@aituber-onair/chat 0.49.0 → 0.51.0

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

@@ -159,6 +159,7 @@ var AITuberOnAirChat = (() => {
     MODEL_GROK_4_20_REASONING: () => MODEL_GROK_4_20_REASONING,
     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_KIMI_K2_5: () => MODEL_KIMI_K2_5,
     MODEL_KIMI_K2_6: () => MODEL_KIMI_K2_6,
     MODEL_KIMI_K2_7_CODE: () => MODEL_KIMI_K2_7_CODE,
@@ -198,6 +199,8 @@ var AITuberOnAirChat = (() => {
     MODEL_OPENAI_GPT_MINI_LATEST: () => MODEL_OPENAI_GPT_MINI_LATEST,
     MODEL_OPENROUTER_AUTO: () => MODEL_OPENROUTER_AUTO,
     MODEL_OPENROUTER_AUTO_BETA: () => MODEL_OPENROUTER_AUTO_BETA,
+    MODEL_OPENROUTER_DEEPSEEK_V4_FLASH: () => MODEL_OPENROUTER_DEEPSEEK_V4_FLASH,
+    MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731: () => MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731,
     MODEL_OPENROUTER_FUSION: () => MODEL_OPENROUTER_FUSION,
     MODEL_PLAMO_2_2_PRIME: () => MODEL_PLAMO_2_2_PRIME,
     MODEL_PLAMO_3_0_PRIME: () => MODEL_PLAMO_3_0_PRIME,
@@ -242,17 +245,25 @@ var AITuberOnAirChat = (() => {
     allowsReasoningNone: () => allowsReasoningNone,
     allowsReasoningXHigh: () => allowsReasoningXHigh,
     buildOpenAICompatibleTools: () => buildOpenAICompatibleTools,
+    buildToolContinuationMessages: () => buildToolContinuationMessages,
+    getChatBackendProviderCapabilities: () => getChatBackendProviderCapabilities,
     getClaudeSupportedReasoningEfforts: () => getClaudeSupportedReasoningEfforts,
+    getDeepSeekSupportedReasoningEfforts: () => getDeepSeekSupportedReasoningEfforts,
     getDefaultClaudeReasoningEffort: () => getDefaultClaudeReasoningEffort,
+    getDefaultDeepSeekReasoningEffort: () => getDefaultDeepSeekReasoningEffort,
     getDefaultGeminiReasoningEffort: () => getDefaultGeminiReasoningEffort,
     getDefaultKimiReasoningEffort: () => getDefaultKimiReasoningEffort,
+    getDefaultOpenRouterReasoningEffort: () => getDefaultOpenRouterReasoningEffort,
     getDefaultReasoningEffortForGPT5Model: () => getDefaultReasoningEffortForGPT5Model,
     getDefaultXaiReasoningEffort: () => getDefaultXaiReasoningEffort,
     getGeminiSupportedReasoningEfforts: () => getGeminiSupportedReasoningEfforts,
     getKimiSupportedReasoningEfforts: () => getKimiSupportedReasoningEfforts,
     getMaxTokensForResponseLength: () => getMaxTokensForResponseLength,
+    getOpenRouterSupportedReasoningEfforts: () => getOpenRouterSupportedReasoningEfforts,
+    getXaiSupportedReasoningEfforts: () => getXaiSupportedReasoningEfforts,
     installGASFetch: () => installGASFetch,
     isClaudeReasoningEffortModel: () => isClaudeReasoningEffortModel,
+    isDeepSeekReasoningEffortModel: () => isDeepSeekReasoningEffortModel,
     isGPT5Model: () => isGPT5Model,
     isGeminiReasoningEffortModel: () => isGeminiReasoningEffortModel,
     isKimiReasoningEffortModel: () => isKimiReasoningEffortModel,
@@ -270,7 +281,9 @@ var AITuberOnAirChat = (() => {
     isZaiToolStreamModel: () => isZaiToolStreamModel,
     isZaiVisionModel: () => isZaiVisionModel,
     normalizeClaudeReasoningEffort: () => normalizeClaudeReasoningEffort,
+    normalizeDeepSeekReasoningEffort: () => normalizeDeepSeekReasoningEffort,
     normalizeGeminiReasoningEffort: () => normalizeGeminiReasoningEffort,
+    normalizeOpenRouterReasoningEffort: () => normalizeOpenRouterReasoningEffort,
     normalizeXaiReasoningEffort: () => normalizeXaiReasoningEffort,
