@contentgrowth/llm-service 1.3.3 → 1.3.5

package/dist/index.cjs

@@ -38,9 +38,11 @@ __export(index_exports, {
   FINISH_REASONS: () => FINISH_REASONS,
   GeminiProvider: () => GoogleProvider,
   GoogleProvider: () => GoogleProvider,
+  LEGACY_MODEL_ALIASES: () => LEGACY_MODEL_ALIASES,
   LLMService: () => LLMService,
   LLMServiceException: () => LLMServiceException,
   MODEL_CONFIGS: () => MODEL_CONFIGS,
+  Modality: () => import_genai2.Modality,
   OpenAIProvider: () => OpenAIProvider,
   TranscriptionService: () => TranscriptionService,
   TranscriptionServiceException: () => TranscriptionServiceException,
@@ -49,6 +51,7 @@ __export(index_exports, {
   extractJsonFromResponse: () => extractJsonFromResponse,
   extractTextAndJson: () => extractTextAndJson,
   handleApiError: () => handleApiError,
+  normalizeModelId: () => normalizeModelId,
   sanitizeError: () => sanitizeError
 });
 module.exports = __toCommonJS(index_exports);
@@ -119,10 +122,27 @@ var DefaultConfigProvider = class extends BaseConfigProvider {
       return null;
     }
   }
+  /**
+   * Resolve the model map for a tenant.
+   * Precedence (highest first):
+   *   1. tenantConfig.model        - single model chosen by the tenant (becomes `default`)
+   *   2. tenantConfig.models[tier] - per-tier overrides supplied by the tenant
+   *   3. MODEL_CONFIGS[provider]   - library defaults
+   */
+  _resolveTenantModels(tenantConfig) {
+    const models = {
+      ...MODEL_CONFIGS[tenantConfig.provider] || {},
+      ...tenantConfig.models || {}
+    };
+    if (typeof tenantConfig.model === "string" && tenantConfig.model.trim()) {
+      models.default = tenantConfig.model;
+    }
+    return normalizeModelMap(models);
+  }
   _buildTenantConfig(tenantConfig, env) {
     return {
       provider: tenantConfig.provider,
-      models: MODEL_CONFIGS[tenantConfig.provider],
+      models: this._resolveTenantModels(tenantConfig),
       // for backward compatibility, api_key is deprecated and shall never be used!!
       apiKey: tenantConfig.apiKey || tenantConfig.api_key,
       baseURL: tenantConfig.baseURL || tenantConfig.base_url || void 0,
@@ -211,39 +231,70 @@ var DefaultConfigProvider = class extends BaseConfigProvider {
 
 // src/llm/config-manager.js
 var MODEL_CONFIGS = {
+  // NOTE: these must be OpenAI *API* model IDs. ChatGPT product names such as
+  // "GPT-5.5 Instant" / "GPT-5.5 Thinking" are not valid API models and return 404.
   openai: {
-    default: "gpt-5.5-instant",
-    edge: "gpt-5.5-thinking",
-    fast: "gpt-5.5-instant",
-    cost: "gpt-5.5-instant",
-    free: "gpt-5.5-instant"
+    default: "gpt-5.6-luna",
+    edge: "gpt-5.6-sol",
+    fast: "gpt-5.6-luna",
+    cost: "gpt-5.6-luna",
+    free: "gpt-5.6-luna"
   },
+  // Gemini Developer API (ai.google.dev). There is no "gemini-3.5-pro"; the Pro tier is gemini-3.1-pro-preview.
   gemini: {
-    default: "gemini-3.5-flash",
-    edge: "gemini-3.5-pro",
-    fast: "gemini-3.5-flash",
-    cost: "gemini-3.5-flash",
-    free: "gemini-3.5-flash",
+    default: "gemini-3.8-flash",
+    edge: "gemini-3.1-pro-preview",
+    fast: "gemini-3.8-flash",
+    cost: "gemini-3.5-flash-lite",
+    free: "gemini-3.5-flash-lite",
     video: "veo",
-    image: "gemini-3.5-pro-image-preview"
+    image: "gemini-3-pro-image"
   },
+  // Vertex AI / Gemini Enterprise Agent Platform
   vertex: {
-    default: "gemini-3.5-flash",
-    edge: "gemini-3.5-pro",
-    fast: "gemini-3.5-flash",
-    cost: "gemini-3.5-flash",
-    free: "gemini-3.5-flash",
+    default: "gemini-3.7-flash",
+    edge: "gemini-3.1-pro-preview",
+    fast: "gemini-3.7-flash",
+    cost: "gemini-3.5-flash-lite",
+    free: "gemini-3.5-flash-lite",
     video: "veo",
-    image: "gemini-3.5-pro-image-preview"
+    image: "gemini-3.1-flash-image"
   },
+  // Anthropic IDs are "claude-<family>-<major>-<minor>", e.g. claude-sonnet-5-5 (not claude-5.5-sonnet).
