Run inference on the open-source GPT-OSS 120B chat model via Jatavo, accepting JSON message payloads and optional streaming.