Haystack 3.5 adds structured agent replies and new routing components
Deepset’s Haystack 3.5.0 is available, adding controls for agent responses and tools for routing data through AI pipelines. GitHub dates the release to November 10, and PyPI lists the corresponding package files on the same day. GitHub release PyPI package
The release introduces a StructuredOutputHook that checks an agent’s final answer against a Pydantic model or JSON schema. If validation fails, it returns the error to the model for another attempt, subject to a retry limit. Release notes
Three new components use OpenAI’s Decisions API to classify documents or route text and images. The image router supports a fallback branch for low-confidence decisions. Release notes
Haystack is an open-source Python framework for assembling retrieval, generation and agent workflows from modular components. The release’s availability is supported by both the project repository and the package registry, rather than a roadmap announcement alone. PyPI package
For developers, response validation can make downstream software easier to integrate with an agent. It establishes whether a response has the expected structure, however, not whether its contents are correct. Production testing still needs examples that pass a schema while containing a wrong classification or unsupported answer.
Sources
- Haystack v3.5.0 release notes (primary; 2026-11-10)
- haystack-ai 3.5.0 package (primary; 2026-11-10)
Sources and citations
Revision history
Published .