openfox 2.0.124 → 2.0.126

package/dist/{chunk-SYTEO5LA.js → chunk-FNXTJLCZ.js}

@@ -3,9 +3,10 @@ import {
 } from "./chunk-GJL3QSMP.js";
 import {
   getBackendCapabilities
-} from "./chunk-AZHWLLPP.js";
+} from "./chunk-PBFGJFGC.js";
 import {
-  ensureVersionPrefix
+  ensureVersionPrefix,
+  stripVersionPrefix
 } from "./chunk-HNCM3D7Y.js";
 import {
   buildNonStreamingCreateParams,
@@ -13,15 +14,199 @@ import {
   getThinking,
   mapFinishReason,
   parseToolArguments
-} from "./chunk-MIPQOHWS.js";
+} from "./chunk-KAZWU73N.js";
 import {
+  ChatHttpClient,
   LLMError,
   OpenAIHttpClient
-} from "./chunk-VPRB5FNR.js";
+} from "./chunk-ZFNBATE3.js";
 import {
   logger
 } from "./chunk-K44MW7JJ.js";
 
+// src/server/llm/ollama-native.ts
+function mapDoneReason(reason) {
+  switch (reason) {
+    case "length":
+      return "length";
+    case "tool_calls":
+      return "tool_calls";
+    case "content_filter":
+      return "content_filter";
+    default:
+      return "stop";
+  }
+}
+function stringifyToolArguments(rawArguments) {
+  if (typeof rawArguments === "string") return rawArguments;
+  if (rawArguments === void 0 || rawArguments === null) return "{}";
+  return JSON.stringify(rawArguments);
+}
+var OLLAMA_THINK_LEVELS = /* @__PURE__ */ new Set(["low", "medium", "high", "max"]);
+function toOllamaThink(effort) {
+  if (effort === "none") return false;
+  if (effort === "xhigh") return "max";
+  if (OLLAMA_THINK_LEVELS.has(effort)) return effort;
+  return true;
+}
+function toNativeMessage(message) {
+  const raw = message;
+  const toolCalls = raw["tool_calls"];
+  if (!Array.isArray(toolCalls) || toolCalls.length === 0) {
+    return raw;
+  }
+  return {
+    ...raw,
+    tool_calls: toolCalls.map((toolCall) => {
+      const fn = toolCall.function;
+      if (!fn || typeof fn.arguments !== "string") return toolCall;
+      try {
+        return {
+          ...toolCall,
+          function: { ...fn, arguments: JSON.parse(fn.arguments) }
+        };
+      } catch {
+        return toolCall;
+      }
+    })
+  };
+}
+function buildOllamaChatRequest(params) {
+  const options = {};
+  if (params.temperature !== void 0) options["temperature"] = params.temperature;
+  if (params.top_p !== void 0) options["top_p"] = params.top_p;
+  if (params.top_k !== void 0) options["top_k"] = params.top_k;
+  if (params.max_tokens !== void 0) options["num_predict"] = params.max_tokens;
+  if (params["num_ctx"] !== void 0) options["num_ctx"] = params["num_ctx"];
+  if (params["stop"]) options["stop"] = params["stop"];
+  if (params["frequency_penalty"] !== void 0) options["frequency_penalty"] = params["frequency_penalty"];
+  if (params["presence_penalty"] !== void 0) options["presence_penalty"] = params["presence_penalty"];
+  if (params["seed"] !== void 0) options["seed"] = params["seed"];
+  const body = {
+    model: params.model,
+    messages: params.messages.map(toNativeMessage),
+    stream: Boolean(params.stream),
+    ...params.tools && params.tools.length > 0 ? { tools: params.tools } : {},
+    ...Object.keys(options).length > 0 ? { options } : {}
+  };
+  if (params["response_format"] && typeof params["response_format"] === "object") {
+    const type = params["response_format"].type;
+    if (type === "json_object" || type === "json") {
+      body["format"] = "json";
+    }
+  }
+  const kwargs = params.chat_template_kwargs;
+  if (kwargs && typeof kwargs["enable_thinking"] === "boolean") {
+    body["think"] = kwargs["enable_thinking"];
+  }
+  if (params.reasoning_effort) {
+    body["think"] = toOllamaThink(params.reasoning_effort);
+  }
+  return body;
+}
+function nativeToolCallToOpenAIResponse(toolCall, index) {
+  const fn = toolCall.function ?? {};
+  return {
+    id: toolCall.id ?? `call_${index}`,
+    type: "function",
+    function: {
+      name: fn.name ?? "",
+      arguments: stringifyToolArguments(fn.arguments)
+    }
+  };
+}
+function nativeToolCallToOpenAIStream(toolCall, index) {
+  const fn = toolCall.function ?? {};
+  return {
+    index: typeof fn.index === "number" ? fn.index : index,
+    id: toolCall.id ?? `call_${index}`,
+    function: {
+      name: fn.name ?? "",
+      arguments: stringifyToolArguments(fn.arguments)
+    }
+  };
+}
+function parseOllamaChatResponse(data) {
+  const msg = data.message ?? {};
+  const toolCalls = (msg.tool_calls ?? []).map(nativeToolCallToOpenAIResponse);
