utilityInference layer 2

packages/inference-service/src/utility.js

@@ -23,7 +23,7 @@ async function utilityComplete({
             model: UTILITY_MODEL,
             messages,
             stream: false,
-            ...UTILITY(json && {format: 'json' }),
+            ...(json && {format: 'json' }),
             options: {
                 temperature: temperature ?? UTILITY.TEMPERATURE,
                 num_predict: maxTokens ?? UTILITY.MAX_TOKENS,

packages/memory-service/src/entities/extraction.js

@@ -1,9 +1,7 @@
 const semantic = require('../semantic')
-const { getEnv, SERVICES, formatEpisodeText, ENTITIES, logger } = require('@nexusai/shared');
+const { getEnv, SERVICES, formatEpisodeText, ENTITIES, logger, utilityInference } = require('@nexusai/shared');
 const { upsertEntity, upsertRelationship, linkEntityToEpisode } = require('./index');
 
-const EXTRACTION_URL = getEnv('EXTRACTION_URL', 'http://localhost:11434');
-const EXTRACTION_MODEL = getEnv('EXTRACTION_MODEL', 'qwen2.5:3b'); // ChatML format — see buildExtractionPrompt
 const EMBEDDING_SERVICE_URL = getEnv('EMBEDDING_SERVICE_URL', SERVICES.EMBEDDING_URL);
 
 const ENTITY_TYPES = ENTITIES.TYPES;
@@ -28,10 +26,9 @@ function mentionedIn(name, haystack) {
   return norm(haystack).includes(norm(name));
 }
 
-// NOTE: This prompt uses ChatML format (<|im_start|> / <|im_end|> tags), which is
+// Returns { system, user } for utilityInference. The model's prompt template
-// specific to qwen-family models. If EXTRACTION_MODEL is changed to a Llama-family
+// (ChatML for qwen, etc.) is applied by Ollama via the inference service's
-// or other model, this format will need to change — most alternatives use either
+// /utility/complete route — no template tags belong in this file.
-// plain text or [INST] / <<SYS>> tags. Silent degradation is likely if mismatched.
 function buildExtractionPrompt(userMessage, aiResponse, knownEntities = []) {
     const knownBlock = knownEntities.length > 0
         ? [
@@ -41,11 +38,9 @@ function buildExtractionPrompt(userMessage, aiResponse, knownEntities = []) {
           ].join('\n')
         : '';
 
-    return [
+    return {
-        '<|im_start|>system',
+        system: 'You are a named entity and relationship extractor. You output only valid JSON.',
-        'You are a named entity and relationship extractor. You output only valid JSON.',
+        user: [
-        '<|im_end|>',
-        '<|im_start|>user',
             'Read the conversation below and extract all named entities and the relationships between them.',
             `Entity types: ${ENTITY_TYPES.join(', ')}`,
             'Use "character" for any fictional, game, or media characters (e.g. characters from anime, games, books, TV shows, movies)',
@@ -65,9 +60,8 @@ function buildExtractionPrompt(userMessage, aiResponse, knownEntities = []) {
             `User: ${userMessage}`,
             `Assistant: ${aiResponse}`,
             '--- END CONVERSATION ---',
-        '<|im_end|>',
+        ].join('\n'),
-        '<|im_start|>assistant',
+    };
-    ].join('\n');
 }
 
 async function embedEntity(entity) {
@@ -91,30 +85,16 @@ async function extractAndStoreEntities(userMessage, aiResponse, episodeId=null,
         // Fetch existing entities to guide the model toward consistent name/type pairs
         const db = require('../db').getDB();
         const knownEntities = db.prepare(`SELECT name, type FROM entities ORDER BY rowid DESC LIMIT 20`).all();
-        const prompt = buildExtractionPrompt(userMessage, aiResponse, knownEntities);
+        const { system, user } = buildExtractionPrompt(userMessage, aiResponse, knownEntities);
 
-
+        const raw = await utilityInference({
-        const res = await fetch(`${EXTRACTION_URL}/api/generate`, {
+            system,
-            method: 'POST',
+            user,
-            headers: { 'Content-Type': 'application/json' },
+            json: true,
-            body: JSON.stringify({
-                model: EXTRACTION_MODEL,
-                prompt: prompt,
-                stream: false,
-                format: 'json',
-                options: {
             temperature: ENTITIES.TEMPERATURE,
-                    num_predict: ENTITIES.NUM_PREDICT,
+            maxTokens: ENTITIES.NUM_PREDICT,
-                },
-            }),
-            signal: AbortSignal.timeout(60_000),
         });
 
-        if (!res.ok) throw new Error(`Ollama responded ${res.status}`);
-
-        const data = await res.json();
-        const raw = data.response?.trim() ?? '';
-
         const jsonMatch = raw.match(/\{[\s\S]*\}/);
         if (!jsonMatch) {
             logger.warn('[entities] No JSON object found in response');

packages/shared/src/index.js

@@ -17,6 +17,7 @@ const {
      } = require('./config/constants');
 const {parseRow, formatEpisodeText, isTrivialTurn} = require('./utils')
 const logger = require('./utils/logger');
+const utilityInference = require('./utils/utilityInference');
 
 module.exports = {
     getEnv, 
@@ -38,4 +39,5 @@ module.exports = {
     logger,
     RETRIEVAL,
     UTILITY,
+    utilityInference,
 };

packages/shared/src/utils/utilityInference.js

@@ -0,0 +1,20 @@
+const { getEnv } = require('../config/env');
+const { SERVICES } = require('../config/constants');
+
+// Client for the inference service's /utility/complete route.
+// Resolved at call time (not module load) so each service's .env applies.
+async function utilityInference({ system, user, json = false, temperature, maxTokens }) {
+    const base = getEnv('INFERENCE_SERVICE_URL', SERVICES.INFERENCE_URL);
+
+    const res = await fetch(`${base}/utility/complete`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({ system, user, json, temperature, maxTokens }),
+    });
+
+    if (!res.ok) throw new Error(`Utility inference error: ${res.status}`);
+    const data = await res.json();
+    return data.text;
+}
+
+module.exports = { utilityInference };