LangChain & LangGraph

Infralo integrates with both LangChain and LangGraph via the LangChain Callbacks framework. This enables seamless tracing of chains, agents, and state graphs.

Setup

First, install the LangGraph extra dependencies:

pip install infralo[langgraph]

To configure tracing, instantiate the InfraloCallbackHandler and pass it to your chain or graph at invocation time.

Usage

LangGraph Example

Create a callback handler instance and pass it inside the invocation config dictionary:

from infralo import Infralo
from infralo.integrations.langgraph import InfraloCallbackHandler
from langgraph.prebuilt import create_react_agent

infralo = Infralo()
agent = create_react_agent(model, tools)

# Invoke within an active Infralo trace
with infralo.start_trace(session_id="langgraph-demo"):
    handler = InfraloCallbackHandler()
    
    result = await agent.ainvoke(
        {"messages": [("user", "Calculate 42 * 7")]},
        config={"callbacks": [handler]}
    )

Plain LangChain Example

The exact same handler works with any standard LangChain LCEL chain:

from infralo import Infralo
from infralo.integrations.langgraph import InfraloCallbackHandler

infralo = Infralo()
chain = prompt | model | parser

# Invoke within an active Infralo trace
with infralo.start_trace(session_id="langchain-demo"):
    handler = InfraloCallbackHandler()
    
    result = chain.invoke(
        {"topic": "observability"},
        config={"callbacks": [handler]}
    )

[!TIP] Handler Lifecycle: It is highly recommended to instantiate a new InfraloCallbackHandler per run (invoke / ainvoke) rather than sharing a single handler globally. This prevents trace leaks and ensures clean execution contexts.

What is Captured

The LangChain/LangGraph integration automatically captures:

• Tool Spans: Executions of functions wrapped with LangChain @tool or subclassing BaseTool.
• Inputs & Outputs: Serialized inputs and output strings/dictionaries.
• Concurrency Safety: Auto-handles parallel executions (e.g. parallel ToolNode structures inside graphs) across different threads and coroutines safely.

Complete Examples

You can view or download the complete, ready-to-run examples:

• langgraph_integration.py — Standard single-agent integration.
• langgraph_complex_nesting.py — Advanced multi-agent delegation and nested state graphs.