checking the network…
Talk to the network.
Your prompt goes to a Sestrian node, which generates from the chain's current weights — the same weights every other node agrees on, at the block height shown with each reply. Nothing here is a hosted API in front of someone else's model.
Expect rough output. This model is being trained from scratch, in public, by whoever shows up — it is around 111M parameters and only a few hundred blocks old. It reads and writes raw bytes, so it will produce half-formed words and wander off. That is what an early model sounds like; watching it stop sounding like that is the point.
One generation at a time, network-wide: if someone else is mid-answer you will be told to try again. Prompts are capped at 1000 characters. Served by api.sestrian.com, which you can query yourself.