Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →To run NemoClaw on a Linux VPS, provision at least 4 vCPUs, 8 GB of RAM and 20 GB of free disk space; 4 or more vCPUs, 16 GB of RAM and 40 GB of free space is a safer production starting point. Install Node.js 22.19 or newer with npm 10 or newer, install and start Docker, obtain an NVIDIA API key, run the current NVIDIA installer, and finish the interactive onboarding wizard. The current Node minimum replaced older 22.16 and Node 20 guidance when NemoClaw 0.0.82 was released on July 14, 2026.
Antihost offers Linux VPS plans with full root access, unlimited traffic on a 1 Gbps port and a 14-day free trial.
What you need before installing NemoClaw
NemoClaw runs several components during setup, including Docker workloads, k3s and the OpenShell gateway. The practical constraint is memory: an undersized VPS can be killed by the operating system while those services start together.
As an Amazon Associate I earn from qualifying purchases.
Free tools Windows power users keep installed
One-click scans. No signup required.
| Resource | Minimum stated by NVIDIA | Recommended starting point | Why it matters |
|---|---|---|---|
| CPU | 4 vCPUs | 4 or more vCPUs | Concurrent containers and gateway processes need parallel CPU capacity. |
| Memory | 8 GB RAM | 16 GB RAM | 8 GB can run out during provisioning when Docker, k3s and OpenShell are active. |
| Disk | 20 GB free | 40 GB free | Images, package caches, logs and model-related files consume space over time. |
These figures are requirements in NVIDIA’s prerequisites documentation; the retrieved page does not state a publication year. Treat them as a floor, not a performance guarantee. If your provider offers only 8 GB, enable swap as a last resort, but provisioning will be slower and swap does not replace physical RAM.
Choose a supported operating-system image
Use a current 64-bit Ubuntu VPS. Third-party setup instructions commonly use Ubuntu 22.04 or 24.04, but NVIDIA’s complete official operating-system list was not available in the material for this guide. If your image is different, confirm support in the current NemoClaw prerequisites page before committing the server.
Check the VPS immediately after connecting
Run these checks over SSH before installing anything:
nproc
free -h
df -h /
nproc should report at least 4, available memory should be 8 GB or more, and the filesystem should have at least 20 GB free. Check the same values after other services are stopped; a provider’s advertised disk size is not the same as free space on the root filesystem.
Decide between Docker and rootless Podman
Docker is NemoClaw’s default Linux runtime. The installer can install Docker, start its service and add your account to the docker group. Podman is an explicit alternative selected with NEMOCLAW_GATEWAY_RUNTIME=podman; it does not require the Docker CLI or daemon.
| Choice | Use it when | Setup implications |
|---|---|---|
| Docker (default) | Your team already uses Docker or you want the path covered by the default installer flow. | Install and start Docker, then verify your login can run it without an error. |
| Rootless Podman | Your security policy avoids a Docker daemon, or Podman is already standard on the VPS. | Install and configure Podman first, then set NEMOCLAW_GATEWAY_RUNTIME=podman for the installer. |
Do not switch runtimes merely to avoid checking prerequisites. Pick the runtime your operations team can monitor, patch and recover.
Install Node.js 22.19 or newer
NemoClaw currently requires Node.js 22.19 or later and npm 10 or later. The minimum took effect in version 0.0.82, according to the July 14, 2026 release notes. Older guides that say Node 20 or Node 22.16 are stale for a current installation.
Install with nvm
- Install or activate nvm using the current instructions for your Linux distribution.
- Install the Node 22 line and make it the active version:
nvm install 22 nvm use 22 - Confirm both versions before running NemoClaw:
node --version npm --version
The Node command must print 22.19 or a later 22.x release (or a newer major release accepted by the current NemoClaw installer), and npm must print 10 or later. If the installer rejects Node, install a newer release through nvm and rerun it rather than forcing an older dependency.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Make the Node installation visible to the installer
nvm and fnm modify shell startup files. A non-interactive SSH session or an already-open terminal may not have the updated PATH. If node works but the NemoClaw command later cannot be found, reload the appropriate startup file:
source ~/.bashrc
# or, for zsh:
source ~/.zshrc
Opening a new SSH session has the same effect. Keep the Node version manager initialization in the account that will run NemoClaw; installing Node for one user and running the installer as another commonly produces PATH and permission confusion.
