What it does
Converts uploaded PDF documents into markdown text that the assistant can read and discuss. Uses Marker, an open-source PDF-to-markdown converter that handles complex layouts, tables, and OCR.
Requirements
Install Marker separately (it’s not bundled with Spaceduck):
pip install marker-pdf # requires Python 3.10+, PyTorch
marker_single is on your PATH, the marker_scan tool is automatically registered at gateway startup.
Marker is GPL-3.0 with Open Rail model weight restrictions. Spaceduck never bundles Marker — it calls marker_single as an external process.
How to use it
- Click the paperclip button in the chat UI (or drag and drop a PDF)
- The file is uploaded to the gateway
- Ask the assistant about the document — it automatically invokes
marker_scan
- The PDF is converted to markdown and the assistant can read and discuss it
Configuration
| Setting | Default | Description |
|---|
timeoutMs | 120,000 ms (2 min) | Subprocess timeout — large PDFs can be slow |
maxOutputChars | 100,000 | Output truncation limit |
pageRange | All pages | Convert specific pages only |
forceOcr | false | Force OCR even on text PDFs |
MARKER_USE_LLM=true as an environment variable to enable LLM-assisted conversion for higher-quality output on complex documents.
Troubleshooting
| Problem | Cause | Fix |
|---|
| Tool not registered | marker_single not on PATH | Run pip install marker-pdf and restart the gateway |
| Timeout on large PDFs | Default 2-minute timeout too short | Increase timeout or use pageRange to convert specific pages |
| Poor OCR quality | Default OCR settings | Try forceOcr: true or enable LLM-assisted conversion |
| Upload rejected | File isn’t a valid PDF | Spaceduck validates magic bytes — only real PDFs are accepted |
