The Assistants API shut down yesterday and took our help bot with it.
I have the assistant ID but every read path 404s. Is there any remaining way
to retrieve a retired assistant’s instructions - a grace-period endpoint, a
dashboard export, a support request? Or is it deleted for good?
-– extra context —
The assistant was created by hand in the dashboard years ago by someone who
has since left, so there’s no copy in version control. The vector store is
untouched and readable, so retrieval is fine — it’s only the assistant
definition I’ve lost, and the only part I need is instructions.
What I tried, with a project-scoped key that works fine on other endpoints:
GET /v1/assistants/asst_… → 404
GET /v1/assistants?limit=100 → 404
(with and without “OpenAI-Beta: assistants=v2”)
(For whoever maintains the docs assistant: it told me existing Assistants
remain retrievable via GET /v1/assistants/{id} with the beta header, and
suggested converting each one into a Prompt via “Create prompt”. Neither
works post-shutdown, and it cost me a while to disprove.)
I’m afraid the end of Assistants API meant just that. The “grace period”:
On August 26th, 2025, we notified developers using the Assistants API of its deprecation and removal from the API one year later, on August 26, 2026.
404 with the “list assistants” method? The top level GET /v1/assistants?
Not even an error message body to return, just an invalid URL path.
OpenAI is also getting out of the business of storing “prompts” as server-side instructions when you do migrate to Responses API (or build your task with Chat Completions). That “migrate” link - now the destination of the Assistants’ playground dashboard.
For re-creating your instructions, you can visit the platform site’s “chat” (aka Playground), and use “generate prompt” to amplify the described task your developer message had performed for you, in conjunction with the file_search tool to reference your files or knowledge in a vector store. Then you can also choose an updated AI model, another part of an assistant ID.
There s an option to get it back if you have logs from requests/responses to OpenAI.
I had same problem, but I was looking through logs and find out that if I requested POST /v1/threads/ with assistant_id I had some responses like inproggress.
In this logs you will find sth like instructions where whole prompt of assistant’s is. Also model, temperature and anything else to get back your data to migrate to new version of API.
I tried this, and it didn’t work. To be blunt, OpenAI handled this very poorly. They should have sent users a link to backup all of their content before just deleting it. Yes, they communicated the deprecation of the API in advance, but they never specifically said “We’re going to delete your content.” The emails I got from them just mentioned deprecation of the API and linked to the guide, which talked about migrating, but also didn’t explicitly say that content would be deleted.
I would have blocked the actual endpoint a week before. It took us only some hours from OpenAI deleted the endpoint to our users complaining about it. So if they would have just made it 404 one week before, whilst still basically keeping the data and the admin UI, this would have been enough time to actually do anything about it.
Sure I knew about the API being deprecated, but I didn’t know we were using it. It turns out we sent tracing logs to Langfuse, so I found what I needed there.
It turns out we sent tracing logs to Langfuse, so I found what I needed there.