# Kite ML

> Kite ML gives robotics teams the tools to augment datasets, train across models, and evaluate policies in simulation — so teams ship robot behavior instead of stitching the stack themselves. Kite is a member of the NVIDIA Inception Program and is backed by Founders Inc. Canopy.

Kite brings these layers together out of the box:
- Pre-supported robots (Unitree G1 humanoid, Unitree Go2 quadruped, Boston Dynamics Spot, SO-100 arm) plus custom URDF upload with kinematic validation.
- A robotics coding agent grounded on the active project's URDF, controllers, sensors, and Kite SDK.
- World generation via the World Labs integration (text-to-3D + image-to-3D scenes with seeded reproducibility).
- Object generation via the SAM 3D pipeline (click-to-segment, lift to physics-ready mesh with inertia + collision primitives).
- Vision-Language-Action (VLA) policy training, starting with Physical Intelligence's π-series (π0, π0-FAST) with PEFT adapters and automatic simulation evaluation.
- Fleet policy evaluation: run a trained policy (ACT, π0, smolVLA, or RL) across many randomized cloud sims, measure success/failure rates with per-episode video proof, and choose the scene from your own project or a built-in benchmark-environment catalog (e.g. SO-Arm Pick & Place, SO-Arm Reach). Runs on GPU MuJoCo (MJX) on Vertex AI.
- A physics + simulation engine built on MuJoCo, extended by Kimodo for motion retargeting, joint-space blending, and quaternion-safe transforms across embodiments.
- Synthetic data generation with scripted task primitives (place, grasp, push, pour, traverse) and per-episode domain randomization; exports to HuggingFace, LeRobot, and raw parquet.
- Dataset Augmentation Studio: import a HuggingFace LeRobot dataset, describe a visual change (lighting, surface, texture, objects), and generate an expanded dataset with new videos (Google Gemini / Gemini Omni) plus motion-preserving joint trajectories, then push it back to the user's own HuggingFace account.
- On-demand cloud compute: CPU processing on the Free plan, GPU compute on the Pro plan (Google Vertex AI custom jobs + Cloud Run GPU).

## Canonical pages

- [Homepage](https://kiteml.com/): product overview, hero demo, capabilities, use cases, pricing teaser, FAQ.
- [Documentation](https://kiteml.com/docs): section-by-section tour of what ships with Kite.
- [Pricing](https://kiteml.com/pricing): Free ($0, one project, CPU), Pro ($100/user/month, unlimited projects + cloud GPUs), Enterprise (custom).

## Sign up and subscribe

- Sign up: https://app.kiteml.com/signup
- Log in: https://app.kiteml.com/login
- Subscribe to Pro (handles auth redirect if needed): https://app.kiteml.com/subscribe?plan=pro

## Partnerships and funding

- NVIDIA Inception Program — member
- Founders Inc. Canopy — investor

## Contact

- Sales / partnerships: raul@kiteml.com