[FEAT]: Additional text splitting / chunking options

What would you like to see?

I would like to request additional text splitting and chunking options in AnythingLLM, ideally exposed as advanced workspace settings and, if possible, configurable per document type. The current chunking works for general use, but for RAG workflows with mixed content such as technical documents, code, and structured reports or operational material, it would be very helpful to support alternatives such as sentence-based splitting, semantic chunking, recursive splitting, and document-type-aware strategies. AnythingLLM already supports text splitting configuration in Docker and desktop, and there are existing feature requests discussing more flexible chunking approaches, including SemanticChunker and SentenceSplitter support. If feasible, I would also appreciate a way to deploy or override this logic in a Docker-based installation, either through environment variables, a plugin-like mechanism, or a custom code path in a self-hosted image. This would make it easier to tune retrieval quality without preprocessing documents externally, and would be especially useful when experimenting with different embedding models and RAG pipelines.

Thank you for considering this feature. It would significantly improve retrieval quality and make AnythingLLM more adaptable for advanced use cases.

WaveQark