LLM inference proxy — send an OpenAI-format chat/completions request and get a response from GPT-4o-mini. No API key needed; pay per call via x402. Supports temperature, top_p, stop, and max_tokens (capped at 4096). Input capped at 16k chars.