UseDesktop
Verified RL environments and trajectory data for CUA
34 followers
Verified RL environments and trajectory data for CUA
34 followers
UseDesktop is a vertically integrated pipeline for CUA data: capture real desktop workflows, normalize traces, build verifiers, run agent evals, and export RL task environments with high-quality trajectories.
Computer-use agents do not need more toy tasks. They need long-horizon workflow data with evidence: where models fail, how success is verified, and whether training improves held-out tasks.
Built for AI labs working on post-training, evaluation, and reinforcement learning for CUA.
Hi,
My name is Seungju from South Korea and I am the founder of UseDesktop.
UseDesktop is the infrastructure for training Desktop Agent.
Desktop Agent lives in your PC and acts based on what you showed them.
It is different compared to other autonomous pro active agents like OpenClaw in the sense that its action is more deterministic making the result more
Accurate
Faster(less latency)
Cheaper(in terms of token)
It is designed privacy-first and you can connect local models to use it.
At UseDesktop, you are able to
Create Desktop Agent that will automate all your works
Train actual finetuned model based on your own data
It was started with the motto of freeing Humans from labors. As We, humans, will gradually become managers managing hundreds of AI Agents that does the labor work for us.
Try it out! and If you need any help, you can reach out to me in the Discord or through my email
seungju@usedesktop.com