SDK Overview

Multi-language SDKs for LLM observability - choose your language

Untrace SDKs

Untrace provides native SDKs for all major programming languages, each designed to integrate seamlessly with your existing codebase and framework of choice.

Supported Languages

JavaScript/TypeScript

Node.js, React, Next.js, Express
Zero-latency instrumentation with decorators and auto-instrumentation

Python

FastAPI, Django, Flask
Async support with context managers and framework integration

Go

Gin, Echo, Fiber
High-performance tracing with context propagation

Rust

Axum, Actix, Tokio
Zero-cost abstractions with async/await support

C#/.NET

ASP.NET Core, Console apps
Built-in DI support and activity-based tracing

Elixir

Phoenix, LiveView, OTP
Actor model integration with supervision trees

Quick Comparison

Language	Package Manager	Key Features	Best For
JavaScript/TypeScript	npm install @untrace/sdk	Decorators, Auto-instrumentation	Web apps, APIs, Full-stack
Python	pip install untrace-sdk	Async/await, Context managers	ML/AI apps, APIs, Data processing
Go	go get github.com/untrace-dev/untrace-sdk-go	Context propagation, Performance	Microservices, APIs, CLI tools
Rust	cargo add untrace-sdk	Zero-cost abstractions, Memory safety	High-performance systems, WebAssembly
C#/.NET	dotnet add package Untrace.Sdk	DI integration, Activities	Enterprise apps, Windows services
Elixir	{:untrace_sdk, "~> 0.1.2"}	OTP integration, Telemetry	Real-time apps, Distributed systems

Common Features

All SDKs provide:

• Zero-latency instrumentation - Minimal performance overhead
• OpenTelemetry compliance - Industry-standard observability
• Automatic provider detection - Works with all major LLM providers
• Token usage tracking - Automatic cost calculation
• Error handling - Comprehensive error capture and reporting
• Custom attributes - Add your own metadata and context
• Sampling support - Control data volume and costs
• Batch processing - Efficient data transmission

Getting Started

1. Choose Your Language

Select the SDK that matches your tech stack:

• Web Development: JavaScript/TypeScript, Python, C#/.NET
• Microservices: Go, Rust, Elixir
• AI/ML: Python, JavaScript/TypeScript
• High Performance: Rust, Go
• Enterprise: C#/.NET, Java (coming soon)

2. Install the SDK

Each language has its own installation method:

JavaScript/TypeScript

npm install @untrace/sdk

3. Initialize and Start Tracing

JavaScript/TypeScript

import { init } from '@untrace/sdk';

const untrace = init({
  apiKey: 'your-api-key',
  serviceName: 'my-app'
});

Framework Integration

Web Frameworks

Framework	JavaScript	Python	Go	Rust	C#	Elixir
React/Next.js	✅	❌	❌	❌	❌	❌
Express	✅	❌	❌	❌	❌	❌
FastAPI	❌	✅	❌	❌	❌	❌
Django	❌	✅	❌	❌	❌	❌
Flask	❌	✅	❌	❌	❌	❌
Gin	❌	❌	✅	❌	❌	❌
Echo	❌	❌	✅	❌	❌	❌
Fiber	❌	❌	✅	❌	❌	❌
Axum	❌	❌	❌	✅	❌	❌
Actix	❌	❌	❌	✅	❌	❌
ASP.NET Core	❌	❌	❌	❌	✅	❌
Phoenix	❌	❌	❌	❌	❌	✅

AI/ML Frameworks

Framework	JavaScript	Python	Go	Rust	C#	Elixir
LangChain	✅	✅	❌	❌	❌	❌
LlamaIndex	✅	✅	❌	❌	❌	❌
OpenAI	✅	✅	✅	✅	✅	✅
Anthropic	✅	✅	✅	✅	✅	✅
Hugging Face	✅	✅	❌	❌	❌	❌

Migration Between Languages

If you’re switching between languages or have a polyglot codebase, Untrace makes it easy to maintain consistent observability:

Consistent Data Model

All SDKs use the same trace data structure:

{
  "event_type": "llm_call",
  "data": {
    "model": "gpt-4",
    "prompt": "Hello world",
    "response": "Hello! How can I help?",
    "tokens_used": 25
  },
  "metadata": {
    "user_id": "user123",
    "session_id": "session456"
  }
}

Cross-Language Correlation

Traces from different languages are automatically correlated using:

• Trace IDs - Unique identifiers for each trace
• Parent-Child relationships - Link related operations
• Workflow IDs - Group related traces across services
• User/Session context - Maintain user journey across languages

Best Practices

1. Choose the Right SDK

• JavaScript/TypeScript: Web applications, full-stack development
• Python: AI/ML applications, data processing, APIs
• Go: Microservices, high-performance APIs, CLI tools
• Rust: High-performance systems, WebAssembly, embedded
• C#/.NET: Enterprise applications, Windows services
• Elixir: Real-time applications, distributed systems

2. Consistent Naming

Use consistent naming conventions across all your services:

// JavaScript
const trace = await untrace.trace("llm_call", {
  model: "gpt-4",
  prompt: message
});

# Python
trace = await client.trace("llm_call", {
    "model": "gpt-4",
    "prompt": message
})

3. Shared Context

Pass context between services using standard headers:

// JavaScript - Set headers
const response = await fetch('/api/process', {
  headers: {
    'X-Trace-ID': traceId,
    'X-User-ID': userId
  }
});

// Go - Read headers
traceID := r.Header.Get("X-Trace-ID")
userID := r.Header.Get("X-User-ID")

Next Steps

Choose Your SDK

Start with the main SDK documentation

Language-Specific Guides

Deep dive into your chosen language

Framework Examples

See real-world implementations

Dashboard Setup

Configure your observability dashboard