@aituber-onair/chat 0.55.0 → 0.57.0

package/dist/umd/aituber-onair-chat.js

@@ -99,6 +99,7 @@ var AITuberOnAirChat = (() => {
     MODEL_CLAUDE_4_OPUS: () => MODEL_CLAUDE_4_OPUS,
     MODEL_CLAUDE_4_SONNET: () => MODEL_CLAUDE_4_SONNET,
     MODEL_CLAUDE_5_1_FABLE: () => MODEL_CLAUDE_5_1_FABLE,
+    MODEL_CLAUDE_5_5_OPUS: () => MODEL_CLAUDE_5_5_OPUS,
     MODEL_CLAUDE_5_FABLE: () => MODEL_CLAUDE_5_FABLE,
     MODEL_CLAUDE_5_OPUS: () => MODEL_CLAUDE_5_OPUS,
     MODEL_CLAUDE_5_SONNET: () => MODEL_CLAUDE_5_SONNET,
@@ -171,6 +172,8 @@ var AITuberOnAirChat = (() => {
     MODEL_GPT_5_MINI: () => MODEL_GPT_5_MINI,
     MODEL_GPT_5_NANO: () => MODEL_GPT_5_NANO,
     MODEL_GPT_6_ASTRA: () => MODEL_GPT_6_ASTRA,
+    MODEL_GPT_6_LUNA: () => MODEL_GPT_6_LUNA,
+    MODEL_GPT_6_SOL: () => MODEL_GPT_6_SOL,
     MODEL_GPT_OSS_20B_FREE: () => MODEL_GPT_OSS_20B_FREE,
     MODEL_GROK_4_1_FAST_NON_REASONING: () => MODEL_GROK_4_1_FAST_NON_REASONING,
     MODEL_GROK_4_1_FAST_REASONING: () => MODEL_GROK_4_1_FAST_REASONING,
@@ -179,6 +182,7 @@ var AITuberOnAirChat = (() => {
     MODEL_GROK_4_3: () => MODEL_GROK_4_3,
     MODEL_GROK_4_5: () => MODEL_GROK_4_5,
     MODEL_GROK_4_6: () => MODEL_GROK_4_6,
+    MODEL_GROK_4_7: () => MODEL_GROK_4_7,
     MODEL_INCEPTION_MERCURY_2_5: () => MODEL_INCEPTION_MERCURY_2_5,
     MODEL_INCLUSIONAI_LING_3_0_FLASH_VL_FREE: () => MODEL_INCLUSIONAI_LING_3_0_FLASH_VL_FREE,
     MODEL_KIMI_K2_5: () => MODEL_KIMI_K2_5,
@@ -248,6 +252,7 @@ var AITuberOnAirChat = (() => {
     MODEL_ZAI_GLM_5_3_FLASH: () => MODEL_ZAI_GLM_5_3_FLASH,
     MistralChatService: () => MistralChatService,
     MistralChatServiceProvider: () => MistralChatServiceProvider,
+    OPENAI_COMPATIBLE_LOCAL_PRESETS: () => OPENAI_COMPATIBLE_LOCAL_PRESETS,
     OPENROUTER_CREDITS_THRESHOLD: () => OPENROUTER_CREDITS_THRESHOLD,
     OPENROUTER_DEPRECATED_MODELS: () => OPENROUTER_DEPRECATED_MODELS,
     OPENROUTER_FREE_DAILY_LIMIT_HIGH_CREDITS: () => OPENROUTER_FREE_DAILY_LIMIT_HIGH_CREDITS,
@@ -258,6 +263,7 @@ var AITuberOnAirChat = (() => {
     OpenAIChatService: () => OpenAIChatService,
     OpenAIChatServiceProvider: () => OpenAIChatServiceProvider,
     OpenAICompatibleChatServiceProvider: () => OpenAICompatibleChatServiceProvider,
+    OpenAICompatibleEndpointError: () => OpenAICompatibleEndpointError,
     OpenRouterChatService: () => OpenRouterChatService,
     OpenRouterChatServiceProvider: () => OpenRouterChatServiceProvider,
     PLAMO_API_BASE_URL: () => PLAMO_API_BASE_URL,
@@ -329,6 +335,7 @@ var AITuberOnAirChat = (() => {
     isZaiReasoningEffortModel: () => isZaiReasoningEffortModel,
     isZaiToolStreamModel: () => isZaiToolStreamModel,
     isZaiVisionModel: () => isZaiVisionModel,
+    listOpenAICompatibleModels: () => listOpenAICompatibleModels,
     normalizeClaudeReasoningEffort: () => normalizeClaudeReasoningEffort,
     normalizeDeepSeekReasoningEffort: () => normalizeDeepSeekReasoningEffort,
     normalizeGeminiReasoningEffort: () => normalizeGeminiReasoningEffort,
@@ -340,11 +347,13 @@ var AITuberOnAirChat = (() => {
     parseOpenAICompatibleToolStream: () => parseOpenAICompatibleToolStream,
     processChatWithOptionalTools: () => processChatWithOptionalTools,
