Get started
We build RL environments for professional software to investigate where models fail and turn those failures into tasks for evaluation and RL training.
Use traces to understand why models fail and design tasks that reduce reward hacking while helping your team hillclimb model performance. Read more.
Ask Claude or Codex to read usedesktop.com/setup.txt and set up your environments, runtime, and SDK.
Preparing
01 / Environments
RL environments
Start with RL environments.
Browse ready-to-run environments with tasks, verifiers, and reset logic.
02 / Runtime
Desktop
Run and trace models.
Use our Desktop SDK or app to run models inside environments and capture traces.
03 / Evaluation
Evals
Evaluate model performance.
Inspect model runs, verifier results, traces, reports, and pass@k summaries.