User input file as persistent data source

Looking for some help. I’ve build an agent for an author to critique a book manuscript and need the agent to use the entire manuscript (150,000 words) for context and continuity when looking at specific chapters.

This means that I need the user to be able to use a user input block to upload updated versions of the whole manuscript and have that chunked and vectorized as source data. I don’t see a way to do this.

I have a way for the user to load the manuscript at runtime and then use a function and external embedding service to chunk/vectorize it but that’s less than ideal. It means that every time the user interacts with the agent he has to upload the entire file.

I’m using Gemini 2.5 for a giant context window of 1m tokens, but it seems like the current set up is problematic.

I’d love for the file to be a chunked and vectored source data file and then have the user use the user input block to update it as needed.

Ideas/guidance?

Thanks!

Hi @jebdasteel,

We don’t currently support dynamically updating Data Sources, but thanks for sharing your use case.

I’ve moved your post to the Feature Requests category.

Thanks, Alex! I’d be very excited to see that.