     refreshOpenRouterFreeModels: () => refreshOpenRouterFreeModels,
+    resolveOpenAICompatibleEndpoint: () => resolveOpenAICompatibleEndpoint,
     resolveVisionModel: () => resolveVisionModel,
     runOnceText: () => runOnceText,
     safeJsonParse: () => safeJsonParse,
     safeParseToolCallInput: () => safeParseToolCallInput,
     screenplayToText: () => screenplayToText,
+    testOpenAICompatibleConnection: () => testOpenAICompatibleConnection,
     textToScreenplay: () => textToScreenplay,
     textsToScreenplay: () => textsToScreenplay
   });
@@ -353,6 +362,8 @@ var AITuberOnAirChat = (() => {
   var ENDPOINT_OPENAI_CHAT_COMPLETIONS_API = "https://api.openai.com/v1/chat/completions";
   var ENDPOINT_OPENAI_RESPONSES_API = "https://api.openai.com/v1/responses";
   var MODEL_GPT_6_ASTRA = "gpt-6-astra";
+  var MODEL_GPT_6_SOL = "gpt-6-sol";
+  var MODEL_GPT_6_LUNA = "gpt-6-luna";
   var MODEL_GPT_5_NANO = "gpt-5-nano";
   var MODEL_GPT_5_MINI = "gpt-5-mini";
   var MODEL_GPT_5 = "gpt-5";
@@ -376,6 +387,8 @@ var AITuberOnAirChat = (() => {
   var MODEL_O1 = "o1";
   var VISION_SUPPORTED_MODELS = [
     MODEL_GPT_6_ASTRA,
+    MODEL_GPT_6_SOL,
+    MODEL_GPT_6_LUNA,
     MODEL_GPT_5_NANO,
     MODEL_GPT_5_MINI,
     MODEL_GPT_5,
@@ -416,19 +429,19 @@ var AITuberOnAirChat = (() => {
     return GPT_5_MODELS.includes(model);
   }
   function isOpenAIReasoningModel(model) {
-    return model === MODEL_GPT_6_ASTRA || isGPT5Model(model);
+    return model === MODEL_GPT_6_ASTRA || model === MODEL_GPT_6_SOL || model === MODEL_GPT_6_LUNA || isGPT5Model(model);
   }
   function isResponsesOnlyGPT5Model(model) {
     return model === MODEL_GPT_5_4_PRO;
   }
   function allowsReasoningXHigh(model) {
-    return model === MODEL_GPT_6_ASTRA || model === MODEL_GPT_5_6 || model === MODEL_GPT_5_6_SOL || model === MODEL_GPT_5_6_TERRA || model === MODEL_GPT_5_6_LUNA || model === MODEL_GPT_5_5 || model === MODEL_GPT_5_4 || model === MODEL_GPT_5_4_MINI || model === MODEL_GPT_5_4_NANO || model === MODEL_GPT_5_4_PRO;
+    return model === MODEL_GPT_6_ASTRA || model === MODEL_GPT_6_SOL || model === MODEL_GPT_6_LUNA || model === MODEL_GPT_5_6 || model === MODEL_GPT_5_6_SOL || model === MODEL_GPT_5_6_TERRA || model === MODEL_GPT_5_6_LUNA || model === MODEL_GPT_5_5 || model === MODEL_GPT_5_4 || model === MODEL_GPT_5_4_MINI || model === MODEL_GPT_5_4_NANO || model === MODEL_GPT_5_4_PRO;
   }
   function allowsReasoningMax(model) {
-    return model === MODEL_GPT_6_ASTRA || model === MODEL_GPT_5_6 || model === MODEL_GPT_5_6_SOL || model === MODEL_GPT_5_6_TERRA || model === MODEL_GPT_5_6_LUNA;
+    return model === MODEL_GPT_6_ASTRA || model === MODEL_GPT_6_SOL || model === MODEL_GPT_6_LUNA || model === MODEL_GPT_5_6 || model === MODEL_GPT_5_6_SOL || model === MODEL_GPT_5_6_TERRA || model === MODEL_GPT_5_6_LUNA;
   }
   function allowsReasoningNone(model) {
-    return model === MODEL_GPT_5_6 || model === MODEL_GPT_5_6_SOL || model === MODEL_GPT_5_6_TERRA || model === MODEL_GPT_5_6_LUNA || model === MODEL_GPT_5_1 || model === MODEL_GPT_5_4 || model === MODEL_GPT_5_5 || model === MODEL_GPT_5_4_MINI || model === MODEL_GPT_5_4_NANO;
