n      gen.outputText += message.text;\n      gen.textChannel.write(message.text);\n    }\n\n    if (message.audio) {\n      const bytes = Buffer.from(message.audio, 'base64');\n      const sampleCount = Math.floor(bytes.byteLength / Int16Array.BYTES_PER_ELEMENT);\n      if (sampleCount > 0) {\n        const pcm = new Int16Array(\n          bytes.buffer.slice(\n            bytes.byteOffset,\n            bytes.byteOffset + sampleCount * Int16Array.BYTES_PER_ELEMENT,\n          ),\n        );\n        const frame = new AudioFrame(\n          pcm,\n          PHONIC_OUTPUT_SAMPLE_RATE,\n          PHONIC_NUM_CHANNELS,\n          sampleCount / PHONIC_NUM_CHANNELS,\n        );\n        gen.audioChannel.write(frame);\n      }\n    }\n  }\n\n  private handleInputText(message: Phonic.InputTextPayload): void {\n    const itemId = shortuuid('PI_');\n    this.emit('input_audio_transcription_completed', {\n      itemId,\n      transcript: message.text,\n      isFinal: true,\n    });\n\n    this._chatCtx.addMessage({\n      role: 'user',\n      content: message.text,\n      id: itemId,\n    });\n  }\n\n  private handleInputSpeechStarted(): void {\n    this.emit('input_speech_started', {});\n    this.closeCurrentGeneration({ interrupted: true });\n  }\n\n  private handleInputSpeechStopped(): void {\n    this.emit('input_speech_stopped', {\n      userTranscriptionEnabled: true,\n    });\n  }\n\n  private startNewAssistantTurn(): void {\n    if (this.currentGeneration) {\n      this.closeCurrentGeneration({ interrupted: true });\n    }\n\n    const responseId = shortuuid('PS_');\n\n    const textChannel = stream.createStreamChannel<string>();\n    const audioChannel = stream.createStreamChannel<AudioFrame>();\n    const functionChannel = stream.createStreamChannel<llm.FunctionCall>();\n    const messageChannel = stream.createStreamChannel<llm.MessageGeneration>();\n\n    messageChannel.write({\n      messageId: responseId,\n      textStream: textChannel.stream(),\n      audioStream: audioChannel.stream(),\n      modalities: Promise.resolve(['audio', 'text']),\n    });\n\n    this.currentGeneration = {\n      responseId,\n      messageChannel,\n      functionChannel,\n      textChannel,\n      audioChannel,\n      