openfox 2.0.123 → 2.0.125

package/CHANGELOG.md

@@ -1,5 +1,38 @@
 # Changelog
 
+## 2.0.125 - 2026-08-21
+
+### Features
+
+- **xAI Grok (SuperGrok) plugin** — new curated plugin logs in with your X (xAI) SuperGrok subscription via OAuth, no API key needed
+
+### Enhancements
+
+- **Small-context models are flagged** — the model picker and edit-model modal warn when a model's context window sits under 16k tokens
+
+### Bug Fixes
+
+- **llama.cpp now honors the reasoning effort** — the chosen level reaches the model (it was silently ignored); 'none' turns thinking off
+- **Symlinked project folders show up in discovery** — directory browser, workspace and plugin listings now include symlinked directories
+- **Ollama requests are fixed end to end** — the native /api/chat endpoint uses the model's full context window so prompts are no longer truncated, unsupported params are dropped, and multi-turn tool calls work again
+- **Thinking mode follows the resolved reasoning effort** — session and sub-agent thinking is no longer force-enabled
+
+## 2.0.124 - 2026-08-19
+
+### Enhancements
+
+- **project_tasks list paginates** — capped at 10 results (25 max) to keep agent context lean
+- **Pinned effort chips light up for any model** — the selector highlights a model's pinned effort even when it isn't the active pick
+
+### Bug Fixes
+
+- **Reasoning effort options come from a curated catalog** — endpoint probing that collapsed or misdetected levels is gone
+- **Qwen effort presets corrected** — qwen3.8 offers none/low/medium/xhigh (default xhigh), qwen3.5/3.6 offer none/high
+- **maxTokens budget accounts for tool results** — tool-result tokens are subtracted so large outputs no longer overstate the budget
+- **Context-overflow retries halve maxTokens** — retry immediately with a halved budget instead of a doomed backoff retry
+- **`..` paths respect shell cd state** — relative traversals no longer trigger false outside-sandbox confirmations
+- **Division expressions no longer flagged as paths** — 'length / 4' no longer trips the root-path safety check
+
 ## 2.0.123 - 2026-08-18
 
 ### Features

package/README.md

@@ -99,6 +99,7 @@ Restart OpenFox after installing or updating a plugin.
 - To authenticate with a ChatGPT Plus or Pro account, you can install the [`openfox-chatgpt`](https://github.com/arthurlacoste/openfox-chatgpt) plugin.
 - To authenticate with a Github copilot account, you can install the [`openfox-github-copilot`](https://github.com/JamesDAdams/openfox-github-copilot) plugin.
 - To authenticate with a Google Antigravity account, you can install the [`openfox-google-antigravity`](https://github.com/JamesDAdams/openfox-google-antigravity) plugin.
+- To authenticate with an X (xAI) SuperGrok subscription, you can install the [`openfox-xai-supergrok`](https://github.com/Olgean-Group/openfox-xai-supergrok) plugin.
 
 ## Screenshots
 

package/dist/CHANGELOG.md

@@ -1,5 +1,38 @@
 # Changelog
 
+## 2.0.125 - 2026-08-21
+
+### Features
+
+- **xAI Grok (SuperGrok) plugin** — new curated plugin logs in with your X (xAI) SuperGrok subscription via OAuth, no API key needed
+
+### Enhancements
+
+- **Small-context models are flagged** — the model picker and edit-model modal warn when a model's context window sits under 16k tokens
+
+### Bug Fixes
+
+- **llama.cpp now honors the reasoning effort** — the chosen level reaches the model (it was silently ignored); 'none' turns thinking off
+- **Symlinked project folders show up in discovery** — directory browser, workspace and plugin listings now include symlinked directories
+- **Ollama requests are fixed end to end** — the native /api/chat endpoint uses the model's full context window so prompts are no longer truncated, unsupported params are dropped, and multi-turn tool calls work again
+- **Thinking mode follows the resolved reasoning effort** — session and sub-agent thinking is no longer force-enabled
+
+## 2.0.124 - 2026-08-19
+
+### Enhancements
+
+- **project_tasks list paginates** — capped at 10 results (25 max) to keep agent context lean
+- **Pinned effort chips light up for any model** — the selector highlights a model's pinned effort even when it isn't the active pick
+
+### Bug Fixes
+
+- **Reasoning effort options come from a curated catalog** — endpoint probing that collapsed or misdetected levels is gone
+- **Qwen effort presets corrected** — qwen3.8 offers none/low/medium/xhigh (default xhigh), qwen3.5/3.6 offer none/high
+- **maxTokens budget accounts for tool results** — tool-result tokens are subtracted so large outputs no longer overstate the budget
+- **Context-overflow retries halve maxTokens** — retry immediately with a halved budget instead of a doomed backoff retry
+- **`..` paths respect shell cd state** — relative traversals no longer trigger false outside-sandbox confirmations
+- **Division expressions no longer flagged as paths** — 'length / 4' no longer trips the root-path safety check
+
 ## 2.0.123 - 2026-08-18
 
