Getting Started

RunForge Desktop is a Windows-native application for creating, monitoring, and inspecting ML training runs. Everything runs locally — no cloud, no telemetry, no accounts.

System requirements

Requirement	Value
OS	Windows 10 (1809+) or Windows 11
Architecture	x64
Runtime	.NET 10 (bundled in MSIX)
Python	3.10+ (for training)
GPU	Optional (CUDA for GPU training)
Disk Space	~100 MB

Install from MSIX release (recommended)

The simplest way to get started:

1. Download the latest .msix file from Releases
2. Double-click to install — Windows handles everything
3. Launch from the Start Menu

The MSIX package is self-contained: it bundles the .NET runtime so you don’t need to install it separately. Installation and uninstallation are clean, isolated, and reversible.

Build from source

If you prefer to build from source, you’ll need the .NET 10 SDK and Visual Studio 2022 (17.12+) with the MAUI workload:

git clone https://github.com/mcp-tool-shop-org/runforge-desktop
cd runforge-desktop
dotnet run --project src/RunForgeDesktop/RunForgeDesktop.csproj

See docs/INSTALL.md for detailed installation options including release builds and MSIX packaging from source.

Your first training run

Once RunForge Desktop is running:

1. Select Workspace — click “Select Workspace” and choose a folder for your ML experiments. This is where all run artifacts will be stored.
2. Start Training — click “Train” to open the training configuration dialog. Choose an epoch preset (Quick, Standard, Extended, or Custom), select GPU or CPU, and adjust advanced settings if needed.
3. Monitor Live — watch training progress with real-time loss charts and streaming logs. You can cancel at any time.
4. Browse Runs — once training completes, browse all runs with filtering by status (Pending, Running, Completed, Failed, Cancelled).
5. Inspect Results — click any run to view its metrics, logs, and artifacts. Open the output folder or copy the training command for reproducibility.

All artifacts are written to the .ml/runs/ directory inside your workspace. Each run gets its own timestamped folder containing run.json (manifest), metrics.jsonl (live metrics), stdout.log, and stderr.log.

Python and PyTorch

RunForge Desktop spawns Python processes to execute training. The bundled runner script uses PyTorch to train a simple CNN on synthetic data.

Minimum requirements:

• Python 3.10 or later (installed from python.org or the Microsoft Store)
• PyTorch (pip install torch) — required for actual training runs

RunForge auto-discovers Python on your system. If it finds the wrong version or you use a virtual environment, go to Settings and set a manual Python path override.

If Python is not detected, the app will show a clear error message explaining what to install and where to get it.