+    return model === MODEL_GPT_6_SOL || model === MODEL_GPT_6_LUNA || model === MODEL_GPT_5_6 || model === MODEL_GPT_5_6_SOL || model === MODEL_GPT_5_6_TERRA || model === MODEL_GPT_5_6_LUNA || model === MODEL_GPT_5_1 || model === MODEL_GPT_5_4 || model === MODEL_GPT_5_5 || model === MODEL_GPT_5_4_MINI || model === MODEL_GPT_5_4_NANO;
   }
   function allowsReasoningMinimal(model) {
     return model === MODEL_GPT_5_NANO || model === MODEL_GPT_5_MINI || model === MODEL_GPT_5;
@@ -446,7 +459,7 @@ var AITuberOnAirChat = (() => {
     return "medium";
   }
   function getDefaultReasoningEffortForOpenAIModel(model) {
-    return model === MODEL_GPT_6_ASTRA ? "low" : getDefaultReasoningEffortForGPT5Model(model);
+    return model === MODEL_GPT_6_ASTRA ? "low" : model === MODEL_GPT_6_SOL ? "medium" : model === MODEL_GPT_6_LUNA ? "low" : getDefaultReasoningEffortForGPT5Model(model);
   }
 
   // src/constants/gemini.ts
@@ -561,6 +574,7 @@ var AITuberOnAirChat = (() => {
   var MODEL_CLAUDE_5_FABLE = "claude-fable-5";
   var MODEL_CLAUDE_5_SONNET = "claude-sonnet-5";
   var MODEL_CLAUDE_5_OPUS = "claude-opus-5";
+  var MODEL_CLAUDE_5_5_OPUS = "claude-opus-5-5";
   var CLAUDE_DEPRECATED_MODELS = [
     MODEL_CLAUDE_3_HAIKU,
     MODEL_CLAUDE_3_5_HAIKU,
@@ -605,7 +619,8 @@ var AITuberOnAirChat = (() => {
     MODEL_CLAUDE_4_8_OPUS,
     MODEL_CLAUDE_5_FABLE,
     MODEL_CLAUDE_5_SONNET,
-    MODEL_CLAUDE_5_OPUS
+    MODEL_CLAUDE_5_OPUS,
+    MODEL_CLAUDE_5_5_OPUS
   ];
   function getClaudeSupportedReasoningEfforts(model) {
     return CLAUDE_REASONING_EFFORTS_BY_MODEL[model] ?? [];
@@ -1054,6 +1069,7 @@ var AITuberOnAirChat = (() => {
   // src/constants/xai.ts
   var ENDPOINT_XAI_CHAT_COMPLETIONS_API = "https://api.x.ai/v1/chat/completions";
   var MODEL_GROK_4_6 = "grok-4.6";
+  var MODEL_GROK_4_7 = "grok-4.7";
   var MODEL_GROK_4_5 = "grok-4.5";
   var MODEL_GROK_4_3 = "grok-4.3";
   var MODEL_GROK_4_20_REASONING = "grok-4.20-0309-reasoning";
@@ -1066,6 +1082,7 @@ var AITuberOnAirChat = (() => {
   ];
   var XAI_VISION_SUPPORTED_MODELS = [
     MODEL_GROK_4_6,
+    MODEL_GROK_4_7,
     MODEL_GROK_4_5,
     MODEL_GROK_4_3,
     MODEL_GROK_4_20_REASONING,
@@ -1078,7 +1095,7 @@ var AITuberOnAirChat = (() => {
     return getXaiSupportedReasoningEfforts(model).length > 0;
   }
   function getXaiSupportedReasoningEfforts(model) {
-    if (model === MODEL_GROK_4_6) {
+    if (model === MODEL_GROK_4_6 || model === MODEL_GROK_4_7) {
       return ["low", "medium", "high", "xhigh"];
     }
     if (model === MODEL_GROK_4_5) {
@@ -1093,6 +1110,9 @@ var AITuberOnAirChat = (() => {
     return getXaiSupportedReasoningEfforts(model).includes("none");
   }
   function getDefaultXaiReasoningEffort(model) {
+    if (model === MODEL_GROK_4_7) {
+      return "low";
+    }
     if (model === MODEL_GROK_4_6 || model === MODEL_GROK_4_5) {
       return "low";
     }
@@ -1811,6 +1831,192 @@ If it's in another language, summarize in that language.
     };
   }
 
+  // src/utils/openaiCompatibleEndpoint.ts
+  var OPENAI_COMPATIBLE_LOCAL_PRESETS = [
+    {
+      id: "ollama",
+      label: "Ollama",
+      baseUrl: "http://localhost:11434/v1",
+      docsUrl: "https://docs.ollama.com/api/openai-compatibility",
+      corsHint: "Set OLLAMA_ORIGINS to allow additional browser origins."