outputText: '',\n    };\n\n    this.emit('generation_created', {\n      messageStream: messageChannel.stream(),\n      functionStream: functionChannel.stream(),\n      userInitiated: false,\n      responseId,\n    });\n  }\n\n  private finishAssistantTurn(): void {\n    this.closeCurrentGeneration({ interrupted: false });\n  }\n\n  private closeCurrentGeneration({ interrupted }: { interrupted: boolean }): void {\n    const gen = this.currentGeneration;\n    if (!gen) return;\n\n    if (gen.outputText) {\n      this._chatCtx.addMessage({\n        role: 'assistant',\n        content: gen.outputText,\n        id: gen.responseId,\n        interrupted,\n      });\n    }\n\n    gen.textChannel.close();\n    gen.audioChannel.close();\n    gen.functionChannel.close();\n    gen.messageChannel.close();\n    this.currentGeneration = undefined;\n  }\n\n  private emitError(error: Error, recoverable: boolean): void {\n    this.emit('error', {\n      timestamp: Date.now(),\n      label: 'phonic_realtime',\n      type: 'realtime_model_error',\n      error,\n      recoverable,\n    } satisfies llm.RealtimeModelError);\n  }\n\n  private *resampleAudio(frame: AudioFrame): Generator<AudioFrame> {\n    if (this.inputResampler) {\n      if (frame.sampleRate !== this.inputResamplerInputRate) {\n        this.inputResampler = undefined;\n        this.inputResamplerInputRate = undefined;\n      }\n    }\n\n    if (\n      this.inputResampler === undefined &&\n      (frame.sampleRate !== PHONIC_INPUT_SAMPLE_RATE || frame.channels !== PHONIC_NUM_CHANNELS)\n    ) {\n      this.inputResampler = new AudioResampler(\n        frame.sampleRate,\n        PHONIC_INPUT_SAMPLE_RATE,\n        PHONIC_NUM_CHANNELS,\n      );\n      this.inputResamplerInputRate = frame.sampleRate;\n    }\n\n    if (this.inputResampler) {\n      for (const resampledFrame of this.inputResampler.push(frame)) {\n        yield resampledFrame;\n      }\n    } else {\n      yield