     parseOpenAICompatibleOneShot: () => parseOpenAICompatibleOneShot,
     parseOpenAICompatibleTextStream: () => parseOpenAICompatibleTextStream,
@@ -589,6 +602,26 @@ var AITuberOnAirChat = (() => {
   var MODEL_KWAIPILOT_KAT_CODER_PRO_V2_5 = "kwaipilot/kat-coder-pro-v2.5";
   var MODEL_XAI_GROK_LATEST = "~x-ai/grok-latest";
   var MODEL_XAI_GROK_4_5 = "x-ai/grok-4.5";
+  var MODEL_OPENROUTER_DEEPSEEK_V4_FLASH = "deepseek/deepseek-v4-flash";
+  var MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731 = "deepseek/deepseek-v4-flash-0731";
+  var DEFAULT_OPENROUTER_REASONING_EFFORTS = [
+    "none",
+    "minimal",
+    "low",
+    "medium",
+    "high"
+  ];
+  var OPENROUTER_DEEPSEEK_V4_FLASH_REASONING_EFFORTS = [
+    "none",
+    "high",
+    "xhigh"
+  ];
+  var OPENROUTER_DEEPSEEK_V4_FLASH_0731_REASONING_EFFORTS = [
+    "none",
+    "low",
+    "high",
+    "max"
+  ];
   var OPENROUTER_FREE_MODELS = [
     MODEL_GPT_OSS_20B_FREE,
     MODEL_ZAI_GLM_4_5_AIR_FREE
@@ -642,6 +675,43 @@ var AITuberOnAirChat = (() => {
       (visionModel) => model.includes(visionModel)
     );
   }
+  function getOpenRouterSupportedReasoningEfforts(model) {
+    const normalizedModel = model.trim();
+    if (normalizedModel === MODEL_OPENROUTER_DEEPSEEK_V4_FLASH) {
+      return OPENROUTER_DEEPSEEK_V4_FLASH_REASONING_EFFORTS;
+    }
+    if (normalizedModel === MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731) {
+      return OPENROUTER_DEEPSEEK_V4_FLASH_0731_REASONING_EFFORTS;
+    }
+    return DEFAULT_OPENROUTER_REASONING_EFFORTS;
+  }
+  function getDefaultOpenRouterReasoningEffort(model) {
+    const normalizedModel = model.trim();
+    return normalizedModel === MODEL_ZAI_GLM_5_2 || normalizedModel === MODEL_OPENROUTER_DEEPSEEK_V4_FLASH || normalizedModel === MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731 ? "none" : void 0;
+  }
+  function normalizeOpenRouterReasoningEffort(model, effort) {
+    const normalizedModel = model.trim();
+    const requested = effort ?? getDefaultOpenRouterReasoningEffort(model);
+    if (!requested) {
+      return void 0;
+    }
+    const supported = getOpenRouterSupportedReasoningEfforts(model);
+    if (supported.includes(requested)) {
+      return requested;
+    }
+    if (normalizedModel === MODEL_OPENROUTER_DEEPSEEK_V4_FLASH) {
+      return requested === "max" ? "xhigh" : "none";
+    }
+    if (normalizedModel === MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731) {
+      if (requested === "minimal" || requested === "medium") {
+        return "low";
+      }
+      if (requested === "xhigh") {
+        return "high";
+      }
+    }
+    return requested;
+  }
 
   // src/constants/zai.ts
   var ENDPOINT_ZAI_CHAT_COMPLETIONS_API = "https://api.z.ai/api/paas/v4/chat/completions";
@@ -672,6 +742,7 @@ 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_5 = "grok-4.5";
   var MODEL_GROK_4_3 = "grok-4.3";
   var MODEL_GROK_4_20_REASONING = "grok-4.20-0309-reasoning";
@@ -679,6 +750,7 @@ var AITuberOnAirChat = (() => {
   var MODEL_GROK_4_1_FAST_REASONING = "grok-4-1-fast-reasoning";
   var MODEL_GROK_4_1_FAST_NON_REASONING = "grok-4-1-fast-non-reasoning";
   var XAI_VISION_SUPPORTED_MODELS = [
+    MODEL_GROK_4_6,
     MODEL_GROK_4_5,
     MODEL_GROK_4_3,
     MODEL_GROK_4_20_REASONING,
@@ -690,13 +762,25 @@ var AITuberOnAirChat = (() => {
     return XAI_VISION_SUPPORTED_MODELS.includes(model);
   }
   function isXaiReasoningEffortModel(model) {
