LangChain Integration
Drop-in callback handler for LangChain and LangChain.js. Every chain step, LLM call, retriever, and tool invocation is traced with no code restructuring.
LangChain is a framework for building LLM-powered applications — chains, agents, retrievers, and tools — available for Python and JavaScript/TypeScript.
Getting Started
Install the SDK
Install @zespan/sdk alongside langchain.
npm install @zespan/sdk langchain @langchain/openaiAdd the callback handler
Pass LumiqCallbackHandler to any chain, agent, or LLM as a callback. No restructuring — it attaches to your existing LangChain code.
import { LumiqCallbackHandler } from '@zespan/sdk';
import { ChatOpenAI } from '@langchain/openai';
import { ConversationChain } from 'langchain/chains';
const handler = new LumiqCallbackHandler({
apiKey: process.env.ZESPAN_API_KEY,
});
const model = new ChatOpenAI({ callbacks: [handler] });
const chain = new ConversationChain({ llm: model });
const result = await chain.invoke({ input: 'Tell me a joke' });View traces in Zespan
Open Trace Explorer. Each chain run appears as a trace with nested spans for every step, LLM call, and tool invocation.
What's captured automatically
- Full chain waterfall: every step, LLM call, retriever, and tool as linked spans
- Agent reasoning: thought, action, and observation per ReAct step
- Token and cost attribution: aggregated across every step in the chain
- Retrieval spans: retrieved documents, scores, and chunk counts for RAG chains
- LangGraph support: each graph node captured as a span
- Error attribution: failures attributed to the correct chain step
FAQ
Does this work with LangGraph?
Yes. LumiqCallbackHandler works with LangGraph workflows. Each node in the graph is captured as a span, and the full execution graph is visible in the trace waterfall.
Do I need to restructure my existing chains?
No. Pass the handler to the callbacks parameter on any existing chain, agent, or LLM. Nothing else changes.
Does this work with LCEL (LangChain Expression Language)?
Yes. Pass the handler to the config parameter when invoking LCEL chains: chain.invoke(input, { callbacks: [handler] }).
Start for free — 10K traces/month, no card needed
LangChain integration works on all plans including the free tier.