OpenLIT — OpenTelemetry LLM Observability

OpenLIT Features

One-Line Auto-Instrumentation

import openlit
openlit.init()  # That's it!

# All these are now automatically traced:
# - OpenAI calls
# - Anthropic calls
# - LangChain chains
# - LlamaIndex queries
# - Vector DB operations
# - Embedding generation

Supported Providers (50+)

Cost Tracking

# OpenLIT automatically calculates costs per request
# Dashboard shows:
# - Total spend per model
# - Cost per request
# - Token usage breakdown (input vs output)
# - Cost trends over time
# - Budget alerts

Guardrails

import openlit

# Add safety guardrails
openlit.init(
    guardrails=[
        openlit.guardrails.ToxicityGuardrail(threshold=0.8),
        openlit.guardrails.PIIGuardrail(),
        openlit.guardrails.PromptInjectionGuardrail(),
    ]
)

OpenTelemetry Native

Send data to any OTLP-compatible backend:

import openlit

# Send to Grafana Tempo
openlit.init(
    otlp_endpoint="http://grafana:4318",
    application_name="my-ai-app",
    environment="production",
)

# Or Jaeger, Datadog, New Relic, etc.

GPU Monitoring

openlit.init(collect_gpu_stats=True)
# Tracks: GPU utilization, memory usage, temperature

FAQ

Q: What is OpenLIT? A: OpenLIT is an OpenTelemetry-native LLM observability platform with 2,300+ GitHub stars. One line of code instruments 50+ providers, tracking costs, latency, tokens, and quality with a built-in dashboard.

Q: How is OpenLIT different from Langfuse or LangSmith? A: OpenLIT is built on OpenTelemetry (open standard), not a proprietary format. Your data can go to any OTLP backend (Grafana, Datadog, Jaeger). Langfuse and LangSmith use their own data formats and dashboards.

Q: Is OpenLIT free? A: Yes, open-source under Apache-2.0. Self-host for free.