Skip to main content
The OpenLIT provider uses instrumentations built and maintained by OpenLIT for comprehensive AI observability across LLMs, vector databases, and AI frameworks.

​
Configuration

apiVersion: openlit.io/v1alpha1
kind: AutoInstrumentation
metadata:
  name: openlit-instrumentation
spec:
  selector:
    matchLabels:
      openlit.io/instrument: "true"
  python:
    instrumentation:
      provider: "openlit"
      version: "latest"
  otlp:
    endpoint: "http://openlit:4318"

​
Provider Specific Features

OpenLIT offers unique capabilities extending the OpenTelemetry Instrumentations:

​
Custom Package Installation

Install additional Python packages alongside OpenLIT instrumentation: 
spec:
  python:
    instrumentation:
      customPackages: "langchain>=0.1.0,chromadb>=0.4.0,pinecone-client>=2.0.0"
Common Package Examples:
  • AI Frameworks: langchain>=0.1.0,llama-index>=0.9.0
  • Vector Databases: chromadb>=0.4.0,pinecone-client>=2.0.0,qdrant-client>=1.0.0
  • LLM Providers: openai>=1.0.0,anthropic>=0.8.0,cohere>=4.0.0
  • Tools: crewai>=0.1.0,autogen>=0.2.0

​
Pricing and Cost Tracking

Automatic cost calculation for LLM operations using built-in pricing data: 
env:
- name: OPENLIT_PRICING_JSON
  value: "https://custom-pricing.json"  # Optional custom pricing
Features:
  • Token-level cost calculation with built-in pricing
  • Model cost comparison across providers
  • Automatic pricing updates

​
GPU and System Metrics

Monitor system resources and GPU utilization: 
env:
- name: OPENLIT_COLLECT_GPU_STATS
  value: "true"
- name: OPENLIT_COLLECT_SYSTEM_METRICS
  value: "true"
Metrics:
  • GPU utilization and memory
  • CPU, memory, disk, network metrics
  • System performance tracking

​
Detailed Tracing Control

Control the granularity of tracing for different use cases: 
env:
- name: OPENLIT_DETAILED_TRACING
  value: "true"  # Component-level tracing for debugging
- name: OPENLIT_DISABLE_METRICS
  value: "false"  # Keep metrics enabled
Tracing modes:
  • Detailed tracing: Component-level spans for debugging (20-50+ spans per query)
  • Workflow tracing: High-level operations only (5-10 spans per query)

​
Instrumentor Control

Disable specific instrumentors if needed: 
env:
- name: OPENLIT_DISABLED_INSTRUMENTORS
  value: "anthropic,cohere"  # Disable specific providers