openfox 2.0.0-beta.11 → 2.0.0-beta.14

package/dist/{chunk-EQQDA4D3.js → chunk-IONKVOSF.js}

@@ -14,12 +14,15 @@ import {
   getMaxPerSession,
   getSessionProcessCount
 } from "./chunk-VRGRAQDG.js";
+import {
+  getProject
+} from "./chunk-XAMAYRDA.js";
 import {
   getCurrentContextWindowId,
   getCurrentWindowMessageOptions,
   getEventStore,
   getRuntimeConfig
-} from "./chunk-NSBYORD5.js";
+} from "./chunk-DHAKJSE7.js";
 import {
   buildContextMessagesFromEventHistory,
   foldContextState,
@@ -46,21 +49,18 @@ import {
   getSetting
 } from "./chunk-RFNEDBVO.js";
 import {
-  getProject
-} from "./chunk-XAMAYRDA.js";
+  getGlobalConfigDir
+} from "./chunk-CQGTEGKL.js";
 import {
   buildModelParams,
   getBackendCapabilities,
   getModelProfile,
   modelSupportsVision,
   streamWithSegments
-} from "./chunk-VIIRNJDT.js";
+} from "./chunk-LXX2CPM5.js";
 import {
   logger
 } from "./chunk-K44MW7JJ.js";
-import {
-  getGlobalConfigDir
-} from "./chunk-CQGTEGKL.js";
 
 // src/server/tools/read.ts
 import { readFile as readFile2, stat, readdir } from "fs/promises";
@@ -2001,12 +2001,12 @@ function matchRetryPatterns(content, thinking, patterns) {
 
