dimensionalOS/dimos

Dimensional is the agentic operating system for physical space. Command humanoids, quadrupeds, drones, and other hardware platforms in natural language and build multi-agent systems that work seamlessly with physical input (cameras, lidar, actuators).

What it solves

DimOS is an operating system and SDK for generalist robotics, designed to remove the complexity of robot programming. It eliminates the need for ROS (Robot Operating System) by allowing developers to build physical applications entirely in Python, providing a standardized way to integrate with various robot manufacturers (humanoids, quadrupeds, drones, and arms).

How it works

DimOS uses a modular architecture where robot subsystems are defined as "Modules" that communicate via standardized messages. These modules are wired together using "Blueprints," which act as instructions for constructing and connecting the robot's data streams. The system is "agent native," meaning LLM-based agents can run as native modules, subscribing to perception streams (like LiDAR and cameras) and controlling motor drivers directly. It also supports multiple transport layers (LCM, SHM, DDS, Zenoh, ROS 2) and provides a CLI for managing agent lifecycles and calling skills via the Model Context Protocol (MCP).

Who it’s for

Robot developers and AI researchers who want to build agentic physical applications without the overhead of ROS, and those who need a cross-platform SDK that works across different hardware platforms like Unitree Go2, G1, and DJI drones.

Highlights

  • Agent-Native Design: Allows "vibecoding" robots using natural language and building multi-agent systems that interact with hardware.
  • Spatial Memory: Implements spatio-temporal RAG and object localization for robots to remember and find objects.
  • Hardware Agnostic: Supports a wide range of platforms including quadrupeds, humanoids, robotic arms, and drones.
  • dimTELE: A hosted remote teleoperation system that allows controlling robots via a browser or Quest headset over WebRTC without needing to open inbound ports.
  • MCP Integration: Uses the Model Context Protocol to expose robot skills as tools that agents can call.

Related

  • Project
    autonomous-ai/autonomous-os

    A customizable operating system for robots that integrates agentic reasoning engines and a modular skill system to enable true physical autonomy.

  • Project
    nutshellai-tech/mobius

    Mobius is a self-evolving Agent OS that connects teams, AI agents, and compute resources into a single workspace, allowing the system to rewrite its own code and UI based on user input.

  • Project
    ROBOTIS-GIT/DynamixelSDK

    A software development kit for controlling Dynamixel actuators and platforms through a standardized packet communication API.

  • Project
    jaylfc/taOS

    A self-hosted AI agent platform that turns spare hardware into a distributed compute cluster, providing a framework-agnostic memory system and a full web-based desktop interface.

  • Project
    OpenMind/OM1

    A modular AI runtime for creating and deploying multimodal agents across digital environments and physical robots, supporting diverse sensors and AI models.