zborek.LangfuseDotnet 0.2.1

WORK IN PROGRESS

LangfuseDotnet

<a href="https://www.nuget.org/packages/zborek.LangfuseDotnet/"> <img src="https://img.shields.io/nuget/v/zborek.LangfuseDotnet" alt="NuGet"> </a>

A .NET client library for Langfuse - an open-source observability and analytics platform for LLM applications. This library enables you to track, monitor, and analyze your AI application performance and behavior.

Features

• Create and manage traces, spans, events, and generations
• Automatic background ingestion of observability data
• Integration with dependency injection
• Support for both .NET 8.0 and 9.0
• Customizable configuration options

Installation

Install the package via NuGet:

dotnet add package zborek.LangfuseDotnet

Getting Started

Configuration

Add Langfuse configuration to your appsettings.json:

{
  "Langfuse": {
    "PublicKey": "YOUR_PUBLIC_KEY",
    "SecretKey": "YOUR_SECRET_KEY",
    "SendLogs": true
  }
}

Registration with DI

// In Program.cs or Startup.cs
builder.Services.AddLangfuse(builder.Configuration);

Usage in Your Application

Inject LangfuseTrace into your service:

public class MyService
{
    private readonly LangfuseTrace _langfuseTrace;

    public MyService(LangfuseTrace langfuseTrace)
    {
        _langfuseTrace = langfuseTrace;
    }
    
    public async Task ProcessAsync(string input)
    {
        _langfuseTrace.Trace.Body.Input = input;
        
        // Create spans to track operations
        using var dataSpan = _langfuseTrace.CreateSpan("GetData");
        var data = await GetDataAsync(input);
        dataSpan.SetOutput(data);
        
        // Track LLM generations
        using var generationSpan = _langfuseTrace.CreateGeneration("LLMProcessing", 
            input: $"Process: {input} with {data}", 
            output: null);
        
        var result = await CallLLMAsync(input, data);
        generationSpan.SetOutput(result);
        
        // Record events
        using var _ = _langfuseTrace.CreateEvent("ProcessingComplete", 
            input: input, 
            output: result);
        
        // Send data to Langfuse
        await _langfuseTrace.IngestAsync();
        
        return result;
    }
}

Core Concepts

• Trace: A complete workflow or user session
• Span: A logical operation or step within a trace
• Generation: A specific LLM call or generation
• Event: A discrete occurrence within your application

Complete Example

See the Examples/Langfuse.Example.WebApi directory for a complete working example showing integration with OpenAI.

API Endpoints in Example Project

• POST /chat: Processes a chat request using direct service implementation
• POST /chatDi: Processes a chat request using dependency injection

Troubleshooting

• Missing ingestion data: Ensure SendLogs is set to true in your configuration
• Authentication errors: Verify your Langfuse API keys
• Data not appearing: Check if IngestAsync() is called after operations

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

License

This project is licensed under the MIT License.

Additional Resources

• Langfuse Documentation
• GitHub Repository