   anthropic: {
-    default: "claude-4.6-sonnet",
-    edge: "claude-4.8-opus",
-    fast: "claude-4.5-haiku",
-    cost: "claude-4.5-haiku",
-    free: "claude-4.5-haiku"
+    default: "claude-sonnet-5-5",
+    edge: "claude-opus-5-5",
+    fast: "claude-haiku-5-5",
+    cost: "claude-haiku-5-5",
+    free: "claude-haiku-5-5"
   }
 };
+var LEGACY_MODEL_ALIASES = {
+  // OpenAI: ChatGPT product names, not API model IDs
+  "gpt-5.5-instant": "gpt-5.6-luna",
+  "gpt-5.5-thinking": "gpt-5.6-sol",
+  // Gemini: never existed
+  "gemini-3.5-pro": "gemini-3.1-pro-preview",
+  "gemini-3.7-pro": "gemini-3.1-pro-preview",
+  "gemini-3.5-pro-image-preview": "gemini-3-pro-image",
+  "gemini-3.7-pro-image-preview": "gemini-3-pro-image",
+  // Anthropic: wrong segment order
+  "claude-4.5-haiku": "claude-haiku-4-5",
+  "claude-4.6-sonnet": "claude-sonnet-4-6",
+  "claude-4.8-opus": "claude-opus-4-8"
+};
+function normalizeModelId(model) {
+  if (typeof model !== "string") return model;
+  const trimmed = model.trim();
+  return LEGACY_MODEL_ALIASES[trimmed] || trimmed;
+}
+function normalizeModelMap(models = {}) {
+  const out = {};
+  for (const [tier, id] of Object.entries(models)) {
+    out[tier] = normalizeModelId(id);
+  }
+  return out;
+}
 var ConfigManager = class {
   /**
    * Set a custom configuration provider.
@@ -382,6 +433,13 @@ var BaseLLMProvider = class {
   async extractWithLLM(fileData, mimeType, prompt, schema, options) {
     throw new Error("This provider does not support extractWithLLM (Multimodal extraction)");
   }
+  /**
+   * Connect to a bidirectional real-time Live session
+   * @param {Object} options - { model, config, callbacks }
+   */
+  async connectLive(options) {
+    throw new Error(`This provider (${this.constructor.name}) does not support live real-time sessions`);
+  }
   /**
    * Helper to get the last 6 digits of the API key for logging.
    * @returns {string} "..." + last 6 chars, or "not_set"
@@ -471,6 +529,7 @@ function extractTextAndJson(input) {
 }
 
 // src/llm/providers/openai-provider.js
+var REASONING_MODEL_PATTERN = /^(o\d|gpt-5|gpt-6)/i;
 var OpenAIProvider = class extends BaseLLMProvider {
   constructor(config) {
     super(config);
@@ -479,13 +538,12 @@ var OpenAIProvider = class extends BaseLLMProvider {
       clientOptions.baseURL = config.baseURL;
     }
     this.client = new import_openai.default(clientOptions);
-    this.models = config.models;
-    this.defaultModel = config.models.default;
+    this.models = normalizeModelMap(config.models);
+    this.defaultModel = this.models.default;
   }
   async chat(userMessage, systemPrompt = "", options = {}) {
     const messages = [{ role: "user", content: userMessage }];
-    const tier = options.tier || "default";
-    const effectiveModel = this._getModelForTier(tier);
+    const effectiveModel = this._resolveModel(options);
     const effectiveMaxTokens = options.maxTokens || this.config.maxTokens;
     const effectiveTemperature = options.temperature !== void 0 ? options.temperature : this.config.temperature;
     const response = await this._chatCompletionWithModel(
@@ -503,22 +561,42 @@ var OpenAIProvider = class extends BaseLLMProvider {
       messages,
       systemPrompt,
       tools,
-      this.defaultModel,
+      this._resolveModel(options),
       this.config.maxTokens,
       this.config.temperature,
       options
     );
   }
+  /**
+   * Per-call model resolution: explicit options.model > options.tier > default
+   */
+  _resolveModel(options = {}) {
+    return normalizeModelId(options.model) || this._getModelForTier(options.tier || "default");
+  }
+  _isReasoningModel(modelName) {
+    return REASONING_MODEL_PATTERN.test(modelName || "");
+  }
   async _chatCompletionWithModel(messages, systemPrompt, tools, modelName, maxTokens, temperature, options = {}) {
