openfox 2.0.0-beta.9 → 2.0.1

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

@@ -1,191 +1,8 @@
-import {
-  logger
-} from "./chunk-K44MW7JJ.js";
-
-// src/server/llm/url-utils.ts
-var VERSION_PREFIX_REGEX = /\/v\d+(\/|$)/;
-function hasVersionPrefix(url) {
-  return VERSION_PREFIX_REGEX.test(url);
-}
-function ensureVersionPrefix(url, defaultVersion = "/v1") {
-  if (hasVersionPrefix(url)) return url;
-  return `${url.replace(/\/+$/, "")}${defaultVersion}`;
-}
-function stripVersionPrefix(url) {
-  return url.replace(/\/v\d+\/?$/, "");
-}
-function buildModelsUrl(baseUrl) {
-  return `${ensureVersionPrefix(baseUrl)}/models`;
-}
-
-// 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
-  }
-};
-function getBackendCapabilities(backend) {
-  return BACKEND_CAPABILITIES[backend];
-}
-async function detectBackend(baseUrl, explicitBackend, silent = false) {
-  if (explicitBackend && explicitBackend !== "unknown") {
-    if (silent) {
-      logger.debug("Using explicit backend", { backend: explicitBackend });
-    } else {
-      logger.info("Using explicit backend", { backend: explicitBackend });
-    }
-    return explicitBackend;
-  }
-  const probeUrl = stripVersionPrefix(baseUrl);
-  try {
-    if (await probeOllama(probeUrl)) {
-      if (silent) {
-        logger.debug("Detected Ollama backend", { url: probeUrl });
-      } else {
-        logger.info("Detected Ollama backend", { url: probeUrl });
-      }
-      return "ollama";
-    }
-    if (await probeLlamaCpp(probeUrl)) {
-      if (silent) {
-        logger.debug("Detected llama.cpp backend", { url: probeUrl });
-      } else {
-        logger.info("Detected llama.cpp backend", { url: probeUrl });
-      }
-      return "llamacpp";
-    }
-    if (await probeSGLang(probeUrl)) {
-      if (silent) {
-        logger.debug("Detected SGLang backend", { url: probeUrl });
-      } else {
-        logger.info("Detected SGLang backend", { url: probeUrl });
-      }
-      return "sglang";
-    }
-    if (await probeOpenAI(baseUrl)) {
-      if (silent) {
-        logger.debug("Detected vLLM backend (OpenAI-compatible)", { url: baseUrl });
-      } else {
-        logger.info("Detected vLLM backend (OpenAI-compatible)", { url: baseUrl });
-      }
-      return "vllm";
-    }
-    if (silent) {
-      logger.debug("Could not detect backend, using unknown", { url: baseUrl });
-    } else {
-      logger.warn("Could not detect backend, using unknown", { url: baseUrl });
-    }
-    return "unknown";
-  } catch (error) {
-    if (silent) {
-      logger.debug("Backend detection failed", {
-        url: baseUrl,
-        error: error instanceof Error ? error.message : String(error)
-      });
-    } else {
-      logger.warn("Backend detection failed", {
-        url: baseUrl,
-        error: error instanceof Error ? error.message : String(error)
-      });
-    }
-    return "unknown";
-  }
-}
-async function probeOllama(baseUrl) {
-  try {
-    const response = await fetch(`${baseUrl}/api/tags`, {
-      signal: AbortSignal.timeout(5e3)
-    });
-    return response.ok;
-  } catch {
-    return false;
-  }
-}
-async function probeLlamaCpp(baseUrl) {
-  try {
-    const response = await fetch(`${baseUrl}/health`, {
-      signal: AbortSignal.timeout(5e3)
-    });
-    if (!response.ok) return false;
-    const data = await response.json();
-    return "slots_idle" in data || "slots_processing" in data;
-  } catch {
-    return false;
-  }
-}
-async function probeSGLang(baseUrl) {
-  try {
-    const response = await fetch(`${baseUrl}/get_model_info`, {
-      signal: AbortSignal.timeout(5e3)
-    });
-    return response.ok;
-  } catch {
-    return false;
-  }
-}
-async function probeOpenAI(baseUrl) {
-  try {
-    const url = ensureVersionPrefix(baseUrl);
-    const response = await fetch(`${url}/models`, {
-      signal: AbortSignal.timeout(5e3)
-    });
-    return response.ok;
-  } catch {
-    return false;
-  }
-}
-function getBackendDisplayName(backend) {
-  switch (backend) {
-    case "vllm":
-      return "vLLM";
-    case "sglang":
-      return "SGLang";
-    case "ollama":
-      return "Ollama";
-    case "llamacpp":
-      return "llama.cpp";
-    case "opencode-go":
-      return "OpenCode Go";
-    case "unknown":
-      return "Unknown";
-  }
-}
-
 // src/server/llm/profiles.ts
 var DEFAULT_PROFILE = {
   name: "default",
   temperature: 0.7,
   topP: 0.9,
-  supportsReasoning: false,
-  reasoningAsContent: false,
   defaultMaxTokens: 4096,
   supportsVision: true
 };
@@ -193,8 +10,6 @@ var MOCK_PROFILE = {
   name: "mock",
   temperature: 0.7,
   topP: 0.9,
-  supportsReasoning: false,
-  reasoningAsContent: false,
   defaultMaxTokens: 1024,
   supportsVision: false
 };
@@ -205,8 +20,6 @@ var MODEL_PROFILES = [
       name: "Mistral",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: false,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -220,8 +33,6 @@ var MODEL_PROFILES = [
       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
     }
@@ -232,8 +43,6 @@ var MODEL_PROFILES = [
       name: "Qwen3",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -244,8 +53,6 @@ var MODEL_PROFILES = [
       name: "Qwen3-VL",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: true
     }
@@ -256,8 +63,6 @@ var MODEL_PROFILES = [
       name: "DeepSeek",
       temperature: 0.6,
       topP: 0.95,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -269,8 +74,6 @@ var MODEL_PROFILES = [
       temperature: 1,
       topP: 0.95,
       topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -282,8 +85,6 @@ var MODEL_PROFILES = [
       temperature: 1,
       topP: 0.95,
       topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -295,8 +96,6 @@ var MODEL_PROFILES = [
       temperature: 1,
       topP: 0.95,
       topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -308,8 +107,6 @@ var MODEL_PROFILES = [
       temperature: 1,
       topP: 0.95,
       topK: 40,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: true
     }
@@ -320,8 +117,6 @@ var MODEL_PROFILES = [
       name: "LLaVA",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: false,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: true
     }
@@ -332,8 +127,6 @@ var MODEL_PROFILES = [
       name: "Llama",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: false,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -344,8 +137,6 @@ var MODEL_PROFILES = [
       name: "Claude",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: false
     }
@@ -356,8 +147,6 @@ var MODEL_PROFILES = [
       name: "Gemma 4",
       temperature: 0.7,
       topP: 0.9,
-      supportsReasoning: true,
-      reasoningAsContent: false,
       defaultMaxTokens: 16384,
       supportsVision: true
     }
@@ -379,6 +168,53 @@ function modelSupportsVision(modelName) {
   return getModelProfile(modelName).supportsVision;
 }
 