-    return model === MODEL_GROK_4_5 || model === MODEL_GROK_4_3;
+    return getXaiSupportedReasoningEfforts(model).length > 0;
+  }
+  function getXaiSupportedReasoningEfforts(model) {
+    if (model === MODEL_GROK_4_6) {
+      return ["low", "medium", "high", "xhigh"];
+    }
+    if (model === MODEL_GROK_4_5) {
+      return ["low", "medium", "high"];
+    }
+    if (model === MODEL_GROK_4_3) {
+      return ["none", "low", "medium", "high"];
+    }
+    return [];
   }
   function isXaiReasoningEffortNoneModel(model) {
-    return model === MODEL_GROK_4_3;
+    return getXaiSupportedReasoningEfforts(model).includes("none");
   }
   function getDefaultXaiReasoningEffort(model) {
-    if (model === MODEL_GROK_4_5) {
+    if (model === MODEL_GROK_4_6 || model === MODEL_GROK_4_5) {
       return "low";
     }
     return isXaiReasoningEffortNoneModel(model) ? "none" : void 0;
@@ -705,10 +789,14 @@ var AITuberOnAirChat = (() => {
     if (!isXaiReasoningEffortModel(model)) {
       return void 0;
     }
-    if (reasoningEffort === "none" && !isXaiReasoningEffortNoneModel(model)) {
-      return "low";
+    if (reasoningEffort === void 0) {
+      return getDefaultXaiReasoningEffort(model);
+    }
+    const supported = getXaiSupportedReasoningEfforts(model);
+    if (supported.includes(reasoningEffort)) {
+      return reasoningEffort;
     }
-    return reasoningEffort ?? getDefaultXaiReasoningEffort(model);
+    return reasoningEffort === "none" ? "low" : "high";
   }
 
   // src/constants/kimi.ts
@@ -753,6 +841,17 @@ var AITuberOnAirChat = (() => {
   var ENDPOINT_DEEPSEEK_CHAT_COMPLETIONS_API = `${DEEPSEEK_API_BASE_URL}/chat/completions`;
   var MODEL_DEEPSEEK_V4_FLASH = "deepseek-v4-flash";
   var MODEL_DEEPSEEK_V4_PRO = "deepseek-v4-pro";
+  var DEEPSEEK_V4_FLASH_REASONING_EFFORTS = [
+    "none",
+    "low",
+    "high",
+    "max"
+  ];
+  var DEEPSEEK_V4_PRO_REASONING_EFFORTS = [
+    "none",
+    "high",
+    "max"
+  ];
   var MODEL_DEEPSEEK_CHAT = "deepseek-chat";
   var MODEL_DEEPSEEK_REASONER = "deepseek-reasoner";
   var DEEPSEEK_SUPPORTED_MODELS = [
@@ -763,6 +862,35 @@ var AITuberOnAirChat = (() => {
     MODEL_DEEPSEEK_CHAT,
     MODEL_DEEPSEEK_REASONER
   ];
+  function getDeepSeekSupportedReasoningEfforts(model) {
+    if (model === MODEL_DEEPSEEK_V4_FLASH) {
+      return DEEPSEEK_V4_FLASH_REASONING_EFFORTS;
+    }
+    if (model === MODEL_DEEPSEEK_V4_PRO) {
+      return DEEPSEEK_V4_PRO_REASONING_EFFORTS;
+    }
+    return [];
+  }
+  function isDeepSeekReasoningEffortModel(model) {
+    return getDeepSeekSupportedReasoningEfforts(model).length > 0;
+  }
+  function getDefaultDeepSeekReasoningEffort(model) {
+    return isDeepSeekReasoningEffortModel(model) ? "none" : void 0;
+  }
+  function normalizeDeepSeekReasoningEffort(model, effort) {
+    const supported = getDeepSeekSupportedReasoningEfforts(model);
+    if (supported.length === 0) {
+      return void 0;
+    }
+    const requested = effort ?? getDefaultDeepSeekReasoningEffort(model);
+    if (requested && supported.includes(requested)) {
+      return requested;
+    }
+    if (model === MODEL_DEEPSEEK_V4_PRO && requested === "low") {
+      return "high";
+    }
+    return getDefaultDeepSeekReasoningEffort(model);
+  }
 
   // src/constants/mistral.ts
   var MISTRAL_API_BASE_URL = "https://api.mistral.ai/v1";
@@ -2188,6 +2316,7 @@ If it's in another language, summarize in that language.
