Hi, I am migrating and agent from code to the agent builder. It needs to process xls or csv files.
However in the preview when I upload one of this files the following error appears
Workflow failed: Invalid file data: ‘input_data.input[0].content[0].file_data’. Expected a base64-encoded data URL with a valid file MIME type (e.g. ‘data:text/plain;base64,SGVsbG8sIFdvcmxkIQ==’), but got unsupported MIME type ‘text/csv’. Please see https://platform.openai.com/docs/assistants/tools/file-search#supported-files for supported file types..
The link redirects to the Assistants API documentation, but I was under the impression that Agent Builder was built on top of the Responses API, not the Assistants API.
How can I enable my agent to process .csv or .xls files?
If this isn’t supported yet, is there an estimated timeline for when it will be?
Thanks in advance!
P.S. You can reproduce this issue by opening any of the templates and trying to upload a .csv or .xls file.
The vector store is a tool, with its own endpoint. As far as I recall it was created for the assistants api but support was later added to responses api, and the bits of documentation references remained there for some reason (probably legacy code in the api server).
Since its primary usage is for non structured text content (RAG), its support is limited. You could upload the csv as a txt file, but it wouldn’t make any difference since chunks are processed as text bits and not by lines or structured columns. You can read more about the updated info on this in the file search section.
A better (but a bit more expensive) alternative, is to use the code interpreter tool. Since it can run python code, it can properly load and understand what a csv/xlsx file is, and even make calculations.
As you or working on a similar problem may I ask if you have got the File Search node to work? My vector stores work when they are tools in Agent nodes but trying to access them through a File Search node just returns an error with no information in it on what went wrong.
Umm, I think you might be mixing up a few concepts.
Uploading files to an LLM is independent from using RAG — you can, and should, be able to process uploaded files in different ways, with RAG being just one of them.
What I’m trying to replicate is a setup where you have a container (or workspace) where users can upload files, and a Code Interpreter tool that processes them.
The issue is that when you use the UI to upload a file that isn’t one of the supported formats listed here — supported file types — the error appears.
Well, I don’t make the concepts. If the API defines what it accepts, how is that an error if you send an unsupported file? The only way I see is that you would have to change the API itself, not a setting.
But of course you are free to have your own interpretation.
I get your point, but that’s not really what I’m arguing.
What I’m trying to understand is how to replicate the same file-handling behavior that already works in Responses — which also exists in Assistants — and both function properly at the API level.
The issue seems specific to the Agent Builder: at the API level, you can already upload files to containers, threads, and other endpoints without any problem — but the Builder currently throws an error when you try to upload formats that aren’t listed as “supported.”
So my question is whether this is a bug, or if file uploads for these formats simply aren’t supported yet in Agent Builder.
It just feels inconsistent between the API and the Builder: ideally, the upload should succeed, and the error should only happen later during processing (depending on which tool is enabled — File Search, Code Interpreter, etc.).
I see. You are referring to the file attach button right? I think that part is still a work in progress.
When using code you might have access to directly refer to files, but apparently not yet in the agent builder / chatkit (only permanent attachments in the UI for now).
Yep, I am referring to the file attach button. Okey so it is not yet completed?. I will have to pospone the migration until this feature is fully added then. I need to manage and work with different types of files and tools.
I hope the OpenAI team puts resources on improving the agent builder, I think it has tons of potencial.