+  const promptTokens = typeof data.prompt_eval_count === "number" ? data.prompt_eval_count : 0;
+  const completionTokens = typeof data.eval_count === "number" ? data.eval_count : 0;
+  const message = {
+    content: msg.content ?? ""
+  };
+  if (msg.reasoning) message["reasoning_content"] = msg.reasoning;
+  if (msg.thinking) message["thinking"] = msg.thinking;
+  if (toolCalls.length > 0) message["tool_calls"] = toolCalls;
+  return {
+    id: `chatcmpl-${Date.now()}`,
+    choices: [
+      {
+        finish_reason: mapDoneReason(data.done_reason),
+        message
+      }
+    ],
+    usage: {
+      prompt_tokens: promptTokens,
+      completion_tokens: completionTokens,
+      total_tokens: promptTokens + completionTokens
+    }
+  };
+}
+function parseOllamaChatChunk(data) {
+  const msg = data.message ?? {};
+  const toolCalls = (msg.tool_calls ?? []).map(nativeToolCallToOpenAIStream);
+  const delta = {};
+  if (msg.content) delta["content"] = msg.content;
+  if (msg.reasoning) delta["reasoning_content"] = msg.reasoning;
+  if (msg.thinking) delta["thinking"] = msg.thinking;
+  if (toolCalls.length > 0) delta["tool_calls"] = toolCalls;
+  const promptTokens = typeof data.prompt_eval_count === "number" ? data.prompt_eval_count : void 0;
+  const completionTokens = typeof data.eval_count === "number" ? data.eval_count : void 0;
+  return {
+    id: `chatcmpl-${Date.now()}`,
+    choices: [
+      {
+        delta,
+        finish_reason: data.done ? mapDoneReason(data.done_reason) : null
+      }
+    ],
+    ...promptTokens !== void 0 ? {
+      usage: {
+        prompt_tokens: promptTokens,
+        completion_tokens: completionTokens ?? 0,
+        total_tokens: promptTokens + (completionTokens ?? 0)
+      }
+    } : {}
+  };
+}
+var OllamaHttpClient = class extends ChatHttpClient {
+  baseURL;
+  constructor(options) {
+    super();
+    this.baseURL = options.baseURL;
+  }
+  buildRequest(params) {
+    return {
+      url: `${this.baseURL}/api/chat`,
+      headers: {
+        "Content-Type": "application/json"
+      },
+      body: JSON.stringify(buildOllamaChatRequest(params))
+    };
+  }
+  parseNonStreaming(data) {
+    return parseOllamaChatResponse(data);
+  }
+  parseStreamLine(trimmed) {
+    try {
+      return parseOllamaChatChunk(JSON.parse(trimmed));
+    } catch (error) {
+      logger.warn("Failed to parse Ollama stream chunk", { data: trimmed, error });
+      return null;
+    }
+  }
+};
+
 // src/server/llm/client.ts
 function extractContentFromBlocks(blocks) {
   let text = "";
@@ -39,12 +224,16 @@ function extractContentFromBlocks(blocks) {
   }
   return { text, thinking };
 }
-function createLLMClient(config, initialBackend = "unknown") {
+function createLLMClient(config, initialBackend = config.llm.backend ?? "unknown") {
   const baseURL = ensureVersionPrefix(config.llm.baseUrl);
   const httpClient = new OpenAIHttpClient({
     baseURL,
     apiKey: config.llm.apiKey ?? "not-needed"
   });
+  const ollamaHttpClient = new OllamaHttpClient({
+    baseURL: stripVersionPrefix(baseURL)
+  });
+  const httpFor = (b) => b === "ollama" ? ollamaHttpClient : httpClient;
   let model = config.llm.model;
   let profile = getModelProfile(model);
   let backend = initialBackend;
@@ -103,7 +292,7 @@ function createLLMClient(config, initialBackend = "unknown") {
           capabilities,
           ...buildExtraParams(resolvedEffort)
         });
-        const httpResponse = await httpClient.createChatCompletion(
+        const httpResponse = await httpFor(backend).createChatCompletion(
           createParams,
           {
             signal: request.signal
@@ -180,7 +369,7 @@ function createLLMClient(config, initialBackend = "unknown") {
           }
         }, 100);
         const streamSignal = request.signal ? AbortSignal.any([request.signal, idleTimeoutController.signal]) : idleTimeoutController.signal;
-        const stream = httpClient.createChatCompletionStream(streamingParams, {
+        const stream = httpFor(backend).createChatCompletionStream(streamingParams, {
           signal: streamSignal
         });
         let fullContent = "";
@@ -317,4 +506,4 @@ function createLLMClient(config, initialBackend = "unknown") {
 export {
   createLLMClient
 };
-//# sourceMappingURL=chunk-SYTEO5LA.js.map
+//# sourceMappingURL=chunk-FNXTJLCZ.js.map