 // src/server/chat/stream-utils.ts
 function buildStreamRequestObject(params) {
-  const { messages, tools, toolChoice, disableThinking, signal, modelSettings } = params;
+  const { messages, tools, toolChoice, reasoningEffort, signal, modelSettings } = params;
   return {
     messages,
     ...tools && { tools },
     ...toolChoice && { toolChoice },
-    disableThinking: disableThinking ?? false,
+    ...reasoningEffort && { reasoningEffort },
     ...signal && { signal },
     ...modelSettings && { modelSettings }
   };
@@ -2060,7 +2060,7 @@ function createEmptyStreamResult(aborted, modelParams, patternMatch) {
   };
 }
 async function* streamLLMPure(options) {
-  const { messageId, systemPrompt, llmClient, messages, tools, toolChoice, signal, disableThinking, retryPatterns } = options;
+  const { messageId, systemPrompt, llmClient, messages, tools, toolChoice, signal, reasoningEffort, retryPatterns } = options;
   const llmMessages = [{ role: "system", content: systemPrompt }, ...messages];
   const profile = getModelProfile(llmClient.getModel());
   const backend = getBackendCapabilities(llmClient.getBackend());
@@ -2093,7 +2093,7 @@ async function* streamLLMPure(options) {
     messages: llmMessages,
     tools,
     toolChoice,
-    disableThinking: disableThinking ?? false,
+    reasoningEffort,
     signal: combinedSignal,
     modelSettings: options.modelSettings
   });
@@ -2648,7 +2648,7 @@ async function executeTools(assistantMsgId, toolCalls, ctx, append) {
         type: "chat.ask_user",
         data: { callId: error.callId, question: error.question }
       });
-      const { awaitAnswer } = await import("./ask-5SSOA6SZ.js");
+      const { awaitAnswer } = await import("./ask-BRLFZKEW.js");
       const answerPromise = awaitAnswer(error.callId);
       if (!answerPromise) {
         throw new Error(`No pending question found for callId: ${error.callId}`);
@@ -2929,7 +2929,7 @@ ${CONTINUE_PROMPT}` : CONTINUE_PROMPT;
     sessionManager.setCurrentContextSize(sessionId, result.usage.promptTokens);
     if (!compacting) {
       const contextState = sessionManager.getContextState(sessionId);
-      const { shouldCompact } = await import("./compactor-TDNRM3A7.js");
+      const { shouldCompact } = await import("./compactor-WL33VHUF.js");
       if (shouldCompact(contextState.currentTokens, contextState.maxTokens, runtimeConfig.context.compactionThreshold)) {
         const compactPromptMsgId = crypto.randomUUID();
         append(
@@ -3051,7 +3051,7 @@ ${COMPACTION_PROMPT}`,
                   stats: stats2
                 })
               );
-              const { createChatDoneMessage: createChatDoneMessage2 } = await import("./protocol-CN24IKQN.js");
+              const { createChatDoneMessage: createChatDoneMessage2 } = await import("./protocol-QQNATDFM.js");
               onMessage(createChatDoneMessage2(assistantMsgId, "complete", stats2));
             }
             break;
@@ -3189,7 +3189,6 @@ function buildAssemblyInput(systemPrompt, baseInput) {
     injectedFiles: baseInput.injectedFiles,
     requestTools: baseInput.requestTools ?? baseInput.promptTools,
     toolChoice: baseInput.toolChoice ?? "auto",
-    disableThinking: baseInput.disableThinking ?? false,
     ...baseInput.customInstructions ? { customInstructions: baseInput.customInstructions } : {},
     ...baseInput.skills && baseInput.skills.length > 0 ? { skills: baseInput.skills } : {}
   });
@@ -3281,7 +3280,7 @@ async function describeImageFromDataUrl(dataUrl, visionModel, options) {
 import { createHash as createHash2 } from "crypto";
 async function loadVisionModelFromGlobalConfig() {
   try {
-    const { loadGlobalConfig, getVisionFallback } = await import("./config-TDVA7MQN.js");
+    const { loadGlobalConfig, getVisionFallback } = await import("./config-MFKBDOOT.js");
     const runtimeConfig = getRuntimeConfig();
     const mode = runtimeConfig.mode ?? "production";
     const globalConfig = await loadGlobalConfig(mode);
@@ -3644,7 +3643,6 @@ async function executeSubAgent(options) {
         injectedFiles: input.injectedFiles,
         requestTools: input.promptTools,
         toolChoice: input.toolChoice,
-        disableThinking: true,
         ...instructionContent ? { customInstructions: instructionContent } : {},
         ...skills.length > 0 ? { skills } : {}
       }),
@@ -3735,7 +3733,7 @@ var callSubAgentTool = {
       };
     }
     try {
-      const { getToolRegistryForAgent: getToolRegistryForAgent2 } = await import("./tools-D3WWIOQD.js");
+      const { getToolRegistryForAgent: getToolRegistryForAgent2 } = await import("./tools-2X2FC6W7.js");
       const toolRegistry = getToolRegistryForAgent2(agentDef);
       const turnMetrics = new TurnMetrics();
       const result = await executeSubAgent({
@@ -4787,4 +4785,4 @@ export {
   getToolRegistryForAgent,
   createToolRegistry
 };
-//# sourceMappingURL=chunk-EQQDA4D3.js.map
+//# sourceMappingURL=chunk-IONKVOSF.js.map

package/dist/{chunk-VIIRNJDT.js → chunk-LXX2CPM5.js}

@@ -2,6 +2,176 @@ import {
   logger
 } from "./chunk-K44MW7JJ.js";
 
