RoboClaw Installation Guide

April 18, 2026 · View on GitHub

This guide is the native host installation path. RoboClaw uses uv as the only supported Python environment and dependency workflow. If you want Docker-based workflows, use:

1. Install uv

Install uv with the official installer:

curl -LsSf https://astral.sh/uv/install.sh | sh

Verify that uv is available:

uv --version

2. Clone RoboClaw

Start from a clean clone and fetch submodules:

git clone --recurse-submodules https://github.com/MINT-SJTU/RoboClaw.git
cd RoboClaw

The embodied engine lives in roboclaw/embodied/engine as a git submodule and is required during installation.

If you already cloned the repository without submodules, run:

git submodule update --init --recursive

3. Sync the Project with uv

RoboClaw pins Python with .python-version and uses uv sync to create .venv and install all default development dependencies:

uv sync

After sync, verify that the roboclaw command is available:

uv run roboclaw --help

Expected result:

  • commands such as onboard, status, agent, and provider are listed

4. Initialize RoboClaw

Run:

uv run roboclaw onboard

This should create ~/.roboclaw/config.json, ~/.roboclaw/workspace/, and the initial workspace scaffold. You can verify it with:

find ~/.roboclaw -maxdepth 4 -type f | sort

You should see at least:

~/.roboclaw/config.json
~/.roboclaw/workspace/AGENTS.md
~/.roboclaw/workspace/HEARTBEAT.md
~/.roboclaw/workspace/SOUL.md
~/.roboclaw/workspace/TOOLS.md
~/.roboclaw/workspace/USER.md
~/.roboclaw/workspace/memory/MEMORY.md

5. Verify Status Output

Run:

uv run roboclaw status

Check that:

  • Config is shown as
  • Workspace is shown as
  • the current Model looks correct
  • provider status matches the actual state of your machine

6. Configure the Model Provider

Before testing roboclaw agent, make sure the model provider is configured.

First run:

uv run roboclaw status

This tells you which providers are already available on the current machine.

Two common cases:

5.1 OAuth provider

If you are using an OAuth-based provider, log in directly.

The current codebase supports:

uv run roboclaw provider login openai-codex
uv run roboclaw provider login github-copilot

5.2 API key provider

If you are using an API-key-based provider, edit:

~/.roboclaw/config.json

Fill in the provider key and default model there.

Common API key providers include:

  • openai
  • anthropic
  • openrouter
  • deepseek
  • gemini
  • zhipu
  • dashscope
  • moonshot
  • minimax
  • siliconflow
  • volcengine
  • azureOpenai
  • custom
  • vllm

Then run:

uv run roboclaw status

Check that:

  • the current Model is correct
  • the provider you want to use is no longer not set

7. Verify the Basic Model Path

Run one minimal message to confirm that RoboClaw can respond:

uv run roboclaw agent -m "hello"

Check that:

  • the agent starts successfully
  • the agent returns a normal reply
  • failures point clearly to model configuration, provider setup, network, or permissions

8. Launch the Web Dashboard

The web dashboard provides a browser-based UI for chatting with RoboClaw.

Install the frontend dependencies:

cd ui
npm install

Production Mode

Build the frontend and start the server:

cd ui && npm run build && cd ..
uv run roboclaw web start

Open http://127.0.0.1:8765 in your browser.

Development Mode (with hot reload)

# Terminal 1: start backend
uv run roboclaw web start

# Terminal 2: start frontend dev server
cd ui
npm run dev

Open http://localhost:5173 in your browser. The Vite dev server proxies /api and /ws to the backend automatically.

Options

uv run roboclaw web start --host 0.0.0.0 --port 9000
FlagDefaultDescription
--host127.0.0.1Bind address
--port8765Port number
--workspace~/.roboclaw/workspaceWorkspace directory
--verboseoffEnable debug logging