I use the Assistant in the retrieval mode, with a couple of PDF files uploaded. When I use the API to run a thread, I will sometimes get the answer back, including the error: “Files you’ve uploaded are not accessible with the myfiles_browser tool.”
If I test the Assistant in the Playground, this error doesn’t occur, or it didn’t happen while I was testing.
This issue doesn’t occur all the time, but it happens often enough to be a problem.
You may find this post interesting Myfiles_browser tool is not operational for these files
It seems to be an ongoing issue for some.
Did you find it charges you every time it errors? I got charged for uploading a 11 page PDF that it ultimately bombed on.
This is the way I look at it, I’m working on cutting edge tech that is in beta, yes I am adding value to OpenAI but I am also getting a lot back and early positioning in the market to be an “expert” pick the superlative of your choice. So I am willing to take a few $ hit here and there when things break, no biggie, I get it, cost of contacting support and all that… not worth the trouble. If I were to get charged $100 or $5000 for an issue, sure I’m going to park my chair at the steps of help.openai.com and wait for a refund.
It’s frustrating when beta stuff bombs out and is not perfect, but, as I say, were getting a lot of useful early adopter knowledge and I’m willing to work in that space with an understanding of the nature of that space.
You are not alone. I think the Files situation is very messy. I have decided to upload via the API, and wait some time before using them. And when I do a thread Run, I explicily tell the assistant to use the files.
Interesting. How do you tell the assitant to use your file? Over the prompt or over the api call with file_ids? or both?
I have uploaded the files using the web UI at https://platform.openai.com/files and specified the file IDs in the message submitted with the thread.