> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/ai-workbench/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/ai-workbench/_mcp/server.

> Detailed instructions for manually installing NVIDIA AI Workbench

# Manually Install on Ubuntu (Remote)

<Note title="Downloading or installing AI Workbench implies consent to the NVIDIA AI Product Agreement.">

By downloading or installing the NVIDIA AI Workbench software, you agree to the terms of the [NVIDIA AI Product Agreement](https://www.nvidia.com/en-us/data-center/products/nvidia-ai-enterprise/eula/) (EULA).
If you do not agree to the terms of the EULA, you are not authorized to download, install, or use NVIDIA AI Workbench.

</Note>

## Overview

**Remote install supports Ubuntu 22.04, 24.04, and 26.04, as well as NVIDIA Jetson Linux.**

Installation on Ubuntu is straightforward, but Jetson Linux requires a few more steps to address how SSH and Docker work.

**Interactively install AI Workbench on a remote system or use NVIDIA Sync to bootstrap it for you.**

Interactive installation involves opening an SSH terminal to the system and running a few commands.
This lets you choose the container runtime and set a few other options.

Using [NVIDIA Sync](https://docs.nvidia.com/dgx/dgx-spark/nvidia-sync.html) automates the remote install and
adds the location to your Desktop App.
It's more streamlined but doesn't have the optionality of the interactive install.

See how to [use NVIDIA Sync here](/ai-workbench/how-to/locations/use-nvidia-sync).

## Key Concepts

**Interactive Install:**

Running commands interactively in an SSH terminal using the CLI TUI, which is a
text based interface that runs in a terminal and provides
features like drop downs and selections through selector keys.

**Non-Interactive Install:**

Running a command manually or in a script with the `--noninteractive` flag.
Lets you preset options.

**AI Workbench Application Folder:**

The destination folder on the remote machine for binaries and program files.
The name is fixed as *.nvwb*, and the default location is in the home directory, *~/.nvwb*.

## Remote Interactive Install

**Prerequisites: Verify the following prior to installation.**

- A remote Ubuntu 22.04, 24.04, or 26.04 system
- SSH access as a non-root user with *sudo* privileges
- `500 MB` of disk space for the application
- `40 GB` of disk space for project containers

**Step One: SSH into the remote machine as non-root user.**

For password based access run:

```bash
ssh <non-root-username>@<remote-host>
```

For key based access run:

```bash
ssh -i <local-path-to-private-key> <non-root-username>@<remote-host>
```

**Step Two: Copy and paste the installation command in the terminal.**

```bash
mkdir -p $HOME/.nvwb/bin && \
curl -L https://workbench.download.nvidia.com/stable/workbench-cli/$(curl -L -s https://workbench.download.nvidia.com/stable/workbench-cli/LATEST)/nvwb-cli-$(uname)-$(uname -m) --output $HOME/.nvwb/bin/nvwb-cli && \
chmod +x $HOME/.nvwb/bin/nvwb-cli && \
sudo -E $HOME/.nvwb/bin/nvwb-cli install
```

to change the application folder path, replace `$HOME` with the desired path in the above commands.

**Step Three: Follow the prompts in the TUI using the arrow keys and Enter to confirm choices.**

1. If prompted, select to install the NVIDIA GPU drivers (will require a reboot)
2. Choose between Docker or Podman as the container runtime

When you approve driver installation, AI Workbench selects a driver that uses the open or proprietary NVIDIA kernel modules, depending on the installed GPU.
AI Workbench does not replace or change an existing driver.
For driver-selection limitations, see [AI Workbench Support Matrix](/ai-workbench/reference/support-matrix).

**Success: The TUI tells you that AI Workbench is installed.**

### Non-Interactive Install

**Prerequisites: Verify the following before the non-interactive install.**

- Same system requirements as the interactive install
- Passwordless *sudo* configured on the remote machine

**Step One: SSH into the remote as a non-root user and curl the AI Workbench CLI.**

We skip explaining the SSH step here but the curl and mod command is below.

```bash
mkdir -p $HOME/.nvwb/bin && \
curl -L https://workbench.download.nvidia.com/stable/workbench-cli/$(curl -L -s https://workbench.download.nvidia.com/stable/workbench-cli/LATEST)/nvwb-cli-$(uname)-$(uname -m) --output $HOME/.nvwb/bin/nvwb-cli && \
chmod +x $HOME/.nvwb/bin/nvwb-cli
```

**Step Two: Run the non-interactive installer with required flags.**

See the command for an example configuration below.

```bash
sudo $HOME/.nvwb/bin/nvwb-cli install \
--noninteractive \
--accept \
--podman \
--drivers \
--uid 1000 \
--gid 1000
```

The arguments in this command assume that machine has an NVIDIA GPU with no drivers installed.
It also selects Podman as the runtime.

When you use `--drivers`, AI Workbench selects a driver that uses the open or proprietary NVIDIA kernel modules, depending on the installed GPU.
AI Workbench does not replace or change an existing driver.
For driver-selection limitations, see [AI Workbench Support Matrix](/ai-workbench/reference/support-matrix).

The non-interactive install will automatically reboot the remote machine if you elect to have the drivers installed.

**Success: AI Workbench is installed on the remote system.**