Schnitt

A lab. We build foundational models that never stop learning.

Every model out there knows the whole internet and nothing about the enterprise it works for. It stops learning the day it ships. Ours never stop. No knowledge cutoff. Every interaction and what happened next goes back into the weights.

Sarathi

Sarathi is our family of models, trained from scratch. Every weight our own.

• Continual learning. Continued pretraining on every interaction and its outcome. No knowledge cutoff.
• RL environments. Reinforcement learning in environments built from real outcomes, so the model is rewarded for what actually happened, not for what a rater liked.

The bet

Self-improving models are the next frontier. We are betting big on RSI.

Access

Sarathi weights are not public. Access for research and enterprise partners is by request.

schnitt.ai · team@schnitt.ai