Variable langChainIntegrationConst

Example

import * as Sentry from '@sentry/node';
import { ChatOpenAI } from '@langchain/openai';

Sentry.init({
  integrations: [Sentry.langChainIntegration()],
  sendDefaultPii: true, // Enable to record inputs/outputs
});

// LangChain calls are automatically instrumented
const model = new ChatOpenAI();
await model.invoke("What is the capital of France?");
Copy

Manual Callback Handler

You can also manually add the Sentry callback handler alongside other callbacks:

Example

import * as Sentry from '@sentry/node';
import { ChatOpenAI } from '@langchain/openai';

const sentryHandler = Sentry.createLangChainCallbackHandler({
  recordInputs: true,
  recordOutputs: true
});

const model = new ChatOpenAI();
await model.invoke(
  "What is the capital of France?",
  { callbacks: [sentryHandler, myOtherCallback] }
);
Copy

Options

• recordInputs: Whether to record input messages/prompts (default: respects sendDefaultPii client option)
• recordOutputs: Whether to record response text (default: respects sendDefaultPii client option)

Default Behavior

By default, the integration will:

• Record inputs and outputs ONLY if sendDefaultPii is set to true in your Sentry client options
• Otherwise, inputs and outputs are NOT recorded unless explicitly enabled

Example

// Record inputs and outputs when sendDefaultPii is false
Sentry.init({
  integrations: [
    Sentry.langChainIntegration({
      recordInputs: true,
      recordOutputs: true
    })
  ],
});

// Never record inputs/outputs regardless of sendDefaultPii
Sentry.init({
  sendDefaultPii: true,
  integrations: [
    Sentry.langChainIntegration({
      recordInputs: false,
      recordOutputs: false
    })
  ],
});
Copy

Supported Events

The integration captures the following LangChain lifecycle events:

• LLM/Chat Model: start, end, error
• Chain: start, end, error
• Tool: start, end, error