Skip to main content

​
Prerequisites

​
Runtime device

  • Windows 10 or Windows 11, x64.
  • A Vulkan-capable GPU with the matching driver. Integrated graphics also work but may be slower.
  • 8B: 8 GB or more of memory; 35B: 16 GB or more.
  • At least 30 GB of free disk space.

​
Build tools

  • Git and PowerShell 7 (the pwsh command).
  • Visual Studio 2022 with the C++ workload.
  • CMake 3.21+ and the Vulkan SDK.
  • Rust stable and Cargo.
  • Node.js LTS and npm.
  • Python 3.10+ with NumPy (used by the app when converting models locally).
Run the commands below in PowerShell. If a command is not recognized after installing a tool, reopen the terminal.
1

Get the source

Clone the full repository; do not download the windows/ folder alone.
2

Prepare Python

By default the app uses the python on your command line to convert models. Confirm Python is available and install NumPy:
If you want to use a different Python environment, set EDGE0_PY to that environment’s python.exe before starting the app.
3

Build the local engine

Run in the Edge0/windows directory:
The first time you run this script, it downloads and builds the engine source, which requires a network connection. The process takes about 10–20 minutes, depending on your computer and network.When it finishes, confirm the engine binary was produced:
A return value of True means the build succeeded. If it returns False, check the build output for errors first.
4

Build and start the desktop app

The build produces:If the version number differs, use the file name that was actually produced. Start the app:
The app automatically uses the engine and conversion tools built in the same repository.
The installer is unsignedThe current installer is unsigned. If a SmartScreen prompt appears, verify the source first, then follow Security & Installation Warnings. Installing and using Edge0 does not require turning off Defender or SmartScreen.
5

Download, convert, and load a model

  1. Open Models.
  2. Select 8B or 35B and click Download.
  3. Wait for the download to finish. The app converts the model on your machine automatically.
  4. When conversion finishes, click Enable.
  5. Once the model is ready, go to the chat page.
The first 8B conversion takes about 2–6 minutes. The model cannot be used until conversion finishes.Files are verified automatically during download, and an interrupted download can be resumed. Models are saved under %USERPROFILE%\.edge0 by default.
6

Complete your first chat

Enter a short message and press Enter to send. The response appears progressively.The local engine is started and managed by the app automatically; you do not need to start any separate service.

​
Environment check and logs

Open Environment Check on the Service page to view:
  • Whether the data directory is writable.
  • Available disk space and memory.
  • Whether the engine binary exists and responds.
  • The status of installed model files.
Logs are under:
  • engine-*.log: engine startup and runtime logs.
  • convert-*.log: model conversion logs.

​
If you move the repository or use another computer

By default the app looks for the engine and conversion scripts in the repository it was built from. The installer contains only the desktop app itself. If you move or delete that repository, or use the installer on another computer, first prepare the engine and conversion tools on that computer, then set the following environment variables before starting the app: For more issues, see Troubleshooting. Before sharing logs, remove any sensitive content.