This demo showcases how to use offline GPT with the help of LM Studio within a .NET 8 Blazor app.

1. Download and install LM Studio.
2. Once LM Studio is installed, download a small model (Phi-3, Qwen, Gemma). If your machine has <8GB VRAM, choose a smaller one.
3. Go to the "Local Server" tab.
4. Select model and start the server.
5. Enable "Cross-Origin-Resource-Sharing (CORS)".
6. Disable "Verbose Server Logging".
7. (Optional) Under "Advanced Settings", set the GPU memory limit to max in "GPU Settings".
8. Stop the server and apply changes if required, then start the server.

1. Ensure you have the latest .NET 8 SDK installed. (Windows Winget: winget install dotnet-sdk-8)
2. Run the application in Visual Studio or use dotnet run in the src/HomeAutomationGpt folder.

From the repository root, run:

• Quickstart (Make) or .NET Aspire:

# Run with Make (client only)
make build
make run

# Or run with .NET Aspire (AppHost)
AI__Provider=Ollama AI__Model=gemma3:1b AI__Endpoint=http://localhost:11434/v1 dotnet run --project src/HomeAutomationGpt.AppHost/HomeAutomationGpt.AppHost.csproj

• macOS/Linux (Docker) or VS Code Dev Container (client only):

make image
make up
# open http://localhost:8080/
# or open in VS Code and use "Dev Containers: Reopen in Container"

• Windows (PowerShell/CMD):

docker build -t home-automation-gpt .\src\HomeAutomationGpt\
docker run -d -p 8080:80 --name home-automation-gpt home-automation-gpt
# open http://localhost:8080/

Open http://localhost:8080/ in your browser.

• Service version selector: Switch between V1–V5 in the UI to compare approaches. V5 is recommended.
• Advanced options: You can optionally edit the system prompt (hidden by default). Leave it off for the default prompts.
• Voice input: Click the “Speak” button to dictate commands (browser’s Web Speech API).
• Switch provider: ./scripts/switch-provider.sh ollama or ./scripts/switch-provider.sh lmstudio. Requires jq. Or use make provider-ollama / make provider-lmstudio.
• AI Trace: Toggle "Show AI trace" in the UI to see system/user prompts, model output, tool calls, and the queued actions.

• "Turn on the lights in the living room."
• "Turn off the lights in the kitchen."
• "Set the temperature to 23 degrees."
• "It's really cold" (Sets A/C to 23, turns on the TV and living room light).
• "Turn off all devices but keep the A/C on."
• "I need more light."
• "I'm feeling cold."