v2.24.0
Major update eliminates boilerplate code and adds direct PDF processing for LLMs.
Deep Dive
Haystack v2.24.0 introduces significant pipeline simplifications, allowing direct connections between components without adapters and automatic type conversions. Rankers now deduplicate documents by ID before ranking. The update adds PDF file support to ChatMessage objects for OpenAI/Azure generators via a new FileContent dataclass. Breaking changes affect users relying on duplicate document IDs in ranking outputs, requiring unique IDs to preserve previous behavior.
Why It Matters
This reduces development complexity for RAG pipelines and enables direct document processing without intermediate conversion steps.