+    },
+    {
+      id: "lmStudio",
+      label: "LM Studio",
+      baseUrl: "http://localhost:1234/v1",
+      docsUrl: "https://lmstudio.ai/docs/developer/openai-compat",
+      corsHint: "Start the server with lms server start --cors for browser access."
+    },
+    {
+      id: "llamaCpp",
+      label: "llama.cpp",
+      baseUrl: "http://127.0.0.1:8080/v1",
+      docsUrl: "https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md",
+      corsHint: "All origins are allowed by default; use --cors-origins to restrict them."
+    },
+    {
+      id: "vllm",
+      label: "vLLM",
+      baseUrl: "http://localhost:8000/v1",
+      docsUrl: "https://docs.vllm.ai/en/latest/serving/online_serving/openai_compatible_server/",
+      corsHint: "All origins are allowed by default; use --allowed-origins to restrict them."
+    }
+  ];
+  var OpenAICompatibleEndpointError = class extends Error {
+    constructor(code, message, status) {
+      super(message);
+      this.name = "OpenAICompatibleEndpointError";
+      this.code = code;
+      this.status = status;
+    }
+  };
+  function resolveOpenAICompatibleEndpoint(input) {
+    let url;
+    try {
+      url = new URL(input.trim());
+    } catch {
+      throw new OpenAICompatibleEndpointError(
+        "invalid-url",
+        "Enter a valid HTTP or HTTPS endpoint URL."
+      );
+    }
+    if (url.protocol !== "http:" && url.protocol !== "https:" || url.search || url.hash || url.username || url.password) {
+      throw new OpenAICompatibleEndpointError(
+        "invalid-url",
+        "Use an HTTP or HTTPS endpoint URL without credentials, a query, or a fragment."
+      );
+    }
+    const path = url.pathname.replace(/\/+$/, "");
+    if (path.endsWith("/api/chat") || path.endsWith("/api/generate")) {
+      throw new OpenAICompatibleEndpointError(
+        "invalid-url",
+        "This looks like Ollama's native API URL. Use the OpenAI-compatible base URL instead, e.g. http://localhost:11434/v1."
+      );
+    }
+    const basePath = !path || path === "/" ? "/v1" : path.endsWith("/chat/completions") ? path.slice(0, -"/chat/completions".length) : path;
+    const baseUrl = `${url.origin}${basePath}`;
+    return {
+      baseUrl,
+      chatCompletionsUrl: `${baseUrl}/chat/completions`,
+      modelsUrl: `${baseUrl}/models`
+    };
+  }
+  async function listOpenAICompatibleModels(options) {
+    const { modelsUrl } = resolveOpenAICompatibleEndpoint(options.endpoint);
+    const timeoutMs = options.timeoutMs ?? 8e3;
+    if (!Number.isFinite(timeoutMs) || timeoutMs <= 0) {
+      throw new RangeError("timeoutMs must be a positive finite number.");
+    }
+    if (options.signal?.aborted) {
+      throw new OpenAICompatibleEndpointError(
+        "aborted",
+        "The models request was aborted by the caller."
+      );
+    }
+    const controller = new AbortController();
+    let abortCause;
+    const onAbort = () => {
+      if (!abortCause) abortCause = "caller";
+      controller.abort();
+    };
+    options.signal?.addEventListener("abort", onAbort, { once: true });
+    if (options.signal?.aborted) onAbort();
+    const timeout = setTimeout(() => {
+      if (!abortCause) abortCause = "timeout";
+      controller.abort();
+    }, timeoutMs);
+    try {
+      const headers = {};
+      if (options.apiKey?.trim()) {
+        headers.Authorization = `Bearer ${options.apiKey.trim()}`;
+      }
+      const response = await (options.fetch ?? fetch)(modelsUrl, {
+        method: "GET",
+        headers,
+        signal: controller.signal
+      });
+      if (abortCause) {
+        throw new OpenAICompatibleEndpointError(
+          "aborted",
+          "The models request was aborted by the caller."
+        );
+      }
+      if (!response.ok) {
+        throw new OpenAICompatibleEndpointError(
+          "http",
+          `Model listing failed with HTTP ${response.status}.`,
+          response.status
+        );
+      }
+      let body;
+      try {
+        body = await response.json();
+      } catch {
+        throw new OpenAICompatibleEndpointError(
+          "invalid-response",
+          "The models endpoint did not return valid JSON."
+        );
+      }
+      if (!body || typeof body !== "object" || !("data" in body) || !Array.isArray(body.data) || !body.data.every(
+        (model) => model && typeof model === "object" && "id" in model && typeof model.id === "string"
+      )) {
+        throw new OpenAICompatibleEndpointError(
+          "invalid-response",
+          "The models response must contain data with model IDs."