 ### Features

package/dist/{auto-config-4BFGLPCP.js → auto-config-454YE6KC.js}

@@ -6,7 +6,7 @@ import {
 } from "./chunk-HNCM3D7Y.js";
 import {
   getCatalogEntry
-} from "./chunk-TPDBUOPN.js";
+} from "./chunk-ISE55FOR.js";
 import {
   logger
 } from "./chunk-K44MW7JJ.js";
@@ -184,64 +184,6 @@ async function probeReasoningInMessages(baseUrl, apiKey, model) {
     return void 0;
   }
 }
-var REASONING_EFFORT_CANDIDATES = ["low", "medium", "high", "xhigh", "max"];
-async function probeReasoningEffort(baseUrl, apiKey, model, effort, signal) {
-  const body = {
-    model,
-    messages: [{ role: "user", content: "say hi in one word" }],
-    max_tokens: 0,
-    ...effort ? { reasoning_effort: effort } : {}
-  };
-  try {
-    const response = await fetch(`${ensureVersionPrefix(baseUrl)}/chat/completions`, {
-      method: "POST",
-      headers: buildAuthHeaders(apiKey),
-      body: JSON.stringify(body),
-      signal
-    });
-    if (!response.ok) {
-      return { effort: effort ?? "baseline", ok: false, promptTokens: 0 };
-    }
-    const data = await response.json();
-    return {
-      effort: effort ?? "baseline",
-      ok: true,
-      promptTokens: data.usage?.prompt_tokens ?? 0
-    };
-  } catch {
-    return { effort: effort ?? "baseline", ok: false, promptTokens: 0 };
-  }
-}
-async function probeReasoningEfforts(baseUrl, apiKey, model) {
-  const timeout = AbortSignal.timeout(15e3);
-  const baseline = await probeReasoningEffort(baseUrl, apiKey, model, void 0, timeout);
-  if (!baseline.ok) {
-    return void 0;
-  }
-  const results = await Promise.all(
-    REASONING_EFFORT_CANDIDATES.map(
-      (effort) => probeReasoningEffort(baseUrl, apiKey, model, effort, timeout).then((r) => ({ ...r, effort }))
-    )
-  );
-  const accepted = results.filter((r) => r.ok);
-  if (accepted.length === 0) {
-    return void 0;
-  }
-  const impactful = accepted.filter((r) => r.promptTokens !== baseline.promptTokens);
-  const pool = impactful.length > 0 ? impactful : accepted;
-  const distinctTokens = new Set(pool.map((r) => r.promptTokens));
-  if (distinctTokens.size <= 1) {
-    return pool.map((r) => r.effort);
-  }
-  const seen = /* @__PURE__ */ new Set();
-  const distinct = [];
-  for (const r of pool) {
-    if (seen.has(r.promptTokens)) continue;
-    seen.add(r.promptTokens);
-    distinct.push(r.effort);
-  }
-  return distinct;
-}
 var STANDARD_PARAMS = ["temperature", "top_p", "max_tokens", "top_k", "reasoning_effort"];
 var PARAM_VALUES = {
   temperature: 0.7,
@@ -340,15 +282,13 @@ async function autoConfig(input) {
       supportsVision
     } = await detectModelInfo(baseUrl, apiKey, backend, model.id);
     const catalog = getCatalogEntry(model.id);
-    const shouldProbeEfforts = backend !== "unknown" || !catalog;
-    const [thinkingConfig, nonThinkingConfig, rejectedParams, reasoningEfforts] = await Promise.all([
+    const reasoningEfforts = catalog?.reasoningEfforts;
+    const [thinkingConfig, nonThinkingConfig, rejectedParams] = await Promise.all([
       probeCombos(baseUrl, apiKey, model.id, THINKING_COMBOS),
       probeCombos(baseUrl, apiKey, model.id, NON_THINKING_COMBOS),
-      probeRejectedParams(baseUrl, apiKey, model.id, backend),
-      shouldProbeEfforts ? probeReasoningEfforts(baseUrl, apiKey, model.id) : Promise.resolve(void 0)
+      probeRejectedParams(baseUrl, apiKey, model.id, backend)
     ]);
     const sendReasoningInMessages = thinkingConfig ? await probeReasoningInMessages(baseUrl, apiKey, model.id) : void 0;
-    const effectiveReasoningEfforts = reasoningEfforts ?? catalog?.reasoningEfforts;
     results.push({
       id: model.id,
       contextWindow,
@@ -358,7 +298,7 @@ async function autoConfig(input) {
       nonThinkingConfig,
       ...sendReasoningInMessages !== void 0 ? { sendReasoningInMessages } : {},
       ...rejectedParams.length > 0 ? { rejectedParams } : {},
-      ...effectiveReasoningEfforts ? { reasoningEfforts: effectiveReasoningEfforts } : {},
+      ...reasoningEfforts ? { reasoningEfforts } : {},
       ...catalog?.defaultReasoningEffort ? { defaultReasoningEffort: catalog.defaultReasoningEffort } : {}
     });
   }
@@ -367,4 +307,4 @@ async function autoConfig(input) {
 export {
   autoConfig
 };
-//# sourceMappingURL=auto-config-4BFGLPCP.js.map
+//# sourceMappingURL=auto-config-454YE6KC.js.map