     var _a, _b, _c, _d, _e;
+    const effectiveMaxTokens = (_a = options.maxTokens) != null ? _a : maxTokens;
     const requestPayload = {
       model: modelName,
-      temperature: (_a = options.temperature) != null ? _a : temperature,
-      max_tokens: (_b = options.maxTokens) != null ? _b : maxTokens,
       messages: [{ role: "system", content: systemPrompt }, ...messages],
       tools,
       tool_choice: tools ? "auto" : void 0
     };
+    if (this._isReasoningModel(modelName)) {
+      requestPayload.max_completion_tokens = effectiveMaxTokens;
+      if (options.reasoningEffort) {
+        requestPayload.reasoning_effort = options.reasoningEffort;
+      }
+      if (options.reasoningEffort === "none" && options.temperature !== void 0) {
+        requestPayload.temperature = options.temperature;
+      }
+    } else {
+      requestPayload.max_tokens = effectiveMaxTokens;
+      requestPayload.temperature = (_b = options.temperature) != null ? _b : temperature;
+    }
     if (options.responseFormat) {
       requestPayload.response_format = this._buildResponseFormat(options);
     }
@@ -657,8 +735,8 @@ var import_genai = require("@google/genai");
 var GoogleProvider = class extends BaseLLMProvider {
   constructor(config) {
     super(config);
-    this.models = config.models;
-    this.defaultModel = config.models.default;
+    this.models = normalizeModelMap(config.models);
+    this.defaultModel = this.models.default;
     this._pendingOperations = /* @__PURE__ */ new Map();
     const httpOptions = config.baseURL ? { baseUrl: config.baseURL } : void 0;
     if (config.provider === "vertex") {
@@ -697,7 +775,7 @@ var GoogleProvider = class extends BaseLLMProvider {
   async chat(userMessage, systemPrompt = "", options = {}) {
     const messages = [{ role: "user", content: userMessage }];
     const tier = options.tier || "default";
-    const effectiveModel = this._getModelForTier(tier);
+    const effectiveModel = normalizeModelId(options.model) || this._getModelForTier(tier);
     const effectiveMaxTokens = options.maxTokens || this.config.maxTokens;
     const effectiveTemperature = options.temperature !== void 0 ? options.temperature : this.config.temperature;
     const response = await this._chatCompletionWithModel(
@@ -712,7 +790,7 @@ var GoogleProvider = class extends BaseLLMProvider {
   }
   async chatCompletion(messages, systemPrompt, tools = null, options = {}) {
     const tier = options.tier || "default";
-    const modelName = this.models[tier] || this.defaultModel;
+    const modelName = normalizeModelId(options.model) || this.models[tier] || this.defaultModel;
     return this._chatCompletionWithModel(
       messages,
       systemPrompt,
@@ -1040,7 +1118,7 @@ ${msg.content}`;
   }
   async imageGeneration(prompt, systemPrompt, options = {}) {
     var _a, _b, _c, _d, _e, _f, _g, _h, _i;
-    const modelName = options.model || this.models.image || "gemini-3-pro-image-preview";
+    const modelName = normalizeModelId(options.model) || this.models.image || "gemini-3-pro-image";
     console.log(`[${this.constructor.name}] Generating image with model: ${modelName}`);
     const hasReferenceImages = options.images && options.images.length > 0;
     const generationConfig = {
@@ -1285,15 +1363,38 @@ ${prompt}` : prompt;
       throw error;
     }
   }
+  /**
+   * Connect to Gemini Live bidirectional real-time audio/text session
+   *
+   * @param {Object} options
+   * @param {string} [options.model] - Model name (e.g. models/gemini-2.5-flash-native-audio-preview-09-2025)
+   * @param {Object} options.config - Live session configuration
+   * @param {Object} options.callbacks - Event callbacks (onopen, onmessage, onerror, onclose)
+   * @returns {Promise<any>} The active Live session
+   */
+  async connectLive({ model, config, callbacks }) {
