Skip to main content

Install Apollia Desktop on Linux

The application is Apollia Desktop. Its installers and the files they put on disk still carry the name of the engine underneath, Apollia OS, which is the name you will read below.

Every file below is attached to each release on the publication page, https://github.com/Apollia-OS/apollia-os/releases.

  • .AppImage (recommended, x86_64): portable desktop application, no installation required.
  • .deb (x86_64): Debian/Ubuntu package (sudo apt install ./Apollia-OS_<version>_amd64.deb).
  • apollia-os-linux-x86-cpu.tar.gz and apollia-os-linux-x86-vulkan.tar.gz: the two command-line bundles for x86_64, one per engine, CPU or Vulkan.
  • apollia-os-linux-arm-cpu.tar.gz: the command-line bundle for aarch64. No desktop installer is published for that architecture.

Requirements​

  • glibc 2.39+ distribution (Ubuntu 24.04, Debian 13, Fedora 40, and so on): the released binaries are built on Ubuntu 24.04 without static linking.
  • 4 GB of free RAM minimum.
  • For GPU inference: a Vulkan 1.3+ driver, Mesa for AMD and Intel, the NVIDIA 550+ driver for NVIDIA.

Installation (AppImage)​

chmod +x Apollia-OS_<version>_amd64.AppImage
./Apollia-OS_<version>_amd64.AppImage

The app starts the daemon in the background. The daemon serves local LLM inference through the embedded llama-server engine and launches the speech-to-text (STT) runner.

One thing to know before you pick this format: an AppImage is a read-only mount that exists only while the app runs. The command line lives inside that mount, and the /usr/local/bin link offered by System > Hardware in the settings, with the developer mode on, points into it, so the link dangles as soon as you quit. For a command line that survives, install the .deb or unpack one of the apollia-os-linux-* archives.

Installation (.deb)​

sudo apt install ./Apollia-OS_<version>_amd64.deb

Then launch the application from your desktop application menu, under the entry Apollia OS: the app starts the daemon itself. The apollia-os command line ships inside the package (/usr/lib/Apollia OS/) but is not on your PATH until you enable it in the app: turn the developer mode on in Settings > Developer, then click Install CLI in System > Hardware, which creates the /usr/local/bin link. The verification commands below assume that link exists.

Verification​

apollia-os --version
apollia-os doctor

doctor checks the data directory, the configuration file, the two databases, the models directory, Python, the sandbox posture and the runtime socket. It does not detect your GPU. The command that reports the detected hardware is a separate one, and it needs the daemon running:

apollia-os model hardware --json

It answers with the total RAM, the CPU and the detected accelerator. Ask for --json: the plain-text rendering of that command prints nothing today.

Closing the app​

The window close button quits Apollia on Linux, and the daemon stops with it, along with the llama-server engine and the apollia-runner-* runner. The tray icon carries three entries, open the window, show the pending approvals, and quit; none of them hides the window while keeping the runtime up, so minimise the window rather than close it when you want Apollia to stay resident. macOS is the exception: there, closing a window leaves the app running behind the menu-bar icon.

GPU acceleration​

Two accelerations, and Linux answers them differently.

Local LLM inference already runs on the GPU. The desktop bundles, AppImage and .deb alike, embed a Vulkan build of the llama-server engine, and the apollia-os-linux-x86-vulkan.tar.gz command-line bundle the Vulkan one. Vulkan drives NVIDIA, AMD and Intel cards alike. Nothing to install and nothing to set: with a working driver the engine uses the card. The -cpu bundle is the one to take on a machine with no graphics driver.

An NVIDIA card can go further. There is one installer per system, and no separate one for NVIDIA cards. When the application detects such a card, its AI screen offers the accelerated engine, Speed up with your NVIDIA card: one click downloads it, after its signature is verified, and the next model load uses it. On Linux that engine relies on the CUDA libraries of the machine.

Dictation runs on the processor in the desktop packages. Speech to text runs in the apollia-runner sidecar, which is built on whisper. The desktop packages ship a single one, apollia-runner-cpu; the Vulkan command-line archive carries apollia-runner-vulkan, whisper on the card, which the daemon uses when the driver provides a Vulkan loader.

Update​

See Update Apollia.

Uninstall​

sudo apt remove apollia-os    # .deb package
# or simply delete the AppImage
rm -rf ~/.apollia              # user data