Confusion around OCR (Mistral) as upload method

[xoChrisCo]

Experimenting with file uploads, I am seeing far better results running Mistral OCR on PDF documents, than when I'm using File Search. Especially for large PDFs or complex PDFs with charts, tables and more.

I see that the documentation says that this feature is currently only available through an agent capability. But a bit further down on the same documentation page, it says that

LibreChat provides two main ways to use OCR functionality:

1. Upload as Text in Chat
2. File Context for Agents

I can only get the 📎 Attach menu to trigger if I enable Code Interpreter. If not, it will just directly show the file picker dialogue window.

I am a little confused, and, I guess I have three questions:

1. Is the only available through an agent capability line outdated documentation?
2. Is the "Upload as text" choice in the 📎 Attach menu the same as the OCR capability in the agent?
3. Why am I only seeing the 📎 Attach menu when Code Interpreter is activated? Is this working as intended?

[danny-avila]

Hi @xoChrisCo 👋

Great questions, I can answer them.

Experimenting with file uploads, I am seeing far better results running Mistral OCR on PDF documents, than when I'm using File Search. Especially for large PDFs or complex PDFs with charts, tables and more.

Yes, using the "as Text" option is best when you need the full context of the file, and/or cutting-edge parsing of tables, etc. I have some ideas on how to improve file search and potentially benefiting from the same type of pluggable OCR. Will be getting help on this soon.

Is the only available through an agent capability line outdated documentation?
Why am I only seeing the 📎 Attach menu when Code Interpreter is activated? Is this working as intended?

Sort of. Technically still true, since enabling "Code Interpreter", also now "Search", activates under-the-hood use of "agents." This is why you see it only with these options selected.

Is the "Upload as text" choice in the 📎 Attach menu the same as the OCR capability in the agent?

Yes. I do want improvements that we're working on soon as some files don't need any OCR as they are text-based.

---

I do apologize for the confusion, as I will make more "under-the-hood" transitions to using Agents so that the menu is more intuitive.

[xoChrisCo]

Thank you so much @danny-avila ! That cleared all confusions I had!

Loving the software!