DimOS, from Dimensional, is an agentic operating system for physical space. It lets you command humanoid robots, quadrupeds, drones and other hardware platforms in natural language, and build multi-agent systems that work directly with physical inputs such as cameras, lidar and actuators.
Robotics software is usually fragmented across drivers, ROS graphs and vendor SDKs. DimOS aims to be a single agent-native layer on top: agents run as native modules that subscribe to embedded streams, from perception and spatial memory down to control loops and motor drivers. It bundles SLAM navigation with dynamic obstacle avoidance and route planning, perception pipelines including detectors, 3D projections, vision-language models and audio processing, and a spatial memory built on spatio-temporal RAG with object localization and permanence. A command-line interface and MCP integration let you inspect and call robot skills directly, and the project supports platforms such as the Unitree Go2 and G1, with MuJoCo-based simulation and session replay for development. Hardware support is tiered from stable to experimental per platform, and the system can run without ROS, which lowers the barrier for Python developers entering robotics. The open-source project has earned more than 4,600 stars on GitHub and was under active development in 2026.
Freemium. The DimOS core is open source and free to run on your own hardware. The team also operates dimTELE, a hosted remote-teleoperation service that lets you drive connected robots from a browser or headset through Dimensional’s console.
Robotics developers and research teams who want to build language-driven, agentic applications on real hardware without stitching together a separate stack for navigation, perception, memory and control.