LangGraph / LangChain - Cascade

Cascade provides auto-instrumentation for LangGraph and LangChain. A single function call translates the framework’s native events into Cascade spans with no changes to your existing agent code.

Setup

from cascade import init_tracing
from cascade.integrations import instrument_langgraph

init_tracing(project="my_langgraph_app")
instrument_langgraph()

# Existing LangGraph code - no changes needed
app = graph.compile()
result = app.invoke({"messages": [...]})

Requires langgraph and langchain-core packages:

pip install langgraph langchain-core

What gets traced automatically

The integration automatically captures:

Graph node executions: chains and their relationships
LLM calls: with model, prompt, completion, and token counts
Tool calls: with input and output values
Retriever calls: with query and retrieved documents
Agent delegations: sub-agent handoffs and orchestration

⌘I