Skip to main content
All CollectionsKeymate Memory Using PDFs with Keymate Memory
Should I load PDF to my Keymate Memory as a file or paste it as an inline prompt text?
Should I load PDF to my Keymate Memory as a file or paste it as an inline prompt text?
Ozgur Ozkan avatar
Written by Ozgur Ozkan
Updated over a week ago

Using PDFs as files or copy-paste with Keymate

At Keymate it had been a long time since we launched PDF feature but we realized there are misconceptions about PDF uploads and this article aims to help you about frequently asked questions.

Misconception 1:
ChatGPT can read 100 pages PDF in one go and answer.

Reality 1:

ChatGPT plus can read max ~5K words (8K context window) in one go even you had a Chat history ChatGPT forgets the history to continue working.

Misconception 2:

ChatGPT can know about your uploaded files.


Reality 2:
PDF plugins develop external web apps and chatGPT can only learn which pdfs are uploaded before by only querying a specific plugin endpoint. ChatGPT doesn't know about your uploaded pdfs until it runs the plugin. So mentioning or querying an uploaded pdf doesn't actually make sense. (Although ChatGPT still can use your uploaded pdf file name and might decide to query the plugin with this information )
If plugin doesn't provide that functionality it doesn't answer.


Misconception 3:
File related operations can only be done by uploading files.

Reality 3:
Input window management is one of the most expensive and hardest thing today. As human you are a tool for AI and you fill the input window. The way you use the ChatGPT and selected plugin(s) affect how you fill that input window.

Most PDF use cases can be solved with higher input window and manually constructed prompt with copied and pasted text chunks from PDF files.

See this example of copying and pasting PDF text: https://webapp.server.searchweb.keymate.ai/share/pREnBSG

This originated from a real story. A lovely realtor user tried to create a market overview from PDF data (which had tables and text etc). Use case is PDF related because text is there, this is understandable but most PDF files are above input context windows.

At Keymate.AI we allow 100 page longs PDFs to be used in personal knowledge base training. Keymate allows you to train your long term memory aka Personal Knowledge Base with PDFs you upload, but it doesn't offer hybrid search which is direct text search. There are great PDF plugins that do direct text search. Keymate only does semantic search today but we are preparing our hybrid text search solution, which will not only use PDFs it will use internet data, your long term memory in hybrid text search ( semantic search + direct text search on multiple sources).

Did this answer your question?