@shogo-ai/voice 1.4.1

package/dist/chunk-ZGYHDC5G.js

@@ -0,0 +1,423 @@
+import {
+  composeExpressivityBlock,
+  stripExpressivityBlock
+} from "./chunk-Q3TJAJ54.js";
+
+// src/prompt.ts
+var DEFAULT_MEMORY_BLOCK = `# Memory
+
+Known context about this user from past conversations:
+{{user_context}}
+
+Memory retrieval is automatic \u2014 the top matching facts are already injected above. Never say "let me check my memory" or "searching memories."
+
+You have one tool:
+- add_memory(fact): save a concise, canonical fact about the user \u2014 preferences, decisions, personal details, follow-ups. Use short normalized phrasings like "favorite color: green" or "lives in Honolulu". Always call it when the user explicitly asks you to remember something.`;
+function stripMemoryBlock(prompt) {
+  if (!prompt) return prompt;
+  return prompt.replace(/\n*<!-- zix-memory-v\d+ -->[\s\S]*$/m, "").replace(/\n*# Memory\s*\n\s*Known context about this user[\s\S]*$/m, "").trimEnd();
+}
+function extractBasePrompt(prompt) {
+  return stripExpressivityBlock(stripMemoryBlock(prompt));
+}
+function composeAgentPrompt(basePrompt, options = {}) {
+  const { expressivity = "subtle", audioTags = null, memoryBlock = DEFAULT_MEMORY_BLOCK } = options;
+  const stripped = extractBasePrompt(basePrompt);
+  const expr = composeExpressivityBlock(expressivity, audioTags);
+  const parts = [stripped];
+  if (expr) parts.push(expr);
+  if (memoryBlock) parts.push(memoryBlock);
+  return parts.join("\n\n");
+}
+
+// src/elevenlabs.ts
+var DEFAULT_ELEVENLABS_BASE_URL = "https://api.elevenlabs.io";
+var CONVAI_SUPPORTED_TTS_MODELS = /* @__PURE__ */ new Set([
+  "eleven_turbo_v2",
+  "eleven_turbo_v2_5",
+  "eleven_flash_v2",
+  "eleven_flash_v2_5",
+  "eleven_multilingual_v2"
+]);
+var CONVAI_TTS_MODEL_FALLBACK = "eleven_turbo_v2_5";
+function resolveConvaiTtsModel(requested) {
+  if (requested && CONVAI_SUPPORTED_TTS_MODELS.has(requested)) return requested;
+  return CONVAI_TTS_MODEL_FALLBACK;
+}
+var ElevenLabsApiError = class extends Error {
+  constructor(message, status, body) {
+    super(message);
+    this.status = status;
+    this.body = body;
+    this.name = "ElevenLabsApiError";
+  }
+};
+var ElevenLabsClient = class {
+  constructor(config) {
+    this.config = config;
+    if (!config.apiKey) throw new Error("ElevenLabsClient: apiKey is required");
+    this.baseUrl = (config.baseUrl ?? DEFAULT_ELEVENLABS_BASE_URL).replace(/\/+$/, "");
+    this.fetchImpl = config.fetch ?? globalThis.fetch;
+    if (typeof this.fetchImpl !== "function") {
+      throw new Error("ElevenLabsClient: global fetch is unavailable; pass config.fetch");
+    }
+  }
+  headers(extra = {}) {
+    return {
+      "xi-api-key": this.config.apiKey,
+      ...extra
+    };
+  }
+  /** Create a new convai agent. Returns the new `agent_id`. */
+  async createAgent(params) {
+    const prompt = composeAgentPrompt(params.systemPrompt, {
+      expressivity: params.expressivity ?? "subtle",
+      audioTags: params.audioTags ?? null,
+      memoryBlock: params.memoryBlock
+    });
+    const ttsConfig = {
+      voice_id: params.voiceId,
+      model_id: resolveConvaiTtsModel(params.ttsModelId)
+    };
+    if (params.voiceSettings) ttsConfig.voice_settings = params.voiceSettings;
+    const body = {
+      name: `Companion-${params.characterName}-for-${params.displayName}`,
+      conversation_config: {
+        agent: {
+          prompt: {
+            prompt,
+            ...params.tools ? { tools: params.tools } : {}
+          },
+          first_message: params.firstMessage,
