Skip to content

1.3. Installation Guide

Backend.AI GO is available for all major desktop operating systems and iOS. Follow the instructions for your specific platform below.

macOS (Apple Silicon Only)

Backend.AI GO is designed specifically for Apple Silicon (M1/M2/M3/M4/M5 chips). Intel-based Macs are not supported.

  1. Download the .dmg file (e.g., backend-ai-go-0.9.0-macos-arm64.dmg).
  2. Double-click the .dmg to mount it.
  3. Drag the Backend.AI GO icon into your Applications folder.
  4. Launch the app from your Applications folder or via Spotlight.

Security Note

On first launch, you may need to right-click the app and select "Open" or go to System Settings > Privacy & Security to allow the application if it's not yet notarized in your version.

Windows

Backend.AI GO supports Windows 10 and 11 (64-bit).

  1. Download the .exe (NSIS) package.
  2. Run the installer and follow the on-screen instructions.
  3. The application will be installed to your Program Files and a shortcut will be created on your Desktop.
  4. Launch Backend.AI GO.

GPU Support

For best performance on Windows, ensure you have the latest NVIDIA drivers installed if you have an NVIDIA GPU.

Linux

We provide packages for both x64 and ARM64 architectures.

Debian/Ubuntu (.deb)

For x64 (Intel/AMD):

  1. Download the .deb package (e.g., backend-ai-go-x.x.x-linux-x64.deb).
  2. Install it via terminal:
    sudo dpkg -i backend-ai-go-x.x.x-linux-x64.deb
    sudo apt-get install -f  # To fix any missing dependencies
    

For ARM64 (Raspberry Pi 5, NVIDIA Jetson, etc.):

  1. Download the ARM64 .deb package (e.g., backend-ai-go-x.x.x-linux-arm64.deb).
  2. Install it via terminal:
    sudo dpkg -i backend-ai-go-x.x.x-linux-arm64.deb
    sudo apt-get install -f  # To fix any missing dependencies
    

Flatpak (x64 only)

Flatpak provides a sandboxed environment that works across most Linux distributions.

  1. Download the .flatpak bundle.
  2. Install it using the following command:
    flatpak install backend-ai-go-x.x.x-linux-x64.flatpak
    
  3. Run the application:
    flatpak run ai.backend.go
    

One-line install (CLI and server)

The packages above install the desktop app. On a headless box, a container, or a CI runner, install the aigo CLI and the headless aigo-server instead:

curl -fsSL https://go.backend.ai/install.sh | sh
irm https://go.backend.ai/install.ps1 | iex

Both binaries go into ~/.aigo/bin and that directory is added to your PATH through a single managed block in your shell startup file. Every archive is verified against the release's SHA256SUMS.txt before anything is written, and there is no flag to skip that check. Re-running the command upgrades in place.

The Windows script installs the CLI only: there is no Windows build of aigo-server, and the desktop app has its own signed installer.

Passing options

A piped script receives its arguments after sh -s --:

curl -fsSL https://go.backend.ai/install.sh | sh -s -- --cli-only -y
Flag Effect
-y, --yes Skip the confirmation prompt
--cli-only Install only the aigo CLI
--system System-wide install with a systemd service (requires root)
--version X.Y.Z Pin an exact release
--channel beta Install the newest prerelease instead of the latest stable
--prefix PATH Install prefix; binaries land in PATH/bin
--no-modify-path Do not touch shell startup files
--dry-run Print every action and change nothing
--uninstall Remove the binaries, the PATH block and the service

iex cannot forward parameters, so the PowerShell script takes them through a script block:

& ([scriptblock]::Create((irm https://go.backend.ai/install.ps1))) -NoPathUpdate

Its parameters are -Version, -Channel, -InstallDir, -NoPathUpdate, -DryRun and -Uninstall.

Server install with systemd

curl -fsSL https://go.backend.ai/install.sh | sudo sh -s -- --system

Installs both binaries into /usr/local/bin, creates the aigo service account, prepares /etc/aigo, /var/lib/aigo, /var/log/aigo and /var/run/aigo, installs the systemd unit, preset and logrotate config, generates /etc/aigo/config.toml if it is missing, and enables the service without starting it. The systemd unit, preset, logrotate config and environment template are checksum-verified against a manifest published alongside them before any of them is written into /etc, the same mandatory, no-skip-path check the release binaries get. Both binaries are also confirmed to actually run before the shell profile or the systemd unit is touched, so a binary that cannot execute on this system aborts cleanly instead of leaving a half-finished install behind it. Start the service when you are ready:

sudo systemctl start aigo-server
sudo systemctl status aigo-server

The installer refuses to run as root in any other mode, and it never calls sudo on your behalf.