+    var _a, _b, _c;
+    if (!((_b = (_a = this.client) == null ? void 0 : _a.live) == null ? void 0 : _b.connect)) {
+      throw new LLMServiceException("The configured Google provider SDK does not support live sessions.", 500);
+    }
+    const resolvedModel = model || ((_c = this.models) == null ? void 0 : _c.live) || this.defaultModel || "models/gemini-2.5-flash-native-audio-preview-09-2025";
+    return this.client.live.connect({
+      model: resolvedModel,
+      config,
+      callbacks
+    });
+  }
 };
 
 // src/llm/providers/anthropic-provider.js
 var import_sdk = __toESM(require("@anthropic-ai/sdk"), 1);
+var SAMPLING_PARAMS_MODEL_PATTERN = /^claude-(3[-.]|(opus|sonnet|haiku)-4(-[0-6])?([-@]\d{8})?$)/i;
+var NON_STREAMING_TIMEOUT_MS = 10 * 60 * 1e3;
 var AnthropicProvider = class extends BaseLLMProvider {
   constructor(config) {
     super(config);
-    this.models = config.models || {};
-    this.defaultModel = this.models.default || "claude-3-5-sonnet-20241022";
+    this.models = normalizeModelMap(config.models || {});
+    this.defaultModel = this.models.default || "claude-sonnet-5-5";
     const clientOptions = { apiKey: config.apiKey || "" };
     if (config.baseURL) {
       clientOptions.baseURL = config.baseURL;
@@ -1303,8 +1404,7 @@ var AnthropicProvider = class extends BaseLLMProvider {
   async chat(userMessage, systemPrompt = "", options = {}) {
     var _a;
     const messages = [{ role: "user", content: userMessage }];
-    const tier = options.tier || "default";
-    const effectiveModel = this._getModelForTier(tier);
+    const effectiveModel = this._resolveModel(options);
     const effectiveMaxTokens = options.maxTokens || this.config.maxTokens || 4096;
     const effectiveTemperature = options.temperature !== void 0 ? options.temperature : (_a = this.config.temperature) != null ? _a : 0.7;
     const response = await this._chatCompletionWithModel(
@@ -1320,8 +1420,7 @@ var AnthropicProvider = class extends BaseLLMProvider {
   }
   async chatCompletion(messages, systemPrompt, tools = null, options = {}) {
     var _a;
-    const tier = options.tier || "default";
-    const effectiveModel = this._getModelForTier(tier);
+    const effectiveModel = this._resolveModel(options);
     const effectiveMaxTokens = options.maxTokens || this.config.maxTokens || 4096;
     const effectiveTemperature = options.temperature !== void 0 ? options.temperature : (_a = this.config.temperature) != null ? _a : 0.7;
     return this._chatCompletionWithModel(
@@ -1335,7 +1434,7 @@ var AnthropicProvider = class extends BaseLLMProvider {
     );
   }
   async _chatCompletionWithModel(messages, systemPrompt, tools, modelName, maxTokens, temperature, options = {}) {
-    var _a, _b, _c, _d;
+    var _a, _b, _c, _d, _e;
     let effectiveSystem = systemPrompt || void 0;
     if (options.responseFormat) {
       const formatType = typeof options.responseFormat === "string" ? options.responseFormat : options.responseFormat.type;
@@ -1355,14 +1454,18 @@ ${jsonPrompt}` : jsonPrompt;
     const requestParams = {
       model: modelName,
       max_tokens: maxTokens,
-      temperature,
       messages: anthropicMessages,
       ...effectiveSystem ? { system: effectiveSystem } : {},
       ...anthropicTools ? { tools: anthropicTools } : {}
     };
+    if (this._supportsSamplingParams(modelName) && temperature !== void 0 && temperature !== null) {
+      requestParams.temperature = temperature;
+    }
     let response;
     try {
-      response = await this.client.messages.create(requestParams);
+      response = await this.client.messages.create(requestParams, {
+        timeout: (_a = this.config.timeout) != null ? _a : NON_STREAMING_TIMEOUT_MS
+      });
     } catch (error) {
       console.error(`[AnthropicProvider] request failed (API Key: ${this._getMaskedApiKey()}):`, error);
