reMarkable PDF OCR Flow - Mistral Handwriting Extraction

Extract handwritten text from PDF files using Mistral's Pixtral vision model in n8n workflows.

Overview

This project provides ready-to-use code for extracting handwritten text from PDFs using Mistral AI's Pixtral model within n8n workflows. Perfect for processing handwritten notes from reMarkable tablets or any other PDF with handwriting.

Features

• ✍️ Handwriting OCR: Extracts handwritten text from PDF pages using Mistral Pixtral 12B
• 📄 Multi-page Support: Processes all pages in a PDF document
• 📝 Dual Output: Returns both JSON (structured data) and Markdown (formatted text)
• 🔄 n8n Integration: Designed specifically for n8n workflows
• 📧 Email Triggered: Works with incoming email attachments
• 🚀 Multiple Implementations: Choose between Python or JavaScript

Use Case

Workflow:

1. Email arrives with PDF attachment containing handwritten notes
2. n8n workflow is triggered
3. PDF is processed through Mistral Pixtral for OCR
4. Extracted text is formatted as JSON and Markdown
5. Result is forwarded to another service (e.g., note-taking app, CRM, etc.)

Prerequisites

1. Mistral API Key

Get your API key from Mistral AI Console

2. n8n Installation

• Self-hosted n8n (recommended) or n8n Cloud
• Version: 1.0.0 or higher

3. Dependencies

For Python Version:

pip install pdf2image requests Pillow

You may also need poppler-utils:

# Ubuntu/Debian
sudo apt-get install poppler-utils

# macOS
brew install poppler

# Docker
# Add to your n8n Dockerfile:
RUN apt-get update && apt-get install -y poppler-utils

For JavaScript Version:

npm install pdf-lib
# or in n8n, these are typically pre-installed

Installation

Option 1: Python Code Node (Recommended)

1. Copy the Python code: Use n8n_mistral_ocr.py
2. Set environment variable in n8n:
   • Go to Settings > Environments
   • Add: MISTRAL_API_KEY=your_api_key_here
3. Install dependencies in your n8n environment
4. Create workflow (see below)

Option 2: JavaScript Code Node

1. Copy the JavaScript code: Use n8n_simple_mistral_ocr.js
2. Set API key in the code or environment variable
3. Create workflow (see below)

n8n Workflow Setup

Basic Workflow Structure

[Email Trigger] → [Extract Attachment] → [Python/JS Code] → [Process Output] → [Send to Service]

Detailed Setup

1. Email Trigger Node

• Node Type: Email Trigger (IMAP)
• Configuration:
  • Set up your email account
  • Filter for emails with attachments
  • Enable "Download Attachments"

2. Code Node (Python or JavaScript)

• Node Type: Code
• Language: Python or JavaScript
• Code: Paste the content from n8n_mistral_ocr.py or n8n_simple_mistral_ocr.js

Environment Variable Configuration:

MISTRAL_API_KEY=your_mistral_api_key_here

3. Output Processing

The code returns:

JSON Output:

{
  "success": true,
  "totalPages": 2,
  "extractedText": "Full extracted text from all pages...",
  "markdown": "# Extracted Handwritten Text\n\n...",
  "pages": [
    {
      "page": 1,
      "text": "Text from page 1...",
      "dimensions": {
        "width": 1654,
        "height": 2339
      }
    }
  ],
  "timestamp": "2025-11-23T12:00:00.000Z",
  "model": "pixtral-12b-2409"
}

Markdown Output:

# Extracted Handwritten Text

## Page 1

[Handwritten text from page 1]

---

## Page 2

[Handwritten text from page 2]

---

*Extracted on: 2025-11-23T12:00:00.000Z*
*Model: Mistral Pixtral 12B*

4. Forward to Service

Use HTTP Request, Webhook, or service-specific nodes to send the extracted text to your destination.

