🧠 SDK
Train, fine-tune, and deploy RL policies with one SDK. Local or cloud training, plug-and-play robot integration, and publish directly to the Model Hub.
Open source. Python-first. Production-ready.
Three commands from install to deployed policy.
Authenticate with your account and connect your robot hardware or simulation environment.
$ ok auth loginUse a pretrained foundation model from the Hub as a starting point, or train a new policy from scratch with your task definition.
$ ok train --base humanoid-walk-v2 --task my_taskPush the trained policy directly to your robot with one command. Monitor performance and iterate without downtime.
$ ok deploy --robot spot --policy my_task-v1Define your robot's task, connect your environment, and launch training. Works with captured real-world scenes or standard simulation environments.
Start from a pretrained policy from the Hub and fine-tune it on your specific hardware, environment, or task. Reduce training time from days to hours.
Train on your own GPU cluster or use our cloud simulation infrastructure. Spin up parallel training runs without managing infrastructure.
Connect your robot, sensors, and constraints in a few lines of Python. The SDK handles the interface layer for all major robot platforms.
Push trained policies to the Model Hub as public or private repositories. Version, document, and share with your team or the community.
Collect new data from real deployments, retrain, and push updates — without manual pipeline management.
Full control over training configs, reward shaping, and architecture. Integrate with your existing MLOps stack.
Spend time on robot behavior, not infrastructure. The SDK abstracts the RL training loop and hardware interface.
Reproduce published results, fork foundation models, and publish new baselines directly to the community Hub.
Deploy training jobs to cloud infrastructure with a single command. CI/CD pipelines for robot policies.
Get early access to the SDK and deploy your first trained policy to real hardware.
Request Early Access