Problem

The workspace knowledgebase only indexes text, markdown, and code files. PDF and DOCX are common document formats.

Research

• PrivateGPT: PDFReader (llama-index), DocxReader. Supports PDF, DOCX, EPUB, PPTX, images (OCR), IPython notebooks
• Khoj: PyMuPDF (fitz) for PDF with pdfplumber fallback, python-docx for DOCX
• LocalGPT: PDFMinerLoader or UnstructuredPDFLoader

Proposed approach

Add two new parser plugins following the existing workspace plugin contract:

PDF parser: plugins/workspace/parsers/pdf/

• Use PyMuPDF (fitz) as primary, pdfplumber as fallback
• Extract text per page with page markers
• Optional dep: pip install pymupdf

DOCX parser: plugins/workspace/parsers/docx/

• Use python-docx library
• Extract paragraph text
• Optional dep: pip install python-docx

Update BINARY_SUFFIXES to exclude .pdf and .docx (binary but parseable). Add workspace-docs optional dep group in pyproject.toml.

Related

Part of workspace foundation (#5840)