ai insight rework #1

docker-compose.yml

@@ -54,6 +54,11 @@ services:
     image: ollama/ollama:latest
     container_name: dashboard-cpsp-llm
     restart: unless-stopped
+    environment:
+      # Plan §6: serialize CPU inference, keep model resident (prefix KV cache reuse).
+      - OLLAMA_NUM_PARALLEL=1
+      - OLLAMA_MAX_LOADED_MODELS=1
+      - OLLAMA_KEEP_ALIVE=24h
     ports:
       - "127.0.0.1:11434:11434"
     volumes: