Run Edge0-35B in Edge0's desktop app on Mac or Windows.
Requirements
| Requirement | Mac | Windows |
|---|---|---|
| Hardware | Apple Silicon Mac, M3 or later | x64 processor and Vulkan-compatible GPU |
| OS | macOS 14 or later | Windows 10 or 11 |
| RAM guidance | No minimum specified in the macOS app README | 16 GB or more for 35B |
| Model download | Approximately 20 GB | Approximately 20 GB |
Both platforms need additional space for development tools and build files. Windows also retains a converted model. The full setup footprint has not been measured here.
Run Edge0-35B on Mac
This follows the macOS app's build instructions. You will build the app in Terminal, then download the model and chat in its interface.
1. Install the tools
| Tool | Requirement |
|---|---|
| Xcode Command Line Tools | Must include a working Metal compiler; also provides Git and Make |
| CMake | Version 3.24 or later; choose the macOS universal package and make cmake available in Terminal |
| Rust | Install through rustup; the repository selects Rust 1.88.0 automatically |
| Node.js | Version 22 or later; choose an LTS macOS ARM64 installer |
| Yarn Classic | Version 1.22; the setup check configures it through Corepack when available |
If Command Line Tools are missing, open Terminal and install them:
xcode-select --install
After installing the tools, reopen Terminal so it picks up their paths.
2. Download Edge0 and check the setup
Download the repository:
git clone https://github.com/Edge0-AI/Edge0.git "$HOME/Edge0"
Enter the macOS directory:
cd "$HOME/Edge0/macos"
Run the prerequisite check:
bash scripts/check-prereqs.sh
Continue when it prints check-prereqs: ok. If it reports that Yarn or Corepack is missing, install the required Yarn version and rerun the check:
npm install --global yarn@1.22.22
3. Build and open the app
make app
This installs the app's dependencies, builds the native engine, and packages the app in macos/dist. Open the current version:
open dist/edge0-0.1.0-arm64.app
If the version changes, use the filename produced by the build. Opening the app also starts its local service.
4. Download the model and chat
- If the welcome screen appears, click Skip for now. Open Models in the sidebar and find edge0-35b.
- Click Download and wait for the download to finish.
- Click Load model and wait for the model to show as loaded.
- Click New chat, type Explain SSD streaming in one sentence., and press Enter.
A generated response confirms that setup worked. Models and app data are stored under ~/.edge0.
Run it again later
In Finder, open the app inside Edge0/macos/dist. Load the downloaded 35B model if needed, then start a chat.
Run Edge0-35B on Windows
The Windows app currently requires building from source. Keep the repository and engine build in place afterward; the app uses them to run the engine and convert models.
1. Install the tools
| Tool | Requirement |
|---|---|
| Git for Windows | x64 Setup, available from PowerShell |
| PowerShell | Stable version 7, Windows x64 MSI |
| Visual Studio 2022 | Community installer in the 17.14 row; select Desktop development with C++ |
| CMake | Stable Windows x64 Installer, version 3.21 or later; add to PATH |
| Vulkan SDK | Windows SDK |
| Rust | Stable 1.88 or later, Windows x64 MSVC toolchain and Cargo |
| Node.js | LTS, Windows x64 installer; includes npm and npx |
| Python | Version 3.10 or later, Windows 64-bit installer; add python.exe to PATH |
| WebView2 | Normally included with Windows 10/11; use the Evergreen Runtime Bootstrapper if missing |
Open PowerShell 7 and check that the command-line tools are available:
Get-Command git, pwsh, cmake, cargo, node, npm.cmd, npx.cmd, python
All eight commands should appear. If one is missing, install it and reopen PowerShell. Confirm Python's version:
python --version
Install NumPy for model conversion:
python -m pip install numpy
Check that it imports:
python -c "import numpy; print(numpy.__version__)"
The engine build also needs a Git name and email for its patches. Check:
git -C "$env:USERPROFILE" var GIT_COMMITTER_IDENT
If it reports an unknown identity, follow Git's identity setup before building.
2. Build the engine
Download the repository:
git clone https://github.com/Edge0-AI/Edge0.git "$env:USERPROFILE\Edge0"
Enter the Windows directory:
Set-Location -LiteralPath "$env:USERPROFILE\Edge0\windows"
Build the engine:
pwsh -File scripts/vendor-build.ps1
After a successful build, check the executable:
Test-Path -LiteralPath '..\wt\win\build-vk\bin\Release\llama-server.exe'
It should return True. If it returns False, resolve the build error before continuing.
3. Build and open the app
Enter the app directory:
Set-Location -LiteralPath "$env:USERPROFILE\Edge0\windows\app"
Install its dependencies:
npm.cmd install
Build the app with Tauri 2, Edge0's Windows desktop framework:
npx.cmd tauri build
After a successful build, check the executable:
Test-Path -LiteralPath '.\src-tauri\target\release\edge0-app.exe'
It should return True. Open the app:
.\src-tauri\target\release\edge0-app.exe
4. Download the model and chat
- Open Models and find 35B.
- Click Download. Wait for both the download and automatic conversion to finish.
- Click Enable and wait for the model to show as resident.
- Click New chat and select 35B if needed.
- Type Explain SSD streaming in one sentence. Press Enter or click Send.
A generated response confirms that setup worked. Conversion is needed because the Windows engine uses GGUF model files.
Run it again later
Open PowerShell 7 and run:
& "$env:USERPROFILE\Edge0\windows\app\src-tauri\target\release\edge0-app.exe"
Enable the existing model in Models if needed, then start a chat.
Troubleshooting
| Problem | Fix |
|---|---|
| The Edge0 folder already exists | If it is your earlier checkout, skip cloning and enter its macos or windows directory. Otherwise, choose an unused path and use it throughout the commands. |
| Mac: prerequisite check fails | Install the tool named in the error, reopen Terminal, and rerun bash scripts/check-prereqs.sh from Edge0/macos. |
| Mac: Metal compiler missing | Check with xcrun --find metal. Install Xcode or Command Line Tools that provide the Metal compiler, then rerun the prerequisite check. |
| Mac: the app opens but chat is unavailable | Open Service and check that the local service is running. In Models, confirm that 35B has finished downloading and is loaded. |
| Windows: command not found | Install the missing tool and reopen PowerShell 7. |
| Windows: unknown Git identity | Complete the Git identity setup linked above, then retry the engine build. |
| Windows: conversion fails or the model will not enable | Open Service and inspect the environment checks. Find engine and converter logs at %USERPROFILE%\.edge0\logs. |
| Not enough disk space | Free space on the model drive. On Windows, both the checkpoint and converted files remain on disk. |
FAQ
Can I use Edge0-35B offline?
Yes. Complete installation, model download, and Windows conversion first. Edge0 then runs locally using the files on your computer.
Do I need an NVIDIA GPU?
No. Windows supports compatible AMD, NVIDIA, and Intel GPUs through Vulkan. The Mac setup uses Apple Silicon's GPU.
Can I use these Mac instructions on an Intel Mac?
No. The macOS app's current instructions require Apple Silicon, M3 or later.
Is Edge0-35B free to use?
Yes. The model is released under Apache 2.0. This local setup does not require a paid cloud API.
Stay connected
Follow us on X (Twitter) or join our Discord for updates and setup help.