Mirrors and air-gapped installs

AIGO_INSTALL_BASE_URL replaces the download origin, in the same shape as the public release URLs (<base>/<tag>/<asset>):

curl -fsSL https://go.backend.ai/install.sh | \
  AIGO_INSTALL_BASE_URL=https://mirror.internal/aigo sh -s -- --version 1.11.1

A mirror has no "latest" redirect to follow, so --version is required when the variable is set. Checksums are still verified against the SHA256SUMS.txt served by the mirror. AIGO_INSTALL_SUPPORT_URL does the same for the systemd files that --system installs.

Uninstall

curl -fsSL https://go.backend.ai/install.sh | sh -s -- --uninstall
curl -fsSL https://go.backend.ai/install.sh | sudo sh -s -- --system --uninstall

This removes the binaries, the managed PATH block, the generated env files and, for --system, the systemd unit, preset and logrotate config. It leaves /etc/aigo, /var/lib/aigo and /var/log/aigo alone and prints the command to remove them.

Supported platforms

Platform Result
macOS 15+ on Apple Silicon Installs aigo and aigo-server
macOS under Rosetta Installs the arm64 build and says so
Intel macOS Aborts: Apple Silicon is required
Linux x86_64 / aarch64, glibc 2.39+ Installs aigo and aigo-server
Linux below glibc 2.39 Aborts naming the detected and required versions
Alpine and other musl systems Aborts: the binaries are glibc builds
Windows x86_64 Installs the CLI through install.ps1

glibc 2.39 is the floor

The Linux binaries are built on Ubuntu 24.04, so they need glibc 2.39 or newer. RHEL 9 ships 2.34, Ubuntu 22.04 ships 2.35 and Debian 12 ships 2.36, all below the floor. Use the container image on those systems.

Required system packages

aigo-server links against shared libraries the release archive does not ship. A desktop distribution usually has them already; a minimal container image, a cloud server image or a CI runner usually does not. Install them before running the server:

sudo apt-get install -y libdrm2 libdrm-amdgpu1 libssl3
sudo dnf install -y libdrm openssl-libs

These are the same package sets the installer names when a binary fails to start.

Library Package (Debian / Ubuntu) Package (RHEL / Fedora) Why
libdrm.so.2, libdrm_amdgpu.so.1 libdrm2, libdrm-amdgpu1 libdrm GPU monitoring links AMD's userspace DRM libraries on every Linux build, including hosts with no AMD GPU
libssl.so.3, libcrypto.so.3 libssl3 openssl-libs x86_64 builds up to and including 1.11.1 only. Later builds use a pure-Rust TLS stack and no longer need these

The OpenSSL packages are harmless to install on any version, so the commands above are safe to run as-is.

The aigo CLI needs none of this. It links only the C runtime, so a CLI-only install (--cli-only) has no package prerequisites beyond glibc 2.39.

The installer checks this for you

The installer runs each binary right after installing it, before it touches the shell profile or, under --system, the systemd unit. If one cannot start it prints the loader's own message, names the packages above and exits non-zero instead of reporting success.

iOS

Coming Soon

The iOS version of Backend.AI GO is currently under development and is not yet available on the App Store. Stay tuned for updates on the release.

Backend.AI GO is available on the App Store for iPhone and iPad.

  1. Open the App Store on your iOS device.
  2. Search for "Backend.AI GO".
  3. Tap Get to download and install.
  4. Launch the app and configure your cloud provider API keys.

iOS Limitations

The iOS version connects to cloud AI providers only. Local model inference is not available on iOS due to platform limitations.

Supported iOS Versions: iOS 18.0 or later

Supported Devices: iPhone and iPad with iOS 18.0+

Requirements

Desktop

Component Minimum Recommended
RAM 8 GB 16 GB or more
Storage 10 GB free space 50 GB+ (for multiple models)
GPU (Optional) 4 GB VRAM 8 GB+ VRAM (NVIDIA/AMD/Metal)

iOS

Component Minimum
iOS Version 18.0
Storage 100 MB
Network Required for cloud providers

Need help? Check our Troubleshooting page.