package/dist/{chunk-RXLAFEZB.js → chunk-H2RZDEVL.js}

@@ -163,7 +163,7 @@ async function runCli(options) {
       break;
     }
     case "provider": {
-      const { runProviderCommand } = await import("./provider-LICG53GG.js");
+      const { runProviderCommand } = await import("./provider-6LUA73HF.js");
       const [, subcommand] = positionals;
       await runProviderCommand(mode, subcommand);
       break;
@@ -200,7 +200,7 @@ async function runCli(options) {
       break;
     }
     case "update": {
-      const { runUpdate } = await import("./update-R2TEHD7V.js");
+      const { runUpdate } = await import("./update-VH75MTCP.js");
       const code = await runUpdate();
       if (code !== 0) {
         process.exit(code);
@@ -213,7 +213,7 @@ async function runCli(options) {
       if (!configExists) {
         await runNetworkSetup(mode);
       }
-      const { runServe } = await import("./serve-75G5Z7UQ.js");
+      const { runServe } = await import("./serve-WJKFUMLI.js");
       const serveOptions = { mode };
       if (values.port) serveOptions.port = parseInt(values.port);
       if (values["no-browser"] === true) serveOptions.openBrowser = false;
@@ -225,4 +225,4 @@ async function runCli(options) {
 export {
   runCli
 };
-//# sourceMappingURL=chunk-RXLAFEZB.js.map
+//# sourceMappingURL=chunk-H2RZDEVL.js.map

