Build with robot data.
Augment teleoperation datasets, train across many VLA and imitation-learning models, and evaluate policies in simulation. One small, predictable REST API — on cloud GPUs you never have to manage.
The platform
Augment
Point Kite at a dataset, describe a visual change in plain language, and get back new episodes in standard Parquet format.
Train
Train across many VLA and imitation-learning policies on the same dataset, on cloud GPUs provisioned on demand.
Evaluate
Roll out policies in simulation and score them against benchmarks — before you ever touch real hardware.
Start building