+        );
+      }
+      if (abortCause) {
+        throw new OpenAICompatibleEndpointError(
+          "aborted",
+          "The models request was aborted by the caller."
+        );
+      }
+      return [...new Set(body.data.map((model) => model.id))];
+    } catch (error) {
+      if (abortCause === "caller") {
+        throw new OpenAICompatibleEndpointError(
+          "aborted",
+          "The models request was aborted by the caller."
+        );
+      }
+      if (abortCause === "timeout") {
+        throw new OpenAICompatibleEndpointError(
+          "timeout",
+          `The models request timed out after ${timeoutMs} ms.`
+        );
+      }
+      if (error instanceof OpenAICompatibleEndpointError) throw error;
+      throw new OpenAICompatibleEndpointError(
+        "network",
+        "Could not reach the models endpoint. In a browser, this is usually a CORS restriction or the server is not running."
+      );
+    } finally {
+      clearTimeout(timeout);
+      options.signal?.removeEventListener("abort", onAbort);
+    }
+  }
+  async function testOpenAICompatibleConnection(options) {
+    const startedAt = Date.now();
+    try {
+      const models = await listOpenAICompatibleModels(options);
+      return {
+        ok: true,
+        latencyMs: Date.now() - startedAt,
+        models,
+        ...options.model !== void 0 ? { modelFound: models.includes(options.model) } : {}
+      };
+    } catch (error) {
+      if (error instanceof RangeError) throw error;
+      return {
+        ok: false,
+        error: error instanceof OpenAICompatibleEndpointError ? error : new OpenAICompatibleEndpointError(
+          "network",
+          "Could not reach the models endpoint. In a browser, this is usually a CORS restriction or the server is not running."
+        )
+      };
+    }
+  }
+
   // src/utils/openaiCompatibleTools.ts
   var buildOpenAICompatibleTools = (tools, format = "chat-completions") => {
     if (tools.length === 0) return [];
@@ -2628,7 +2834,8 @@ If it's in another language, summarize in that language.
         MODEL_CLAUDE_5_1_FABLE,
         MODEL_CLAUDE_5_FABLE,
         MODEL_CLAUDE_5_SONNET,
-        MODEL_CLAUDE_5_OPUS
+        MODEL_CLAUDE_5_OPUS,
+        MODEL_CLAUDE_5_5_OPUS
       ];
     }
     /**
@@ -4335,6 +4542,7 @@ If it's in another language, summarize in that language.
   var GeminiNanoChatService = class {
     constructor(options = {}) {
       this.provider = "gemini-nano";
+      this.consumedTranscript = [];
       this.expectedInputLanguages = options.expectedInputLanguages ?? [
         "ja",
         "en"
@@ -4344,6 +4552,7 @@ If it's in another language, summarize in that language.
         ...prompt
       }));
       this._responseLength = options.responseLength;
+      this.sessionMode = options.sessionMode ?? "stateless";
     }
     getModel() {
       return MODEL_GEMINI_NANO;
@@ -4351,23 +4560,33 @@ If it's in another language, summarize in that language.
     getVisionModel() {
       return MODEL_GEMINI_NANO;
     }
+    /** Destroy any persistent Prompt API sessions held by this service. */
+    dispose() {
+      this.destroyPersistentSessions();
+    }
     /**
      * Process chat messages using Gemini Nano.
-     * Non-streaming: calls onPartialResponse once with the full response,
-     * then calls onCompleteResponse.
+     * Streams response deltas when Chrome exposes promptStreaming(), then calls
+     * onCompleteResponse with the complete response.
      */
     async processChat(messages, onPartialResponse, onCompleteResponse) {
-      const response = await this.generateResponse(messages);
-      onPartialResponse(response);
+      const response = await this.generateResponse(
+        messages,
+        true,
+        onPartialResponse
+      );
       await onCompleteResponse(response);
     }
     async processVisionChat(_messages, _onPartialResponse, _onCompleteResponse) {
       throw new Error("Gemini Nano does not support vision capabilities.");
     }
-    async chatOnce(messages, _stream = false, onPartialResponse = () => {
+    async chatOnce(messages, stream = false, onPartialResponse = () => {
     }, _maxTokens) {
-      const response = await this.generateResponse(messages);
-      onPartialResponse(response);
+      const response = await this.generateResponse(
+        messages,
+        stream,
+        onPartialResponse
+      );
       return {
         blocks: [{ type: "text", text: response }],
         stop_reason: "end"
@@ -4380,7 +4599,7 @@ If it's in another language, summarize in that language.
     /**
      * Core logic: extract system prompt, manage session, call prompt().
