Tracenet

🌟 Overview

Tracenet is a language-agnostic tracing middleware designed specifically for AI/ML applications. It provides automatic instrumentation for popular AI frameworks while offering powerful manual instrumentation capabilities when needed.

Why Tracenet?

Modern LLM applications face several critical challenges:

• Complexity: LLM interactions are complex, involving multiple steps, retries, and chain-of-thought processes
• Observability Gap: Traditional APM tools don't understand LLM-specific concepts like prompt engineering, token usage, or completion quality
• Integration Overhead: Manually instrumenting each LLM interaction is time-consuming and error-prone

Tracenet solves these challenges by providing:

• Zero-Config Auto-Instrumentation: Automatically captures LLM interactions, tokens, latency, and costs
• AI-Native Design: Purpose-built for LLM applications with deep understanding of AI patterns
• Universal Integration: Works with any LLM framework or provider while maintaining consistent observability

Unlike general-purpose tracing tools, Tracenet is specifically designed for LLM applications, offering:

• Native understanding of LLM concepts (prompts, completions, tokens)
• Automatic framework detection for popular LLM libraries
• Built-in support for common AI patterns and architectures

Key Benefits

• 🚀 Zero-Config Setup: Just import and go - automatic framework detection and configuration
• 🔄 Language Agnostic: First-class support for both Python and TypeScript
• 🎯 AI-First Design: Built specifically for tracing AI/ML applications
• 📊 Rich Observability: Detailed tracing for both automatic and manual instrumentation
• 🔌 Extensible: Plugin architecture for custom tracing backends

Architecture

The following diagram illustrates Tracenet's architecture and integration points:

graph TD
    A[Your Application] --> B[Tracenet Middleware]
    B --> C{Framework Detection}
    
    C -->|Auto-Detect| D[Native Integrations]
    D --> D1[OpenAI SDK]
    D --> D2[LangChain]
    D --> D3[CrewAI]
    D --> D4[Google ADK]
    D --> D5[Other Frameworks...]
    
    C -->|Manual| E[Manual Instrumentation]
    E --> E1[Decorators]
    E --> E2[Context Managers]
    E --> E3[Direct API]
    
    B --> F{Tracing Backend}
    F -->|Default| G[Langfuse]
    F -->|Extensible| H[Custom Backends]
    
    style A fill:#f9f,stroke:#333,stroke-width:2px
    style B fill:#bbf,stroke:#333,stroke-width:2px
    style C fill:#dfd,stroke:#333,stroke-width:2px
    style F fill:#dfd,stroke:#333,stroke-width:2px
    style G fill:#fdd,stroke:#333,stroke-width:2px
    style H fill:#fdd,stroke:#333,stroke-width:2px

🚀 Quick Start

Python

# Just import the package - it automatically sets up tracing
import tracenet

# Your existing code will now be traced automatically!
from openai import OpenAI
client = OpenAI()

response = client.chat.completions.create(
    model="gpt-3.5-turbo",
    messages=[{"role": "user", "content": "Hello!"}]
)
# The API call is automatically traced!

TypeScript

// Import the package
import { tracenet } from '@stackgen-ai/tracenet';

// Your existing code will now be traced automatically!
import OpenAI from 'openai';
const client = new OpenAI();

const response = await client.chat.completions.create({
    model: "gpt-3.5-turbo",
    messages: [{ role: "user", content: "Hello!" }]
});
// The API call is automatically traced!

📦 Installation

Python

pip install tracenet

TypeScript/JavaScript

npm install @stackgen-ai/tracenet
# or
yarn add @stackgen-ai/tracenet

✨ Features

Framework Support

Manual Instrumentation

Both Python and TypeScript support:

• Function/Method Tracing
• Context Managers/Spans
• LLM Generation Tracking
• Custom Attributes
• Error Handling
• Async Operations

⚙️ Configuration

Environment Variables

Langfuse Backend Configuration

📚 API Reference

Python API

Automatic Tracing

import tracenet  # Automatically sets up tracing

Manual Instrumentation

from tracenet import trace, start_span, start_generation

# Function decorator
@trace(name="my_function")
def my_function(arg1, arg2):
    return arg1 + arg2

# Context manager
with start_span("operation_name", tags=["tag1"]) as span:
    result = operation()
    span.update(output=result)

# LLM Generation tracking
with start_generation("text_gen", model="gpt-4") as span:
    response = llm.generate("prompt")
    span.update(output=response)

TypeScript API

Automatic Tracing

import { tracenet } from '@stackgen-ai/tracenet';  // Automatically sets up tracing

Manual Instrumentation

import { trace, startSpan, startGeneration } from '@stackgen-ai/tracenet';

// Function decorator
@trace({ name: "myFunction" })
myFunction(arg1: string, arg2: string): string {
    return arg1 + arg2;
}

// Context manager
const span = await startSpan("operationName", { tags: ["tag1"] });
try {
    const result = await operation();
    span.update({ output: result });
} finally {
    await span.end();
}

// LLM Generation tracking
const genSpan = await startGeneration("textGen", { model: "gpt-4" });
try {
    const response = await llm.generate("prompt");
    genSpan.update({ output: response });
} finally {
    await genSpan.end();
}

🔍 Tracing Flow

The following diagram shows how Tracenet handles different types of traces:

sequenceDiagram
    participant App as Your Application
    participant TN as Tracenet
    participant Backend as Tracing Backend
    
    Note over App,Backend: Automatic Framework Detection
    App->>TN: Import tracenet
    TN->>TN: Detect frameworks
    TN->>TN: Configure integrations
    
    Note over App,Backend: Manual Instrumentation
    App->>TN: @trace decorator
    TN->>Backend: Start span
    App->>TN: Execute function
    TN->>Backend: Update span
    TN->>Backend: End span
    
    Note over App,Backend: Context Managers
    App->>TN: start_span()
    TN->>Backend: Create span
    App->>TN: Operation execution
    App->>TN: span.update()
    TN->>Backend: Update span data
    TN->>Backend: Close span

📖 Examples

For detailed examples, check out our example repositories:

• Python Examples
• TypeScript Examples

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

Development Setup

1. Clone the repository:

git clone https://github.com/stackgenhq/tracenet
cd tracenet

2. Install dependencies:

# Python
pip install -e ".[dev]"

# TypeScript
npm install

3. Run tests:

# Python
pytest

# TypeScript
npm test

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Langfuse for the amazing tracing backend
• OpenTelemetry for inspiration on observability patterns
• All our contributors