Standard — Generic GGUF model runner via local LLAMA-Server process. All AI computation happens in the external LLAMA-Server process. If it OOMs the Tomcat JVM stays alive — only the LLAMA-Server dies.