Embedded local LLM inference via ONNX Runtime — zero external processes, inference runs entirely inside the JVM on CPU, no data leaves the process. Uses platform threads only, so it runs on Java 17+.