Skip to content

Multi-Agent Reference Workflow

AgenticLens includes a verified offline LangGraph supervisor workflow based on the orchestration patterns published by the official LangGraph repositories.

LangGraph Supervisor

The workflow uses a compiled LangGraph StateGraph. A supervisor delegates work to research, calculation, and review agents. AgenticLens records delegation, retrieval, tool, and final-response spans during the actual graph execution.

uv sync --extra langgraph
uv run python examples/reference_workflows/langgraph_supervisor.py
uv run agenticlens inspect examples/reference_workflows/artifacts/langgraph-supervisor.json

The workflow is deterministic and does not require a provider account or API key. Its saved reference artifact is maintained at:

examples/reference_workflows/artifacts/langgraph-supervisor.json

Optional Dependency

LangGraph remains optional. Installing the AgenticLens core package does not install it.

uv sync --extra multi-agent

The multi-agent extra currently contains the verified LangGraph dependency.

Instrumentation Coverage

Current instrumentation is explicit:

  • trace() records the workflow boundary
  • delegation spans record supervisor transfers
  • retrieval and tool spans record specialist work
  • the final-response span records completion
  • framework identity is stored on the run

Automatic conversion of every framework-native event into AgenticLens spans remains roadmap work.

Official Sources