utilityInference

packages/shared/src/config/constants.js

@@ -105,6 +105,14 @@ const RETRIEVAL = {
     KEYWORD_WEIGHT:     0.5,    // Weight applied to keyword (SQLite) results, 0 = disables, set >0 to enable and tune balance between semantic vs keyword matches
 }
 
+const UTILITY = {
+    DEFAULT_URL:   'http://localhost:11434', // Ollama host for background/utility tasks
+    DEFAULT_MODEL: 'qwen2.5:3b',
+    TEMPERATURE:   0.2,      // precise, low-creativity default for extraction/summaries
+    MAX_TOKENS:    1200,
+    TIMEOUT_MS:    120_000,  // extraction previously used 60s; summaries had none — standardized
+};
+
 module.exports = {
     QDRANT,
     COLLECTIONS,
@@ -119,4 +127,5 @@ module.exports = {
     SUMMARIES,
     ENTITIES,
     RETRIEVAL,
+    UTILITY,
 };

packages/shared/src/index.js

@@ -1,5 +1,20 @@
 const {getEnv} = require('./config/env');
-const {QDRANT, COLLECTIONS, EPISODIC, SERVICES, OLLAMA, PORTS, LLAMACPP, INFERENCE_DEFAULTS, SQLITE, ORCHESTRATION, SUMMARIES, ENTITIES, RETRIEVAL } = require('./config/constants');
+const { 
+        QDRANT, 
+        COLLECTIONS, 
+        EPISODIC, 
+        SERVICES, 
+        OLLAMA, 
+        PORTS, 
+        LLAMACPP, 
+        INFERENCE_DEFAULTS, 
+        SQLITE, 
+        ORCHESTRATION, 
+        SUMMARIES, 
+        ENTITIES, 
+        RETRIEVAL,
+        UTILITY
+     } = require('./config/constants');
 const {parseRow, formatEpisodeText, isTrivialTurn} = require('./utils')
 const logger = require('./utils/logger');
 
@@ -22,4 +37,5 @@ module.exports = {
     ENTITIES,
     logger,
     RETRIEVAL,
+    UTILITY,
 };