Add Natural Language To Any Application With Our Haystack NLP Framework

Pick and deploy a model. Build, evaluate and launch a fully-fledged, composable NLP system with Haystack. Transform the way users search. Add question answering for your documents. Scale it to millions of users with powerful Haystack pipelines and versatile building blocks for common NLP tasks.

# Build a simple Question Answering pipeline
p = ExtractiveQAPipeline(
query="What is the outlook for the US market?"
# Returned answer:
[{'answers': '+22% growth in revenue',
"For the US market, we expect a strong Q4
with +22% growth in revenue.",
'probability': 0.989,
QueryWhat is the outlook for the US Market?
Answer“For the US market, we expect a strong Q4 with +22% growth in revenue.”

Start exploring Haystack!

Check on GitHub
  • 3.8k


  • 100


  • Latest Models

    Deploy any Transformer model from Hugging Face's Model Hub, experiment, find the one that works.

  • Flexible Document Store

    Use Haystack question answering system on top of Elasticsearch, OpenSearch, or plain SQL.

  • Vector Databases

    Boost neural search performance with Milvus, FAISS, or Weaviate vector databases, and dense passage retrieval.

  • Scalable

    Build language-aware applications that can scale to millions of documents — powered by neural search and question answering.

  • End-to-end

    Building blocks for the entire product development cycle such as file converters, indexing functions, models, labeling tools, domain adaptation modules, REST API.

  • Pipelines

    It's not one-size-fits-all! Combine nodes into flexible and scalable pipelines and launch a powerful question answering system.

When to add neural search and question answering to your applications

From an idea to a natural language search — address the varying demands of business users with ease. Below are some examples of use case-driven applications for Haystack.

Ready to get started?

Start learning Haystack

Frequently asked questions

  • Haystack is a flexible open source Python framework for building question answering and neural search systems on top of large document collections. It enables the developers to apply Transformer models to real world applications, for example, semantic search, open-domain question answering, information extraction, summarization, answer generation, and chatbots. Haystack is being developed by deepset.

  • To quote Wikipedia: "Question answering (QA) is a computer science discipline within the fields of information retrieval and natural language processing (NLP), which is concerned with building systems that automatically answer questions posed by humans in a natural language."

  • The Transformer is a deep learning model initially open sourced by Google in late 2018. It sparked a wave of interest from researchers and developers alike because of its groundbreaking performance. Thanks to a community of researchers and NLP practitioners, many new variants of the Transformer model appeared.

  • Haystack is designed to be a very practical, down-to-earth open source NLP framework. A decent knowledge of Python would be a prerequisite, and some fundamentals of machine learning too. We work on lowering the barrier of entry to NLP, and we always appreciate sharing your experiences with us. For inference, you'd also need a GPU-enabled system.

  • Our community consists of NLP researchers, experts, but also of NLP practitioners, and even full-stack developers who build question answering with Haystack. We welcome everyone who's interested in natural language processing and is solving a real-world use case involving question answering and neural search.

  • Haystack framework is completely open source, licensed under Apache License 2.0. As deepset-the-company, we've already helped the largest European companies and public sector organizations to instrument neural search, and we're also building an enterprise SaaS product to complement Haystack. We'd be very happy to discuss this in more detail.