Quickstart

Install DimOS, replay a robot session with no hardware, then run a real Unitree Go2.

In fifteen minutes you will: install DimOS, replay a recorded Unitree Go2 session in the Rerun viewer with no hardware, and then run the same stack on a real Go2.

This page is geared toward the Unitree Go2. If you were given a Galaxea A1Z arm instead, start with the A1Z guide.

If you use coding agents (Claude Code, OpenClaw, or similar), point them at AGENTS.md in the repo.

System requirements

Component	Minimum	Recommended
OS	Ubuntu 22.04, macOS 12.6+	Ubuntu 24.04
Python	3.10	3.12
RAM	16 GB	32 GB+
Disk	10 GB SSD	25 GB+ SSD
CPU	8-core Intel / AMD	12+ cores
GPU (optional)	NVIDIA RTX 3000+ (8 GB VRAM)	RTX 4070+ (12 GB+ VRAM)

GPU is required only for perception, VLMs, and AI features. Optional for basic robot control and everything on this page.

Step 1: Install

DimOS supports Python 3.10 to 3.12 and targets 3.12. The examples use uv; plain python -m venv and pip work too.

# fetch large files on-demand instead of pulling all of them immediately
export GIT_LFS_SKIP_SMUDGE=1
git clone https://github.com/dimensionalOS/dimos.git
cd dimos
uv venv --python 3.12
source .venv/bin/activate
uv pip install --pre -e '.[base,unitree]'

This is an editable install from the repo: you get the dimos CLI for everything below, and you can open the code and write your own skills and blueprints right away. Extras keep installs lean: base is the runtime, modules, transports, and CLI; unitree adds WebRTC and skills for Go2 / G1 (real or replayed).

Other install options

One-command interactive install (system deps + Python env):

curl -fsSL https://raw.githubusercontent.com/dimensionalOS/dimos/main/scripts/install.sh | bash

Or install system dependencies yourself with the guide for your OS: Ubuntu (primary tested path), macOS (Homebrew-based, less mature than Linux).

Step 2: Replay a recorded session (no hardware)

dimos --replay run unitree-go2

A Rerun viewer window opens showing the robot’s camera, the live-built map, and the planned paths from a recorded office navigation session, with SLAM, costmap, and A-star planning running for real against the recorded sensors. On first run the window may stay black briefly while roughly 75 MB of recorded data downloads.

While it runs, try the ops commands from a second terminal:

dimos status      # what is running
dimos log -f      # follow logs
dimos stop        # shut the stack down

Step 3: Real robot

For a physical Unitree Go2 over WebRTC, set the robot’s IP:

export ROBOT_IP=<YOUR_ROBOT_IP>

Newer Go2 firmware encrypts the connection. If you were given an AES key, set it the same way (robots on older firmware only need ROBOT_IP):

export UNITREE_AES_128_KEY=<YOUR_AES_KEY>

Then run the Go2 blueprint:

dimos run unitree-go2

Do not skip the platform guide - network setup, latency, and safety habits matter: Unitree Go2. For other hardware, see Platforms in the sidebar, including the Galaxea A1Z arm.

More things to run

Command	What it does
dimos --replay run unitree-go2	Quadruped navigation replay: SLAM, costmap, A-star planning
dimos --replay --replay-db go2_bigoffice run unitree-go2-memory	Replay a long session while recording it into a queryable SQLite store (memory2)
dimos --simulation run unitree-go2-agentic	Quadruped agent + MCP server in simulation
dimos --simulation run unitree-g1-sim	Humanoid in MuJoCo simulation
dimos --replay run drone-basic	Drone video + telemetry replay
dimos --replay run drone-agentic	Drone + LLM agent with flight skills (replay)
dimos run demo-camera	Webcam demo, no robot needed
dimos run keyboard-teleop-xarm7	Keyboard teleop with a mock xArm7 (dimos[manipulation] extra)

dimos list prints every available blueprint. Blueprint reference: Blueprints.

What next?

How DimOS fits together

The mental model: modules, streams, and blueprints behind every workflow.

Agents

Natural language control: skills, MCP, and how to add your own.

Platforms

Bring-up guides: Unitree Go2, G1, and manipulator arms.

Memory

The three memory systems and what each is for.