Hi, I previously created a workflow using a Webhook (POST) and I’m trying to upload documents for processing. However, I’m encountering some limitations.
My documents include multiple file types such as PDF, Word, XLSX, and ZIP. I understand that ZIP files may not be supported, but I noticed that the workflow can only handle up to 2 attachments per upload. If I upload more than that in one request, the process fails.
However, if I upload the files two at a time, or combine multiple documents into a single file, there is no issue.
May I check if there is any known limitation on:
-
Number of attachments per webhook request
-
File size or file-type constraints
-
Payload or memory limits that could cause this behavior
My objective is to allow n8n to ingest and learn from the uploaded documents, and then automatically respond to client enquiries from the website based on those documents.
Appreciate your clarification. Thank you.
There’s no hard 2-attachment limit in n8n. The real constraint is total payload size, which defaults to 16MB for all files combined. If you’re self-hosting, increase it via the N8N_PAYLOAD_SIZE_MAX environment variable (and adjust your reverse proxy like NGINX). On n8n Cloud, consider uploading large batches to external storage first and sending URLs to your webhook instead.
For file types: PDF, DOCX, and XLSX work fine as binaries. ZIP files aren’t auto-extracted - you’ll need to unzip them before upload or add an unzip step in your workflow.
To handle multiple files, configure your Webhook node’s Binary Property (e.g., name it “data”) to accept multipart/form-data. Files arrive as file0, file1, etc. Use a Code node or Split In Batches to process each file individually.
For your knowledge base workflow, you’ll need to extract text from those documents before feeding to AI. I use Ainoflow Convert for multi-format OCR and Ainoflow Storage with Ainoflow Files, to persist the extracted content as your knowledge base. Then you need vector database to build your rag, most common choice is Supabase, which combines relational data and vector. When do search find references in vector DB. Load files if needed then feed found content to AI agent to build answers for user queries.
1 Like