Write in Python
One codebase for humanoids, quadrupeds, drones, and arms.
Laptop first
Develop against recorded sessions and simulation before any hardware.
Agent native
LLM agents are modules: they see streams, call skills, and take natural language.
No robot or GPU needed to try it. The quickstart replays a recorded session on your laptop in a few minutes.
See it in action
Agents on real hardware, driven by natural language.Drone follow
“Follow the next white car that comes through the intersection” — MAVLink drone agent on X.
Office patrol
“Patrol the office… if you see someone in a hoodie, follow them” — autonomous Go2 agent on X.
Capabilities at a glance
Navigation & mapping
Map a space while driving, avoid obstacles, and plan routes — native modules or ROS.
Remote teleoperation
dimTELE — low-latency WebRTC from a browser, phone, or VR headset. No port forwarding.
Agents
Natural-language control, with every skill also exposed as an MCP tool.
Spatial memory
Remember what was seen and where — then navigate back to it.
Manipulation
Motion planning and teleop for arms like xArm7, A-750, and OpenArm.
Perception
Detect objects in camera, place them in 3D, with VLMs and audio built in.
Start here
Quickstart
Replay a real robot session on your laptop. No robot needed.
Platforms
Unitree Go2 and G1 setup for simulation and real hardware.