Get Nook
  1. Nook
  2. Labs

Nook Labs

Experiments with AI inference on the computers people actually own. We publish the setup, the numbers and the runs that went wrong, so you can check our work and run it yourself.

How we run an experiment

Every note answers one question, on hardware we name, in a way you can repeat.

01

Real hardware, named

The card, the driver, the engine and its version, the model file down to its quantisation. A number without its setup is a rumour.

02

Every run counts

The failures and the “no difference” results go up too, and we say how many runs a number comes from.

03

Run it yourself

Commands, settings and test files wherever we can share them. Get a different number? Send it and we add it to the note.

Run it on your own GPU.

Nook runs open models on your own computer: no account, no upload, and it keeps itself up to date.