I am building a Custom GPT that needs to fetch the report of a client, once the client provides the contract number. To do this, I created an action that communicates with our API in our server and retrieve a Shareable link from SharePoint after the client provides the contract number. The problem comes when the GPT needs to ingest the content of the PDF using the Shareable link provided by the server. The GPT successfully provides the link to the user to download the file, but it doesn’t ingest the content inside of the PDF. It says that due to security reasons it cannot access to externals URL’s. I am not interested to make that the clients download the link and then upload it in the GPT because that is too cumbersome for our clients using their mobile phones. Any idea how to solve this problem?
It’s very surprising the process of linking Custom GPT:s and SharePoint documents isn’t more straightforward, given the partnership of OpenAI and Microsoft.
No. but my feeling is that it is not allowed by OpenAI. You can download and reintroduce the PDF but not directly load the PDF from the source to make the GPT report the results. I work around is to have the API connected to an LLM that reads an summarizes the text and send the results of the summarization directly as API response. Of course that defeats the purpose of using the power of GPT4 in the first place. AS mention before, other workaround is ask the user to reload the download document, which is very cumbersome if you have users that don’t know how to safe files in their phones and retrieve them to reintroduce them to the chat. Honestly, We got tired of trying and now we are using an open source model for it.