     const dec = new TextDecoder();
     const textBlocks = [];
     const toolCallsMap = /* @__PURE__ */ new Map();
+    const providerItems = /* @__PURE__ */ new Map();
     let responseStatus;
     let incompleteDetails;
     let usage;
@@ -2215,7 +2344,7 @@ If it's in another language, summarize in that language.
               onPartial,
               textBlocks,
               toolCallsMap,
-              options,
+              providerItems,
               (metadata) => {
                 if (metadata.responseStatus !== void 0) {
                   responseStatus = metadata.responseStatus;
@@ -2242,21 +2371,31 @@ If it's in another language, summarize in that language.
         type: "tool_use",
         id: tool.id,
         name: tool.name,
-        input: tool.input || {}
+        input: safeParseToolCallInput(tool.arguments, options.onJsonError)
       })
     );
+    const providerContent = Array.from(providerItems.values());
+    const text = StreamTextAccumulator.getFullText(textBlocks);
     return {
       blocks: [...textBlocks, ...toolBlocks],
       stop_reason: toolBlocks.length ? "tool_use" : "end",
       truncated: responseStatus === "incomplete",
       response_status: responseStatus,
       incomplete_details: incompleteDetails,
-      usage
+      usage,
+      ...providerContent.length > 0 ? {
+        assistant_message: {
+          role: "assistant",
+          content: text,
+          provider_content: providerContent
+        }
+      } : {}
     };
   }
-  function handleResponsesSSEEvent(eventType, data, onPartial, textBlocks, toolCallsMap, options, onMetadata) {
+  function handleResponsesSSEEvent(eventType, data, onPartial, textBlocks, toolCallsMap, providerItems, onMetadata) {
     switch (eventType) {
       case "response.output_item.added":
+        rememberProviderItem(providerItems, data.item);
         if (data.item?.type === "message" && Array.isArray(data.item.content)) {
           data.item.content.forEach((c) => {
             if (c.type === "output_text" && c.text) {
@@ -2266,12 +2405,41 @@ If it's in another language, summarize in that language.
           });
         } else if (data.item?.type === "function_call") {
           toolCallsMap.set(data.item.id, {
-            id: data.item.id,
+            id: data.item.call_id ?? data.item.id,
             name: data.item.name,
-            input: safeParseToolCallInput(
-              data.item.arguments,
-              options.onJsonError
-            )
+            arguments: data.item.arguments ?? ""
+          });
+        }
+        break;
+      case "response.function_call_arguments.delta": {
+        const tool = toolCallsMap.get(data.item_id);
+        if (tool && typeof data.delta === "string") {
+          tool.arguments += data.delta;
+        }
+        const item = providerItems.get(data.item_id);
+        if (item && typeof data.delta === "string") {
+          item.arguments = `${item.arguments ?? ""}${data.delta}`;
+        }
+        break;
+      }
+      case "response.function_call_arguments.done": {
+        const tool = toolCallsMap.get(data.item_id);
+        if (tool && typeof data.arguments === "string") {
+          tool.arguments = data.arguments;
+        }
+        const item = providerItems.get(data.item_id);
+        if (item && typeof data.arguments === "string") {
+          item.arguments = data.arguments;
+        }
+        break;
+      }
+      case "response.output_item.done":
+        rememberProviderItem(providerItems, data.item);
+        if (data.item?.type === "function_call") {
+          toolCallsMap.set(data.item.id, {
+            id: data.item.call_id ?? data.item.id,
+            name: data.item.name,
+            arguments: data.item.arguments ?? ""
           });
         }
         break;
@@ -2306,6 +2474,11 @@ If it's in another language, summarize in that language.
