@core-ai/langfuse

Langfuse middleware for @core-ai/core-ai.

• Langfuse
• Langfuse docs

Installation

npm install @core-ai/langfuse @langfuse/tracing @langfuse/otel @opentelemetry/sdk-node

@langfuse/tracing and @langfuse/otel are peer dependencies. @opentelemetry/sdk-node is required to register the LangfuseSpanProcessor.

Setup

Create an instrumentation file and import it at your application's entry point:

import { NodeSDK } from '@opentelemetry/sdk-node';
import { LangfuseSpanProcessor } from '@langfuse/otel';

export const sdk = new NodeSDK({
    spanProcessors: [new LangfuseSpanProcessor()],
});

sdk.start();

Usage

import { wrapChatModel } from '@core-ai/core-ai';
import { createOpenAI } from '@core-ai/openai';
import { createLangfuseMiddleware } from '@core-ai/langfuse';

const openai = createOpenAI({ apiKey: process.env.OPENAI_API_KEY });

const model = wrapChatModel({
    model: openai.chatModel('gpt-5-mini'),
    middleware: createLangfuseMiddleware({ recordContent: true }),
});

recordContent is disabled by default. Enable it only when you want prompts and outputs attached to Langfuse observations.