      */
-    async generateResponse(messages) {
+    async generateResponse(messages, stream, onPartialResponse) {
       const api = getLanguageModelAPI();
       if (!api) {
         throw new Error(
@@ -4388,11 +4607,8 @@ If it's in another language, summarize in that language.
         );
       }
       const modelOptions = this.getModelOptions();
-      const availability = await api.availability(modelOptions);
-      if (availability !== "available" && availability !== "downloadable") {
-        throw new Error(
-          `Gemini Nano Prompt API is not ready in this environment. LanguageModel.availability() returned "${availability}". Expected "available" or "downloadable".`
-        );
+      if (this.sessionMode === "stateless") {
+        await this.ensureAvailability(api, modelOptions);
       }
       const systemMessages = messages.filter((m) => m.role === "system");
       const systemPrompt = systemMessages.map((m) => m.content).join("\n");
@@ -4402,14 +4618,31 @@ If it's in another language, summarize in that language.
         throw new Error("No user message found in the provided messages.");
       }
       const lastUserMessage = conversationMessages[lastUserMessageIndex];
+      const contextHistory = conversationMessages.slice(0, lastUserMessageIndex);
+      if (this.sessionMode === "persistent") {
+        return this.generatePersistentResponse(
+          api,
+          systemPrompt,
+          contextHistory,
+          lastUserMessage.content,
+          modelOptions,
+          stream,
+          onPartialResponse
+        );
+      }
       const session = await this.createSession(
         api,
         systemPrompt,
-        conversationMessages.slice(0, lastUserMessageIndex),
+        contextHistory,
         modelOptions
       );
       try {
-        return await session.prompt(lastUserMessage.content);
+        return await this.promptSession(
+          session,
+          lastUserMessage.content,
+          stream,
+          onPartialResponse
+        );
       } finally {
         try {
           session.destroy();
@@ -4417,6 +4650,241 @@ If it's in another language, summarize in that language.
         }
       }
     }
+    async generatePersistentResponse(api, systemPrompt, contextHistory, userPrompt, modelOptions, stream, onPartialResponse) {
+      const sessionKey = this.getPersistentSessionKey(systemPrompt, modelOptions);
+      if (this.persistentSessionKey !== sessionKey) {
+        this.destroyPersistentSessions();
+      }
+      if (!this.baseSession && !this.liveSession) {
+        await this.ensureAvailability(api, modelOptions);
+        await this.createPersistentBaseSession(
+          api,
+          systemPrompt,
+          contextHistory,
+          modelOptions
+        );
+        this.persistentSessionKey = sessionKey;
+      }
+      let session = await this.getPersistentLiveSession(
+        api,
+        systemPrompt,
+        contextHistory,
+        modelOptions
+      );
+      const reusedLiveSession = session === this.liveSession && this.isHistoryCompatible(contextHistory, this.consumedTranscript);
+      this.persistentSessionKey = sessionKey;
+      try {
+        const response = await this.promptSession(
+          session,
+          userPrompt,
+          stream,
+          onPartialResponse
+        );
+        this.recordConsumedTranscript(
+          contextHistory,
+          userPrompt,
+          response,
+          reusedLiveSession
+        );
+        return response;
+      } catch (error) {
+        if (!this.isQuotaExceededError(error)) {
+          this.destroyPersistentLiveSession();
+          throw error;
+        }
+        this.destroyPersistentLiveSession();
+        session = await this.getPersistentLiveSession(
+          api,
+          systemPrompt,
+          [],
+          modelOptions
+        );
+        this.persistentSessionKey = sessionKey;
+        try {
+          const response = await this.promptSession(
+            session,
+            userPrompt,
+            stream,
+            onPartialResponse
+          );
+          this.recordConsumedTranscript([], userPrompt, response, false);
+          return response;
+        } catch (retryError) {
+          this.destroyPersistentLiveSession();
+          throw retryError;
+        }
+      }
+    }
+    async ensureAvailability(api, modelOptions) {
+      const availability = await api.availability(modelOptions);
+      if (availability !== "available" && availability !== "downloadable") {
+        throw new Error(
+          `Gemini Nano Prompt API is not ready in this environment. LanguageModel.availability() returned "${availability}". Expected "available" or "downloadable".`
+        );
+      }
+    }
+    async createPersistentBaseSession(api, systemPrompt, contextHistory, modelOptions) {
+      this.baseSession = await this.createSession(
+        api,
+        systemPrompt,
+        [],
+        modelOptions
+      );
+      if (contextHistory.length > 0) {
+        this.liveSession = await this.createSession(
+          api,
+          systemPrompt,
+          contextHistory,
+          modelOptions
+        );
+        this.consumedTranscript = contextHistory.map((message) => ({
+          ...message
+        }));
+      } else if (typeof this.baseSession.clone === "function") {
