Dimensional (dimos) is an Agentic Operating System for physical space, positioned as a modern operating system for the general robotics field and committed to establishing the next-generation SDK standard, and has already integrated with most robot manufacturers.
In terms of use, Dimensional lets users command humanoid robots, quadruped robots, drones, and other hardware platforms in natural language, and build multi-agent systems that seamlessly collaborate with physical inputs (cameras, lidar, actuators). Its core goal is to make robot development like "vibecode," completing programming in natural language.
Main capabilities include: no ROS required, building physical applications with only Python, and running on any humanoid, quadruped, or drone; agent-native support, with agents running as native modules that can subscribe to any embedded data stream, from perception (lidar, cameras) and spatial memory, to control loops and motor drives; support for local and hosted multi-agent systems; providing capabilities such as navigation and mapping, agent control, and MCP; providing Agent CLI and MCP interfaces, allowing agents such as OpenClaw and Claude Code to directly participate in building Dimensional applications; providing dimTELE remote teleoperation, remotely operating DimOS robots from any browser or Quest headset via WebRTC, with the robot actively dialing out to a hosted broker, without the need to open inbound ports on the robot's network, and able to work behind a home router, over Wi-Fi, wired LAN, or cellular network; supporting the use of DimOS as a library, with modules communicating through standardized messages, and Blueprints used to describe how modules are built and connected, able to connect data streams by (name, type) via autoconnect(...) and return a Blueprint, with blueprints composable and remappable, and overriding transport when autoconnect() fails due to variable name conflicts or In[]/Out[] message type conflicts; the transport layer supports Zenoh, LCM, DDS, and ROS 2; the Web side provides cockpit, web SDK, and relay; in terms of multilingual support, Python serves as the glue and prototyping language, while also supporting multiple languages through LCM interoperability.
In terms of usage, it is recommended to use the official installer for guided installation to set up system dependencies, Python 3.12, and dimOS; you can view scripts/install.sh --help for non-interactive and advanced options; manual installation is also possible, according to Ubuntu, macOS, or Nix system packages and Python steps, then activate the environment using the command printed by the installer and run. Regarding Agent CLI and MCP, the dimos CLI manages the full lifecycle—running blueprints, checking status, interacting with agents, and invoking skills via MCP. dimTELE usage steps: open the Dimensional console, log in and create an API key (API keys → Create key); run the teleoperation blueprint on the robot, passing the key as TRANSPORTS__BROKER__API_KEY, for example: TRANSPORTS__BROKER__API_KEY=<your-api-key> dimos run teleop-hosted-go2-transport; the robot will appear under Available Robots, and clicking Connect allows you to drive the robot from the browser.
Applicable scenarios include: physical application development for hardware platforms such as humanoids, quadrupeds, and drones; building multi-agent systems; remote robot teleoperation; natural-language-based robot programming and prototyping; robot applications that need to collaborate with physical inputs such as cameras, lidar, and actuators.
In terms of limitations, it is currently in the Pre-Release Beta stage. NixOS/general Linux has not been CI-tested. Full system requirements, tested configurations, and dependency tiers should be referenced in docs/requirements.md. Other specific limitations are not yet provided.