+          language: params.language ?? "en"
+        },
+        tts: ttsConfig
+      },
+      platform_settings: {
+        overrides: {
+          conversation_config_override: {
+            agent: { prompt: { prompt: true }, first_message: true, language: true },
+            tts: { voice_id: true },
+            conversation: { text_only: true }
+          }
+        },
+        dynamic_variables: {
+          dynamic_variable_placeholders: {
+            user_context: "No prior memories yet.",
+            character_name: "",
+            ...params.dynamicVariablePlaceholders ?? {}
+          }
+        }
+      }
+    };
+    const res = await this.fetchImpl(`${this.baseUrl}/v1/convai/agents/create`, {
+      method: "POST",
+      headers: this.headers({ "content-type": "application/json" }),
+      body: JSON.stringify(body)
+    });
+    if (!res.ok) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(`createAgent failed: ${res.status}`, res.status, text);
+    }
+    const data = await res.json();
+    return data.agent_id;
+  }
+  /** PATCH any subset of agent settings. Idempotent; only sends the keys you supply. */
+  async patchAgent(agentId, params) {
+    const body = {};
+    if (params.displayName !== void 0 || params.characterName !== void 0) {
+      body.name = `Companion-${params.characterName ?? ""}-for-${params.displayName ?? ""}`.replace(
+        /-for-$/,
+        ""
+      );
+    }
+    const conv = {};
+    const agent = {};
+    const prompt = {};
+    if (params.systemPrompt !== void 0) {
+      prompt.prompt = composeAgentPrompt(params.systemPrompt, {
+        expressivity: params.expressivity ?? "subtle",
+        audioTags: params.audioTags ?? null,
+        memoryBlock: params.memoryBlock
+      });
+    }
+    if (params.tools) prompt.tools = params.tools;
+    if (Object.keys(prompt).length) agent.prompt = prompt;
+    if (params.firstMessage !== void 0) agent.first_message = params.firstMessage;
+    if (Object.keys(agent).length) conv.agent = agent;
+    const tts = {};
+    if (params.voiceId !== void 0) tts.voice_id = params.voiceId;
+    if (params.voiceSettings) tts.voice_settings = params.voiceSettings;
+    if (params.ttsModelId !== void 0) tts.model_id = resolveConvaiTtsModel(params.ttsModelId);
+    if (Object.keys(tts).length) conv.tts = tts;
+    if (Object.keys(conv).length) body.conversation_config = conv;
+    if (params.enableUserContextOverride) {
+      body.platform_settings = {
+        overrides: {
+          conversation_config_override: {
+            agent: { prompt: { prompt: true }, first_message: true, language: true },
+            tts: { voice_id: true },
+            conversation: { text_only: true }
+          },
+          custom_llm_extra_body: false,
+          enable_conversation_initiation_client_data_from_webhook: false
+        },
+        dynamic_variables: {
+          dynamic_variable_placeholders: {
+            user_context: "No prior memories yet.",
+            character_name: ""
+          }
+        }
+      };
+    }
+    if (!Object.keys(body).length) return;
+    const res = await this.fetchImpl(`${this.baseUrl}/v1/convai/agents/${encodeURIComponent(agentId)}`, {
+      method: "PATCH",
+      headers: this.headers({ "content-type": "application/json" }),
+      body: JSON.stringify(body)
+    });
+    if (!res.ok) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(`patchAgent failed: ${res.status}`, res.status, text);
+    }
+  }
+  /**
+   * Delete a convai agent. Best-effort: swallow errors so teardown during a
+   * 500 response doesn't cascade into a second failure.
