AiSphere Desktop is a versatile, privacy-focused local AI application manager. It empowers users to discover, install, and manage a wide array of AI tools—including Large Language Models (LLMs), autonomous agents, and datasets—directly on their own hardware.
By leveraging local runtimes (containers, singularity or natives/conda), AiSphere ensures that your data and models never leave your machine, providing a secure and high-performance environment for modern AI workflows.
Below are some highlights of the AiSphere Desktop interface:
A unified view of your local AI ecosystem, showing active applications and system health.
Discover new AIs to install locally, categorized by type (LLM, Agent, Tool) with detailed hardware requirement checks.
Manage running instances with built-in terminals and one-click access to local web services.
Automated checks and installation guides for required runtimes like Podman and Apptainer.
- Local Execution: Run powerful AI models like Ollama, DeepSeek-R1, and custom agents without cloud dependencies.
- Multi-Runtime Support: Native integration with Podman and Apptainer (Singularity) to handle various container formats and isolation requirements.
- Integrated Workspace:
- Interactive Shell: Directly attach to running containers.
- Real-time Logs: Monitor installation and execution processes with a resizable terminal.
- Internal Web View: Access web-based AI interfaces (like Open WebUI) directly inside the desktop app.
- MCP tool calling: Use compatible local models with attached MCP servers. The built-in filesystem and image tools are restricted to a folder selected separately for each workspace.
- Hardware Monitoring: Real-time insights into your system's CPU, GPU (NVIDIA/Apple Silicon/Linux), Memory, and Storage availability.
- Global App Store: An "Explore" section to discover new AI tools tailored for your specific hardware architecture.
- App management: Select multiple catalog apps to install, or select installed apps to start, stop, or uninstall them in batches. Each installed instance gets a unique ID, and a catalog entry can only be installed once.
- Runtime setup: When an Explore app needs a runtime that is not installed, AiSphere can install it and continue the app installation. Configure Settings → Runtime auto-install to ask every time (default), install automatically, or never install automatically.
More are on the way...vLLM & SGLang...
To begin your local AI journey, ensure you have a supported container runtime installed via the Runtimes tab, then head over to Explore to install your first AI application.
On Windows, AiSphere prepares Docker's WSL 2 prerequisites before installing Docker Desktop. Approve the Windows elevation prompt when asked. If Windows reports that a restart is required, restart your computer and retry the Docker installation. Hardware virtualization must be enabled in BIOS/UEFI.
Install AiSphere LAN NAS from Explore, then open it in My Apps and choose Manage share. Select the folder to share and start the service. AiSphere shows an address for each available private IPv4 network; open one of those addresses in a browser on another device to browse, download, and upload files.
Password authentication is enabled by default. Each time the service starts, AiSphere generates a new password; use the username aisphere when the browser prompts for credentials. Authentication can be disabled in the share settings, but then any device on the local network can read and upload files in the selected folder. The service uses HTTP without encryption, so use it only on a trusted network. The server stops when AiSphere quits.
For a Docker-based alternative, install File Browser NAS from Explore. Its files, settings, and database are persisted in its AiSphere app data folder; the web UI is available at http://localhost:8093 after starting the app. Docker must be installed and running.
Open Runtimes → Docker → Docker Hub registry mirrors to add your own pull-through mirrors. Each mirror can be enabled or disabled, and the list order is the order AiSphere tries. When an app needs a public Docker Hub image, AiSphere pulls through the enabled mirrors in sequence and stops at the first success; if every mirror fails, it pulls directly from Docker Hub. Images hosted on other registries (for example ghcr.io) are always pulled directly. This setting only affects images pulled by AiSphere and does not change Docker Desktop's global configuration.
AiSphere can serve its desktop interface to other devices on your network. In
resources/config/config.json, set networkAccess.enabled to true and choose
an available TCP port between 1024 and 65535, then restart AiSphere. The default
configuration keeps network access disabled and uses port 3210.
Open Settings in the desktop app to find the LAN address and copy the authentication token. Enter the token in the browser sign-in page on the other device. Anyone with this token can control AiSphere and run apps on the host; use this only on a trusted local network and keep the token private. The browser server uses HTTP, not HTTPS. Your operating-system firewall may need to allow AiSphere on the selected port and private networks.
The repository ships self-contained scripts for Windows 11. They bootstrap a portable
Node.js and Yarn into .tools/ when no suitable system Node.js is available, so the
only prerequisite is Windows PowerShell 5.1+ (included with Windows 11) and an internet
connection for the first run.
From the repository root, double-click the .cmd files or run them from a terminal:
| Command | What it does |
|---|---|
scripts\windows\setup.cmd |
Downloads Node.js/Yarn if needed, installs dependencies, and prepares the icons. Run this first. |
scripts\windows\dev.cmd |
Starts Vite and launches Electron in development mode (hot reload). |
scripts\windows\build.cmd |
Builds the renderer/main process and packages a Windows installer. |
scripts\windows\build.cmd fast-debug-build |
Quickly updates the unpacked app while preserving its local database, apps, downloads, logs, and installed runtimes. |
Build options (PowerShell):
# Build the NSIS installer for x64 (default)
.\scripts\windows\build.ps1 -Arch x64
# Build an unpacked application folder instead (faster, no installer)
.\scripts\windows\build.ps1 -Kind dir
# Refresh the unpacked app for debugging without deleting its local data.
# Close AiSphere first, then run it from release\win-unpacked\aisphere.exe.
.\scripts\windows\build.cmd fast-debug-build
# Build for ARM64
.\scripts\windows\build.ps1 -Arch arm64
# Reinstall dependencies and/or skip the TypeScript check
.\scripts\windows\setup.ps1 -ForceInstall
.\scripts\windows\build.ps1 -SkipTypeCheckOutputs are written to release/:
release\aisphere-<version>-win-<arch>.exe– the installer.release\win-<arch>-unpacked\aisphere.exe– the unpacked app, runnable without installing.
Notes:
- Dependencies are installed with
--ignore-scripts; the scripts then run only the installers that are actually needed (Electron download, node-pty, esbuild). This keeps the build working without Visual Studio C++ build tools or Python. - Closing the main window hides AiSphere to the system tray instead of quitting. Click the tray icon (or launch the app again) to reopen the window; use Quit AiSphere in the tray menu to exit completely.
- By default, quitting AiSphere stops its running apps and managed Docker/Podman engines. Configure Settings → When quitting AiSphere to ask each time, stop automatically, or leave them running. System-installed engines are never stopped.
.tools/is ignored by git and can be deleted at any time to force a clean bootstrap.- Portable-tool versions can be overridden with the
AISPHERE_NODE_VERSIONenvironment variable (defaultv24.22.2).
- zegang.luo@qq.com
- You are welcomed to join AiSphere to make everyone enjoy AI
AiSphere Desktop - Your Versatile Local AI