Internal error while accessing files

Good afternoon. I work with the Chart Assistant API via platform.openai. and the Make service. The point is that our company has a Telegram bot that works via the Make service and we use the OpenAI module (GPT Assistant). Everything was fine until about May 6-7, then the bot started to say that it didn’t know the answer to any request (even the simplest one, which it always answered). I went further, turned on the 4o model (to be sure; before that there was 4o-mini) and here’s the most interesting thing - the API response is: “Unfortunately, I can’t perform a search at the moment due to an internal error. Please try reuploading the files or give me more information so that I can help.” or “Unfortunately, an error occurred while trying to search for information. Try asking a different question or rephrase it. Thank you.” I realized that this is not a Make service error, but an OpenAI error. Studying the problem, I saw that people periodically encounter such an error periodically and it is not clear what to do with it.

Similar issues:
For some reason I can’t attach links to them…

I tried creating new assistants, new storages, but it still can’t extract information from files. This is critical for our company.

I ask for your help or advice in solving this problem. All the options I have tried do not work. I also wrote in the chat on the help.open.ai website, but there is no answer…

1 Like

When using gpt-4o mini, the name of the internal search tool is myfiles_browser, so if you are telling the AI anything different, you are going to break it.

I tried to get an error. But did not. Only frustration that there is no way to attach a prior file ID to a vector store in the platform site’s UI.

Also get a nice citation in a JSON…

The next step into self-help on this deprecation-by-indifference endpoint, increasingly: issue a new API key. Create a new project. Ensure all data is in that scope. Test between files purposes [assistants | user_data]. Send the correct org and project headers in the API request also, or ensure they are never there in env variables. Try in the playground. Ask an AI assistant, one you’ve given “you are a developer’s debugger” instructions, what’s the called tool and the returned error in the thread from the tool use attempt.

You’ll see in the screenshot I have a bunch more instructions to overcome “files were uploaded by a user” as how file search behaves.

Thank you for your answer! I will add that everything works correctly in Playground, but it is the use of the API through, let’s say, an external service that causes an error. Will this option that you suggested help me?

You’ll need to take a close look at the API parameters being sent, and make your own API calls locally. It may be that something that you thought was set being overridden in the run call, such as someone mistakenly sending “None” or null for a run tool or the tool_choice parameter.