Before The Machines Soared, They Trained.

Our Vision

AI should be able to learn from experience.

We build the reinforcement learning infrastructure so AI - no matter the form factor - can improve on its own. Developers play a game of whack-a-mole when it comes to improving AI agents, where each edge case demands a manual adjustment to prompts or context. This creates a natural bottleneck.

And this challenge is exacerbated when the models used aren’t improving from task-to-task. So when you deal with a multi-step use case, the failure rate compounds quickly - and even if the task is solved, the latency is often high.

Osmosis is the first reinforcement fine-tuning platform designed to solve these practical, real-world challenges that teams building AI agents face. And because it’s your own data, we believe you should have full ownership of the fine-tuned models - serve with us, or export to self-host.

Our vision is to help developers deploy AI agents that actually learn and work.

Our Team

Our reason for working on reinforcement learning is simple:

We believe it is the most important challenge when it comes to unlocking AI's true capabilities. We want to help create the first AI agents that truly learn from experience.

Osmosis was founded in 2024 by Kasey and Andy (friends since middle school). Kasey is a repeat founder who sold his first company in undergrad, while Andy was the youngest tech lead at TikTok on the real-time recommendations/model serving team.

In order to do great work, you need to work with great people. If you want to work on the bleeding-edge of technology and help shape how AI agents improve, we'd love to talk.