Connect models.

Run them as one.

We build a communication protocol that lets AI models run inference together, communicating in latent space instead of passing text.

↓

Problem

Every model is good at different things, but there is no way to get the strengths of several models into one answer.

The usual way up is a bigger model at a bigger price. Chaining models does not help either: they exchange finished text and lose the computation behind it.

Solution

We connect models behind the text layer.

Connected models exchange internal representations in latent space and run inference together. One input, one output, the strengths of each model in it.

Team

15 researchers, engineers and mathematicians.

Previously: Google X, NVIDIA, Perplexity, FAR.AI.

Talk to us.