feat: Nvidia API direkt-Anbindung mit Thinking-Mode Support

Neues Server-Setting NvidiaApiKey – wenn gesetzt, wird direkt
integrate.api.nvidia.com verwendet statt g4f.space Proxy.
Thinking-Modelle (kimi-k2, qwq, deepseek-r1) erhalten automatisch
chat_template_kwargs und reasoning-Output wird als klappbare
Denkprozess-Bubble im Chat angezeigt.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

wwwroot/js/chat.js

@@ -28,6 +28,18 @@ function addChainMetaBubble(d, streamed) {
 
 function handleStreamEvent(eventType, data) {
   switch (eventType) {
+    case 'thinking_start':
+      startThinkingBubble();
+      break;
+
+    case 'thinking_delta':
+      appendToThinking(data.text);
+      break;
+
+    case 'thinking_end':
+      endThinkingBubble();
+      break;
+
     case 'text_start':
       if (!_streamDiv) startStreamBubble();
       break;

wwwroot/js/history.js

@@ -259,6 +259,24 @@ function _renderMessagesIntoFragment(fragment, msgs, imageMessageIds, opts) {
         }
         break;
       }
+      case 'thinking': {
+        if (msg.role === 'assistant' && msg.content) {
+          var thinkDiv = document.createElement('div');
+          thinkDiv.className = 'tool-box thinking-box' + (archived ? ' archived' : '');
+          var thinkHeader = document.createElement('div');
+          thinkHeader.className = 'tool-box-header clickable';
+          thinkHeader.innerHTML = '<span class="tool-name">Denkprozess</span>';
+          var thinkContent = document.createElement('div');
+          thinkContent.className = 'thinking-content markdown';
+          thinkContent.innerHTML = renderMarkdown(msg.content);
+          thinkContent.hidden = true;
+          thinkHeader.addEventListener('click', function() { thinkContent.hidden = !thinkContent.hidden; });
+          thinkDiv.appendChild(thinkHeader);
+          thinkDiv.appendChild(thinkContent);
+          fragment.appendChild(thinkDiv);
+        }
+        break;
+      }
       case 'tool_use':
         if (msg.toolName === 'ask_user_choice') {
           _renderChoiceBubbleInFragment(fragment, msgs, _msgIdx);

wwwroot/js/stream-ui.js

@@ -52,6 +52,57 @@ function endStreamBubble() {
   _streamText = '';
 }
 
+// ── Thinking Bubble (Nvidia Reasoning) ──
+
+let _thinkingDiv = null;
+let _thinkingText = '';
+
+function startThinkingBubble() {
+  _thinkingDiv = document.createElement('div');
+  _thinkingDiv.className = 'tool-box thinking-box streamed';
+  var header = document.createElement('div');
+  header.className = 'tool-box-header clickable';
+  header.innerHTML = '<span class="tool-name">Denkprozess</span>';
+  var content = document.createElement('div');
+  content.className = 'thinking-content markdown';
+  content.hidden = true;
+  content.innerHTML = '<p><span class="cursor"></span></p>';
+  header.addEventListener('click', function() { content.hidden = !content.hidden; });
+  _thinkingDiv.appendChild(header);
+  _thinkingDiv.appendChild(content);
+  _thinkingText = '';
+  chatEl().appendChild(_thinkingDiv);
+  scrollDown();
+}
+
+function appendToThinking(text) {
+  if (!_thinkingDiv) startThinkingBubble();
+  _thinkingText += text;
+  var contentEl = _thinkingDiv.querySelector('.thinking-content');
+  contentEl.innerHTML = renderMarkdown(_thinkingText);
+  var cursor = document.createElement('span');
+  cursor.className = 'cursor';
+  var lastBlock = contentEl.lastElementChild;
+  if (lastBlock) lastBlock.appendChild(cursor);
+  else contentEl.appendChild(cursor);
+  scrollDown();
+}
+
+function endThinkingBubble() {
+  if (_thinkingDiv) {
+    var cursor = _thinkingDiv.querySelector('.cursor');
+    if (cursor) cursor.remove();
+    var contentEl = _thinkingDiv.querySelector('.thinking-content');
+    if (!_thinkingText.trim()) _thinkingDiv.remove();
+    else {
+      contentEl.innerHTML = renderMarkdown(_thinkingText);
+      contentEl.hidden = true;
+    }
+  }
+  _thinkingDiv = null;
+  _thinkingText = '';
+}
+
 // ── Review Bubbles (Streaming) ──
 
 let _reviewBubble = null;