If your code already calls an OpenAI-compatible endpoint, you can point it at Cluster without rewriting any of it.
pip install cluster-sdk, change base_url, keep every call you already wrote.
Behind that one URL: 500+ models across chat, embeddings, image, TTS, STT and reranking. Multi-provider, so a deprecated model is a routing change instead of a rewrite.
The part that matters if you are building agents is x402. Client sends a request with no payment attached. Server returns 402 Payment Required with the amount, the recipient wallet and the network. Client pays, retries, gets the response. No account, no API key, no prepaid balance sitting somewhere.
Which means an agent that spins up for forty seconds can call a model and pay for that one call. Nothing provisioned in advance, nothing reserved, no human approving anything mid-run.
Swap the base URL and see what breaks.