package/dist/{chunk-W4SM4FFZ.js → chunk-HOSFRPPI.js}

@@ -131,7 +131,8 @@ function resolveSessionProvider(session, providerManager, effective) {
     providerId: provider.id,
     baseUrl: provider.url,
     ...provider.apiKey ? { apiKey: provider.apiKey } : {},
-    model: effectiveModel
+    model: effectiveModel,
+    ...provider.backend ? { backend: provider.backend } : {}
   };
 }
 async function generateSessionNameForSession(sessionId, userMessage, deps, signal) {
@@ -161,14 +162,17 @@ async function generateSessionNameForSession(sessionId, userMessage, deps, signa
       client = deps.getLLMClient();
       client.setModel(providerConfig.model);
     } else if (providerConfig) {
-      const { createLLMClient } = await import("./client-XHJHOFMU.js");
-      client = createLLMClient({
-        llm: {
-          baseUrl: providerConfig.baseUrl,
-          model: providerConfig.model,
-          ...providerConfig.apiKey ? { apiKey: providerConfig.apiKey } : {}
-        }
-      });
+      const { createLLMClient } = await import("./client-YHYIWZY7.js");
+      client = createLLMClient(
+        {
+          llm: {
+            baseUrl: providerConfig.baseUrl,
+            model: providerConfig.model,
+            ...providerConfig.apiKey ? { apiKey: providerConfig.apiKey } : {}
+          }
+        },
+        providerConfig.backend ?? "unknown"
+      );
     } else if (deps.getLLMClient) {
       client = deps.getLLMClient();
     } else {
@@ -212,4 +216,4 @@ export {
   needsNameGeneration,
   generateSessionNameForSession
 };
-//# sourceMappingURL=chunk-W4SM4FFZ.js.map
+//# sourceMappingURL=chunk-HOSFRPPI.js.map

package/dist/{chunk-MIPQOHWS.js → chunk-KAZWU73N.js}

@@ -617,11 +617,35 @@ async function buildChatCompletionCreateParams(model, request, profile, capabili
     ;
     params["top_k"] = topK;
   }
+  if (capabilities.supportsNumCtx && request.modelSettings?.numCtx) {
+    ;
+    params["num_ctx"] = request.modelSettings.numCtx;
+  }
   const resolvedEffort = reasoningEffort ?? request.reasoningEffort;
   const queryParams = request.modelSettings?.queryParams;
   const hasQueryParams = queryParams && Object.keys(queryParams).length > 0;
   const hasExplicitModelSettings = hasQueryParams || !!request.modelSettings?.chatTemplateKwargs;
-  if (hasQueryParams) {
+  if (capabilities.routesEffortViaChatTemplateKwargs) {
+    let userKwargs = request.modelSettings?.chatTemplateKwargs;
+    if (hasQueryParams) {
+      const { chat_template_kwargs: userKwargsFromQP, ...restQueryParams } = queryParams;
+      Object.assign(params, restQueryParams);
+      if (userKwargsFromQP) {
+        userKwargs = userKwargsFromQP;
+      }
+    }
+    const kwargs = { ...userKwargs ?? {} };
+    if (resolvedEffort === "none") {
+      delete kwargs["reasoning_effort"];
+      kwargs["enable_thinking"] = false;
+    } else if (resolvedEffort && kwargs["reasoning_effort"] === void 0) {
+      kwargs["reasoning_effort"] = resolvedEffort;
+    }
+    if (Object.keys(kwargs).length > 0) {
+      ;
+      params["chat_template_kwargs"] = kwargs;
+    }
+  } else if (hasQueryParams) {
     Object.assign(params, queryParams);
     if (resolvedEffort) {
       ;
@@ -711,4 +735,4 @@ export {
   buildStreamingCreateParams,
   mapFinishReason
 };
-//# sourceMappingURL=chunk-MIPQOHWS.js.map
+//# sourceMappingURL=chunk-KAZWU73N.js.map

package/dist/{chunk-62GUO2EG.js → chunk-LJX7KXRU.js}

@@ -6,10 +6,10 @@ import {
 } from "./chunk-GJL3QSMP.js";
 import {
   getBackendCapabilities
-} from "./chunk-AZHWLLPP.js";
+} from "./chunk-PBFGJFGC.js";
 import {
   buildModelParams
-} from "./chunk-MIPQOHWS.js";
+} from "./chunk-KAZWU73N.js";
 import {
   logger
 } from "./chunk-K44MW7JJ.js";
@@ -518,4 +518,4 @@ export {
   createChatDoneEvent,
   consumeStreamGenerator
 };
-//# sourceMappingURL=chunk-62GUO2EG.js.map
+//# sourceMappingURL=chunk-LJX7KXRU.js.map