+   */
+  async deleteAgent(agentId) {
+    try {
+      await this.fetchImpl(`${this.baseUrl}/v1/convai/agents/${encodeURIComponent(agentId)}`, {
+        method: "DELETE",
+        headers: this.headers()
+      });
+    } catch {
+    }
+  }
+  /** Fetch a short-lived signed URL the browser can use to start a session. */
+  async getSignedUrl(agentId) {
+    const res = await this.fetchImpl(
+      `${this.baseUrl}/v1/convai/conversation/get-signed-url?agent_id=${encodeURIComponent(agentId)}`,
+      { headers: this.headers() }
+    );
+    if (!res.ok) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(`getSignedUrl failed: ${res.status}`, res.status, text);
+    }
+    const data = await res.json();
+    return data.signed_url;
+  }
+  /**
+   * One-shot TTS. Returns the raw audio bytes plus the model id that was
+   * actually used (some models aren't available for every account, so we
+   * fall back to `eleven_turbo_v2_5` on the first failure).
+   */
+  async textToSpeech(params) {
+    const modelId = params.modelId && params.modelId.length > 0 ? params.modelId : "eleven_v3";
+    const accept = params.accept ?? "audio/mpeg";
+    const attempt = async (m) => {
+      const res2 = await this.fetchImpl(
+        `${this.baseUrl}/v1/text-to-speech/${encodeURIComponent(params.voiceId)}`,
+        {
+          method: "POST",
+          headers: this.headers({ "content-type": "application/json", accept }),
+          body: JSON.stringify({
+            text: params.text,
+            model_id: m,
+            voice_settings: params.voiceSettings
+          })
+        }
+      );
+      return res2;
+    };
+    let res = await attempt(modelId);
+    let used = modelId;
+    if (!res.ok && modelId !== "eleven_turbo_v2_5") {
+      res = await attempt("eleven_turbo_v2_5");
+      used = "eleven_turbo_v2_5";
+    }
+    if (!res.ok) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(`textToSpeech failed: ${res.status}`, res.status, text);
+    }
+    const audio = await res.arrayBuffer();
+    const contentType = res.headers.get("content-type") ?? accept;
+    return { audio, modelId: used, contentType };
+  }
+  /** Probe whether a given voice id still exists on the caller's account. */
+  async voiceExists(voiceId) {
+    const res = await this.fetchImpl(
+      `${this.baseUrl}/v1/voices/${encodeURIComponent(voiceId)}`,
+      { headers: this.headers() }
+    );
+    return res.ok;
+  }
+  /**
+   * Register a Twilio phone number with an ElevenLabs agent. EL becomes
+   * the SIP target for inbound calls to that number — Twilio bridges
+   * the call via EL's native integration and we never have to run a
+   * Media Streams ↔ convai WebSocket bridge ourselves.
+   *
+   * Returns the ElevenLabs `phone_number_id` (distinct from the Twilio
+   * `sid`) which is the handle used by `outboundCall` + `deletePhoneNumber`.
+   */
+  async createPhoneNumberTwilio(params) {
+    const body = {
+      phone_number: params.phoneNumber,
+      label: params.label ?? params.phoneNumber,
+      provider: "twilio",
+      sid: params.twilioAccountSid,
+      token: params.twilioAuthToken
+    };
+    const res = await this.fetchImpl(
+      `${this.baseUrl}/v1/convai/phone-numbers`,
+      {
+        method: "POST",
+        headers: this.headers({ "content-type": "application/json" }),
+        body: JSON.stringify(body)
+      }
+    );
+    if (!res.ok) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(
+        `createPhoneNumberTwilio failed: ${res.status}`,
+        res.status,
+        text
+      );
+    }
+    const data = await res.json();
+    const id = data.phone_number_id ?? data.id;
+    if (!id) {
+      throw new ElevenLabsApiError(
+        "createPhoneNumberTwilio returned no id",
+        500,
+        JSON.stringify(data)
+      );
+    }
+    const patchRes = await this.fetchImpl(
+      `${this.baseUrl}/v1/convai/phone-numbers/${encodeURIComponent(id)}`,
+      {
+        method: "PATCH",
+        headers: this.headers({ "content-type": "application/json" }),
+        body: JSON.stringify({ agent_id: params.agentId })
+      }
+    );
+    if (!patchRes.ok) {
+      const text = await patchRes.text();
+      try {
+        await this.deletePhoneNumber(id);
+      } catch {
+      }
+      throw new ElevenLabsApiError(
+        `assignAgentToPhoneNumber failed: ${patchRes.status}`,
+        patchRes.status,
+        text
+      );
+    }
+    return { phoneNumberId: id };
+  }
+  /** Detach a previously-registered phone number from ElevenLabs. */
+  async deletePhoneNumber(phoneNumberId) {
+    const res = await this.fetchImpl(
+      `${this.baseUrl}/v1/convai/phone-numbers/${encodeURIComponent(phoneNumberId)}`,
+      { method: "DELETE", headers: this.headers() }
+    );
+    if (!res.ok && res.status !== 404) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(
+        `deletePhoneNumber failed: ${res.status}`,
+        res.status,
+        text
+      );
+    }
+  }
+  /**
+   * Place an outbound PSTN call via a previously-registered EL phone
+   * number. EL drives Twilio (or a SIP trunk) under the hood — we just
+   * hand it the destination number and the agent to bridge to. Returns
+   * the Twilio `callSid` + EL `conversationId` for metering / tracing.
