Omium instruments LangGraph at runtime. Once enabled, your calls to
invoke(), ainvoke(), stream(), and astream() are traced without rewriting your graph.Quickstart
Install dependencies:What Omium captures
| Signal | Notes |
|---|---|
| Execution ID | One per graph run (invoke / ainvoke). |
| Inputs & outputs | Initial state and final state for the run. |
| Timing | Total duration and step timings (where available). |
| Errors | Exceptions and failure context for debugging. |
| Streaming metadata | For streaming runs, counts and progress events. |
Instrumentation lifecycle
Enable instrumentation
Callinstrument_langgraph() once at process start (before you begin running graphs):
Disable instrumentation
Temporarily turn it off (for debugging or A/B testing):Async and streaming
Omium supports both sync and async execution.Configuration
Most teams start withomium init and defaults. For explicit configuration:
Troubleshooting
Traces not appearing
- Ensure
omium.init()is called beforeinstrument_langgraph(). - Confirm your API key and URL with Configure.
- Verify connectivity to
https://api.omium.ai.
ImportError: LangGraph is not installed
Next steps
CrewAI
Auto-instrument
kickoff() and multi-agent runs.Python SDK
Full SDK reference for tracing and checkpoints.