Install and test Docker
The NemoClaw installer can handle Docker on Linux, but preparing Docker yourself makes failures easier to diagnose and lets you test permissions before onboarding. On Ubuntu, install the distribution package or follow Docker’s current official installation instructions, then start the service:
sudo apt update
sudo apt install -y docker.io
sudo systemctl enable --now docker
Add the account that will run NemoClaw to Docker’s group and refresh group membership:
Recommended Free Tools
sudo usermod -aG docker "$USER"
newgrp docker
docker run --rm hello-world
The hello-world container should complete without a permission error. If you prefer not to preinstall Docker, allow the NemoClaw installer to do it; Docker setup may require elevated privileges even though the NemoClaw installer itself does not require root. If the installer says it cannot restore Docker access, run newgrp docker and retry.
Podman preparation
For a Podman deployment, install Podman using your distribution’s package manager, verify that your account can run podman, and export the runtime selection before launching the installer:
export NEMOCLAW_GATEWAY_RUNTIME=podman
Do not set this variable when you want the default Docker path. Mixing a Podman selection with Docker-only troubleshooting commands can make a healthy installation appear broken.
Obtain the NVIDIA credential
Default cloud inference requires an NVIDIA API key. Create one through build.nvidia.com; keys begin with nvapi-. Copy it into a password manager or another secure secret store. Do not paste it into shell history, commit it to a repository, or place it in a public cloud-init file.
Cloud inference does not require a GPU in the VPS according to the available setup guidance. A GPU becomes relevant when you deliberately choose local inference, which has substantially different capacity, driver and model-storage requirements.
Run the official NemoClaw installer
NVIDIA’s installer command and flags can change with the fast-moving release line, so copy the current Linux command from NVIDIA’s NemoClaw installation documentation rather than using a command copied from an older blog post. Run it as the normal account that will operate NemoClaw, not as a root shell.
Interactive installation
- Make sure
node --version,npm --versionanddocker run --rm hello-worldsucceed in the same terminal. - Paste the current official installer command and allow it to install or configure Docker if prompted.
- When the installer asks for Docker-group access, follow its instruction. If it asks you to run
newgrp docker, do so, then rerun the installer from the new shell. - Keep the terminal open until the installer hands control to onboarding or reports completion.
Scripted or image-build installation
Set both required variables before invoking the official installer:
export NEMOCLAW_NON_INTERACTIVE=1
export NEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1
Then run the current installer command supplied by NVIDIA. The first variable suppresses interactive prompts; the second records acceptance of the third-party software terms needed by the automated path. Review the command’s output and logs in your image-build system, because a successful shell exit alone does not prove that onboarding or the gateway started.
Complete the onboarding wizard
Onboarding is interactive unless you deliberately used the non-interactive variables. Follow this order:
- Select OpenClaw as the agent. This is the agent choice presented by the current wizard.
- Select an inference provider and model. Choose the provider that matches the credential and latency requirements of your application.
- Enter the provider credential. Paste the NVIDIA key when using the default cloud provider. Avoid echoing it in diagnostic commands.
- Skip optional web search and messaging integrations on the first run. You can add them later after the core gateway is working.
- Accept the suggested network-policy tier. Start with the wizard’s suggested tier instead of broadening access during initial troubleshooting.
If your SSH connection drops or you close the wizard, resume rather than starting a second installation:
nemoclaw onboard --resume
When onboarding completes, record which provider, model and policy tier you selected. Those choices explain later connectivity or capability differences and make a rebuild reproducible.
Verify the installation
First make sure the shell can resolve the command:
command -v nemoclaw
nemoclaw --help
If command -v returns nothing immediately after an nvm or fnm installation, reload ~/.bashrc or ~/.zshrc, or open a fresh terminal. Do not reinstall solely because the current shell has an old PATH.
Use the CLI reference that ships with your NemoClaw release for the exact version-check command; that command was not confirmed in the available documentation excerpt. A useful verification pass should establish that the CLI is on PATH, onboarding reports completion, the selected runtime is reachable, and the gateway can start without an authentication error.
Basic operational checks
- Run
docker pswhen using Docker and confirm the expected NemoClaw-related containers remain up. - Run the corresponding
podman pscheck when using Podman. - Check
free -hwhile the gateway is active; sustained near-zero available memory indicates that the VPS is too small. - Review the installer and gateway logs for provider authentication failures before changing network policies.
Troubleshoot the common failures
“Node version is unsupported”
Install a current Node 22 release through nvm, activate it with nvm use 22, verify npm 10 or newer, and rerun the installer. Do not rely on a system Node binary that appears earlier in PATH.