+   *
+   * As of 2026-Q2 the EL API uses provider-scoped routes:
+   *   - Twilio numbers → POST /v1/convai/twilio/outbound-call
+   *   - SIP numbers    → POST /v1/convai/sip-trunk/outbound-call
+   * and `agent_phone_number_id` is a body field (no longer a path
+   * segment). The previous `/v1/convai/phone-numbers/{id}/outbound-call`
+   * route now returns 404. Callers that don't know the provider can
+   * look it up from `GET /v1/convai/phone-numbers/{id}` (`provider:
+   * "twilio" | "sip_trunk"`); most hosted setups are Twilio, so that's
+   * the default here.
+   */
+  async outboundCall(params) {
+    const provider = params.provider ?? "twilio";
+    const body = {
+      agent_id: params.agentId,
+      agent_phone_number_id: params.phoneNumberId,
+      to_number: params.toNumber
+    };
+    if (params.dynamicVariables) {
+      body.conversation_initiation_client_data = {
+        dynamic_variables: params.dynamicVariables
+      };
+    }
+    const res = await this.fetchImpl(
+      `${this.baseUrl}/v1/convai/${provider}/outbound-call`,
+      {
+        method: "POST",
+        headers: this.headers({ "content-type": "application/json" }),
+        body: JSON.stringify(body)
+      }
+    );
+    if (!res.ok) {
+      const text = await res.text();
+      throw new ElevenLabsApiError(
+        `outboundCall failed: ${res.status}`,
+        res.status,
+        text
+      );
+    }
+    const data = await res.json();
+    const callSid = data.call_sid ?? data.callSid;
+    const conversationId = data.conversation_id ?? data.conversationId;
+    if (!callSid || !conversationId) {
+      throw new ElevenLabsApiError(
+        "outboundCall missing callSid or conversationId",
+        500,
+        JSON.stringify(data)
+      );
+    }
+    return { callSid, conversationId };
+  }
+};
+var MEMORY_CLIENT_TOOLS = [
+  {
+    type: "client",
+    name: "add_memory",
+    description: "Save a concise canonical fact about the user (preferences, decisions, personal details, follow-ups). Use whenever the user shares something worth remembering.",
+    expects_response: true,
+    parameters: {
+      type: "object",
+      properties: {
+        fact: {
+          type: "string",
+          description: 'A short normalized fact, e.g. "prefers window seats", "favorite color: green". Max 500 chars.'
+        }
+      },
+      required: ["fact"]
+    }
+  }
+];
+
+export {
+  DEFAULT_MEMORY_BLOCK,
+  stripMemoryBlock,
+  extractBasePrompt,
+  composeAgentPrompt,
+  DEFAULT_ELEVENLABS_BASE_URL,
+  CONVAI_SUPPORTED_TTS_MODELS,
+  CONVAI_TTS_MODEL_FALLBACK,
+  resolveConvaiTtsModel,
+  ElevenLabsApiError,
+  ElevenLabsClient,
+  MEMORY_CLIENT_TOOLS
+};
+//# sourceMappingURL=chunk-ZGYHDC5G.js.map