DiffFind combines traditional lexical diffing with semantic search and semantic comparison for text and documents.
For Semantic Find, I use semantic retrieval followed by LLM reranking and extraction. In my testing, this has worked significantly better than simply giving a document to an LLM and asking it to search.
It currently supports text and PDF documents, plus a REST API for programmatic use.
I'd love feedback from HN, particularly on the semantic search/comparison approach and where you see it being useful.