+        this.liveSession = await this.baseSession.clone();
+      } else {
+        this.liveSession = this.baseSession;
+      }
+    }
+    async getPersistentLiveSession(api, systemPrompt, contextHistory, modelOptions) {
+      if (this.liveSession && this.isHistoryCompatible(contextHistory, this.consumedTranscript)) {
+        return this.liveSession;
+      }
+      if (this.liveSession && this.liveSession !== this.baseSession) {
+        this.destroyPersistentLiveSession();
+      }
+      if (contextHistory.length === 0 && this.baseSession) {
+        if (typeof this.baseSession.clone === "function") {
+          this.liveSession = await this.baseSession.clone();
+        } else {
+          if (this.consumedTranscript.length > 0) {
+            this.destroyPersistentSessions();
+            await this.createPersistentBaseSession(
+              api,
+              systemPrompt,
+              [],
+              modelOptions
+            );
+            return this.liveSession;
+          }
+          this.liveSession = this.baseSession;
+        }
+        return this.liveSession;
+      }
+      if (this.baseSession && typeof this.baseSession.clone === "function") {
+        this.liveSession = await this.createSession(
+          api,
+          systemPrompt,
+          contextHistory,
+          modelOptions
+        );
+        return this.liveSession;
+      }
+      this.destroyPersistentSessions();
+      await this.ensureAvailability(api, modelOptions);
+      await this.createPersistentBaseSessionWithHistory(
+        api,
+        systemPrompt,
+        contextHistory,
+        modelOptions
+      );
+      return this.liveSession;
+    }
+    async createPersistentBaseSessionWithHistory(api, systemPrompt, contextHistory, modelOptions) {
+      this.baseSession = await this.createSession(
+        api,
+        systemPrompt,
+        contextHistory,
+        modelOptions
+      );
+      this.liveSession = this.baseSession;
+      this.consumedTranscript = contextHistory.map((message) => ({
+        ...message
+      }));
+    }
+    recordConsumedTranscript(contextHistory, userPrompt, response, reusedLiveSession) {
+      const priorTranscript = reusedLiveSession ? this.consumedTranscript : contextHistory;
+      this.consumedTranscript = [
+        ...priorTranscript,
+        { role: "user", content: userPrompt },
+        { role: "assistant", content: response }
+      ];
+    }
+    getPersistentSessionKey(systemPrompt, modelOptions) {
+      return JSON.stringify({
+        systemPrompt,
+        modelOptions,
+        initialPrompts: this.initialPrompts,
+        responseLength: this._responseLength
+      });
+    }
+    isHistoryCompatible(left, right) {
+      if (left.length === 0 || right.length === 0) {
+        return left.length === 0 && right.length === 0;
+      }
+      return this.isSuffix(left, right) || this.isSuffix(right, left);
+    }
+    isSuffix(candidate, full) {
+      if (candidate.length > full.length) return false;
+      const offset = full.length - candidate.length;
+      return candidate.every(
+        (message, index) => message.role === full[offset + index].role && message.content === full[offset + index].content
+      );
+    }
+    isQuotaExceededError(error) {
+      return typeof error === "object" && error !== null && "name" in error && error.name === "QuotaExceededError";
+    }
+    destroyPersistentLiveSession() {
+      if (this.liveSession) {
+        this.destroySession(this.liveSession);
+        if (this.liveSession === this.baseSession) {
+          this.baseSession = void 0;
+        }
+      }
+      this.liveSession = void 0;
+      this.consumedTranscript = [];
+    }
+    destroyPersistentSessions() {
+      if (this.liveSession && this.liveSession !== this.baseSession) {
+        this.destroySession(this.liveSession);
+      }
+      if (this.baseSession) {
+        this.destroySession(this.baseSession);
+      }
+      this.baseSession = void 0;
+      this.liveSession = void 0;
+      this.persistentSessionKey = void 0;
+      this.consumedTranscript = [];
+    }
+    destroySession(session) {
+      try {
+        session.destroy();
+      } catch {
+      }
+    }
+    async promptSession(session, prompt, stream, onPartialResponse) {
+      if (stream && typeof session.promptStreaming === "function") {
+        const reader = session.promptStreaming(prompt).getReader();
+        let response2 = "";
+        let streamMode = "unknown";
+        while (true) {
+          const { done, value } = await reader.read();
+          if (done) break;
+          const chunk = String(value ?? "");
+          if (streamMode === "unknown" && response2 !== "") {
+            streamMode = chunk.startsWith(response2) ? "cumulative" : "delta";
+          }
+          const isCumulative = streamMode === "cumulative";
+          const delta = isCumulative ? chunk.slice(response2.length) : chunk;
+          response2 = isCumulative ? chunk : response2 + chunk;
+          if (delta) onPartialResponse(delta);
+        }
+        return response2;
+      }
+      const response = await session.prompt(prompt);
+      onPartialResponse(response);
+      return response;
+    }
     /**
      * Create a new LanguageModel session with structured initial prompts.
