Haystack — AI Orchestration for Search & RAG

Core Concepts

Pipelines

Compose components into DAGs — each component does one thing well:

pipe.add_component("converter", HTMLToDocument())
pipe.add_component("splitter", DocumentSplitter(split_length=200))
pipe.add_component("embedder", OpenAIDocumentEmbedder())
pipe.add_component("writer", DocumentWriter(document_store=store))
pipe.connect("converter", "splitter")
pipe.connect("splitter", "embedder")
pipe.connect("embedder", "writer")

Component Library

50+ built-in components: document converters, splitters, embedders, retrievers, rankers, generators, and more.

Document Stores

Pluggable storage: Elasticsearch, Qdrant, Pinecone, Weaviate, ChromaDB, pgvector, and in-memory.

Evaluation

Built-in evaluation metrics for RAG quality: faithfulness, context relevance, answer relevance.

FAQ

Q: What is Haystack? A: An open-source AI orchestration framework by deepset for building production RAG, search, and agent pipelines with modular components. 25K+ GitHub stars.

Q: How is Haystack different from LangChain? A: Haystack uses a strict pipeline/DAG model with typed components, making it more predictable for production. LangChain is more flexible but less structured.