infertrace 0.1.0

Developer-first ML & LLM observability SDK. OpenTelemetry for Machine Learning.

🔭 InferTrace

Developer-first ML & LLM observability SDK. OpenTelemetry for Machine Learning.

Why InferTrace?

The modern AI stack is complex enough. Existing observability tools often require deploying heavy infrastructure, managing Docker containers, setting up Kubernetes clusters, or paying exorbitant SaaS fees.

InferTrace is different. We believe ML observability should be as simple as a pip install. It works immediately, locally, and effortlessly. No containers to spin up, no infrastructure to manage—just actionable insights in seconds.

Quick Install

pip install infertrace

With optional provider integrations:

pip install "infertrace[openai]"     # OpenAI integration
pip install "infertrace[anthropic]"  # Anthropic integration
pip install "infertrace[ollama]"     # Ollama integration

5-Minute Quick Start

1. Manual Tracking (3 lines)

from infertrace import Monitor

# Initialize monitor
monitor = Monitor()

# Track any ML operation
monitor.track_request(
    model="custom-model-v1",
    prompt="Hello",
    response="World",
    latency_ms=150
)

2. Auto-Instrumentation (3 lines)

from infertrace import Monitor
import openai

# Initialize monitor and patch OpenAI client
monitor = Monitor()
monitor.instrument_openai()

# All OpenAI calls are now automatically tracked!
client = openai.Client()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Tell me a joke."}]
)

3. Function Decorator

from infertrace import Monitor

monitor = Monitor()

@monitor.track(model="text-classifier")
def classify_text(text: str) -> str:
    # Your ML code here
    return "positive"

4. Launch Dashboard

Visualize your data instantly with the built-in local dashboard:

infertrace dashboard

Key Features

• 🔭 Auto-Instrumentation: Drop-in monitoring for major LLM SDKs with zero code changes to your application logic.
• 📡 Provider-Agnostic: A unified event model that standardizes metrics across OpenAI, Anthropic, local models, and custom APIs.
• 📊 Beautiful Dashboard: A sleek, local-first web UI to visualize requests, latencies, tokens, and drift in real-time.
• ⚡ Zero Config: SQLite storage backend by default means it works out of the box. No database setup required.
• 🔌 Plugin System: Easily extensible architecture to write custom provider adapters or storage backends.
• 📈 Analytics & Metrics: Track cost, latency, token usage, and error rates effortlessly.
• 📋 Reports: Generate sharable markdown or HTML reports of your system's health.
• 🛠️ CLI: Powerful command-line interface to manage data, generate reports, and launch the dashboard.
• 🎯 DataDriftGuard Integration: Ready to integrate with advanced drift detection algorithms out of the box.
• 🔒 Local-First: Your data never leaves your machine unless you explicitly configure a remote backend. Privacy by design.

CLI Commands

# Initialize configuration
infertrace init

# Launch local Streamlit dashboard
infertrace dashboard --port 8501

# Run DataDriftGuard data drift analysis
infertrace drift --baseline baseline.csv --incoming incoming.csv --save

# Generate telemetry reports
infertrace report --format html

# Export telemetry data
infertrace export --format csv

# System health & dependency check
infertrace doctor

Author

Suryanandan Babbar
Email: [e-mail]
GitHub: https://github.com/suryanandanbabbar/infertrace

License

This project is licensed under the MIT License.