Rêwî

Rêwî

A new kind of AI, not another large language model.

A ground-up architecture and model that runs on roughly a tenth of the resources today's LLMs need, and answers in a fraction of the time.

~10× leanerResource use
up to 1000× fasterResponse time

A note on accuracy: Rêwî is early. It does not yet match the accuracy of today's large language models. The efficiency and speed above come first, and reaching parity on accuracy is still ahead of us.

Inside Rêwî: a living map of concepts, beliefs, and the reasoning it has learned.
Its mind: concepts, beliefs, and the reasoning it builds as it learns.
Talking with Rêwî: it answers from what it knows and can show how it thought.
In conversation: it reasons from what it knows, and says so when something's beyond it.
Rêwî's lab: it practises on its own, reading and quizzing itself.
The lab: it practises on its own, reading and quizzing itself to get sharper.
Rêwî's identity: sliders for its character, judgement, and what it trusts.
Its character: shape how it behaves, what it trusts, and when it is sure enough to speak.
Rêwî's tools: the sources it can look things up in when it does not know.
Its tools: where it looks things up when it does not know, each treated as a source to check.
Coming soon
← Samalstudios