I’m experimenting with RAG on a large number of PDF files which are already hosted on the cloud.
To share these with my agents, currently it seems I need to use multiple instances of the PDFSearchTool and pass those to my agents. To do this, I need to download the PDF to my local file system.
I receive the files as a filestream, saving them to my local filesystem feels unnecessary. Is it possible to use a filestream directly or to be able to do so, would I be going down the path of a custom tool?
Hi @seanrobbins you would need to create a custom tool for sure due to the following
- The current implementation of our RAG Tools are tightly coupled with embedchain’s App class which expects file paths
- The PDFEmbedchainAdapter passes arguments directly to embedchain’s add method, which doesn’t support filestreams as far as I am aware.
Custom tool definitely the way to go though
1 Like