Haystack

Haystack suits developers assembling a document-grounded application from explicit components. Indexing, retrieval, routing and generation can form a pipeline, while agents add tool use where the task needs it. The framework allows model and data-store choices; a managed enterprise offering is separate from operating the free software on your own infrastructure.

A practical workflow

1. Install the framework and select a supported document store and model.
2. Build a small indexing pipeline for permitted documents.
3. Connect retrieval and generation components for one question type.
4. Inspect returned evidence and test missing-document and failure cases.

Illustrative input

This example describes a task, not a tested output. Adjust the source material and settings to your own situation.

Answer a question from a small policy collection and return the document passages that support the answer; report when no passage matches.

Result check: Check retrieval selected the relevant version and ensure the generated answer does not add an unsupported policy.

Access and alternatives

The open-source framework can be used without a software subscription; enterprise platform and support access are paid. Self-hosted deployment still needs storage, compute and possibly paid models. No free hosted inference is implied by the framework license.

Choose another workflow when its emphasis fits better: RAGFlow for a document-retrieval application; Dify for a more visual application-building workflow.