         break;
     }
   }
+  function rememberProviderItem(items, item) {
+    if (!item || typeof item !== "object") return;
+    const key = typeof item.id === "string" && item.id ? item.id : `output-item-${items.size}`;
+    items.set(key, { ...item });
+  }
   function extractResponsesMetadata(data, fallbackStatus) {
     const response = data?.response ?? data;
     return {
@@ -2328,7 +2501,7 @@ If it's in another language, summarize in that language.
         if (outputItem.type === "function_call") {
           blocks.push({
             type: "tool_use",
-            id: outputItem.id,
+            id: outputItem.call_id ?? outputItem.id,
             name: outputItem.name,
             input: safeParseToolCallInput(
               outputItem.arguments,
@@ -2344,7 +2517,16 @@ If it's in another language, summarize in that language.
       truncated: data?.status === "incomplete",
       response_status: data?.status,
       incomplete_details: data?.incomplete_details ?? null,
-      usage: data?.usage
+      usage: data?.usage,
+      ...Array.isArray(data?.output) && data.output.length > 0 ? {
+        assistant_message: {
+          role: "assistant",
+          content: blocks.filter(
+            (block) => block.type === "text"
+          ).map((block) => block.text).join(""),
+          provider_content: data.output
+        }
+      } : {}
     };
   }
 
@@ -2513,6 +2695,14 @@ If it's in another language, summarize in that language.
     if (provider === "plamo" && reasoning_effort) {
       body.reasoning_effort = reasoning_effort;
     }
+    if (provider === "deepseek" && reasoning_effort) {
+      body.thinking = {
+        type: reasoning_effort === "none" ? "disabled" : "enabled"
+      };
+      if (reasoning_effort !== "none") {
+        body.reasoning_effort = reasoning_effort;
+      }
+    }
     const toolDefinitions = buildOpenAIToolsDefinition({
       tools,
       mcpServers,
@@ -2555,10 +2745,22 @@ If it's in another language, summarize in that language.
     }
   }
   function cleanMessagesForResponsesAPI(messages) {
-    return messages.map((msg) => {
-      const role = msg.role === "tool" ? "user" : msg.role;
+    return messages.flatMap((msg) => {
+      if (msg.role === "assistant" && msg.provider_content?.length) {
+        return msg.provider_content;
+      }
+      if (msg.role === "tool" && msg.tool_call_id) {
+        return [
+          {
+            type: "function_call_output",
+            call_id: msg.tool_call_id,
+            output: msg.content
+          }
+        ];
+      }
+      const items = [];
       const cleanMsg = {
-        role
+        role: msg.role === "tool" ? "user" : msg.role
       };
       if (typeof msg.content === "string") {
         cleanMsg.content = msg.content;
@@ -2580,7 +2782,19 @@ If it's in another language, summarize in that language.
       } else {
         cleanMsg.content = msg.content;
       }
-      return cleanMsg;
+      const hasContent = typeof cleanMsg.content !== "string" || cleanMsg.content.length > 0;
+      if (hasContent) items.push(cleanMsg);
+      if (msg.role === "assistant" && msg.tool_calls) {
+        items.push(
+          ...msg.tool_calls.map((call) => ({
+            type: "function_call",
+            call_id: call.id,
+            name: call.function.name,
+            arguments: call.function.arguments
+          }))
+        );
+      }
+      return items;
     });
   }
   function cleanMessagesForMistralChatCompletions(messages) {
@@ -2800,7 +3014,11 @@ If it's in another language, summarize in that language.
 
   // src/services/providers/deepseek/DeepSeekChatService.ts
   var DeepSeekChatService = class extends OpenAIChatService {
-    constructor(apiKey, model = MODEL_DEEPSEEK_V4_FLASH, visionModel = model, tools, endpoint = ENDPOINT_DEEPSEEK_CHAT_COMPLETIONS_API, responseLength) {
+    constructor(apiKey, model = MODEL_DEEPSEEK_V4_FLASH, visionModel = model, tools, endpoint = ENDPOINT_DEEPSEEK_CHAT_COMPLETIONS_API, responseLength, reasoningEffort) {
+      const normalizedReasoningEffort = normalizeDeepSeekReasoningEffort(
+        model,
+        reasoningEffort
+      );
       super(
         apiKey,
         model,
@@ -2810,11 +3028,16 @@ If it's in another language, summarize in that language.
