1

prisma/config.ts

@@ -14,12 +14,6 @@ export const MODELS: { value: ModelOption; label: string; desc: string; provider
     desc: 'Deep reasoning, complex tasks, higher intelligence.',
     provider: 'google'
   },
-  {
-    value: 'deepseek-reasoner',
-    label: 'DeepSeek R1',
-    desc: 'State-of-the-art open reasoning model.',
-    provider: 'deepseek'
-  },
   {
     value: 'custom',
     label: 'Custom Model',

prisma/services/deepThink/openaiClient.ts

@@ -56,7 +56,8 @@ export const generateContent = async (
   const requestOptions: any = {
     model: config.model,
     messages,
-    temperature: config.temperature,
+    // Clamp temperature to 1.0 max for compatibility with strict providers (NVIDIA, vLLM, etc.)
+    temperature: typeof config.temperature === 'number' ? Math.min(config.temperature, 1.0) : undefined,
   };
 
   if (config.responseFormat === 'json_object') {
@@ -68,7 +69,7 @@ export const generateContent = async (
     const message = response.choices[0]?.message;
     const content = message?.content || '';
     
-    // Check for DeepSeek native reasoning field
+    // Check for native reasoning_content field (DeepSeek/NVIDIA style)
     const reasoningContent = (message as any)?.reasoning_content;
 
     if (reasoningContent && config.thinkingConfig?.includeThoughts) {
@@ -108,7 +109,8 @@ export async function* generateContentStream(
   const requestOptions: any = {
     model: config.model,
     messages,
-    temperature: config.temperature,
+    // Clamp temperature to 1.0 max for compatibility with strict providers
+    temperature: typeof config.temperature === 'number' ? Math.min(config.temperature, 1.0) : undefined,
     stream: true,
   };
 
@@ -119,29 +121,32 @@ export async function* generateContentStream(
   let currentThought = '';
 
   for await (const chunk of (stream as any)) {
-    const delta = chunk.choices[0]?.delta?.content || '';
+    const delta = chunk.choices[0]?.delta;
-    // Support DeepSeek native reasoning field
-    const reasoningDelta = (chunk.choices[0]?.delta as any)?.reasoning_content || '';
-
-    // If we have native reasoning content, yield it immediately as thought
-    if (reasoningDelta) {
-      yield { text: '', thought: reasoningDelta };
-    }
-
     if (!delta) continue;
 
-    accumulatedText += delta;
+    const content = delta.content || '';
+    // Check for native reasoning_content field (DeepSeek/NVIDIA style)
+    const reasoning = delta.reasoning_content || '';
+
+    // If native reasoning field exists, emit it immediately
+    if (reasoning && config.thinkingConfig?.includeThoughts) {
+      yield { text: '', thought: reasoning };
+    }
+
+    if (content) {
+      accumulatedText += content;
 
       if (config.thinkingConfig?.includeThoughts) {
-      if (delta.includes('<thinking>')) {
+        // Fallback to tag parsing if reasoning_content wasn't provided but tags exist
+        if (content.includes('<thinking>')) {
           inThinking = true;
           continue;
         }
 
         if (inThinking) {
-        if (delta.includes('</thinking>')) {
+          if (content.includes('</thinking>')) {
             inThinking = false;
-          const parts = delta.split('</thinking>', 2);
+            const parts = content.split('</thinking>', 2);
             currentThought += parts[0];
 
             if (currentThought.trim()) {
@@ -153,17 +158,19 @@ export async function* generateContentStream(
               yield { text: parts[1], thought: '' };
             }
           } else {
-          currentThought += delta;
+            currentThought += content;
-          if (currentThought.length > 100) {
+            // Emit thought chunks periodically so it doesn't hang
+            if (currentThought.length > 50) {
               yield { text: '', thought: currentThought };
               currentThought = '';
             }
           }
         } else {
-        yield { text: delta, thought: '' };
+          yield { text: content, thought: '' };
         }
       } else {
-      yield { text: delta, thought: '' };
+        yield { text: content, thought: '' };
+      }
     }
   }
 

prisma/services/deepThink/prompts.ts

@@ -5,11 +5,11 @@ export const MANAGER_SYSTEM_PROMPT = `You are the "Dynamic Planning Engine". You
 
 Your job is to create SUPPLEMENTARY experts
 
-For each expert, you must assign a specific 'temperature' (0.0 to 2.0) based on the nature of their task:
+For each expert, you must assign a specific 'temperature' (0.0 to 1.0) based on the nature of their task:
 
-*   High temperature (1.0 - 2.0) 
+*   High temperature (0.7 - 1.0) - For creative, brainstorming, or open-ended tasks.
-*   Low temperature (0.0 - 0.4) 
+*   Low temperature (0.0 - 0.3) - For code, math, logic, or factual tasks.
-*   Medium temperature (0.4 - 1.0)`;
+*   Medium temperature (0.3 - 0.7) - For balanced analysis and general explanation.`;
 
 export const MANAGER_REVIEW_SYSTEM_PROMPT = `
 You are the "Quality Assurance & Orchestration Engine".