frame;\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAIA,oBAOO;AACP,sBAA2C;AAE3C,oBAA6B;AAG7B,MAAM,2BAA2B;AACjC,MAAM,4BAA4B;AAClC,MAAM,sBAAsB;AAC5B,MAAM,wBAAwB;AAC9B,MAAM,gBAAgB;AACtB,MAAM,kBAAkB;AAwBjB,MAAM,sBAAsB,kBAAI,cAAc;AAAA;AAAA,EAEnD;AAAA,EAEA,IAAI,QAAgB;AAClB,WAAO,KAAK,SAAS;AAAA,EACvB;AAAA,EAEA,YACE,UAkEI,CAAC,GACL;AACA,UAAM;AAAA,MACJ,mBAAmB;AAAA,MACnB,eAAe;AAAA,MACf,mBAAmB;AAAA;AAAA;AAAA,MAGnB,yBAAyB;AAAA,MACzB,aAAa;AAAA,IACf,CAAC;AAED,UAAM,SAAS,QAAQ,UAAU,QAAQ,IAAI;AAC7C,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI,MAAM,mEAAmE;AAAA,IACrF;AAEA,SAAK,WAAW;AAAA,MACd;AAAA,MACA,OAAO,QAAQ;AAAA,MACf,aAAa,QAAQ;AAAA,MACrB,SAAS,QAAQ;AAAA,MACjB,gBAAgB,QAAQ;AAAA,MACxB,wBAAwB,QAAQ;AAAA,MAChC,WAAW,QAAQ;AAAA,MACnB,YAAY,QAAQ;AAAA,MACpB,aAAa,QAAQ;AAAA,MACrB,iBAAiB,QAAQ;AAAA,MACzB,yBAAyB,QAAQ;AAAA,MACjC,gBAAgB,QAAQ;AAAA,MACxB,iBAAiB,QAAQ;AAAA,MACzB,2BAA2B,QAAQ;AAAA,MACnC,aAAa,QAAQ,eAAe;AAAA,MACpC,OAAO,QAAQ,SAAS;AAAA,MACxB,SAAS,QAAQ;AAAA,IACnB;AAAA,EACF;AAAA;AAAA;AAAA;AAAA,EAKA,UAA2B;AACzB,WAAO,IAAI,gBAAgB,IAAI;AAAA,EACjC;AAAA,EAEA,MAAM,QAAuB;AAAA,EAAC;AAChC;AAcO,MAAM,wBAAwB,kBAAI,gBAAgB;AAAA,EAC/C,SAA0B,CAAC;AAAA,EAC3B,WAAW,kBAAI,YAAY,MAAM;AAAA,EAEjC;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EAEA;AAAA,EACA;AAAA,EAEA;AAAA,EACA;AAAA,EACA,aAAS,mBAAI;AAAA,EACb,SAAS;AAAA,EACT,aAAa;AAAA,EACb;AAAA,EACA;AAAA,EACA;AAAA,EAER,YAAY,eAA8B;AACxC,UAAM,aAAa;AACnB,SAAK,UAAU,cAAc;AAE7B,SAAK,2BAA2B,MAAM;AAAA,IAAC;AACvC,SAAK,oBAAoB,IAAI,QAAc,CAAC,YAAY;AACtD,WAAK,2BAA2B;AAAA,IAClC,CAAC;AAED,SAAK,SAAS,IAAI,2BAAa;AAAA,MAC7B,QAAQ,KAAK,QAAQ;AAAA,MACrB,SAAS,KAAK,QAAQ;AAAA,IACxB,CAAC;AACD,SAAK,UAAU,IAAI;AAAA,MACjB;AAAA,MACA;AAAA,MACC,2BAA2B,wBAAyB;AAAA,IACvD;AACA,SAAK,cAAc,KAAK,QAAQ,EAAE,MAAM,CAAC,UAAmB;AAC1D,YAAM,kBAAkB,iBAAiB,QAAQ,QAAQ,IAAI,MAAM,OAAO,KAAK,CAAC;AAChF,WAAK,UAAU,iBAAiB,KAAK;AAAA,IACvC,CAAC;AAAA,EACH;AAAA,EAEA,IAAI,UAA2B;AAC7B,WAAO,KAAK,SAAS,KAAK;AAAA,EAC5B;AAAA,EAEA,IAAI,QAAyB;AAC3B,WAAO,EAAE,GAAG,KAAK,OAAO;AAAA,EAC1B;AAAA,EAEA,MAAM,mBAAmB,cAAqC;AAC5D,QAAI,KAAK,YAAY;AACnB,WAAK,OAAO;AAAA,QACV;AAAA,MACF;AACA;AAAA,IACF;AACA,SAAK,QAAQ,eAAe;AAC5B,SAAK,yBAAyB;AAAA,EAChC;AAAA,EAEA,MAAM,cAAc,UAA0C;AAC5D,SAAK,OAAO,KAAK,8DAA8D;AAAA,EACjF;AAAA,EAEA,MAAM,YAAY,OAAuC;AACvD,QAAI,OAAO,KAAK,KAAK,EAAE,SAAS,GAAG;AACjC,WAAK,OAAO,KAAK,yDAAyD;AAAA,IAC5E;AAAA,EACF;AAAA,EAEA,cAAc,UAAwD;AACpE,SAAK,OAAO,KAAK,8DAA8D;AAAA,EACjF;AAAA,EAEA,UAAU,OAAyB;AACjC,QAAI,KAAK,QAAQ;AACf;AAAA,IACF;AAEA,eAAW,kBAAkB,KAAK,cAAc,KAAK,GAAG;AACtD,iBAAW,SAAS,KAAK,QAAQ,MAAM,eAAe,KAAK,MAAqB,GAAG;AACjF,cAAM,QAAQ,OAAO,KAAK,MAAM,KAAK,QAAQ,MAAM,KAAK,YAAY,MAAM,KAAK,UAAU;AACzF,cAAM,UAAoC;AAAA,UACxC,MAAM;AAAA,UACN,OAAO,MAAM,SAAS,QAAQ;AAAA,QAChC;AAEA,YAAI,CAAC,KAAK,QAAQ;AAChB;AAAA,QACF;AACA,aAAK,OAAO,eAAe,OAAO;AAAA,MACpC;AAAA,IACF;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,cAAc,eAA6D;AAC/E,UAAM,IAAI;AAAA,MACR;AAAA,IACF;AAAA,EACF;AAAA,EAEA,MAAM,cAA6B;AACjC,SAAK,OAAO,KAAK,4DAA4D;AAAA,EAC/E;AAAA,EACA,MAAM,aAA4B;AAChC,SAAK,OAAO,KAAK,2DAA2D;AAAA,EAC9E;AAAA,EAEA,MAAM,YAA2B;AAC/B,SAAK,OAAO,KAAK,0DAA0D;AAAA,EAC7E;AAAA,EAEA,MAAM,SAAS,UAA+E;AAC5F,SAAK,OAAO,KAAK,yDAAyD;AAAA,EAC5E;AAAA,EAEA,MAAM,QAAuB;AA9S/B;AA+SI,SAAK,SAAS;AACd,SAAK,yBAAyB;AAC9B,SAAK,uBAAuB,EAAE,aAAa,MAAM,CAAC;AAClD,eAAK,WAAL,mBAAa;AACb,UAAM,KAAK;AACX,UAAM,MAAM,MAAM;AAAA,EACpB;AAAA,EAEA,MAAc,UAAyB;AACrC,SAAK,SAAS,MAAM,KAAK,OAAO,cAAc,QAAQ;AAAA,MACpD,mBAAmB,KAAK,QAAQ,YAAY;AAAA,IAC9C,CAAC;AAED,QAAI,KAAK,QAAQ;AACf,WAAK,OAAO,MAAM;AAClB;AAAA,IACF;AAEA,SAAK,OAAO;AAAA,MAAG;AAAA,MAAW,CAAC,YACzB,KAAK,oBAAoB,OAAsB;AAAA,IACjD;AACA,SAAK,OAAO,GAAG,SAAS,CAAC,UAAiB,KAAK,UAAU,OAAO,KAAK,CAAC;AACtE,SAAK,OAAO,GAAG,SAAS,CAAC,UAA6B;AACpD,WAAK,uBAAuB,EAAE,aAAa,MAAM,CAAC;AAClD,UAAI,CAAC,KAAK,UAAU,MAAM,SAAS,iBAAiB;AAClD,aAAK,UAAU,IAAI,MAAM,sCAAsC,MAAM,QAAQ,EAAE,EAAE,GAAG,KAAK;AAAA,MAC3F;AAAA,IACF,CAAC;AAED,UAAM,KAAK,OAAO,YAAY;AAC9B,UAAM,KAAK;AACX,QAAI,KAAK,OAAQ;AACjB,SAAK,aAAa;AAClB,SAAK,OAAO,WAAW;AAAA,MACrB,MAAM;AAAA,MACN,OAAO,KAAK,QAAQ;AAAA,MACpB,OAAO,KAAK,QAAQ;AAAA,MACpB,SAAS,KAAK,QAAQ;AAAA,MACtB,iBAAiB,KAAK,QAAQ;AAAA,MAC9B,0BAA0B,KAAK,QAAQ;AAAA,MACvC,eAAe,KAAK,QAAQ;AAAA,MAC5B,UAAU,KAAK,QAAQ;AAAA,MACvB,cAAc;AAAA,MACd,eAAe;AAAA,MACf,sBAAsB,KAAK,QAAQ;AAAA,MACnC,aAAa,KAAK,QAAQ;AAAA,MAC1B,OAAO,KAAK,QAAQ;AAAA,MACpB,kBAAkB,KAAK,QAAQ;AAAA,MAC/B,6BAA6B,KAAK,QAAQ;AAAA,MAC1C,mBAAmB,KAAK,QAAQ;AAAA,MAChC,oBAAoB,KAAK,QAAQ;AAAA,MACjC,+BAA+B,KAAK,QAAQ;AAAA,IAC9C,CAAC;AAAA,EACH;AAAA,EAEQ,oBAAoB,SAA4B;AACtD,QAAI,KAAK,QAAQ;AACf;AAAA,IACF;AAEA,YAAQ,QAAQ,MAAM;AAAA,MACpB,KAAK;AACH,aAAK,sBAAsB;AAC3B;AAAA,MACF,KAAK;AACH,aAAK,oBAAoB;AACzB;AAAA,MACF,KAAK;AACH,aAAK,iBAAiB,OAAO;AAC7B;AAAA,MACF,KAAK;AACH,aAAK,gBAAgB,OAAO;AAC5B;AAAA,MACF,KAAK;AACH,aAAK,yBAAyB;AAC9B;AAAA,MACF,KAAK;AACH,aAAK,yBAAyB;AAC9B;AAAA,MACF,KAAK;AACH,aAAK,UAAU,IAAI,MAAM,QAAQ,MAAM,OAAO,GAAG,KAAK;AACtD;AAAA,MACF,KAAK;AACH,aAAK;AAAA,UACH,IAAI;AAAA,YACF;AAAA,UACF;AAAA,UACA;AAAA,QACF;AACA;AAAA,MACF,KAAK;AACH,aAAK;AAAA,UACH,IAAI;AAAA,YACF;AAAA,UACF;AAAA,UACA;AAAA,QACF;AACA;AAAA,MACF,KAAK;AACH,aAAK,iBAAiB,QAAQ;AAC9B,aAAK,OAAO,KAAK,sCAAsC,KAAK,cAAc,EAAE;AAC5E;AAAA,MACF,KAAK;AAAA,MACL,KAAK;AAAA,MACL,KAAK;AAAA,MACL,KAAK;AAAA,MACL,KAAK;AAAA,MACL,KAAK;AAAA,MACL;AACE;AAAA,IACJ;AAAA,EACF;AAAA,EAEQ,iBAAiB,SAAiD;AAKxE,UAAM,MAAM,KAAK;AACjB,QAAI,CAAC,IAAK;AAEV,QAAI,QAAQ,MAAM;AAChB,UAAI,cAAc,QAAQ;AAC1B,UAAI,YAAY,MAAM,QAAQ,IAAI;AAAA,IACpC;AAEA,QAAI,QAAQ,OAAO;AACjB,YAAM,QAAQ,OAAO,KAAK,QAAQ,OAAO,QAAQ;AACjD,YAAM,cAAc,KAAK,MAAM,MAAM,aAAa,WAAW,iBAAiB;AAC9E,UAAI,cAAc,GAAG;AACnB,cAAM,MAAM,IAAI;AAAA,UACd,MAAM,OAAO;AAAA,YACX,MAAM;AAAA,YACN,MAAM,aAAa,cAAc,WAAW;AAAA,UAC9C;AAAA,QACF;AACA,cAAM,QAAQ,IAAI;AAAA,UAChB;AAAA,UACA;AAAA,UACA;AAAA,UACA,cAAc;AAAA,QAChB;AACA,YAAI,aAAa,MAAM,KAAK;AAAA,MAC9B;AAAA,IACF;AAAA,EACF;AAAA,EAEQ,gBAAgB,SAAwC;AAC9D,UAAM,aAAS,yBAAU,KAAK;AAC9B,SAAK,KAAK,uCAAuC;AAAA,MAC/C;AAAA,MACA,YAAY,QAAQ;AAAA,MACpB,SAAS;AAAA,IACX,CAAC;AAED,SAAK,SAAS,WAAW;AAAA,MACvB,MAAM;AAAA,MACN,SAAS,QAAQ;AAAA,MACjB,IAAI;AAAA,IACN,CAAC;AAAA,EACH;AAAA,EAEQ,2BAAiC;AACvC,SAAK,KAAK,wBAAwB,CAAC,CAAC;AACpC,SAAK,uBAAuB,EAAE,aAAa,KAAK,CAAC;AAAA,EACnD;AAAA,EAEQ,2BAAiC;AACvC,SAAK,KAAK,wBAAwB;AAAA,MAChC,0BAA0B;AAAA,IAC5B,CAAC;AAAA,EACH;AAAA,EAEQ,wBAA8B;AACpC,QAAI,KAAK,mBAAmB;AAC1B,WAAK,uBAAuB,EAAE,aAAa,KAAK,CAAC;AAAA,IACnD;AAEA,UAAM,iBAAa,yBAAU,KAAK;AAElC,UAAM,cAAc,qBAAO,oBAA4B;AACvD,UAAM,eAAe,qBAAO,oBAAgC;AAC5D,UAAM,kBAAkB,qBAAO,oBAAsC;AACrE,UAAM,iBAAiB,qBAAO,oBAA2C;AAEzE,mBAAe,MAAM;AAAA,MACnB,WAAW;AAAA,MACX,YAAY,YAAY,OAAO;AAAA,MAC/B,aAAa,aAAa,OAAO;AAAA,MACjC,YAAY,QAAQ,QAAQ,CAAC,SAAS,MAAM,CAAC;AAAA,IAC/C,CAAC;AAED,SAAK,oBAAoB;AAAA,MACvB;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA;AAAA,MACA,YAAY;AAAA,IACd;AAEA,SAAK,KAAK,sBAAsB;AAAA,MAC9B,eAAe,eAAe,OAAO;AAAA,MACrC,gBAAgB,gBAAgB,OAAO;AAAA,MACvC,eAAe;AAAA,MACf;AAAA,IACF,CAAC;AAAA,EACH;AAAA,EAEQ,sBAA4B;AAClC,SAAK,uBAAuB,EAAE,aAAa,MAAM,CAAC;AAAA,EACpD;AAAA,EAEQ,uBAAuB,EAAE,YAAY,GAAmC;AAC9E,UAAM,MAAM,KAAK;AACjB,QAAI,CAAC,IAAK;AAEV,QAAI,IAAI,YAAY;AAClB,WAAK,SAAS,WAAW;AAAA,QACvB,MAAM;AAAA,QACN,SAAS,IAAI;AAAA,QACb,IAAI,IAAI;AAAA,QACR;AAAA,MACF,CAAC;AAAA,IACH;AAEA,QAAI,YAAY,MAAM;AACtB,QAAI,aAAa,MAAM;AACvB,QAAI,gBAAgB,MAAM;AAC1