ElBruno.FoundryLocalMonitor 0.5.30

ElBruno.FoundryLocalMonitor

Windows systray monitor for Foundry Local — detects model load/unload events from any app using Foundry Local, regardless of language or SDK.

⚠️ Windows only. This tool uses WPF and the Windows Notification Area. It cannot run on macOS or Linux.

Package

Package	NuGet	Downloads	Description
ElBruno.FoundryLocalMonitor			Windows systray monitor for Foundry Local

Features

• 🔔 Toast notifications when models load/unload (Windows 10/11 Action Center)
• 📊 Compact mini status window (always-on-top)
• 🖥️ Full dashboard: Status, Loaded Models, and Available Models tabs
• 📁 Click any process path to open its folder in Explorer
• ⚙️ Systray icon with context menu
• 🔍 Automatic multi-app discovery — detects Foundry Local running in C#, Python, Node.js, Aspire, or any other context without manual configuration
• 🔕 Smart notifications — suppresses noisy SDK-internal events by default; only alerts for meaningful model changes

UI Overview

Status tab

Shows all discovered Foundry Local instances on the machine — each endpoint as a card with its URL, process name, port, PID, and process path. Click 📂 to open the process folder in Explorer.

Loaded Models tab

Groups discovered instances by OS process (PID). Each card shows:

• Process name, PID, and type badge (sdk proxy or daemon)
• All ports this process listens on
• Process executable path with 📂 folder shortcut
• Every model currently loaded, with a device badge and the source port

Device badge colours:

The right-hand column shows the source port (e.g. :55588) so you can tell which endpoint serves each model when a proxy listens on multiple ports.

Available Models tab

Lists all models downloaded to your local Foundry cache and ready to load into memory.

How Discovery Works

The monitor does not rely on the foundry CLI alone. It runs a parallel HTTP port scan across all active localhost listeners and identifies any endpoint serving the Foundry OpenAI-compatible API:

All 127.0.0.1 listeners  →  parallel GET /v1/models  →  group by PID  →  per-process cards
      (kernel call, ~1ms)       (800ms timeout each)      (merge ports)

This means the monitor detects:

• Models loaded via foundry model load (CLI / daemon path)
• Models loaded via the C# FoundryLocalManager SDK (port 55588 by default)
• Models loaded via the Python foundry-local-sdk (port 55589 by default)
• Models exposed through .NET Aspire proxy ports
• Any other app on any dynamic port that serves /v1/models

→ Full details: docs/discovery.md

Install

dotnet tool install -g ElBruno.FoundryLocalMonitor
foundrylocalmon

Requirements

• Windows 10/11
• .NET 10 runtime
• Foundry Local installed

Samples

Both samples run together in the multi-client E2E test to verify the monitor detects events from two simultaneous apps.

E2E Testing

# Single-client test (C# only)
cd tests/e2e
.\Run-E2ETest.ps1 -LaunchMonitor

# Multi-client test (C# + Python simultaneously)
.\Run-E2EMultiClient.ps1 -LaunchMonitor

See tests/e2e/README.md for full test documentation.

Build from source

dotnet restore
dotnet build
dotnet test

Publishing

GitHub Release or manual workflow_dispatch triggers publish. The workflow requires the release environment and the NUGET_USER secret, and uses OIDC Trusted Publishing (NUGET_API_KEY is not used).

License

MIT

👋 About the Author

Made with ❤️ by Bruno Capuano.

🙏 Acknowledgments

• Foundry Local — runtime foundation
• Hardcodet.NotifyIcon.Wpf — tray icon support