         [],
         responseLength,
         void 0,
-        void 0,
+        normalizedReasoningEffort,
         false,
         "deepseek",
         false
       );
+      if (tools?.length && normalizedReasoningEffort !== void 0 && normalizedReasoningEffort !== "none") {
+        throw new Error(
+          "DeepSeek thinking with tools is not supported yet. Use reasoning_effort: 'none' for tool calling."
+        );
+      }
     }
   };
 
@@ -2830,7 +3053,8 @@ If it's in another language, summarize in that language.
         options.visionModel ?? model,
         tools,
         this.resolveEndpoint(options),
-        options.responseLength
+        options.responseLength,
+        options.reasoning_effort
       );
     }
     getProviderName() {
@@ -3006,6 +3230,7 @@ If it's in another language, summarize in that language.
       currentRole = null;
     };
     for (const msg of messages) {
+      if (msg.role === "system") continue;
       const role = mapRoleToGemini(msg.role);
       if (msg.tool_calls) {
         pushCurrent();
@@ -3049,6 +3274,7 @@ If it's in another language, summarize in that language.
       currentRole = null;
     };
     for (const msg of messages) {
+      if (msg.role === "system") continue;
       const role = mapRoleToGemini(msg.role);
       if (msg.tool_calls) {
         pushCurrent();
@@ -3103,6 +3329,18 @@ If it's in another language, summarize in that language.
     pushCurrent();
     return geminiMessages;
   }
+  function extractGeminiSystemInstruction(messages) {
+    const parts = messages.flatMap((message) => {
+      if (message.role !== "system") return [];
+      if (typeof message.content === "string") {
+        return message.content ? [{ text: message.content }] : [];
+      }
+      return message.content.flatMap(
+        (block) => block.type === "text" && block.text ? [{ text: block.text }] : []
+      );
+    });
+    return parts.length > 0 ? { parts } : void 0;
+  }
   function createFunctionCallPart(call, options) {
     options.callIdMap?.set(call.id, call.function.name);
     const context = options.functionCallContextMap?.get(call.id);
@@ -3445,6 +3683,8 @@ If it's in another language, summarize in that language.
           maxOutputTokens: maxTokens !== void 0 ? maxTokens : getMaxTokensForResponseLength(this.responseLength)
         }
       };
+      const systemInstruction = extractGeminiSystemInstruction(messages);
+      if (systemInstruction) body.systemInstruction = systemInstruction;
       const thinkingConfig = this.resolveThinkingConfig(model);
       if (thinkingConfig) {
         body.generationConfig.thinkingConfig = thinkingConfig;
@@ -3776,7 +4016,7 @@ If it's in another language, summarize in that language.
       const api = getLanguageModelAPI();
       if (!api) {
         throw new Error(
-          "Gemini Nano is not available in this environment. Chrome 138+ with Prompt API enabled is required."
+          "Gemini Nano is not available in this environment. Chrome 148+ on a supported desktop device is required for web pages."
         );
       }
       const modelOptions = this.getModelOptions();
@@ -4606,7 +4846,6 @@ If it's in another language, summarize in that language.
   // src/services/providers/openrouter/OpenRouterChatService.ts
   var isOpenRouterAutoModel = (model) => [MODEL_OPENROUTER_AUTO, MODEL_OPENROUTER_AUTO_BETA].includes(model.trim());
   var isTokenLimitUnsupportedModel = (model) => [MODEL_GPT_OSS_20B_FREE, MODEL_ZAI_GLM_5_2].includes(model.trim());
-  var getDefaultReasoningEffort = (model) => model.trim() === MODEL_ZAI_GLM_5_2 ? "none" : void 0;
   var ensureAutoRouterOutput = (model, hasVisibleOutput) => {
     if (isOpenRouterAutoModel(model) && !hasVisibleOutput) {
       throw new Error(
@@ -4818,16 +5057,20 @@ If it's in another language, summarize in that language.