B,QAAI,eAAe,MAAM;AACzB,SAAK,oBAAoB;AAAA,EAC3B;AAAA,EAEQ,UAAU,OAAc,aAA4B;AAC1D,SAAK,KAAK,SAAS;AAAA,MACjB,WAAW,KAAK,IAAI;AAAA,MACpB,OAAO;AAAA,MACP,MAAM;AAAA,MACN;AAAA,MACA;AAAA,IACF,CAAkC;AAAA,EACpC;AAAA,EAEA,CAAS,cAAc,OAA0C;AAC/D,QAAI,KAAK,gBAAgB;AACvB,UAAI,MAAM,eAAe,KAAK,yBAAyB;AACrD,aAAK,iBAAiB;AACtB,aAAK,0BAA0B;AAAA,MACjC;AAAA,IACF;AAEA,QACE,KAAK,mBAAmB,WACvB,MAAM,eAAe,4BAA4B,MAAM,aAAa,sBACrE;AACA,WAAK,iBAAiB,IAAI;AAAA,QACxB,MAAM;AAAA,QACN;AAAA,QACA;AAAA,MACF;AACA,WAAK,0BAA0B,MAAM;AAAA,IACvC;AAEA,QAAI,KAAK,gBAAgB;AACvB,iBAAW,kBAAkB,KAAK,eAAe,KAAK,KAAK,GAAG;AAC5D,cAAM;AAAA,MACR;AAAA,IACF,OAAO;AACL,YAAM;AAAA,IACR;AAAA,EACF;AACF;","names":[]}

package/dist/realtime/realtime_model.d.cts

@@ -0,0 +1,156 @@
+import type { APIConnectOptions } from '@livekit/agents';
+import { llm } from '@livekit/agents';
+import { AudioFrame } from '@livekit/rtc-node';
+import type { Phonic } from 'phonic';
+import type { Voice } from './api_proto.js';
+export interface RealtimeModelOptions {
+    apiKey: string;
+    model: string;
+    phonicAgent?: string;
+    voice?: Voice | string;
+    welcomeMessage?: string;
+    generateWelcomeMessage?: boolean;
+    project?: string;
+    connOptions: APIConnectOptions;
+    baseUrl?: string;
+    languages?: string[];
+    audioSpeed?: number;
+    phonicTools?: string[];
+    boostedKeywords?: string[];
+    generateNoInputPokeText?: boolean;
+    noInputPokeSec?: number;
+    noInputPokeText?: string;
+    noInputEndConversationSec?: number;
+    /** Set by `updateInstructions` via `voice.Agent` rather than the RealtimeModel constructor */
+    instructions?: string;
+}
+export declare class RealtimeModel extends llm.RealtimeModel {
+    /** @internal */
+    _options: RealtimeModelOptions;
+    get model(): string;
+    constructor(options?: {
+        /**
+         * Phonic API key. If not provided, will attempt to read from PHONIC_API_KEY environment variable
+         */
+        apiKey?: string;
+        /**
+         * The name of the model to use. Defaults to 'merritt'
+         */
+        model?: Phonic.ConfigPayload['model'] | string;
+        /**
+         * Phonic agent to use for the conversation. Options explicitly set here will override the agent settings.
