SDK overview

OpenTelemetry-native AI observability for Python, TypeScript, and Go — with zero code changes required.

The OpenLIT SDK is an open-source, OpenTelemetry-native observability library that instruments your AI applications automatically. You add two lines of code (or zero, using the CLI), and OpenLIT begins capturing distributed traces, token usage, latency, cost, and more across every LLM call, agent workflow, and vector database operation.

How it works

OpenLIT patches supported libraries at import time. When your code calls OpenAI, LangChain, Pinecone, or any other supported integration, OpenLIT automatically creates spans with rich AI-specific attributes — no manual instrumentation required.

You can also run with zero code changes using the CLI:

openlit-instrument python your_app.py

Key capabilities

Capability	Description
Distributed tracing	End-to-end traces across HTTP → framework → LLM → agent → tool
Token and cost tracking	Per-call token usage and USD cost from a built-in pricing database
GPU monitoring	Utilization, memory, temperature, and power for NVIDIA and AMD GPUs
Guardrails	Real-time prompt injection, sensitive topic, and topic restriction detection
Evaluations	Programmatic hallucination, bias, and toxicity scoring
Metrics	Latency histograms, error rates, and throughput via OpenTelemetry metrics

Supported languages

Python

Complete AI observability with automatic dependency detection. Instruments LLMs, frameworks, vector databases, and GPUs.

TypeScript / JavaScript

Full LLM monitoring for Node.js applications with distributed tracing, metrics, and cost optimization.

Go

OpenTelemetry-native instrumentation for Go. Wrap your OpenAI and Anthropic clients for automatic tracing, token tracking, and cost monitoring.

Supported integrations

LLM providers

OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Ollama, Groq, Cohere, Mistral, and more

AI and agentic frameworks

LangChain, LlamaIndex, CrewAI, mem0, AG2, DSPy, Agno, and more

Vector databases

ChromaDB, Pinecone, Qdrant, Milvus, Weaviate, and more

GPU hardware

NVIDIA and AMD GPUs via OpenTelemetry metrics

HTTP frameworks and clients

FastAPI, Flask, Django, Requests, HTTPX, aiohttp, urllib, and more

Get started

Quickstart

Get your first trace in under five minutes

Configuration

All openlit.init() parameters and environment variables

Integrations

60+ integrations with automatic instrumentation

Destinations

Send telemetry to Datadog, Grafana, New Relic, Dash0, and more