+// src/server/llm/profiles.ts
+var DEFAULT_PROFILE = {
+  name: "default",
+  temperature: 0.7,
+  topP: 0.9,
+  defaultMaxTokens: 4096,
+  supportsVision: true
+};
+var MOCK_PROFILE = {
+  name: "mock",
+  temperature: 0.7,
+  topP: 0.9,
+  defaultMaxTokens: 1024,
+  supportsVision: false
+};
+var MODEL_PROFILES = [
+  {
+    pattern: "mistral",
+    profile: {
+      name: "Mistral",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "qwen3-coder-next",
+    profile: {
+      name: "Qwen3-Coder-Next",
+      // Per Qwen docs: "temperature=1.0, top_p=0.95, top_k=40"
+      temperature: 1,
+      topP: 0.95,
+      topK: 40,
+      // "This model supports only non-thinking mode and does not generate <think></think> blocks"
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "qwen3",
+    profile: {
+      name: "Qwen3",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "qwen3-vl",
+    profile: {
+      name: "Qwen3-VL",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: true
+    }
+  },
+  {
+    pattern: "deepseek",
+    profile: {
+      name: "DeepSeek",
+      temperature: 0.6,
+      topP: 0.95,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "minimax-m2.5",
+    profile: {
+      name: "MiniMax-M2.5",
+      temperature: 1,
+      topP: 0.95,
+      topK: 40,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "minimax-m2.7",
+    profile: {
+      name: "MiniMax-M2.7",
+      temperature: 1,
+      topP: 0.95,
+      topK: 40,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "minimax",
+    profile: {
+      name: "MiniMax",
+      temperature: 1,
+      topP: 0.95,
+      topK: 40,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "minimax-m3",
+    profile: {
+      name: "MiniMax-M3",
+      temperature: 1,
+      topP: 0.95,
+      topK: 40,
+      defaultMaxTokens: 16384,
+      supportsVision: true
+    }
+  },
+  {
+    pattern: "llava",
+    profile: {
+      name: "LLaVA",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: true
+    }
+  },
+  {
+    pattern: "llama",
+    profile: {
+      name: "Llama",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "claude",
+    profile: {
+      name: "Claude",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: false
+    }
+  },
+  {
+    pattern: "gemma-4",
+    profile: {
+      name: "Gemma 4",
+      temperature: 0.7,
+      topP: 0.9,
+      defaultMaxTokens: 16384,
+      supportsVision: true
+    }
+  }
+];
+function getModelProfile(modelName) {
+  const lowerName = modelName.toLowerCase();
+  if (lowerName.includes("mock")) {
+    return MOCK_PROFILE;
+  }
+  for (const { pattern, profile } of MODEL_PROFILES) {
+    if (lowerName.includes(pattern.toLowerCase())) {
+      return profile;
+    }
+  }
+  return DEFAULT_PROFILE;
+}
+function modelSupportsVision(modelName) {
+  return getModelProfile(modelName).supportsVision;
+}
+
 // src/server/llm/url-utils.ts
 var VERSION_PREFIX_REGEX = /\/v\d+(\/|$)/;
 function hasVersionPrefix(url) {
@@ -21,33 +191,27 @@ function buildModelsUrl(baseUrl) {
 // src/server/llm/backend.ts
 var BACKEND_CAPABILITIES = {
   vllm: {
-    supportsReasoningField: true,
     supportsChatTemplateKwargs: true,
     supportsTopK: true
   },
   sglang: {
-    supportsReasoningField: true,
     supportsChatTemplateKwargs: true,
     supportsTopK: true
   },
   ollama: {
-    supportsReasoningField: false,
     supportsChatTemplateKwargs: false,
     supportsTopK: false
   },
   llamacpp: {
-    supportsReasoningField: false,
     supportsChatTemplateKwargs: false,
     supportsTopK: true
   },
   "opencode-go": {
-    supportsReasoningField: false,
     supportsChatTemplateKwargs: false,
     supportsTopK: true
   },
   unknown: {
     // Assume vLLM-like for unknown backends
-    supportsReasoningField: true,
     supportsChatTemplateKwargs: true,
     supportsTopK: true
   }
@@ -179,206 +343,6 @@ function getBackendDisplayName(backend) {
   }
 }
 
