AI assistants file limits

I’m working on setting up a super flexible AI buddy that can tweak its hosted files on the fly. The idea is to make a smart system that keeps getting better based on how it performs and what my company needs. When I see that my AI sidekick needs more training, I’m planning to smoothly add more info to make it even better.

Right now, openai has a rule that says we can only have up to 20 files. This rule is a bit of a headache, especially since I’m planning to fine-tune the AI’s performance regularly. Because this project is always changing and needs constant upgrades, it’s important to figure out how to handle this file limit. Finding the right balance between making the AI awesome and dealing with the 20-file cap takes some careful thinking and considering different options to make sure this whole thing is a success.

2 Likes

I like your idea!

It’s 20 per Assistant - but the documents themselves don’t really have a limit.

1 Like

Just as a heads up, the assistants API has limits for file size at 512MB / 2 million tokens per file.

2 Likes

With 20 x 512mb ie 10gb and 40 m tokens I would hope that is not a problem for him :slight_smile:

1 Like

I’d like to delve deeper into this issue.
The concern doesn’t arise from storage capacity constraints but rather from the inability to integrate new information or fine-tune the model using existing files.
In essence, the challenge surfaces when attempting to input additional data, directly tied to the limit on the total number of usable files.
Editing existing files is not possible, only adding new ones.
Therefore, even if the initial 20 files are only 1kb each, a problem arises.

Ok - but you can change the files that are attached to an Assistant. You can have hundreds of files uploaded - and you can remove / add what files are connected to an Assistant. So your ‘retune by changing the dataset’ is certainly possible?

1 Like

To change files you’d need to know which ones to use, however there is no retrieval for stored files, only for the attached ones if I understood this assistants API correctly

You can list the Assistant files:
https://platform.openai.com/docs/api-reference/assistants/listAssistantFiles

You can also list ‘all’ files
https://platform.openai.com/docs/api-reference/files/list

The only thing that I would love to have is a way of knowing from the ‘all files’ if a file is attached to either an Assistant or Thread. Or if it is an ‘orphan’.
Would love a ‘delete all orphans’ options too :slight_smile:

So to confirm is it 20 files total or 20 per assistant? I had 1 assistant with 20 files then started a new assistant and could not attach a file, until I deleted 1 from the first assistant, so wanted to clarify.

Related question, does the limit increase if you change your account to a Team account?

its 20 per assistant from this page: https://platform.openai.com/docs/assistants/how-it-works/creating-assistants

" You can attach a maximum of 20 files per Assistant, and they can be at most 512 MB each. The size of all the files uploaded by your organization should not exceed 100 GB. You can request an increase in this storage limit using our help center. In addition to the 512 MB file size limit, each file can only contain 2,000,000 tokens. Assistant or Message creation will fail if any attached files exceed the token limit."

1 Like