OpenAI-compatible vision (image understanding) from a private, self-hosted local vision-language model. Pay per call in USDC on Base via x402 - no API key or account. Send a public image URL plus a prompt (or OpenAI image_url messages); receive a standard chat.completion JSON answering about the image. Up to 4 images per call (more are clamped). Same privacy-first operator as our Private LLM Inference chat tiers.