package/dist/{backend-67OLCS5O.js → backend-4YPEC3J2.js}

@@ -1,10 +1,10 @@
 import {
   getBackendCapabilities,
   getBackendDisplayName
-} from "./chunk-AZHWLLPP.js";
+} from "./chunk-PBFGJFGC.js";
 import "./chunk-YLJ4XMA6.js";
 export {
   getBackendCapabilities,
   getBackendDisplayName
 };
-//# sourceMappingURL=backend-67OLCS5O.js.map
+//# sourceMappingURL=backend-4YPEC3J2.js.map

package/dist/{chat-handler-UWZYFGRY.js → chat-handler-HJ2RBGNV.js}

@@ -2,25 +2,25 @@ import {
   buildRunChatTurnParams,
   finalizeTurnCompletion,
   generateSessionNameForSession
-} from "./chunk-W4SM4FFZ.js";
+} from "./chunk-HOSFRPPI.js";
 import {
   runChatTurn
-} from "./chunk-V2UJP5CX.js";
-import "./chunk-4L7UMEHD.js";
-import "./chunk-PZUKG3OZ.js";
+} from "./chunk-NBOSTFTS.js";
+import "./chunk-O5A3XKE5.js";
+import "./chunk-KETVPTMI.js";
 import "./chunk-ZIL6VKLU.js";
-import "./chunk-62GUO2EG.js";
+import "./chunk-LJX7KXRU.js";
 import "./chunk-GIJUWPRQ.js";
 import "./chunk-EHFIUXWZ.js";
 import "./chunk-RV5G7QQU.js";
 import "./chunk-I72ACQ6R.js";
 import "./chunk-JE7LW7Y6.js";
 import "./chunk-OAN4BXEW.js";
-import "./chunk-3NMLSANH.js";
+import "./chunk-Y3BSBKHY.js";
 import "./chunk-J7GAZGZT.js";
-import "./chunk-SS6KV6DI.js";
+import "./chunk-CTHRGR4O.js";
 import "./chunk-7YP4O4XE.js";
-import "./chunk-INOO465O.js";
+import "./chunk-KD4W3I6C.js";
 import {
   getEventStore
 } from "./chunk-YJD5A375.js";
@@ -34,21 +34,21 @@ import {
 import "./chunk-66MG44TF.js";
 import "./chunk-JURM3RPZ.js";
 import "./chunk-LCLH6ZUL.js";
-import "./chunk-H4JRR5KV.js";
+import "./chunk-T3JVYO3Z.js";
 import "./chunk-K2QKX4ZJ.js";
 import "./chunk-J2GP3J3X.js";
 import "./chunk-KBESUA3W.js";
 import "./chunk-OSXPQSLW.js";
 import "./chunk-3B232XG4.js";
-import "./chunk-SYTEO5LA.js";
+import "./chunk-FNXTJLCZ.js";
 import "./chunk-GJL3QSMP.js";
-import "./chunk-AZHWLLPP.js";
+import "./chunk-PBFGJFGC.js";
 import "./chunk-HNCM3D7Y.js";
-import "./chunk-MIPQOHWS.js";
-import "./chunk-VPRB5FNR.js";
+import "./chunk-KAZWU73N.js";
+import "./chunk-ZFNBATE3.js";
 import "./chunk-N2Q3KYWM.js";
 import "./chunk-OYBUJIC3.js";
-import "./chunk-TPDBUOPN.js";
+import "./chunk-ISE55FOR.js";
 import "./chunk-K44MW7JJ.js";
 import "./chunk-KSDWQ6IZ.js";
 import "./chunk-KSXCOHHW.js";
@@ -174,4 +174,4 @@ export {
   startChatSession,
   stopSessionExecution
 };
-//# sourceMappingURL=chat-handler-UWZYFGRY.js.map
+//# sourceMappingURL=chat-handler-HJ2RBGNV.js.map