+         */
+        phonicAgent?: string;
+        /**
+         * Voice ID for agent outputs
+         */
+        voice?: Voice;
+        /**
+         * Welcome message for the agent to say when the conversation starts. Ignored when generateWelcomeMessage is true
+         */
+        welcomeMessage?: string;
+        /**
+         * When true, the welcome message will be automatically generated and welcomeMessage will be ignored
+         */
+        generateWelcomeMessage?: boolean;
+        /**
+         * Project name to use for the conversation. Defaults to `main`
+         */
+        project?: string;
+        /**
+         * ISO 639-1 language codes the agent should recognize and speak
+         */
+        languages?: string[];
+        /**
+         * Audio playback speed
+         */
+        audioSpeed?: number;
+        /**
+         * Phonic tool names available to the assistant
+         */
+        phonicTools?: string[];
+        /**
+         * Keywords to boost in speech recognition
+         */
+        boostedKeywords?: string[];
+        /**
+         * Auto-generate poke text when user is silent
+         */
+        generateNoInputPokeText?: boolean;
+        /**
+         * Seconds of silence before sending poke message
+         */
+        noInputPokeSec?: number;
+        /**
+         * Poke message text (ignored when generateNoInputPokeText is true)
+         */
+        noInputPokeText?: string;
+        /**
+         * Seconds of silence before ending conversation
+         */
+        noInputEndConversationSec?: number;
+        /**
+         * Connection options for the API connection
+         */
+        connOptions?: APIConnectOptions;
+        baseUrl?: string;
+    });
+    /**
+     * Create a new realtime session
+     */
+    session(): RealtimeSession;
+    close(): Promise<void>;
+}
+/**
+ * Realtime session for Phonic (https://docs.phonic.co/)
+ */
+export declare class RealtimeSession extends llm.RealtimeSession {
+    private _tools;
+    private _chatCtx;
+    private options;
+    private bstream;
+    private inputResampler?;
+    private inputResamplerInputRate?;
+    private currentGeneration?;
+    private conversationId?;
+    private client;
+    private socket?;
+    private logger;
+    private closed;
+    private configSent;
+    private instructionsReady;
+    private resolveInstructionsReady;
+    private connectTask;
+    constructor(realtimeModel: RealtimeModel);
+    get chatCtx(): llm.ChatContext;
+    get tools(): llm.ToolContext;
+    updateInstructions(instructions: string): Promise<void>;
+    updateChatCtx(_chatCtx: llm.ChatContext): Promise<void>;
+    updateTools(tools: llm.ToolContext): Promise<void>;
+    updateOptions(_options: {
+        toolChoice?: llm.ToolChoice | null;
+    }): void;
+    pushAudio(frame: AudioFrame): void;
+    generateReply(_instructions?: string): Promise<llm.GenerationCreatedEvent>;
+    commitAudio(): Promise<void>;
+    clearAudio(): Promise<void>;
+    interrupt(): Promise<void>;
+    truncate(_options: {
+        messageId: string;
+        audioEndMs: number;
+        audioTranscript?: string;
+    }): Promise<void>;
+    close(): Promise<void>;
+    private connect;
+    private handleServerMessage;
+    private handleAudioChunk;
+    private handleInputText;
+    private handleInputSpeechStarted;
+    private handleInputSpeechStopped;
+    private startNewAssistantTurn;
+    private finishAssistantTurn;
+    private closeCurrentGeneration;
+    private emitError;
+    private resampleAudio;
+}
+//# sourceMappingURL=realtime_model.d.ts.map