--- Here’s a practical **Windows 11 + NVIDIA GPU + 64 GB RAM** checklist for setting up **OpenClaw in WSL2**, which is the route current Windows guidance and community writeups keep converging on. The key idea is: **Windows hosts the GPU driver, WSL2 hosts Linux, and Docker Desktop bridges the container layer**.[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI)skywork+2[youtube](https://www.youtube.com/watch?v=uqc_54fbsPU) ## Checklist ## Windows prep - Update Windows 11 fully. - Install the latest NVIDIA Windows driver with WSL/CUDA support. - Reboot. - Open PowerShell as Administrator.[forecr](https://www.forecr.io/blogs/installation/nvidia-docker-installation-for-ubuntu-in-wsl-2)[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI) ## Install WSL2 ``` powershell wsl --install wsl --update wsl --set-default-version 2 wsl --install -d Ubuntu-24.04 wsl --status ``` WSL2 is the required Linux layer for GPU-backed Docker workflows on Windows 11.[developer.nvidia](https://forums.developer.nvidia.com/t/run-cuda-inside-docker-wsl-on-windows-11-hyper-v-vm-with-gpu-patritioning/308212)[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI) ## Open Ubuntu and prep Linux ``` bash sudo apt update && sudo apt upgrade -y sudo apt install -y curl git ca-certificates gnupg lsb-release ``` ## Confirm GPU visibility in WSL ``` bash nvidia-smi ``` If this works, WSL can see the NVIDIA GPU through the Windows driver stack.[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI)[forecr](https://www.forecr.io/blogs/installation/nvidia-docker-installation-for-ubuntu-in-wsl-2) ## Install Docker Desktop - Install Docker Desktop in Windows. - Turn on **Use the WSL 2 based engine**. - Enable integration for your Ubuntu distro. - Restart Docker Desktop if needed.[developer.nvidia](https://forums.developer.nvidia.com/t/run-cuda-inside-docker-wsl-on-windows-11-hyper-v-vm-with-gpu-patritioning/308212)[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI) ## Verify Docker ``` bash docker version docker run --rm hello-world ``` ## Enable NVIDIA GPU passthrough for Docker in WSL ``` bash curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list sudo apt update sudo apt install -y nvidia-container-toolkit sudo nvidia-ctk runtime configure --runtime=docker ``` Then restart WSL: ``` powershell wsl --shutdown ``` Test GPU containers: ``` bash docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi ``` This is the important pass/fail test for NVIDIA GPU passthrough in Docker on WSL2.[forecr](https://www.forecr.io/blogs/installation/nvidia-docker-installation-for-ubuntu-in-wsl-2)[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI)[developer.nvidia](https://forums.developer.nvidia.com/t/run-cuda-inside-docker-wsl-on-windows-11-hyper-v-vm-with-gpu-patritioning/308212) ## Install OpenClaw In WSL Ubuntu, run the official Linux installer command from OpenClaw’s docs or site. Current Windows guides consistently place this step inside WSL, not native Windows.[skywork](https://skywork.ai/skypage/en/openclaw-windows-install-guide/2036741351875907584)[youtube](https://www.youtube.com/watch?v=uqc_54fbsPU) Typical pattern: ``` bash curl -fsSL <OPENCLAW_INSTALL_URL> | bash ``` If the installer asks about Docker, use the Docker Desktop-backed WSL integration you already enabled.[cybernative](https://cybernative.ai/t/openclaw-on-windows-11-home-wsl2-a-beginner-setup-that-won-t-eat-your-machine/33992?tl=en)[youtube](https://www.youtube.com/watch?v=uqc_54fbsPU) ## Start OpenClaw Use the command OpenClaw’s installer or docs provide for startup. If you’re following the current Windows/WSL guidance, keep the service and agent runtime inside WSL rather than crossing back into Windows filesystem paths too early.[youtube](https://www.youtube.com/watch?v=uqc_54fbsPU)[cybernative](https://cybernative.ai/t/openclaw-on-windows-11-home-wsl2-a-beginner-setup-that-won-t-eat-your-machine/33992?tl=en) ## Lock down WSL - Avoid working out of `/mnt/c` for agent data. - Keep project files inside the Linux filesystem. - If OpenClaw’s docs mention systemd or WSL config, enable them in `/etc/wsl.conf` and restart WSL.[cybernative](https://cybernative.ai/t/openclaw-on-windows-11-home-wsl2-a-beginner-setup-that-won-t-eat-your-machine/33992?tl=en)[youtube](https://www.youtube.com/watch?v=uqc_54fbsPU) ## Best-practice notes Your 64 GB RAM is more than enough for the platform and local models; the real issue is GPU routing and container reachability, not memory. Community reports on OpenClaw/NemoClaw-style setups on Windows show that WSL2 + Docker works best when the model and the agent runtime stay in the same WSL environment.[github](https://github.com/NVIDIA/NemoClaw/issues/208)[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI)cybernative+1 If you want, I can turn this into a **single copy-paste runbook** with placeholders like `<OPENCLAW_INSTALL_URL>` filled in once you share the exact OpenClaw project link. --- > turn this into a single copy-paste runbook with placeholders like <OPENCLAW_INSTALL_URL> filled in once you share the exact OpenClaw project link. Absolutely — here is a **single copy-paste Windows 11 WSL2 runbook** using the official OpenClaw install script URL and the current docs’ Linux/WSL2 flow. I’m assuming you want OpenClaw on **Windows 11 + WSL2 Ubuntu + Docker Desktop + NVIDIA GPU**.openclaw+2[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI) ## Copy-paste runbook ## 1) Windows PowerShell (Admin) ``` powershell wsl --install wsl --update wsl --set-default-version 