package/dist/{chunk-ZKMZI4ZG.js → chunk-7DO4JMGT.js}

@@ -1,14 +1,14 @@
 import {
   runAgentTurn
-} from "./chunk-V2UJP5CX.js";
+} from "./chunk-NBOSTFTS.js";
 import {
   executeSubAgent,
   getToolRegistryForAgent
-} from "./chunk-4L7UMEHD.js";
+} from "./chunk-O5A3XKE5.js";
 import {
   TurnMetrics,
   createMessageStartEvent
-} from "./chunk-62GUO2EG.js";
+} from "./chunk-LJX7KXRU.js";
 import {
   checkAborted,
   decodeUtf8,
@@ -27,7 +27,7 @@ import {
   loadProjectWorkflows,
   loadUserWorkflows,
   normalizeWorkflowScope
-} from "./chunk-SS6KV6DI.js";
+} from "./chunk-CTHRGR4O.js";
 import {
   getCurrentContextWindowId,
   getEventStore
@@ -44,7 +44,7 @@ import {
 } from "./chunk-KBESUA3W.js";
 import {
   LLMError
-} from "./chunk-VPRB5FNR.js";
+} from "./chunk-ZFNBATE3.js";
 import {
   logger
 } from "./chunk-K44MW7JJ.js";
@@ -1002,4 +1002,4 @@ export {
   abortRunnerRun,
   launchWorkflowRun
 };
-//# sourceMappingURL=chunk-ZKMZI4ZG.js.map
+//# sourceMappingURL=chunk-7DO4JMGT.js.map

package/dist/{chunk-SS6KV6DI.js → chunk-CTHRGR4O.js}

@@ -1,6 +1,6 @@
 import {
   parseDefaultModelSelection
-} from "./chunk-H4JRR5KV.js";
+} from "./chunk-T3JVYO3Z.js";
 import {
   deleteItemFromDir,
   getDefaultIds,
@@ -1099,4 +1099,4 @@ export {
   isTaskGateError,
   createTasksService
 };
-//# sourceMappingURL=chunk-SS6KV6DI.js.map
+//# sourceMappingURL=chunk-CTHRGR4O.js.map

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-GZUD5E6E.js

@@ -0,0 +1,7 @@
+// src/constants.ts
+var VERSION = "2.0.125";
+
+export {
+  VERSION
+};
+//# sourceMappingURL=chunk-GZUD5E6E.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-TPDBUOPN.js → chunk-ISE55FOR.js}

@@ -33,9 +33,16 @@ var CATALOG = [
     entry: { reasoningEfforts: ["none", "low", "high", "max"], defaultReasoningEffort: "high" }
   },
   {
-    // Qwen3.6 / 3.7 / 3.8 families: low/medium/high is the common denominator.
-    pattern: /^qwen3\.(6|7|8)(\b|-)/i,
-    entry: { reasoningEfforts: ["low", "medium", "high"], defaultReasoningEffort: "medium" }
+    // Qwen3.8: official reasoning_effort values are low/medium/xhigh
+    // (xhigh is the model default); 'none' disables thinking on setups
+    // that honor it.
+    pattern: /^qwen3\.8(\b|-)/i,
+    entry: { reasoningEfforts: ["none", "low", "medium", "xhigh"], defaultReasoningEffort: "xhigh" }
+  },
+  {
+    // Qwen3.5 / 3.6: on/off thinking only — none (off) or high (thinking on).
+    pattern: /^qwen3\.(5|6)(\b|-)/i,
+    entry: { reasoningEfforts: ["none", "high"], defaultReasoningEffort: "high" }
   },
   {
     // Z.ai GLM-5.3: only max/high/low are accepted.
@@ -79,4 +86,4 @@ export {
   getCatalogEntry,
   getCatalogDefaultEffort
 };
-//# sourceMappingURL=chunk-TPDBUOPN.js.map
+//# sourceMappingURL=chunk-ISE55FOR.js.map