its unlikely that it would happen with

[sonic chasm]

So for now, there is no way to upload a file directly via api?

[shut ridge]

nope, and for assistants the file search document explains it quite good why vector stores are used

[sonic chasm]

How does vision work? Does vision not accept files?

[shut ridge]

its takes image url or base64

https://platform.openai.com/docs/guides/vision/quick-start

[sonic chasm]

Thanks. One more thing - do you know how 4o handles PDFs and such in ChatGPT? Is it always converting to text or using OCR? Just trying to understand why ChatGPT enables file uploading but API doesn’t

[shut ridge]

chatgpt just uses similar vector store thing in the background as large files do not fit the context

[sonic chasm]

Thanks. You mentioned vision can handle image urls, can you upload image based PDF urls?

[shut ridge]

nope