package/dist/{chunk-AZHWLLPP.js → chunk-PBFGJFGC.js}

@@ -2,39 +2,57 @@
 var BACKEND_CAPABILITIES = {
   vllm: {
     supportsChatTemplateKwargs: true,
-    supportsTopK: true
+    supportsTopK: true,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   },
   sglang: {
     supportsChatTemplateKwargs: true,
-    supportsTopK: true
+    supportsTopK: true,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   },
   openai: {
     supportsChatTemplateKwargs: false,
-    supportsTopK: false
+    supportsTopK: false,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   },
   anthropic: {
     supportsChatTemplateKwargs: false,
-    supportsTopK: false
+    supportsTopK: false,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   },
   ollama: {
     supportsChatTemplateKwargs: false,
-    supportsTopK: false
+    supportsTopK: false,
+    supportsNumCtx: true,
+    routesEffortViaChatTemplateKwargs: false
   },
   llamacpp: {
     supportsChatTemplateKwargs: false,
-    supportsTopK: true
+    supportsTopK: true,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: true
   },
   lmstudio: {
     supportsChatTemplateKwargs: false,
-    supportsTopK: true
+    supportsTopK: true,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   },
   "opencode-go": {
     supportsChatTemplateKwargs: false,
-    supportsTopK: true
+    supportsTopK: true,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   },
   unknown: {
     supportsChatTemplateKwargs: true,
-    supportsTopK: true
+    supportsTopK: true,
+    supportsNumCtx: false,
+    routesEffortViaChatTemplateKwargs: false
   }
 };
 function getBackendCapabilities(backend) {
@@ -67,4 +85,4 @@ export {
   getBackendCapabilities,
   getBackendDisplayName
 };
-//# sourceMappingURL=chunk-AZHWLLPP.js.map
+//# sourceMappingURL=chunk-PBFGJFGC.js.map

package/dist/{chunk-A6EZBX62.js → chunk-T3JVYO3Z.js}

@@ -3,13 +3,13 @@ import {
 } from "./chunk-K2QKX4ZJ.js";
 import {
   createLLMClient
-} from "./chunk-SYTEO5LA.js";
+} from "./chunk-FNXTJLCZ.js";
 import {
   getModelProfile
 } from "./chunk-GJL3QSMP.js";
 import {
   getBackendCapabilities
-} from "./chunk-AZHWLLPP.js";
+} from "./chunk-PBFGJFGC.js";
 import {
   buildModelsUrl,
   ensureVersionPrefix,
@@ -17,7 +17,7 @@ import {
 } from "./chunk-HNCM3D7Y.js";
 import {
   resolveAttachmentsInMessages
-} from "./chunk-MIPQOHWS.js";
+} from "./chunk-KAZWU73N.js";
 import {
   getCatalogEntry,
   resolveEffortForModel
@@ -456,7 +456,10 @@ function createProviderManager(config, options = {}) {
   function createClientForProvider(provider, model, reasoningEffort) {
     const resolvedModel = resolveProviderModel(provider, model);
     const transport = options.adapters?.getTransport(resolveTransportAdapter(provider));
-    return transport ? createTransportLLMClient(provider, resolvedModel, transport, reasoningEffort) : createLLMClient(createConfigForProvider(provider, resolvedModel, reasoningEffort));
+    return transport ? createTransportLLMClient(provider, resolvedModel, transport, reasoningEffort) : createLLMClient(
+      createConfigForProvider(provider, resolvedModel, reasoningEffort),
+      provider.backend
+    );
   }
   async function fetchProviderModels(provider) {
     const transport = options.adapters?.getTransport(resolveTransportAdapter(provider));
@@ -761,6 +764,8 @@ function createProviderManager(config, options = {}) {
       if (model["maxTokens"] !== void 0) baseSettings["maxTokens"] = model["maxTokens"];
       if (model["supportsVision"] !== void 0) baseSettings["supportsVision"] = model["supportsVision"];
       if (model["omitParams"] !== void 0) baseSettings["omitParams"] = model["omitParams"];
+      if (typeof model["contextWindow"] === "number" && model["contextWindow"] > 0)
+        baseSettings["numCtx"] = model["contextWindow"];
       const rawQueryParams = mode === "thinking" ? model.thinkingQueryParams : model.nonThinkingQueryParams;
       if (rawQueryParams) {
         return { ...baseSettings, queryParams: JSON.parse(rawQueryParams) };
@@ -776,9 +781,14 @@ function createProviderManager(config, options = {}) {
       if (fallbackRawQP) {
         return { ...baseSettings, queryParams: JSON.parse(fallbackRawQP) };
       }
-      if (model["omitParams"] !== void 0) return baseSettings;
+      if (Object.keys(baseSettings).length > 0) return baseSettings;
       return void 0;
     },
+    resolveModelEffort(providerId, modelId, explicitEffort) {
+      const provider = providers.find((p) => p.id === providerId);
+      if (!provider) return void 0;
+      return resolveModelThinkingConfig(provider, modelId, explicitEffort).reasoningEffort;
+    },
     async refreshProviderModels(providerId) {
       const provider = providers.find((p) => p.id === providerId);
       if (!provider) {
@@ -841,4 +851,4 @@ export {
   parseDefaultModelSelection,
   createProviderManager
 };
-//# sourceMappingURL=chunk-A6EZBX62.js.map
+//# sourceMappingURL=chunk-T3JVYO3Z.js.map

