The models are here. So are the users who pay.
One balance across every vendor: chat, image, video and speech from a single key, with China's SOTA models on day one. Drop-in for fal / OpenAI / Anthropic apps — change the base URL and key. Usage-based billing with an itemized ledger.
- OpenAI / Anthropic SDKs: change base_url and key, nothing else.
- fal clients: Python sets FAL_RUN_HOST / FAL_QUEUE_RUN_HOST; JavaScript has no host env var — use requestMiddleware.
- Model catalog (JSON): /catalog/models; one model: /catalog/models/{id}.
- Sign in with GateMux: users pay from their own balance, developers earn a share.
Machine-readable: /llms.txt, /models.md (append .md to any docs URL for raw markdown)