       throw error;
@@ -1395,9 +1498,9 @@ ${jsonPrompt}` : jsonPrompt;
       _rawFinishReason: rawFinishReason,
       _responseFormat: options.responseFormat,
       usage: {
-        prompt_tokens: ((_a = response.usage) == null ? void 0 : _a.input_tokens) || 0,
-        completion_tokens: ((_b = response.usage) == null ? void 0 : _b.output_tokens) || 0,
-        total_tokens: (((_c = response.usage) == null ? void 0 : _c.input_tokens) || 0) + (((_d = response.usage) == null ? void 0 : _d.output_tokens) || 0)
+        prompt_tokens: ((_b = response.usage) == null ? void 0 : _b.input_tokens) || 0,
+        completion_tokens: ((_c = response.usage) == null ? void 0 : _c.output_tokens) || 0,
+        total_tokens: (((_d = response.usage) == null ? void 0 : _d.input_tokens) || 0) + (((_e = response.usage) == null ? void 0 : _e.output_tokens) || 0)
       },
       model: modelName
     };
@@ -1511,6 +1614,15 @@ ${jsonPrompt}` : jsonPrompt;
   _getModelForTier(tier) {
     return this.models[tier] || this.defaultModel;
   }
+  /**
+   * Per-call model resolution: explicit options.model > options.tier > default
+   */
+  _resolveModel(options = {}) {
+    return normalizeModelId(options.model) || this._getModelForTier(options.tier || "default");
+  }
+  _supportsSamplingParams(modelName) {
+    return SAMPLING_PARAMS_MODEL_PATTERN.test(modelName || "");
+  }
   _shouldAutoParse(options) {
     if (!options.responseFormat) return false;
     const formatType = typeof options.responseFormat === "string" ? options.responseFormat : options.responseFormat.type;
@@ -1617,7 +1729,7 @@ var LLMService = class {
       return { tools: param1, options: param2 || {} };
     }
     if (typeof param1 === "object") {
-      const optionsKeys = ["responseFormat", "responseSchema", "temperature", "maxTokens", "tier", "autoParse", "strictSchema", "schemaName"];
+      const optionsKeys = ["responseFormat", "responseSchema", "temperature", "maxTokens", "tier", "model", "reasoningEffort", "autoParse", "strictSchema", "schemaName"];
       const hasOptionsKeys = optionsKeys.some((key) => key in param1);
       if (hasOptionsKeys) {
         return { tools: null, options: param1 };
@@ -1804,6 +1916,21 @@ var LLMService = class {
     }
     return provider.extractWithLLM(fileData, mimeType, prompt, schema, options);
   }
+  /**
+   * Connect to a bidirectional real-time Live session
+   * 
+   * @param {Object} params - { model, config, callbacks }
+   * @param {string|null} [tenantId] - Tenant ID
+   * @returns {Promise<any>} The active Live session
+   */
+  async connectLive(params, tenantId = null) {
+    var _a;
+    const provider = await this._getProvider(tenantId);
+    if (typeof provider.connectLive !== "function") {
+      throw new LLMServiceException(`Provider ${((_a = provider.config) == null ? void 0 : _a.provider) || provider.constructor.name} does not support live real-time sessions`, 400);
+    }
+    return provider.connectLive(params);
+  }
 };
 var LLMServiceException = class extends Error {
   constructor(message, statusCode = 500, details = null) {
@@ -2138,6 +2265,9 @@ function createSpeechHandler(app, getConfig) {
   });
   return app;
 }
+
+// src/index.js
+var import_genai2 = require("@google/genai");
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   AnthropicProvider,
@@ -2147,9 +2277,11 @@ function createSpeechHandler(app, getConfig) {
   FINISH_REASONS,
   GeminiProvider,
   GoogleProvider,
+  LEGACY_MODEL_ALIASES,
   LLMService,
   LLMServiceException,
   MODEL_CONFIGS,
+  Modality,
   OpenAIProvider,
   TranscriptionService,
   TranscriptionServiceException,
@@ -2158,6 +2290,7 @@ function createSpeechHandler(app, getConfig) {
   extractJsonFromResponse,
   extractTextAndJson,
   handleApiError,
+  normalizeModelId,
   sanitizeError
 });
 //# sourceMappingURL=index.cjs.map