Assistant API v Assistant Playground - very different output

I’ve got a simple assistant that takes an image of a customer order (and extracted text from the order from extract) and processes into JSON.

It works well however I’ve now added a file with all our product codes for the assistant to check and confirm the codes match / find the best match.

It works perfectly in the playground, but when I use the same assistant via the API, it doesn’t appear to me using the retrieval and checking the file.

Out of the numerous tests via API it has completed the task successfully once - so I guess it does have access to it.

I’m using the same prompt in both the playground and API.

Any ideas?

CHeck the following

  • Are you using the same temperature settings?
  • Are you using the latest assistant api?
  • Are you passing the file for each runs?
  • I am not sending temp settings, I assume these will be the same as the assistant set in OpenAI platform?
  • yes I believe it’s the latest assistant api
  • no I am not passing the file. It’s saved as a vector store under file search - do I need to pass it via the api each time?