+// src/server/llm/backend.ts
+var BACKEND_CAPABILITIES = {
+  vllm: {
+    supportsChatTemplateKwargs: true,
+    supportsTopK: true
+  },
+  sglang: {
+    supportsChatTemplateKwargs: true,
+    supportsTopK: true
+  },
+  ollama: {
+    supportsChatTemplateKwargs: false,
+    supportsTopK: false
+  },
+  llamacpp: {
+    supportsChatTemplateKwargs: false,
+    supportsTopK: true
+  },
+  "opencode-go": {
+    supportsChatTemplateKwargs: false,
+    supportsTopK: true
+  },
+  unknown: {
+    supportsChatTemplateKwargs: true,
+    supportsTopK: true
+  }
+};
+function getBackendCapabilities(backend) {
+  return BACKEND_CAPABILITIES[backend];
+}
+function getBackendDisplayName(backend) {
+  switch (backend) {
+    case "vllm":
+      return "vLLM";
+    case "sglang":
+      return "SGLang";
+    case "ollama":
+      return "Ollama";
+    case "llamacpp":
+      return "llama.cpp";
+    case "opencode-go":
+      return "OpenCode Go";
+    case "unknown":
+      return "Unknown";
+  }
+}
+
 // src/server/llm/client-pure.ts
 function buildModelParams(params) {
   return {
@@ -408,7 +244,14 @@ 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 +260,7 @@ 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 +276,7 @@ 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 +297,7 @@ 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 +322,10 @@ 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 +345,34 @@ 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 +391,6 @@ 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 +470,9 @@ 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,
@@ -649,17 +488,13 @@ async function* streamWithSegments(client, request) {
 export {
   getModelProfile,
   modelSupportsVision,
-  ensureVersionPrefix,
-  stripVersionPrefix,
-  buildModelsUrl,
   getBackendCapabilities,
-  detectBackend,
   getBackendDisplayName,
   buildModelParams,
+  getThinking,
   buildNonStreamingCreateParams,
   buildStreamingCreateParams,
   mapFinishReason,
-  extractThinking,
   streamWithSegments
 };
-//# sourceMappingURL=chunk-VIIRNJDT.js.map
+//# sourceMappingURL=chunk-FFEAEPJB.js.map

