PDF to Markdown Converter

How to use

1. Navigate to the Issues tab (or click the button above).
2. Click New Issue.
3. Drag and drop your .pdf file into the issue description box.
4. (Optional) Add slash commands to the issue description to customize the extraction.
5. Click Submit new issue.

Within a few minutes, a GitHub Action bot will comment on your issue containing direct download links for the raw .md file and a .zip archive containing the Markdown and any extracted images.

Advanced Configuration (Slash Commands)

You can completely customize how the pipeline processes your PDF by dropping these slash commands anywhere in your issue body.

Engine Selection

The pipeline supports three distinct architectural paths. If no engine is specified, the system defaults to marker.

• /engine=marker (Default): Lightning-fast extraction using the PDF's native digital text layer. Zero LLM calls. Best for clean, natively digital PDFs.
• /engine=marker-llm: Extracts text rapidly using Marker, but routes the output through a cloud LLM to perfectly format complex tables, fix math, and clean up the structural Markdown.
• /engine=vision: Bypasses Marker entirely. Converts PDF pages into images and uses a multimodal Vision AI to holistically read and transcribe the page. Best for heavily corrupted PDFs, scanned documents, or complex visual math.

Modifiers & Overrides

• /force-ocr: Tells Marker to ignore the PDF's digital text layer and boot up its neural network to visually read every character. Use only for image-only scans. (Warning: Runs locally on GitHub Actions CPU and takes ~6 minutes per page. Do not use with /engine=vision.)
• /page-range=X-Y: Processes only a specific chunk of the document (e.g., /page-range=1-20). The outputs are dynamically named so you can process large books in safe increments without overwriting previous chunks.
• /timeout=X: Sets the maximum execution time in minutes before the engine gracefully stops and packages whatever pages it has finished (e.g., /timeout=30). Default: 15.
• /model=X: Overrides the OpenRouter LLM used by the vision or marker-llm engines (e.g., /model=qwen/qwen-2.5-72b-instruct:free).

Example Workflows

1. The Fast Default (Clean Digital PDFs)
Just drop the PDF in the issue. No commands needed. Runs standard Marker text extraction in seconds.

2. The "Bad Scan" Fixer

/engine=marker-llm
/force-ocr
/page-range=1-5

Forces Marker to visually read a scanned document, then uses an LLM to clean up the messy OCR formatting. Limited to 5 pages to avoid the 15-minute CPU timeout.

3. The Complex Math & Layout Reader

/engine=vision
/page-range=10-25
/timeout=30

Takes pictures of pages 10 through 25 and lets the default Vision AI perfectly transcribe the LaTeX math and tables, giving it an extended 30-minute window to finish the job.

Maintenance

Processed files are stored in the conversions/ directory and are automatically purged after 30 days to optimize repository storage.