      * System instructions, few-shot examples, and conversation history retain
@@ -4511,7 +4979,8 @@ If it's in another language, summarize in that language.
         expectedInputLanguages: options.expectedInputLanguages,
         expectedOutputLanguages: options.expectedOutputLanguages,
         initialPrompts: options.initialPrompts,
-        responseLength: options.responseLength
+        responseLength: options.responseLength,
+        sessionMode: options.sessionMode
       });
     }
     getProviderName() {
@@ -5061,6 +5530,9 @@ If it's in another language, summarize in that language.
       let shouldUseResponsesAPI = false;
       if (mcpServers.length > 0 || modelName === MODEL_GPT_6_ASTRA || visionModel === MODEL_GPT_6_ASTRA) {
         shouldUseResponsesAPI = true;
+      } else if (modelName === MODEL_GPT_6_SOL || visionModel === MODEL_GPT_6_SOL || modelName === MODEL_GPT_6_LUNA || visionModel === MODEL_GPT_6_LUNA) {
+        const preference = optimizedOptions.gpt5EndpointPreference || "responses";
+        shouldUseResponsesAPI = preference !== "chat" || optimizedOptions.reasoning_effort !== "none";
       } else if (isResponsesOnlyGPT5Model(modelName)) {
         shouldUseResponsesAPI = true;
       } else if (isOpenAIReasoningModel(modelName)) {
@@ -5107,6 +5579,8 @@ If it's in another language, summarize in that language.
         MODEL_GPT_5_5,
         MODEL_GPT_5_6,
         MODEL_GPT_6_ASTRA,
+        MODEL_GPT_6_SOL,
+        MODEL_GPT_6_LUNA,
         MODEL_GPT_5_6_SOL,
         MODEL_GPT_5_6_TERRA,
         MODEL_GPT_5_6_LUNA,
@@ -6035,6 +6509,7 @@ If it's in another language, summarize in that language.
     getSupportedModels() {
       return [
         MODEL_GROK_4_6,
+        MODEL_GROK_4_7,
         MODEL_GROK_4_5,
         MODEL_GROK_4_3,
         MODEL_GROK_4_20_REASONING,
@@ -6553,7 +7028,7 @@ If it's in another language, summarize in that language.
         mcp: MCP_SUPPORTED_PROVIDERS.has(providerName),
         jsonMode: JSON_MODE_SUPPORTED_PROVIDERS.has(providerName),
         responseLength: true,
-        reasoningEffort: providerName === "openai" && model === MODEL_GPT_6_ASTRA ? ["low", "medium", "high", "xhigh", "max"] : providerName === "claude" && model ? [...getClaudeSupportedReasoningEfforts(model)] : providerName === "gemini" && model ? [...getGeminiSupportedReasoningEfforts(model)] : providerName === "kimi" && model ? [...getKimiSupportedReasoningEfforts(model)] : providerName === "deepseek" && model ? [...getDeepSeekSupportedReasoningEfforts(model)] : providerName === "openrouter" && model ? [...getOpenRouterSupportedReasoningEfforts(model)] : providerName === "xai" && model ? [...getXaiSupportedReasoningEfforts(model)] : providerName === "zai" && model ? [...getZaiSupportedReasoningEfforts(model)] : REASONING_EFFORT_BY_PROVIDER[providerName] ?? []
+        reasoningEffort: providerName === "openai" && model === MODEL_GPT_6_ASTRA ? ["low", "medium", "high", "xhigh", "max"] : providerName === "openai" && model === MODEL_GPT_6_SOL ? ["none", "low", "medium", "high", "xhigh", "max"] : providerName === "openai" && model === MODEL_GPT_6_LUNA ? ["none", "low", "medium", "high", "xhigh", "max"] : providerName === "claude" && model ? [...getClaudeSupportedReasoningEfforts(model)] : providerName === "gemini" && model ? [...getGeminiSupportedReasoningEfforts(model)] : providerName === "kimi" && model ? [...getKimiSupportedReasoningEfforts(model)] : providerName === "deepseek" && model ? [...getDeepSeekSupportedReasoningEfforts(model)] : providerName === "openrouter" && model ? [...getOpenRouterSupportedReasoningEfforts(model)] : providerName === "xai" && model ? [...getXaiSupportedReasoningEfforts(model)] : providerName === "zai" && model ? [...getZaiSupportedReasoningEfforts(model)] : REASONING_EFFORT_BY_PROVIDER[providerName] ?? []
       };
     }
     /**