Example destinations:

• Notion (create a new page)
• Airtable (add record)
• Google Docs (append to document)
• Slack (send message)
• Email (send formatted email)

Usage Examples

Example 1: Email → PDF OCR → Notion

[Email Trigger]
  ↓
[Code: Mistral OCR]
  ↓
[Notion: Create Page]
  - Title: Email subject
  - Content: {{ $json.markdown }}

Example 2: Email → PDF OCR → Slack

[Email Trigger]
  ↓
[Code: Mistral OCR]
  ↓
[Slack: Send Message]
  - Channel: #notes
  - Message:
    New handwritten note:
    {{ $json.extractedText }}

Example 3: Email → PDF OCR → Webhook

[Email Trigger]
  ↓
[Code: Mistral OCR]
  ↓
[HTTP Request]
  - Method: POST
  - URL: https://your-service.com/api/notes
  - Body: {{ $json }}

Configuration Options

Adjusting OCR Quality

In the code, you can modify:

Python:

# Line 34: Adjust DPI for quality vs speed
images = convert_from_bytes(pdf_bytes, dpi=300)  # Higher = better quality, slower

JavaScript:

// Adjust the prompt for specific extraction needs
text: 'Extract all handwritten text, including margin notes and annotations...'

Handling Multi-page PDFs

Both implementations automatically process all pages. The results include:

• Per-page extracted text
• Combined text from all pages
• Page numbers and dimensions

API Costs

Mistral Pixtral 12B Pricing (as of 2025):

• Input: ~$0.15 per 1M tokens
• Output: ~$0.45 per 1M tokens

Estimated cost per page:

• ~$0.001 - $0.005 per page (depending on image size and text length)

Troubleshooting

Error: "MISTRAL_API_KEY environment variable not set"

Solution: Set the environment variable in n8n settings or directly in the code.

Error: "No PDF binary data found"

Solution: Check that your email trigger is configured to download attachments. Verify the binary data path in the code matches your workflow.

Error: "Mistral API error: 401"

Solution: Invalid API key. Verify your Mistral API key is correct and active.

Error: "pdf2image dependencies not found"

Solution: Install poppler-utils on your system (see Prerequisites).

Poor OCR Quality

Solutions:

• Increase DPI in convert_from_bytes() (Python) from 300 to 400-600
• Ensure source PDF has good resolution
• Try adjusting the Mistral prompt to be more specific about what to extract

Timeout Errors

Solutions:

• Process large PDFs in batches
• Increase n8n workflow timeout settings
• Consider splitting multi-page PDFs

Advanced Usage

Custom Prompts

Modify the extraction prompt for specific use cases:

# For forms
"Extract all handwritten text from this form, organized by field labels."

# For mathematical notation
"Extract all handwritten mathematical expressions and equations, preserving notation."

# For multi-language
"Extract all handwritten text. This may contain multiple languages including English and German."

Batch Processing

To process multiple PDFs in one workflow, use n8n's loop or batch processing nodes before the code node.

Output Formatting

Customize the markdown format in the format_as_markdown() function:

def format_as_markdown(results):
    markdown = "# Meeting Notes\n\n"
    markdown += f"**Date**: {datetime.now().strftime('%Y-%m-%d')}\n\n"
    # ... add your custom formatting
    return markdown

Files in This Repository

• n8n_mistral_ocr.py - Python implementation (recommended)
• n8n_simple_mistral_ocr.js - Simplified JavaScript implementation
• n8n_mistral_pdf_ocr.js - Full-featured JavaScript implementation
• README.md - This file
• example_workflow.json - Example n8n workflow (coming soon)
• package.json - Node.js dependencies

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

License

MIT License - feel free to use in your projects!

Support

For issues and questions:

• Open an issue on GitHub
• Check n8n community forums
• Review Mistral AI documentation

Credits

• Mistral AI: For the Pixtral vision model
• n8n: For the workflow automation platform
• Inspired by various OCR and document processing workflows

Changelog

v1.0.0 (2025-11-23)

• Initial release
• Python and JavaScript implementations
• Support for multi-page PDFs
• JSON and Markdown output formats
• n8n workflow integration

---

Happy OCR'ing! ✍️→📄→✨