package/dist/{chunk-EZUR7OEP.js → chunk-HPCGVAS4.js}

@@ -1,9 +1,3 @@
-import {
-  detectModel
-} from "./chunk-UKTPL5ZG.js";
-import {
-  detectBackend
-} from "./chunk-VIIRNJDT.js";
 import {
   getGlobalConfigPath
 } from "./chunk-CQGTEGKL.js";
@@ -13,26 +7,6 @@ import { z } from "zod";
 import { randomUUID } from "crypto";
 import { readFile, writeFile, mkdir, access } from "fs/promises";
 import { dirname } from "path";
-var SMART_DEFAULTS = ["http://localhost:8000", "http://localhost:11434", "http://localhost:8080"];
-async function trySmartDefaults(_mode) {
-  const results = await Promise.all(
-    SMART_DEFAULTS.map(async (url) => {
-      try {
-        const [backend, model] = await Promise.all([
-          detectBackend(url, void 0, true),
-          detectModel(url, 1, true)
-          // Only 1 retry attempt
-        ]);
-        if (backend !== "unknown" && model) {
-          return { url, backend, model };
-        }
-      } catch {
-      }
-      return null;
-    })
-  );
-  return results.find((r) => r !== null) || null;
-}
 async function configFileExists(mode) {
   const configPath = getGlobalConfigPath(mode);
   try {
@@ -42,17 +16,7 @@ async function configFileExists(mode) {
     return false;
   }
 }
-var backendSchema = z.enum([
-  "auto",
-  "vllm",
-  "sglang",
-  "ollama",
-  "llamacpp",
-  "openai",
-  "anthropic",
-  "opencode-go",
-  "unknown"
-]);
+var backendSchema = z.enum(["vllm", "sglang", "ollama", "llamacpp", "openai", "anthropic", "opencode-go", "unknown"]);
 var modelConfigSchema = z.object({
   id: z.string(),
   contextWindow: z.number(),
@@ -122,10 +86,17 @@ var configSchema = z.object({
 var oldLlmSchema = z.object({
   url: z.string().url().default("http://localhost:8000/v1"),
   model: z.string().default("auto"),
-  backend: z.enum(["auto", "vllm", "sglang", "ollama", "llamacpp"]).default("auto"),
+  backend: z.enum(["vllm", "sglang", "ollama", "llamacpp", "unknown"]).default("unknown"),
   maxContext: z.number().default(2e5),
-  disableThinking: z.boolean().default(false),
+  // Deprecated: use reasoningEffort instead
+  disableThinking: z.boolean().optional(),
+  reasoningEffort: z.string().optional(),
   apiKey: z.string().optional()
+}).transform((data) => {
+  if (data.disableThinking && !data.reasoningEffort) {
+    return { ...data, reasoningEffort: "none", disableThinking: void 0 };
+  }
+  return { ...data, disableThinking: void 0 };
 });
 var oldConfigSchema = z.object({
   llm: oldLlmSchema,
@@ -141,7 +112,8 @@ function migrateConfig(raw) {
     const obj = raw;
     let migrationOccurred = false;
     const providers = obj.providers.map((p) => {
-      const { model, maxContext, models: existingModels, ...rest } = p;
+      const { model, maxContext, models: existingModels, backend, ...rest } = p;
+      const normalizedBackend = backend === "auto" ? "unknown" : backend;
       let models = existingModels ?? [];
       if (maxContext !== void 0 && (existingModels === void 0 || existingModels.length === 0)) {
         migrationOccurred = true;
@@ -155,6 +127,7 @@ function migrateConfig(raw) {
       }
       return {
         ...rest,
+        backend: normalizedBackend,
         models
       };
     });
@@ -189,6 +162,11 @@ function migrateConfig(raw) {
     };
   }
   if (typeof raw === "object" && raw !== null && "llm" in raw) {
+    const rawObj = raw;
+    const llm = rawObj["llm"];
+    if (llm && llm["backend"] === "auto") {
+      llm["backend"] = "unknown";
+    }
     const oldConfig = oldConfigSchema.parse(raw);
     const providerId = randomUUID();
     const models = [
@@ -415,9 +393,8 @@ function mergeConfigs(...configs) {
       llm: {
         url: "http://localhost:8000/v1",
         model: "auto",
-        backend: "auto",
-        maxContext: 2e5,
-        disableThinking: false
+        backend: "unknown",
+        maxContext: 2e5
       },
       server: { port: 10369, host: "127.0.0.1", openBrowser: true },
       logging: { level: "error" }
@@ -427,7 +404,6 @@ function mergeConfigs(...configs) {
 }
 
 export {
-  trySmartDefaults,
   configFileExists,
   getVisionFallback,
   migrateConfig,
@@ -443,4 +419,4 @@ export {
   activateProvider,
   mergeConfigs
 };
-//# sourceMappingURL=chunk-EZUR7OEP.js.map
+//# sourceMappingURL=chunk-HPCGVAS4.js.map