-// src/server/llm/profiles.ts
-var DEFAULT_PROFILE = {
-  name: "default",
-  temperature: 0.7,
-  topP: 0.9,
-  supportsReasoning: false,
-  reasoningAsContent: false,
-  defaultMaxTokens: 4096,
-  supportsVision: true
-};
-var MOCK_PROFILE = {
-  name: "mock",
-  temperature: 0.7,
-  topP: 0.9,
-  supportsReasoning: false,
-  reasoningAsContent: false,
-  defaultMaxTokens: 1024,
-  supportsVision: false
-};
-var MODEL_PROFILES = [
-  {
-    pattern: "mistral",
-    profile: {
-      name: "Mistral",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: false,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "qwen3-coder-next",
-    profile: {
-      name: "Qwen3-Coder-Next",
-      // Per Qwen docs: "temperature=1.0, top_p=0.95, top_k=40"
-      temperature: 1,
-      topP: 0.95,
-      topK: 40,
-      // "This model supports only non-thinking mode and does not generate <think></think> blocks"
-      supportsReasoning: false,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "qwen3",
-    profile: {
-      name: "Qwen3",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "qwen3-vl",
-    profile: {
-      name: "Qwen3-VL",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: true
-    }
-  },
-  {
-    pattern: "deepseek",
-    profile: {
-      name: "DeepSeek",
-      temperature: 0.6,
-      topP: 0.95,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "minimax-m2.5",
-    profile: {
-      name: "MiniMax-M2.5",
-      temperature: 1,
-      topP: 0.95,
-      topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "minimax-m2.7",
-    profile: {
-      name: "MiniMax-M2.7",
-      temperature: 1,
-      topP: 0.95,
-      topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "minimax",
-    profile: {
-      name: "MiniMax",
-      temperature: 1,
-      topP: 0.95,
-      topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "minimax-m3",
-    profile: {
-      name: "MiniMax-M3",
-      temperature: 1,
-      topP: 0.95,
-      topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: true
-    }
-  },
-  {
-    pattern: "llava",
-    profile: {
-      name: "LLaVA",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: false,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: true
-    }
-  },
-  {
-    pattern: "llama",
-    profile: {
-      name: "Llama",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: false,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "claude",
-    profile: {
-      name: "Claude",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: false
-    }
-  },
-  {
-    pattern: "gemma-4",
-    profile: {
-      name: "Gemma 4",
-      temperature: 0.7,
-      topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
-      defaultMaxTokens: 16384,
-      supportsVision: true
-    }
-  }
-];
-function getModelProfile(modelName) {
-  const lowerName = modelName.toLowerCase();
-  if (lowerName.includes("mock")) {
-    return MOCK_PROFILE;
-  }
-  for (const { pattern, profile } of MODEL_PROFILES) {
-    if (lowerName.includes(pattern.toLowerCase())) {
-      return profile;
-    }
-  }
-  return DEFAULT_PROFILE;
-}
-function modelSupportsVision(modelName) {
-  return getModelProfile(modelName).supportsVision;
-}
-
 // src/server/llm/client-pure.ts
 function buildModelParams(params) {
   return {
@@ -408,7 +372,14 @@ function convertToolCalls(toolCalls) {
     }
   }));
 }
-function buildAssistantMessage(msg) {
+function getThinking(msg, override) {
+  if (override) {
+    const val = msg[override];
+    if (val) return val;
+  }
+  return msg["reasoning"] ?? msg["reasoning_content"] ?? msg["thinking"];
+}
+function buildAssistantMessage(msg, thinkingField) {
   const result = {
     role: "assistant",
     content: msg.content || null
@@ -417,7 +388,7 @@ function buildAssistantMessage(msg) {
     result["tool_calls"] = convertToolCalls(msg.toolCalls);
   }
   if (msg.thinkingContent) {
-    result["reasoning_content"] = msg.thinkingContent;
+    result[thinkingField ?? "reasoning"] = msg.thinkingContent;
   }
   return result;
 }
@@ -433,7 +404,7 @@ function convertAttachmentSync(attachment, modelSupportsVision2) {
     text: `[Image: ${attachment.filename || "image"}] (vision not supported, cannot describe)`
   };
 }
-function convertMessages(messages, modelSupportsVision2) {
+function convertMessages(messages, modelSupportsVision2, thinkingField) {
   const filtered = messages.filter((msg) => {
