Skip to main content

v1.3.0 - 2024-01-31

  • Completely redesign chunking system mechanism at the parsing phase (hierarchical aggregations, better control to get 256 tokens)
  • Switched to Pinecone serverless for performance improvements
  • Add TOC section for each source in the citations panel as breadcrumbs
  • EXPERIMENTAL - Figure retrieval system within the answer (caption pictures with GPT4 Vision and indexing the caption in the retrieval system)