Stream a model's answer to the browser token by token. Air Pipe reads the provider's token stream and forwards it to a realtime channel, so a chat UI renders text while the model is still writing. Works with any OpenAI-compatible provider.