       } else if (tokenLimit) {
         body.max_tokens = tokenLimit;
       }
-      const defaultReasoningEffort = getDefaultReasoningEffort(model);
-      if (this.reasoning_effort !== void 0 || this.includeReasoning !== void 0 || this.reasoningMaxTokens || defaultReasoningEffort) {
+      const defaultReasoningEffort = getDefaultOpenRouterReasoningEffort(model);
+      const reasoningEffort = normalizeOpenRouterReasoningEffort(
+        model,
+        this.reasoning_effort
+      );
+      if (reasoningEffort !== void 0 || this.includeReasoning !== void 0 || this.reasoningMaxTokens || defaultReasoningEffort) {
         body.reasoning = {};
-        if (this.reasoning_effort && this.reasoning_effort !== "none") {
-          const effort = this.reasoning_effort === "minimal" ? "low" : this.reasoning_effort;
+        if (reasoningEffort) {
+          const effort = reasoningEffort === "minimal" ? "low" : reasoningEffort;
           body.reasoning.effort = effort;
-        } else if (this.reasoning_effort === void 0 && defaultReasoningEffort) {
+        } else if (defaultReasoningEffort) {
           body.reasoning.effort = defaultReasoningEffort;
         }
-        if (this.reasoning_effort === "none" || this.includeReasoning !== true) {
+        if (reasoningEffort === "none" || this.includeReasoning !== true) {
           body.reasoning.exclude = true;
         }
         if (this.reasoningMaxTokens) {
@@ -4970,6 +5213,9 @@ If it's in another language, summarize in that language.
         // xAI models
         MODEL_XAI_GROK_LATEST,
         MODEL_XAI_GROK_4_5,
+        // DeepSeek models
+        MODEL_OPENROUTER_DEEPSEEK_V4_FLASH_0731,
+        MODEL_OPENROUTER_DEEPSEEK_V4_FLASH,
         // Other models
         MODEL_MOONSHOTAI_KIMI_LATEST,
         MODEL_MOONSHOTAI_KIMI_K3,
@@ -5377,6 +5623,7 @@ If it's in another language, summarize in that language.
      */
     getSupportedModels() {
       return [
+        MODEL_GROK_4_6,
         MODEL_GROK_4_5,
         MODEL_GROK_4_3,
         MODEL_GROK_4_20_REASONING,
@@ -5675,21 +5922,120 @@ If it's in another language, summarize in that language.
     new PlamoChatServiceProvider()
   ];
 
-  // src/services/ChatServiceFactory.ts
-  var TOOL_SUPPORTED_PROVIDERS = /* @__PURE__ */ new Set([
+  // src/backend.ts
+  var KNOWN_CHAT_PROVIDERS = /* @__PURE__ */ new Set([
+    "claude",
+    "claude-agent-sdk",
+    "codex-sdk",
+    "copilot-sdk",
+    "deepseek",
+    "gemini",
+    "gemini-nano",
+    "kimi",
+    "mistral",
     "openai",
     "openai-compatible",
     "openrouter",
-    "gemini",
-    "claude",
-    "zai",
+    "plamo",
+    "sakana",
     "xai",
-    "kimi",
+    "zai"
+  ]);
+  var TOOL_SUPPORTED_PROVIDERS = /* @__PURE__ */ new Set([
+    "claude",
     "deepseek",
+    "gemini",
+    "kimi",
     "mistral",
+    "openai",
+    "openai-compatible",
+    "openrouter",
+    "plamo",
     "sakana",
-    "plamo"
+    "xai",
+    "zai"
   ]);
+  var NON_STREAMING_PROVIDERS = /* @__PURE__ */ new Set(["codex-sdk"]);
+  function getChatBackendProviderCapabilities(provider) {
+    if (!KNOWN_CHAT_PROVIDERS.has(provider)) return void 0;
+    return {
+      text: true,
+      streaming: !NON_STREAMING_PROVIDERS.has(provider),
+      tools: TOOL_SUPPORTED_PROVIDERS.has(provider)
+    };
+  }
+  function buildToolContinuationMessages(input) {
+    const isClaude = input.provider === "claude";
+    const messages = input.messages.filter(
+      (message) => !isEmptyClaudeAssistantMessage(isClaude, message)
+    );
+    const toolUses = input.completion.blocks.filter(
+      (block) => block.type === "tool_use"
+    );
+    const assistantMessage = input.completion.assistant_message ?? buildAssistantToolCall(isClaude, toolUses);
+    if (!isEmptyClaudeAssistantMessage(isClaude, assistantMessage)) {
+      messages.push({ ...assistantMessage });
+    }
+    messages.push(...buildToolResultMessages(isClaude, input.toolResults));
+    return messages;
+  }