     return !(msg.role === "assistant" && !msg.content?.trim() && (!msg.toolCalls || msg.toolCalls.length === 0));
   });
@@ -454,7 +425,7 @@ function convertMessages(messages, modelSupportsVision2) {
       };
     }
     if (msg.role === "assistant") {
-      return buildAssistantMessage(msg);
+      return buildAssistantMessage(msg, thinkingField);
     }
     if (msg.role === "user" && msg.attachments && msg.attachments.length > 0) {
       const content = buildAttachmentContent(msg.content, msg.attachments, modelSupportsVision2);
@@ -479,10 +450,10 @@ function convertTools(tools) {
     }
   }));
 }
-async function buildChatCompletionCreateParams(model, request, profile, capabilities, disableThinking, isStreaming) {
+async function buildChatCompletionCreateParams(model, request, profile, capabilities, reasoningEffort, isStreaming, thinkingField) {
   const userVisionOverride = request.modelSettings?.supportsVision;
   const modelSupportsVision2 = userVisionOverride ?? profile.supportsVision ?? false;
-  const convertedMessages = convertMessages(request.messages, modelSupportsVision2);
+  const convertedMessages = convertMessages(request.messages, modelSupportsVision2, thinkingField);
   const temperature = request.modelSettings?.temperature ?? request.temperature ?? profile.temperature;
   const maxTokens = request.modelSettings?.maxTokens ?? request.maxTokens ?? profile.defaultMaxTokens;
   const topP = request.modelSettings?.topP ?? profile.topP;
@@ -502,23 +473,34 @@ async function buildChatCompletionCreateParams(model, request, profile, capabili
     ;
     params["top_k"] = topK;
   }
-  const shouldDisableThinking = disableThinking || request.disableThinking;
-  if (capabilities.supportsChatTemplateKwargs && profile.supportsReasoning && shouldDisableThinking) {
+  const resolvedEffort = reasoningEffort ?? request.reasoningEffort;
+  if (resolvedEffort) {
     ;
-    params["chat_template_kwargs"] = { enable_thinking: false };
+    params["reasoning_effort"] = resolvedEffort;
+    const chatTemplateKwargs = request.modelSettings?.chatTemplateKwargs;
+    if (chatTemplateKwargs) {
+      ;
+      params["chat_template_kwargs"] = chatTemplateKwargs;
+    } else if (capabilities.supportsChatTemplateKwargs) {
+      ;
+      params["chat_template_kwargs"] = {
+        enable_thinking: resolvedEffort !== "none"
+      };
+    }
   }
   const modelParams = buildModelParams({ temperature, topP, topK, maxTokens });
   return { params, modelParams };
 }
 async function buildCreateParamsFromInput(input, isStreaming) {
-  const { model, request, profile, capabilities, disableThinking } = input;
+  const { model, request, profile, capabilities, reasoningEffort, thinkingField } = input;
   return buildChatCompletionCreateParams(
     model,
     request,
     profile,
     capabilities,
-    !!disableThinking,
-    isStreaming
+    reasoningEffort,
+    isStreaming,
+    thinkingField
   );
 }
 var buildNonStreamingCreateParams = (input) => buildCreateParamsFromInput(input, false);
@@ -537,20 +519,6 @@ function mapFinishReason(reason) {
       return "stop";
   }
 }
-function extractThinking(content) {
-  const thinkRegex = /<think>([\s\S]*?)<\/think>/g;
-  let thinkingContent = "";
-  let cleanContent = content;
-  let match;
-  while ((match = thinkRegex.exec(content)) !== null) {
-    thinkingContent += match[1];
-    cleanContent = cleanContent.replace(match[0], "");
-  }
-  return {
-    content: cleanContent.trim(),
-    thinkingContent: thinkingContent.trim() || null
-  };
-}
 
 // src/server/llm/streaming.ts
 async function* streamWithSegments(client, request) {
@@ -630,10 +598,9 @@ async function* streamWithSegments(client, request) {
   const ttft = ((firstTokenTime ?? endTime) - startTime) / 1e3;
   const completionTime = (endTime - (firstTokenTime ?? startTime)) / 1e3;
   const { promptTokens, completionTokens } = response.usage;
-  const effectiveThinkingContent = thinkingContent || content === "" && response.reasoning_content || "";
   return {
     content,
-    thinkingContent: effectiveThinkingContent,
+    thinkingContent,
     toolCalls,
     response,
     segments,
@@ -656,10 +623,10 @@ export {
   detectBackend,
   getBackendDisplayName,
   buildModelParams,
+  getThinking,
   buildNonStreamingCreateParams,
   buildStreamingCreateParams,
   mapFinishReason,
-  extractThinking,
   streamWithSegments
 };
-//# sourceMappingURL=chunk-VIIRNJDT.js.map
+//# sourceMappingURL=chunk-LXX2CPM5.js.map