Summary

Add a PDF community source so Coral can query text, tables, images, links, annotations, form fields, and document metadata from local PDF documents through SQL. The source uses a PyMuPDF converter script to extract page-level and document-level data into JSONL files.

Why this source

PDF is one of the most common document formats for resumes, reports, contracts, and research papers. Coral does not currently include a PDF source under sources/core or sources/community. Adding PDF support enables users to search across document collections, extract structured data from tables, find hyperlinks and annotations, and analyze form fields—all through Coral SQL.

Provider docs

• PyMuPDF: https://pymupdf.readthedocs.io/
• OCRmyPDF (optional, for scanned PDFs): https://ocrmypdf.readthedocs.io/

Source shape

Source scope

• Uses a Python converter script (scripts/pdf-to-jsonl.py) that reads PDFs via PyMuPDF and writes JSONL files.
• The manifest reads from ~/.coral/pdf/pages.jsonl and ~/.coral/pdf/documents.jsonl.
• Provides read-only access to extracted PDF content and metadata.
• No required filters — all columns are queryable directly.
• Includes a converter script with support for OCR (via ocrmypdf), recursive directory scanning, and custom output paths.
• Binary image data is not stored — only references (xref, dimensions).
• Table extraction uses PyMuPDF's built-in table detection.
• Scanned/image-only PDFs will have empty text without the --ocr flag.

Validation plan

• coral source lint
• coral source add --file
• coral source test
• coral.tables introspection
• coral.columns introspection for both tables
• Live pages query
• Live documents query