n8n RAG: Adding PDF & DOCX for Better Document Processing

Recently, I came across a great YouTube tutorial from Cole Medin on building a RAG system with n8n, Google Drive, OpenAI, and Supabase. The workflow was solid for Google Docs, but I quickly realized it wouldn’t work with my existing Word documents or PDFs. Always up for a challenge, I decided to enhance Cole’s original workflow to support PDF and DOCX documents.

The Problem: PDFs and Word Docs Are Binary Blobs

If you try to use the original workflow with PDF or DOCX files, you’ll end up with something like this in your database:

PK DciZ word/numbering.xmlí�[nã6 †WÐ= zOt±,;Æ8ƒ` )Z E�N @K´MD" R²"n¡ }k_»¶¤ÔÕ·±*y¢zRÿOŠ ùÿä¡ŽùÑ�!¿{ÿ …ƒ5•Š >3ì[Ë Pî‹€ñåÌøåÓãÍÄ ¨„ð€„‚ÓñB•ñþþ›w›)O£9•ºß@[p5�ü±J'xjšÊ_Ñˆ¨[ S BF$Ñ/åÒŒˆ|Jã *D1IØœ…,y1

Yikes! That’s raw binary data, not the text content that our RAG system needs. The AI can’t extract any meaningful information from this jumble.

The Solution: File Type Detection and Conversion

To fix this, I needed to add a few key improvements to the workflow:

Update the Set File ID node
Add a Switch node to detect file types
Implement ConvertAPI integration to transform PDFs/DOCXs to plain text
Use a Code node to ensure filenames have proper extensions

Here’s how I implemented each part:

1. Add mime_type to the Set File ID Node

First, I updated the Set File ID node to set the mime_type. This is used to drive the logic in the new Switch node below.

2. Detecting File Types with the Switch Node

Next, I added a Switch node after downloading the file from Google Drive. This node checks the MIME type (mime_type) of the file and routes it accordingly:

application/pdf → PDF conversion path
application/vnd.openxmlformats-officedocument.wordprocessingml.document → DOCX conversion path
application/vnd.google-apps.document → Google Docs path (the original flow)

This routing ensures each file type gets handled appropriately. If you had more file types that you wanted to support, you could easily add them here.

3. Converting Files with ConvertAPI

For both PDF and DOCX paths, I integrated with ConvertAPI, a service that handles file conversions elegantly. You’ll need to:

Sign up for a free ConvertAPI account (you get 250 free conversions)
Copy your secret key from the dashboard
Add the ConvertAPI (HTTP Request) nodes as shown in my workflow
Configure your Credential for ConvertAPI with the secret from #2

For PDFs:

POST https://v2.convertapi.com/convert/pdf/to/txt

For DOCX:

POST https://v2.convertapi.com/convert/docx/to/txt

The beauty of this approach is that ConvertAPI handles all the complex conversion logic, including OCR for scanned PDFs if needed.

4. Adding File Extensions with a Code Node

I discovered a quirk: ConvertAPI expects files to have proper extensions in their filenames. Since my Google Docs file names did not include a file extension, I added a Code node with this simple JavaScript:

$input.item.binary.data.fileName = `${$input.item.binary.data.fileName}.${$input.item.binary.data.fileExtension}`; 
return $input.item;

This ensures the filename has an extension before we send it to ConvertAPI. Ideally, you’d check to see if a file extension exists before adding one, but I know my Google Docs did not include one.

The Complete Workflow

After these modifications, the workflow now does the following:

Detects when a file is created or updated in Google Drive
Identifies the file type
Converts PDFs and DOCX files to plain text
Extracts the text content
Stores it in Supabase with proper embeddings
Makes it available for the RAG AI agent to query

Check out the full workflow in the screenshot:

Why This Matters

Having your RAG system work with PDFs and Word documents is game-changing for several reasons:

Most business knowledge is locked in these formats
Legacy documents don’t need to be converted manually
You can ingest reports, contracts, manuals, and more
The system becomes practical for real-world use

Additional Considerations

While implementing this enhanced RAG workflow with PDF and DOCX support, here are some important factors to keep in mind:

Cost Considerations

ConvertAPI Pricing: The free tier includes 250 conversions, which is perfect for testing and small implementations. Beyond that, pricing starts at $30/month for 1,000 conversions and scales up based on volume.
OpenAI Costs: Remember that each document processed generates embedding costs. For large document collections, these can add up quickly. Consider batching updates or implementing selective processing.
Storage Costs: As your document database grows, Supabase storage and database costs may increase. Monitor your usage to avoid unexpected bills.

Technical Limitations

Complex Document Formatting: While this solution works well for text-focused documents, highly formatted content with tables, charts, or complex layouts may lose structural information during conversion.
Image-Heavy PDFs: If your PDFs primarily contain images or scanned text, ConvertAPI’s OCR capabilities will help, but may not be perfect. Consider testing with your specific documents.
Large Files: Very large documents (over 25MB) might encounter timeout issues, and you may need to implement chunking strategies.
Language Support: Non-English documents may have varying degrees of success depending on character sets and the OCR capabilities of ConvertAPI.

Security Considerations

Data Privacy: This workflow sends document content to external services (ConvertAPI and OpenAI). Ensure this complies with your organization’s data policies.
API Key Security: Store your ConvertAPI and OpenAI credentials securely. In n8n, use the credentials vault rather than hardcoding keys.
Access Controls: Set appropriate permissions in Google Drive and Supabase to limit who can access the source documents and extracted content.
Data Retention: Consider how long you need to keep both the original documents and their processed text. Implement a cleanup strategy for outdated material.

By addressing these considerations upfront, you’ll build a more robust, scalable, and secure document processing workflow that can handle real-world challenges.

Getting Started Yourself

If you want to implement this enhanced workflow:

You can download my modified workflow JSON from GitHub
Import it into your n8n instance
Set up your credentials for Google Drive, OpenAI, Supabase, and ConvertAPI
Configure the folder path in Google Drive where your documents live
Start adding files and chatting with your data!

If you’re new to n8n, I highly recommend using n8n Cloud for the easiest setup. Their free trial is perfect for testing, and you can upgrade as your needs grow.

I’d love to hear how you adapt this approach for your own needs! Have you found ways to handle other file types? Created interesting agent prompts? Drop me a comment below and let me know!