OM1 is OpenMind’s open-source AI runtime for robots — a modular hardware abstraction layer that lets the same AI “brain” perceive, reason and act across very different robot bodies. Large language and vision models handle understanding and planning, while OM1’s modular architecture connects them to sensors, speech and actuators on quadrupeds, humanoids, drones and other machines. The project has about 3,000 stars on GitHub and is actively developed in 2026.
Robotics software is usually welded to one chassis; OM1 instead separates the agent into composable modules — inputs from cameras, microphones, lidar and other sensors, LLM/VLM-powered cognition, and action outputs — coordinated over ROS 2 and Zenoh messaging. You configure an agent in files rather than code: pick a persona, choose models (local or cloud), attach skills, and deploy the same configuration to different hardware by swapping the hardware-specific blocks. Developers can add new sensors, end-effectors or model providers as modules, and the runtime handles the plumbing between perception, reasoning and motion. OpenMind positions OM1 as the open foundation of a broader robot ecosystem, but the repository stands alone as a practical way to put modern multimodal AI in charge of real robots.
OM1 is free and open source under the MIT license. Costs are limited to your robot hardware and any cloud model APIs you choose to plug in.
OM1 is for robotics developers, researchers and hobbyists who want to give physical robots modern AI perception and reasoning without rewriting the stack for every new machine.