How to use custom chunking with external API (NLM Ingestor) in OpenWebUI?

[drspam1991]

Description

I want to use a custom chunking method for document processing in OpenWebUI. Specifically, I have an external service (nlm-ingestor) that extracts semantic chunks from PDFs very accurately. I want to integrate this service with OpenWebUI's document handling pipeline.

Use Case

Currently, OpenWebUI supports document processing with built-in chunking, but I need to override this with my own chunking logic. I chunk the input document like this:

from llmsherpa.readers import LayoutPDFReader

llmsherpa_api_url = "http://localhost:5010/api/parseDocument?renderFormat=all"
pdf_path = "/home/user/sample.pdf"

pdf_reader = LayoutPDFReader(llmsherpa_api_url)
doc = pdf_reader.read_pdf(pdf_path)

chunks = [chunk.to_text() for chunk in doc.chunks()]

I want to inject these extracted chunks into OpenWebUI so that they can be indexed and retrieved via its RAG system.

[samuelbailor]

I'm trying to do a similar things but with CSV and Excel file types. I believe I have a better chunker for Tabular data than what's available via OpenWebUI. I can't find any examples in this codebase for how to take over certain stages of the RAG pipeline, like the chunker/parser.