🧭 CozeLoop LangChain Integration

Official LangChain integration for CozeLoop - seamlessly report traces from your LangChain and LangGraph applications.

Features

• CozeloopCallbackHandler: Automatically capture and report traces to CozeLoop
• Support for both LangChain and LangGraph
• W3C trace context propagation for distributed tracing

Installation

npm install @cozeloop/langchain
# or
pnpm install @cozeloop/langchain

Configuration

Environment Variables

The following environment variables can be used to configure the integration:

| Variable | Description | Example | |----------|-------------|---------| | COZELOOP_WORKSPACE_ID | Workspace ID for resource association | '7487806534651887643' | | COZELOOP_API_TOKEN | API token for authentication. See Authentication Guide | 'pat_xxxx' | | COZELOOP_OTLP_ENDPOINT | Trace reporting endpoint | 'https://api.coze.cn/v1/loop/opentelemetry/v1/traces' |

Usage

CozeloopCallbackHandler

The callback handler integrates with LangChain and LangGraph to automatically capture traces.

Note: Since trace exporting is asynchronous, proper cleanup is required:

• CLI scripts: Call await callback.flush() before exit to ensure all traces are exported.
• Server applications: Call await callback.shutdown() during graceful shutdown to release resources.

Initialization

import { CozeloopCallbackHandler } from '@cozeloop/langchain';

const callback = new CozeloopCallbackHandler({
  // Span exporter configuration
  spanExporter: {
    workspaceId: 'xxx',
    token: 'pat_xxx',
    traceEndpoint: 'https://api.coze.cn/v1/loop/opentelemetry/v1/traces',
  },
  // Optional: Filter specific trace types
  // ignoreAgent: false,
  // ignoreChain: false,
  // ignoreCustomEvent: false,
  // ignoreLLM: false,
  // ignoreRetriever: false,
  // ignorePrompt: false,

  // Optional: Propagate trace context from upstream services
  // propagationHeaders: {
  //   traceparent: '00-b3691bfe8af1415029177821d4114cef-ddd0307891d51ce3-01',
  //   tracestate: '',
  // },
});

// Access W3C propagation headers for downstream services
// const headers = callback.w3cPropagationHeaders;

With LangChain

import { CozeloopCallbackHandler } from '@cozeloop/langchain';
import { ChatPromptTemplate } from '@langchain/core/prompts';

const callback = new CozeloopCallbackHandler({ /* options */ });

const prompt = ChatPromptTemplate.fromTemplate('What is 1 + {number}?');
const model = new CustomLLM({});
const chain = prompt.pipe(model);

const resp = await chain.invoke(
  { number: 1 },
  {
    runName: 'SuperChain',
    callbacks: [callback],
  },
);

// Ensure traces are exported before script exits
await callback.flush();

With LangGraph

import { CozeloopCallbackHandler } from '@cozeloop/langchain';
import { createReactAgent } from '@langchain/langgraph/prebuilt';

const callback = new CozeloopCallbackHandler({ /* options */ });

const agent = createReactAgent({
  llm: model,
  tools: [tool1, tool2],
});

const resp = await agent.invoke(
  {
    messages: [{ role: 'user', content: 'Hello!' }],
  },
  { callbacks: [callback] },
);

// Ensure traces are exported before script exits
await callback.flush();