📝 2. Release Notes (v0.5.1)
🔧 Highlights:
Image Enrichment Agent: GPT-4o-based insights added to image-linked chunks
Image Indexing: New image_index.faiss and image_metadata.jsonl support multi-modal retrieval
DOCX & PPTX Loaders: Now extract and deduplicate all inline, floating, and sibling images
Chunker: Allows image-only chunks to pass min-token filter
Pipeline Updates: step_enrich() and step_index_images() added with CLI parity
Late Fusion Retrieval: Combines text + image-based matches in RetrievalManager