Inference console
Send a prompt. Watch it stream.
Pick a model, send a request, and the router streams a model's tokens back from a live GPU worker. This is the product, not a video of it.
Inference console
Checking network…
Your streamed response appears here, token by token.
⌘/Ctrl + Enter to run · streamed over SSE · prompts are not stored
When the live network is offline you'll see a labeled preview — real tokens stream the moment a worker is online.
01
Route
Your prompt goes to the router, which picks an eligible GPU worker by model, availability, and measured speed.
02
Stream
The worker runs the model and streams tokens straight back to you over SSE, one at a time.
03
Settle
Usage is metered per token; the worker earns for the compute, and your prompt is discarded. Nothing is stored.