OpenInference LangChain Instrumentation

Python auto-instrumentation library for LangChain.

These traces are fully OpenTelemetry compatible and can be sent to an OpenTelemetry collector for viewing, such as arize-phoenix.

Compatibility

This instrumentation works with:

LangChain 1.x (langchain>=1.0.0): Modern agent framework built on LangGraph
LangChain Classic (langchain-classic>=1.0.0): Legacy chains and tools (formerly langchain 0.x)
All LangChain partner packages (langchain-openai, langchain-anthropic, langchain-google-vertexai, etc.)

The instrumentation hooks into langchain-core, which is the shared foundation used by all LangChain packages.

Installation

For LangChain 1.x (Recommended for New Projects)

pip install openinference-instrumentation-langchain langchain langchain-openai

For LangChain Classic (Legacy Applications)

pip install openinference-instrumentation-langchain langchain-classic langchain-openai

For Both (Migration Scenarios)

pip install openinference-instrumentation-langchain langchain langchain-classic langchain-openai

Quickstart

Example with LangChain 1.x (New Agent Framework)

Install packages needed for this demonstration.

pip install openinference-instrumentation-langchain langchain langchain-openai arize-phoenix opentelemetry-sdk opentelemetry-exporter-otlp

Start the Phoenix app in the background as a collector. By default, it listens on http://localhost:6006. You can visit the app via a browser at the same address.

The Phoenix app does not send data over the internet. It only operates locally on your machine.

python -m phoenix.server.main serve

The following Python code sets up the LangChainInstrumentor to trace langchain and send the traces to Phoenix at the endpoint shown below.

from langchain.agents import create_agent
from langchain_openai import ChatOpenAI
from openinference.instrumentation.langchain import LangChainInstrumentor
from opentelemetry import trace as trace_api
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk import trace as trace_sdk
from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor

endpoint = "http://127.0.0.1:6006/v1/traces"
tracer_provider = trace_sdk.TracerProvider()
trace_api.set_tracer_provider(tracer_provider)
tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
tracer_provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))

LangChainInstrumentor().instrument()

To demonstrate tracing, we'll create a simple agent. First, configure your OpenAI credentials.

import os

os.environ["OPENAI_API_KEY"] = "<your openai key>"

Now we can create an agent and run it.

def get_weather(city: str) -> str:
    """Get the weather for a city."""
    return f"The weather in {city} is sunny!"

model = ChatOpenAI(model="gpt-4")
agent = create_agent(model, tools=[get_weather])
result = agent.invoke({"messages": [{"role": "user", "content": "What's the weather in Paris?"}]})
print(result)

Example with LangChain Classic (Legacy Chains)

For legacy applications using LangChain Classic:

from langchain_classic.chains import LLMChain
from langchain_core.prompts import PromptTemplate
from langchain_openai import OpenAI

# ... (same instrumentation setup as above)

prompt_template = "Tell me a {adjective} joke"
prompt = PromptTemplate(input_variables=["adjective"], template=prompt_template)
llm = LLMChain(llm=OpenAI(), prompt=prompt, metadata={"category": "jokes"})
completion = llm.predict(adjective="funny", metadata={"variant": "funny"})
print(completion)