utility inference layer 3

packages/orchestration-service/src/services/summarization.js

@@ -1,7 +1,5 @@
-const { getEnv, SERVICES, SUMMARIES, logger } = require('@nexusai/shared');
+const { getEnv, SERVICES, SUMMARIES, logger, utilityInference } = require('@nexusai/shared');
 
-const EXTRACTION_URL  = getEnv('EXTRACTION_URL', 'http://localhost:11434');
-const EXTRACTION_MODEL = getEnv('EXTRACTION_MODEL', 'qwen2.5:3b');
 const MEMORY_URL      = getEnv('MEMORY_SERVICE_URL', SERVICES.MEMORY_URL);
 
 const THRESHOLD_TOKENS  = parseInt(getEnv('SUMMARY_THRESHOLD_TOKENS', SUMMARIES.THRESHOLD_TOKENS));
@@ -35,41 +33,20 @@ Do not include greetings, sign-offs, or filler. Output only the summary text.
 Conversation:
 ${context}`;
 
-    return [
-        '<|im_start|>user',   // ChatML for qwen2.5
-        instruction,
-        '<|im_end|>',
-        '<|im_start|>assistant',
-    ].join('\n');
+    // No ChatML wrapper — the model's own prompt template is applied server-side
+    // by the inference service's /utility/complete route (Ollama /api/chat).
+    return instruction;
 }
 
 async function generateSummary(episodes, existingSummary = null) {
-    const prompt = buildSummaryPrompt(episodes, existingSummary);
+    const user = buildSummaryPrompt(episodes, existingSummary);
 
-    const res = await fetch(`${EXTRACTION_URL}/api/generate`, {
-        method: 'POST',
-        headers: { 'Content-Type': 'application/json' },
-        body: JSON.stringify({
-            model: EXTRACTION_MODEL,
-            prompt,
-            stream: false,
-            options: {
-                temperature: 0.2,   // slightly higher than entities — summaries benefit from some fluency
-                num_predict: 500,   // generous but bounded — keeps summaries from running long
-            },
-        }),
+    const content = await utilityInference({
+        user,
+        temperature: 0.2,   // slightly higher than entities — summaries benefit from some fluency
+        maxTokens: 500,     // generous but bounded — keeps summaries from running long
     });
 
-    if (!res.ok) throw new Error(`Ollama responded ${res.status}`);
-    const data = await res.json();
-
-
-    const raw = data.response?.trim() ?? '';
-    // Strip any leaked ChatML tokens Qwen echoes back
-    const content = raw
-        .replace(/<\|im_start\|>.*?<\|im_end\|>/gs, '')
-        .replace(/<\|im_start\|>|<\|im_end\|>|<\|im_sep\|>/g, '')
-        .trim();
     return content;
 }
 
@@ -148,4 +125,4 @@ async function triggerSummary(session) {
     );
 }
 
-module.exports = { triggerSummary, maybeSummarize };
+module.exports = { triggerSummary, maybeSummarize };