package/dist/{chunk-FAL4WO4V.js → chunk-TP2PGZSE.js}

@@ -105,7 +105,7 @@ function resolveAgentDef(sessionManager, sessionId) {
 }
 async function buildCachedPrompt(sessionManager, sessionId, agentDef, modelName) {
   const { instructionContent, skills } = await loadSessionContext(sessionManager, sessionId);
-  const { getToolRegistryForAgent } = await import("./tools-G44LBKD4.js");
+  const { getToolRegistryForAgent } = await import("./tools-AOLICK5U.js");
   const tools = getToolRegistryForAgent(agentDef, sessionId).definitions;
   const toolFingerprint = getToolFingerprint(tools);
   const allAgents = await loadAllAgentsDefault(sessionManager.getProjectWorkdir(sessionId));
@@ -124,7 +124,7 @@ async function buildCachedPrompt(sessionManager, sessionId, agentDef, modelName)
 async function computeSessionHash(sessionManager, sessionId, modelName) {
   const { instructionContent, skills } = await loadSessionContext(sessionManager, sessionId);
   const agentDef = await resolveAgentDef(sessionManager, sessionId);
-  const { getToolRegistryForAgent } = await import("./tools-G44LBKD4.js");
+  const { getToolRegistryForAgent } = await import("./tools-AOLICK5U.js");
   const tools = getToolRegistryForAgent(agentDef, sessionId).definitions;
   const toolFingerprint = getToolFingerprint(tools);
   return computeDynamicContextHash(instructionContent, skills, toolFingerprint, modelName);
@@ -148,4 +148,4 @@ export {
   computeSessionHash,
   applyDynamicContext
 };
-//# sourceMappingURL=chunk-FAL4WO4V.js.map
+//# sourceMappingURL=chunk-TP2PGZSE.js.map

package/dist/{chunk-JEZN37KV.js → chunk-UAAFZEIS.js}

@@ -1,6 +1,6 @@
 import {
   createMessageStartEvent
-} from "./chunk-62GUO2EG.js";
+} from "./chunk-LJX7KXRU.js";
 import {
   COMPACTION_PROMPT
 } from "./chunk-GIJUWPRQ.js";
@@ -36,4 +36,4 @@ export {
   COMPACTION_MAX_RATIO,
   shouldCompact
 };
-//# sourceMappingURL=chunk-JEZN37KV.js.map
+//# sourceMappingURL=chunk-UAAFZEIS.js.map