Innate is the open-source SDK for training, simulating, and deploying Vision-Language-Action models. Plug in your data. Get a robot that works.
Five stages. One config file. Robot arms do the heavy lifting so you don't have to.
LeRobot, ROS2, HDF5 — any format
inx connect lerobot
Canonical Episode schema
Episode → Steps
MuJoCo, Isaac Sim backends
inx sim --backend mujoco
LoRA fine-tuning, any VLA
inx train --config inx.yaml
Export & run on real hardware
inx export --format onnx
LeRobot, ROS2 bags, HDF5, RLDS — Airbyte-style connectors normalize everything to one canonical schema. No more format wrestling.
Swap between MuJoCo and Isaac Sim with one config line. Evaluate your policy in simulation before touching hardware.
LoRA fine-tuning for SmolVLA, OpenVLA, and more. Local-first experiment tracking. No vendor lock-in, no cloud required.
Join the waitlist. Help us decide what to build first.