I’m having issues with the Assistants API. It keeps returning “Sorry, something went wrong” and “server_error”. Is there any way I can prevent this from happening or any workaround? I’ve read some posts on here that other people are running into the same issues but I haven’t seen any potential fixes / workarounds for it…
Yep same here, and yet https://status.openai.com/ shows all green. Extremely frustrating openai, at least some transparency or an acknowledgement of the problem would be better than nothing
edit: fixed for me by replacing restrictive api key with a full access api key.
Same here, with just Hello World! as one message, in a new thread and with an assistant with the instruction: "You’re helping debugging why the api calls to you are not working. Please share everything , all technical details you know about the interaction and the communication aspects. "
When the same thread is run in the playground, it runs fine.
I have it currently in php (using openai-php lib), curl and dart. No idea how to debug this. Tried different models as well.
[model] => gpt-3.5-turbo
[instructions] => You're helping debugging why the api calls to you are not working. Please share everything , all technical details you know about the interaction and the communication aspects.
[tools] => Array
(
)
[file_ids] => Array
(
)
[metadata] => Array
(
)
[usage] => Array
(
[prompt_tokens] => 0
[completion_tokens] => 0
[total_tokens] => 0
)
For anyone else experiencing this issue recently with Assistants and threads I was able to resolve by updating the API Token from a Project token to a User token.
Would be nice to get something a bit more meaningful back in the future so we don’t have to guess
Hello there people, I’m also still facing the same issue.
A thread got completely unusable. I’ve tried to disable AdBlocker in my browser or change my API token to an organization one, as mentioned above.
Deleting the first errored message, which sometimes solve the issue, also did not help this time.
I have a solution in place that if nothing works after some attempts, the system creates a new thread with the same messages. But unfortunately a thread can only be created with a maximum of 32 messages, which would be nice to know beforehand but is not in the docs.
A more meaningful error would be great, years as a dev, and I’m still not a very good clairvoyant.
The “server_error” with the Assistants API is likely a backend issue. You can implement retry logic, ensure you’re within API rate limits, and monitor the API status for updates. If it persists, contact the support team for assistance.