May 17, 2024

Haystack <> Langfuse Integration

OSS observability and analytics for the popular RAG application framework.

We’re excited to highlight a new Langfuse integration with Haystack!

This integration allows you to easily trace your Haystack pipelines in the Langfuse UI. We’ve previously launched integrations with popular tools that devs love — including LlamaIndex, LangChain and LiteLLM — and we’re excited to be continuing that with Haystack.

Thanks to the team at deepset for developing the integration. We’re excited to see how you use it!

What’s New

The langfuse-haystack package integrates tracing capabilities into Haystack (2.x) pipelines using Langfuse. You can then add LangfuseConnector as a tracer to automatically trace the operations and data flow within the pipeline.

What is Haystack?

Haystack is the open-source Python framework developed by deepset. Its modular design allows users to implement custom pipelines to build production-ready LLM applications, like retrieval-augmented generative (RAG) pipelines and state-of-the-art search systems. It integrates with Hugging Face Transformers, Elasticsearch, OpenSearch, OpenAI, Cohere, Anthropic and others, making it an extremely popular framework for teams of all sizes.

RAG has proven to be a pragmatic and efficient way of working with LLMs. The integration of custom data sources through RAG can significantly enhance the quality of an LLM’s response, improving user experience. Haystack is a lightweight and powerful tool to build data-augmented LLM applications; you can read more about their approach to the building blocks of pipelines here.

Haytsack recently introduced Haystack 2.0 with a new architecture and modular design. Building on top of Haystack makes applications easier to understand, maintain and extend.

How Can Langfuse Help?

Langfuse tracing can be helpful for Haystack pipelines in the following ways:

Capture comprehensive details of each execution trace in a beautiful UI dashboard
- Latency
- Token usage
- Cost
- Scores
Capture the full context of the execution
Monitor and score traces
Build fine-tuning and testing datasets

Langfuse integration with a tool like Haystack can help monitor model performance, can assist with pinpointing areas for improvement, or create datasets from your pipeline executions for fine-tuning and testing.

Overview

Quickstart

Here are the steps to get started! First, add your API keys. You can find your Langfuse public and private keys on the dashboard. Make sure to set HAYSTACK_CONTENT_TRACING_ENABLED to "True".

import os
 
# Get keys for your project from the project settings page
# https://cloud.langfuse.com
os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-..."
os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-..."
os.environ["LANGFUSE_HOST"] = "https://cloud.langfuse.com" # 🇪🇺 EU region
# os.environ["LANGFUSE_HOST"] = "https://us.cloud.langfuse.com" # 🇺🇸 US region
os.environ["HAYSTACK_CONTENT_TRACING_ENABLED"] = "True"
 
# Your openai key
os.environ["OPENAI_API_KEY"] = "sk-proj-..."

Here’s how you add LangfuseConnector as a tracer to the pipeline:

from datasets import load_dataset
from haystack import Document, Pipeline
from haystack.components.builders import PromptBuilder
from haystack.components.embedders import SentenceTransformersDocumentEmbedder, SentenceTransformersTextEmbedder
from haystack.components.generators import OpenAIGenerator
from haystack.components.retrievers import InMemoryEmbeddingRetriever
from haystack.document_stores.in_memory import InMemoryDocumentStore
from haystack_integrations.components.connectors.langfuse import LangfuseConnector
 
basic_rag_pipeline = Pipeline()
    # Add components to your pipeline
    basic_rag_pipeline.add_component("tracer", LangfuseConnector("Basic RAG Pipeline"))
    basic_rag_pipeline.add_component(
        "text_embedder", SentenceTransformersTextEmbedder(model="sentence-transformers/all-MiniLM-L6-v2")
    )
    basic_rag_pipeline.add_component("retriever", retriever)
    basic_rag_pipeline.add_component("prompt_builder", prompt_builder)
    basic_rag_pipeline.add_component("llm", OpenAIGenerator(model="gpt-3.5-turbo", generation_kwargs={"n": 2}))