Skip to content
Docs

Desktop Quickstart

A native application with a tray icon and no container runtime. The console, API and model registry are the same as the container’s.

From github.com/IterateAI/lifeboat-releases:

Desktop builds are cut per platform, so the newest version differs between them — and the newest build for yours is often not on the latest release. Take the newest file bearing your platform’s name from all releases.

Platform File
Windows 10 (build 17763) / 11, x64 Lifeboat-<version>-setup.exe
macOS, Apple Silicon (M1–M4), 13+ Lifeboat-<version>-macos-arm64.dmg
macOS, Intel, 13+ Lifeboat-<version>-macos-x86_64.dmg
Linux, Debian/Ubuntu x64 lifeboat-desktop_<version>_amd64.deb
Linux, Debian/Ubuntu arm64 lifeboat-desktop_<version>_arm64.deb
Linux, any distro, x64 / arm64 Lifeboat-<version>-linux-x86_64.tar.gz / -aarch64.tar.gz

Each release also carries SHA256SUMS.

Windows — download Lifeboat-<version>-setup.exe and double-click it. It installs for you rather than for the machine, so there is no administrator prompt: the app goes to %LOCALAPPDATA%\Programs\Lifeboat and your models to %LOCALAPPDATA%\Lifeboat. Start Lifeboat when I sign in and Create a desktop shortcut are separate checkboxes — neither implies the other. The installer and both executables are Authenticode-signed.

The tray needs the .NET 8 Desktop Runtime, which the installer checks for but cannot install; without it the app installs and then appears to do nothing when launched:

Terminal window
winget install --id Microsoft.DotNet.DesktopRuntime.8 -e

lifeboat-core.exe runs headless and does not need it.

macOS — open the .dmg and drag Lifeboat to the Applications shortcut in the window, then double-click it. Both builds are signed with an Apple Developer ID, notarized by Apple and stapled, so there is no security warning and no right-click dance, and it works offline because the ticket travels with the file.

Linux, Debian or Ubuntu:

Terminal window
sudo apt install ./lifeboat-desktop_<version>_amd64.deb

Use apt install ./file.deb rather than dpkg -i: the tray binding and the Vulkan loader are recommended packages, and dpkg does not pull them in.

Linux, anything else — the tarball unpacks to the same layout, rooted at /:

Terminal window
sudo tar -C / -xzf Lifeboat-<version>-linux-x86_64.tar.gz

Either way you get /opt/lifeboat and two commands on PATH: lifeboat-core (the server and CLI) and lifeboat-tray.

Linux needs glibc 2.31 or newer (Debian 11+, Ubuntu 20.04+, RHEL 9+), python3 >= 3.10 and gir1.2-gtk-3.0. Install gir1.2-ayatanaappindicator3-0.1 for the tray icon on GNOME — without it the icon silently does not render. lifeboat-core runs headless with none of it.

The tray menu has Open Console, Show Log, Check for Updates… and Quit. There is no Dock icon and no window by design.

The console is at http://127.0.0.1:30800 — note this differs from the container’s 8001. On first run it asks you to create an account.

Terminal window
shasum -a 256 -c SHA256SUMS --ignore-missing # macOS
sha256sum -c SHA256SUMS --ignore-missing # Linux

macOS and Windows builds are code-signed, so the operating system checks them for you. Linux has no equivalent; check the sum yourself, and take SHA256SUMS from the release page over HTTPS rather than from wherever the binary came from.

Vulkan on Windows and Linux, Metal on Apple Silicon. Vulkan is one binary that offloads to NVIDIA, AMD and Intel alike, which is why there is no separate CUDA download.

Platform GPU path Covers
Windows Vulkan NVIDIA, AMD, Intel (including integrated)
Linux Vulkan NVIDIA, AMD, Intel (including integrated)
macOS, Apple Silicon Metal the built-in GPU
macOS, Intel Metal, if the GPU supports it the built-in or discrete GPU; falls back to the CPU

The backend is loaded at runtime, so a machine with no Vulkan driver or no capable GPU simply uses the CPU — nothing fails, it is just slower. On Linux install libvulkan1 plus your vendor’s driver (mesa-vulkan-drivers covers AMD and Intel; NVIDIA’s own driver includes it) if lifeboat-core doctor reports no GPU and you expect one.

NPUs and Intel XPU are not used. An “AI PC” NPU is idle under Lifeboat: there is no inference path to it, and it shares system memory, so it would not lift the bandwidth ceiling that governs decode speed anyway.

The GGUF engine on every platform, and MLX on Apple Silicon, which is the fast path there. The tensor engine is not included, so the GPU optimization layer and speculative decoding do not apply — see Choose your deployment.

The model catalogue is filtered to formats your machine can serve, in preference order, and a model it could never load is refused before the download starts. See Model formats.

Under the user’s application-support directory — ~/Library/Application Support/Lifeboat on macOS, %LOCALAPPDATA%\Lifeboat on Windows, ~/.local/share/lifeboat on Linux. Registry, licence, audit log and logs live there; back it up if the deployment matters.