Skip to main content
Use Lemma’s LangChain integration when your app runs chains, chat models, retrievers, tools, or agents through LangChain callbacks. The integration creates one Lemma trace for the root run, records LLM calls as generations, tools as tool calls, retrievers as spans, and nested chains as child spans. You do not need to wrap LangChain calls in lemma.trace(). Pass the callback handler through LangChain’s callbacks option.

​
TypeScript

Install LangChain and the Lemma SDK: 
npm install @uselemma/tracing @langchain/core @langchain/openai
Pass langChain() as a callback handler: 
import { ChatOpenAI } from "@langchain/openai";
import { HumanMessage } from "@langchain/core/messages";
import { langChain } from "@uselemma/tracing";

const model = new ChatOpenAI({
  model: "gpt-4o",
  callbacks: [
    langChain({
      apiKey: process.env.LEMMA_API_KEY,
      projectId: process.env.LEMMA_PROJECT_ID,
      agentName: "support-agent",
    }),
  ],
});

export async function callAgent(userMessage: string) {
  const response = await model.invoke([new HumanMessage(userMessage)]);
  return response.content;
}

​
Python

Install the optional extra: 
pip install "uselemma-tracing[langchain]" langchain-openai
Pass langchain() as a callback handler: 
import os

from langchain_openai import ChatOpenAI
from uselemma_tracing import langchain

model = ChatOpenAI(
    model="gpt-4o",
    callbacks=[
        langchain(
            api_key=os.environ["LEMMA_API_KEY"],
            project_id=os.environ["LEMMA_PROJECT_ID"],
            agent_name="support-agent",
        )
    ],
)


def call_agent(user_message: str):
    response = model.invoke(user_message)
    return response.content

​
What Lemma records

LangChain callbackLemma record
Root chain start/endOne Lemma trace named from agentName or the LangChain run name
LLM or chat model start/endGeneration with input messages, output text, model, timing, and parent ID
Tool start/endTool call with name, input, output or error, timing, and parent ID
Retriever start/endSpan with query input, output, timing, and parent ID
Nested chain start/endChild span preserving LangChain parent run IDs
Use langChain({ recordInputs: false, recordOutputs: false }) in TypeScript or langchain(record_inputs=False, record_outputs=False) in Python to avoid sending prompts, tool inputs, tool outputs, or model output text.

​
Debugging

Enable debug mode while developing to confirm that spans arrive live and stay nested under the expected parent: 
import { enableDebugMode } from "@uselemma/tracing";

enableDebugMode();

from uselemma_tracing import enable_debug_mode

enable_debug_mode()