“nemoclaw: command not found”
This is usually a shell PATH issue after nvm or fnm changes. Run source ~/.bashrc or source ~/.zshrc, then open a new terminal if necessary. Confirm with command -v nemoclaw before attempting another install.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsDocker permission denied
Check that the service is running with sudo systemctl status docker, then check group membership with groups. If you just added yourself to the Docker group, run newgrp docker or reconnect over SSH. The error normally comes from Docker setup or group propagation, not from the NemoClaw installer logic itself.
Processes are killed or the gateway disappears
Check memory first. During setup, Docker, k3s and OpenShell can run at the same time; an 8 GB machine has little headroom. Move to a 16 GB VPS where possible. Swap may prevent an immediate out-of-memory kill, but it makes provisioning and runtime workloads slower.
Onboarding stopped halfway
Reconnect to the server, verify the runtime and Node PATH, and run nemoclaw onboard --resume. Starting a second independent onboarding flow can leave credentials and policy choices inconsistent.
Cloud inference authentication fails
Confirm that the key came from build.nvidia.com, begins with nvapi-, and was entered for the same provider selected in onboarding. Replace a revoked or mistyped key rather than weakening network-policy settings.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteA dashboard or HTTPS endpoint is unreachable
First establish that the core CLI and gateway work locally on the VPS. A third-party guide suggests opening ports 80 and 443 in the cloud firewall before installing Caddy, but that advice is not confirmed by the official material available here. Treat those ports as an exposure decision: verify the current NemoClaw and VPS-provider documentation, restrict source addresses where possible, and do not publish an unauthenticated dashboard just to test connectivity.
Cloud inference or local inference?
| Inference mode | VPS hardware implication | Best fit | Main trade-off |
|---|---|---|---|
| Cloud (default) | Works on a modern VPS without a local GPU. | Fastest setup and modest server sizes. | Requires an NVIDIA API key and depends on network access and provider availability. |
| Local | Requires a suitable GPU, drivers, VRAM and additional disk capacity; exact NemoClaw GPU requirements are not stated in the available material. | Teams that need local execution or have an existing GPU host. | More hardware, driver and model-management work. |
Choose local inference only after confirming the model’s hardware requirements for the specific release. A CPU-only VPS is not a substitute for a GPU when a local model requires one.
Production hardening after the first successful run
- Use a dedicated non-root account for NemoClaw and keep SSH administration separate.
- Allow SSH only from trusted addresses where your provider’s firewall supports source restrictions.
- Keep the API key in a secret manager or protected environment configuration, never in a public repository or shared terminal transcript.
- Apply Ubuntu, Node.js, Docker or Podman and NemoClaw updates on a maintenance schedule. Check NVIDIA release notes before upgrading because the Node minimum and installer behavior are volatile.
- Monitor RAM, disk growth and container restarts. The recommended 40 GB free-disk target leaves room for image layers and logs but does not eliminate the need for cleanup.
- Back up configuration and document the selected runtime, provider, model and network-policy tier so the VPS can be rebuilt without guessing.
When to resize the VPS
Resize before failures become data-loss events. Move from the minimum tier to 16 GB RAM when onboarding repeatedly triggers out-of-memory kills, when available memory stays low under normal gateway traffic, or when you add optional integrations. Increase disk beyond 40 GB when container images, logs or local model files consume the reserve. Add CPU only when sustained load, rather than memory pressure, is the bottleneck.
Quick Recap
A repeatable installation checklist
- Ubuntu VPS selected and supported by the current NVIDIA documentation.
- At least 4 vCPUs, 8 GB RAM and 20 GB free disk verified; 16 GB RAM and 40 GB free disk preferred.
- Node.js 22.19 or newer and npm 10 or newer active in the installer shell.
- Docker installed and tested, or Podman selected explicitly with
NEMOCLAW_GATEWAY_RUNTIME=podman. - NVIDIA API key obtained from build.nvidia.com and stored securely.
- Current NVIDIA installer command copied from its documentation.
NEMOCLAW_NON_INTERACTIVE=1andNEMOCLAW_ACCEPT_THIRD_PARTY_SOFTWARE=1set only for scripted installs.- OpenClaw, provider, model, credential and suggested network-policy tier completed in onboarding.
nemoclaw onboard --resumeavailable if the first onboarding session is interrupted.- CLI PATH, runtime containers, memory and logs checked after installation.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
