LLM inference API: OpenAI-compatible chat completions, send messages, get choices and usage. Cheap LLM inference for agents (Llama 3.1 8B), no key, no streaming.