Under one assistant, is context shared between threads?

I’m creating an app using the Assistant’s API and multiple users will use that app. I am thinking of using 1 “master” assistant with each user having their own thread. For privacy and UX concerns, it is critical that context is NOT shared between threads, that is, each thread is completely isolated such that stuff that a user writes in their own thread isn’t shared with other threads (the threads of other users). If this happened, user data would be mixed and this is NOT acceptable.

If I want to make sure that only user A’s messages are used as context for User A and only user B’s messages used as context for user B, do I need to create dedicated thread(s) for each user’s messages or dedicated assistants for each user? Obviously the latter would be a lot more expensive if I’m not mistaken.

So my question is, is it correct that context is limited to each thread? It’s important to note that users will be uploading individual files to their own thread, like (anonymised) blood test results, and other personal medical files, which is why it’s important this information isn’t shared between threads.

I’ve heard that multiple threads can receive knowledge from an Assistant, which makes it sound like all threads could be sharing the Assistant’s context? Is this true? I’m a bit confused.

Much thanks.

1 Like

Pretty sure that context is not shared between threads… It just uses whatever it has in the thread… You might want to test, though…


Would love if someone could chime in with a definitive answer. The privacy per user is very important so it’s critical I get a confident answer.


Threads are independent. The AI is only informed by conversations and functions that happened before by that session.

However, personal and health information is inherently troublesome. Files and “uploading” means uploading to an organization’s common file store. Simply then attach to another assistant and ask about their contents. A little glitch in your database and files are disassociated from the correct user? The same problems of OpenAI a year ago (“oops, sorry that you could see other people’s conversations”) cannot be ruled out. One organization member password breach and you are notifying everybody under disclosure laws.

And then you are asking a word generator with randomness of word selection to provide that information and give analysis…

1 Like

What do you mean by this? By “session” do you mean thread?

What do you mean by this? I understand you’re saying use a different assistant with that thread that contains the sensitive file. Are you saying if I make this mistake it’s problematic? Not sure what you’re driving home here.