My chatgpt api call fails when I try and test it. It works fine when initialising but it doesn’t work when I use it in my bubble app. I have attached a screenshot of what I have entered in the bubble api connector and my in app workflows.
You’ve got confused where you format as JSON safe. You format the keys (the message content) as JSON-safe. Not the entire JSON body.
{“key1”: “Example”} formatted as JSON-safe returns “{\”key1\”: \”Example\”} which is obviously not what you want.
So is it just the dynamic text I replace with Arbitrary text formatted as JSON safe?
Thanks for all your help so far!
I didn’t look at all the screen shots. I just read the error message which indicates it is not formatted as proper JSON.
I formatted everything as JSON safe and it returned the same error
Formatted everything JSON related as JSON safe - for clarity
The way you are formatting looks like it is incorrect.
You can verify this by taking the values generated through your setup and put them into a tool like JSON formatter (google search test JSON format) and see if there are errors in the formatting.
You have to save the “message content” as a JSON safe, not the complete string.
I guess the structure of the json should look like this:
[{"role": "user", "content": "Text1"},{"role": "assistant", "content": "output1"},{"role": "user", "content": "Text2"},{"role": "assistant", "content": "output2"},....]
But if you now do the json safe after the search the following happens with the sting above.
"[{"role\" \"user\", \"content\": \"Text1\"},{\"role\": \"assistant\", \"content\": \"output1\"},{\"role\": \"user\", \"content\": \"Text2\"},{\"role\": \"assistant\", \"content\": \"output2\"},....]"
The API cannot work with this and gives you the 400 error because of the incorrect structure.
Interesting, I am not a coder but I tried the workflows with no JSON safes before and it didn’t work either
Is this what you are suggesting?
It found no errors when I ran it
Can you save and show the input, which you send as prompt. A small example should be enough. That makes the analysis a bit easier 