2 wsl --install -d Ubuntu-24.04 wsl --status ``` ## 2) Open Ubuntu in WSL, then run ``` bash sudo apt update && sudo apt upgrade -y sudo apt install -y curl git ca-certificates gnupg lsb-release ``` ## 3) Verify GPU is visible in WSL ``` bash nvidia-smi ``` ## 4) Install Docker Desktop in Windows - Install Docker Desktop. - Enable **Use the WSL 2 based engine**. - Enable integration for Ubuntu. - Restart Docker Desktop if needed. ## 5) Install NVIDIA container tooling in WSL ``` bash curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg curl -s -L https://nvidia.github.io/libnvidia-container/stable/deb/nvidia-container-toolkit.list | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list sudo apt update sudo apt install -y nvidia-container-toolkit sudo nvidia-ctk runtime configure --runtime=docker ``` ## 6) Restart WSL ``` powershell wsl --shutdown ``` ## 7) Test Docker GPU passthrough ``` bash docker run --rm --gpus all nvidia/cuda:12.4.1-base-ubuntu22.04 nvidia-smi ``` ## 8) Install OpenClaw ``` bash curl -fsSL https://openclaw.ai/install.sh | bash ``` If you want to skip onboarding for a moment: ``` bash curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard ``` The official docs also list npm and GitHub install options if you prefer those paths.openclaw+1 ## 9) Start OpenClaw onboarding ``` bash openclaw onboard --install-daemon ``` ## 10) Verify OpenClaw ``` bash openclaw --version openclaw status ``` ## NVIDIA GPU notes The important part is that **GPU support comes from the Windows NVIDIA driver into WSL2, then into Docker through the NVIDIA container runtime**. If `nvidia-smi` works in WSL and the CUDA container test works, your GPU passthrough is good.[youtube](https://www.youtube.com/watch?v=CO43b6XWHNI)[forecr](https://www.forecr.io/blogs/installation/nvidia-docker-installation-for-ubuntu-in-wsl-2) ## If something fails - If `nvidia-smi` fails, update the Windows NVIDIA driver and reboot. - If Docker GPU tests fail, re-run the `nvidia-ctk runtime configure` step and restart WSL. - If OpenClaw install fails, try the npm path from the docs: ``` bash npm install -g openclaw@latest openclaw onboard --install-daemon ``` OpenClaw’s docs and GitHub also show a source install path if needed.github+1 If you want, I can turn this into a **one-screen checklist with checkboxes only**, or a **version that also includes Ollama in the same runbook**. --- > How do I configure OpenClaw agents after installation After installation, you usually configure OpenClaw agents in three layers: **global defaults, per-agent definitions, and verification**. The current docs describe `openclaw onboard`, `openclaw config set`, `openclaw agents add`, `openclaw agents list`, and `openclaw config get` as the main workflow.meta-intelligence+2 ## Checklist - Finish onboarding first: `openclaw onboard`.meta-intelligence+1 - Set your model provider keys in config or environment variables.getopenclaw+1 - Set global defaults for model, temperature, max tokens, workspace, and tools.getopenclaw+1 - Create one or more agents with `openclaw agents add`.meta-intelligence+1 - Give each agent a clear identity file and workspace scope.getopenclaw+1 - Verify your agents with `openclaw agents list --verbose` and `openclaw config get ...`.meta-intelligence+1 - Test a new agent with `--dry-run` before using it live.[meta-intelligence](https://www.meta-intelligence.tech/en/insight-openclaw-agents-guide) ## Set globals These commands set the default behavior that new agents inherit: ``` bash openclaw config set agents.defaults.model.primary claude-sonnet-4 --global openclaw config set agents.defaults.model.fallback gpt-4.1-mini --global openclaw config set agents.defaults.model.temperature 0.5 --global openclaw config set agents.defaults.model.maxTokens 64000 --global ``` The docs say OpenClaw stores config in `~/.openclaw/openclaw.json`, and global defaults live under `agents.defaults`.getopenclaw+1 ## Add agents Use `openclaw agents add` to register a new agent. The documented flags include `--model`, `--identity`, `--workspace`, `--tools`, `--template`, `--description`, `--max-tokens`, `--temperature`, and `--global`.meta-intelligence+1 Example: ``` bash openclaw agents add architect \ --model claude-opus-4-6 \ --workspace ~/projects/my-app \ --description "System architect — design decisions and technology selection" ``` You can repeat that pattern for developer, reviewer, or other specialized roles.ai2sql+1 ## Identity file Each agent can have an identity file, often `agent.md`, that defines its persona, rules, and behavior. The docs also mention `SOUL.md` as a customization option in some setups, but the basic idea is the same: keep the agent’s instructions close to the agent folder.getopenclaw+2 A typical structure looks like this: ``` ~/.openclaw/agents/ architect/ agent.md reviewer/ agent.md ``` ## Verify config Use these checks after creating agents: ``` bash openclaw agents list --verbose openclaw config get agents.registered.architect.model openclaw agents test architect --dry-run ``` The docs say `agents list` shows the registered agents, `config get` confirms the stored model settings, and dry-run testing helps catch bad config before launch.meta-intelligence+1 ## If you want a simple starting profile For a first setup, use one default generalist agent plus one specialist: - `default` for general tasks. - `coder` for implementation. - `reviewer` for quality control. That matches the multi-agent patterns described in the current OpenClaw guides.lumadock+2 ---