PDF support

This would be great to have

And add support for .xlsx as well. A lot of table and structure is available in that format.

Yes, would love to put PDF and implement code from that paper.

I think this is crucial to implement because people often implement things based on specifications or written documentation of some sort and you want the model to have the right context.

👆🏻

We need this. Codex examines pdfs either via OCR extraction or it has to convert the pages to images. Its methodology is inconsistent and it takes a long time. I don't understand why it doesn't treat them like file inputs, which the OpenAI API supports:

https://platform.openai.com/docs/guides/pdf-files

+1

Would be great to have native support for PDFs

I second this

+1.

+1. All my docs are in pdf form and there are often issues

Oh man, this is mandatory. ChatGPT can do this, why not Codex (MacOS app)???

@pheuter @tomaszdudek7 @Chris-H0 @Suzuhana @mmmmie @eugene-kim @shera-amit and others You should try our fork of Codex that we developed for our internal R&D and recently open-sourced: https://github.com/Agents2AgentsAI/ata

It handles pdf files and urls natively, with extra tools to use for reading papers and doing research.

could you add this feature? thank you.

This feature is crucial for doing scientific work in Codex. I still struggle with PDFs in Codex. Being able to interpret diagrams and understand the sources I’m working with without preprocessing PDFs would be amazing.

It’s been almost a year, but there still hasn’t been any follow-up on this. PDF support in Codex is really needed.

The fact this issue is still open is astonishing.

I gave Codex a try today since there was a 1 month free trial. My project folder contained PDFs and I watched it try to pull the OCR first. Then it grabbed images for pages with handwritten text. The reasoning UI was so clunky as I watched it try to use tesseract and other tools to do this in the reasoning dialog. At the end, I didn't have confidence that it successfully got all the information from the PDFs in my project.

I paid $20 for Claude Code. The UI was cleaner, with just one line for a PDF read, and response quicker.

No regrets.

Tried Codex to code a firmware function to read IIC registers according to the IC spec which is in pdf. It's struggling with reading the simple pdf doc. It's unbelievable that such thing is not supported natively.