Integrate your Haystack RAG pipelines with ZenML

Haystack provides a component-based framework for Retrieval-Augmented Generation (RAG); integrating Haystack with ZenML wraps your retrievers, prompt builders, and LLM calls inside reproducible pipelines with artifact tracking, observability, and a clean path from local experiments to production.

Features with ZenML

End-to-end RAG orchestration. Run complete Haystack retrieval and generation pipelines inside ZenML.
‍Unified artifact tracking. Automatically log queries, retrieved documents, and answers as ZenML artifacts.
‍Custom evaluation workflows. Add eval steps to measure answer quality, latency, or retrieval accuracy.
‍Flexible retriever integration. Swap in-memory retrievers for vector databases or cloud backends within the same pipeline.
‍Production-ready pipelines. Scale Haystack RAG from local dev to production orchestrators like Kubernetes or Airflow with no code changes.

Main Features

Retrieval-Augmented Generation. Build RAG flows with retriever, prompt builder, and LLM.
‍Component architecture. Swap or configure components without changing your pipeline shape.
‍In-memory store. Use a simple in-memory document store and retriever for quick starts.

How to use ZenML with

Haystack

from zenml import ExternalArtifact, pipeline, step
from haystack_agent import pipeline as haystack_pipeline

@step
def run_haystack(question: str) -> str:
    # Execute the Haystack pipeline and extract the first LLM reply
    result = haystack_pipeline.run(
        {"retriever": {"query": question},
         "prompt_builder": {"question": question}},
        include_outputs_from={"llm"},
    )
    replies = result.get("llm", {}).get("replies", [])
    return replies[0] if replies else "No response generated"

@pipeline
def haystack_rag_pipeline() -> str:
    q = ExternalArtifact(value="What city is home to the Eiffel Tower?")
    return run_haystack(q.value)

if __name__ == "__main__":
    print(haystack_rag_pipeline())

Additional Resources

ZenML Agent Framework Integrations (GitHub)

ZenML Documentation

Haystack Documentation