+  function isEmptyClaudeAssistantMessage(isClaude, message) {
+    if (!isClaude || message.role !== "assistant") {
+      return false;
+    }
+    const content = message.content;
+    return Array.isArray(content) && content.length === 0;
+  }
+  function buildAssistantToolCall(isClaude, toolUses) {
+    if (isClaude) {
+      return {
+        role: "assistant",
+        content: toolUses.map((toolUse) => ({
+          type: "tool_use",
+          id: toolUse.id,
+          name: toolUse.name,
+          input: toolUse.input ?? {}
+        }))
+      };
+    }
+    return {
+      role: "assistant",
+      content: [],
+      tool_calls: toolUses.map((toolUse) => ({
+        id: toolUse.id,
+        type: "function",
+        function: {
+          name: toolUse.name,
+          arguments: JSON.stringify(toolUse.input ?? {})
+        }
+      }))
+    };
+  }
+  function buildToolResultMessages(isClaude, toolResults) {
+    if (isClaude) {
+      return toolResults.map(
+        (result) => ({
+          role: "user",
+          content: [
+            {
+              type: "tool_result",
+              tool_use_id: result.tool_use_id,
+              content: result.content
+            }
+          ]
+        })
+      );
+    }
+    return toolResults.map(
+      (result) => ({
+        role: "tool",
+        tool_call_id: result.tool_use_id,
+        content: result.content
+      })
+    );
+  }
+
+  // src/services/ChatServiceFactory.ts
   var MCP_SUPPORTED_PROVIDERS = /* @__PURE__ */ new Set(["openai", "gemini", "claude"]);
   var JSON_MODE_SUPPORTED_PROVIDERS = /* @__PURE__ */ new Set([
     "openai",
@@ -5701,11 +6047,12 @@ If it's in another language, summarize in that language.
     openai: ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
     claude: ["low", "medium", "high", "xhigh", "max"],
     gemini: ["minimal", "low", "medium", "high"],
-    openrouter: ["none", "minimal", "low", "medium", "high"],
+    openrouter: ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
     kimi: ["low", "high", "max"],
     mistral: ["low", "medium", "high"],
     plamo: ["none", "medium"],
-    xai: ["none", "low", "medium", "high"]
+    xai: ["none", "low", "medium", "high", "xhigh"],
+    deepseek: ["none", "low", "high", "max"]
   };
   var ChatServiceFactory = class {
     /**
@@ -5754,18 +6101,19 @@ If it's in another language, summarize in that language.
         return void 0;
       }
       const vision = model ? this.getVisionSupportLevelForModel(providerName, model) : provider.getVisionSupportLevel();
+      const backendCapabilities = getChatBackendProviderCapabilities(providerName);
       return {
         provider: providerName,
         models: provider.getSupportedModels(),
         defaultModel: typeof provider.getDefaultModel === "function" ? provider.getDefaultModel() : void 0,
         text: true,
-        streaming: true,
+        streaming: backendCapabilities?.streaming ?? true,
         vision,
-        tools: TOOL_SUPPORTED_PROVIDERS.has(providerName),
+        tools: backendCapabilities?.tools ?? false,
         mcp: MCP_SUPPORTED_PROVIDERS.has(providerName),
         jsonMode: JSON_MODE_SUPPORTED_PROVIDERS.has(providerName),
         responseLength: true,
-        reasoningEffort: providerName === "claude" && model ? [...getClaudeSupportedReasoningEfforts(model)] : providerName === "kimi" && model ? [...getKimiSupportedReasoningEfforts(model)] : REASONING_EFFORT_BY_PROVIDER[providerName] ?? []
+        reasoningEffort: providerName === "claude" && model ? [...getClaudeSupportedReasoningEfforts(model)] : providerName === "kimi" && model ? [...getKimiSupportedReasoningEfforts(model)] : providerName === "deepseek" && model ? [...getDeepSeekSupportedReasoningEfforts(model)] : providerName === "openrouter" && model ? [...getOpenRouterSupportedReasoningEfforts(model)] : providerName === "xai" && model ? [...getXaiSupportedReasoningEfforts(model)] : REASONING_EFFORT_BY_PROVIDER[providerName] ?? []
       };
     }
     /**