"You Must provide a model Parameter" trying to connect to OpenAI's API with the same http request that works in Postman

[tregan]

Hello, 

I am perplexed by this issue and am starting to wonder if my company's IT security policies are somehow blocking my entire json payload from making it to the Open AI API. Trying to follow this video to connect the Open API chat endpoint with a Teams channel and literally ready to pull my hair out over the number of failures on something that is supposed to be simple. I even ran this by a developer who could not see any issues and suggested that maybe this is just somehow getting blocked by security.

Interestingly, I am able to run the same http request to the same endpoint in Postman and I get the expected response from the Open AI chat endpoint, so I know the authorization, body and endpoint are good to go from a syntax perspective. Why do I keep getting this message about providing a model parameter when I have clearly already provided it in the json payload? What am I missing here? Is my company security/IT policy somehow blocking the content of my payload from actually reaching the OpenAI API? I know they do have some security protocols setup and I see warning messages when visiting the OpenAI site so clearly something is at least happening there, not sure if its related here though. Hoping someone can explain this to me and/or point me in the right direction, been struggling with this for days at this point!

Thanks in advance!

Successful request from Postman:

 Failed Flow in Power Automate with the same HTTP request - you can even see that Power Automate is picking up the content of my teams chat properly in the channel from the error message, but for some reason its not able to send even the model parameter to the Open AI API, which says to me that its somehow getting blocked due to the same thing working in Postman. Either that, or there is something else that Postman is doing in its header that I am not doing in mine? HALP! 🙂 

[Expiscornovus]

Hi @tregan,

I see you used HTML tags in the content property of the Power Automate HTTP action test. I believe that content property needs to be valid json. That HTML might turn it into invalid json. It is worth double checking and testing with regular text instead.

Can you try the same content like in your Postman test, without HTML. I would suggest to convert the Body Content to text with a HTML to text action before you use it in your HTTP action to see if that makes a difference?

[tregan]

Hey @Expiscornovus thanks for the quick reply and suggestion - looks like the step you suggested did actually make the conversion, I can see the content value is no longer in html in the payload from the error msg (though it did add some random \n chars at the end for some reason), but no dice - still the exact same error, like its not even getting to the content, it can't even get past model number, which is why I feel like the whole thing is just somehow getting blocked. Any other ideas given these results?

[wskinnermctc]

Looks like "application" is spelled incorrectly as "appliication"

I don't know if that is breaking it?

[Expiscornovus]

Hi @wskinnermctc,

Good catch, I did not notice that typo. Yes, that might be breaking it 🙂

[tregan]

Great catch - thanks so much, this resolved the connection issue!

I am now able to get Flowbot to respond in the channel...kind of.  It says the flow ran successfully, which is technically true - but the only thing now is...the reply output seems to be either blank or collapsed somehow in Teams! See screencap below - any ideas? Is this a formatting thing in my last step? I tried to click/expand these replies, no dice. Curious!

I am so close to the finish line here I can taste it - any additional insights on what may be happening here?

[wskinnermctc]

It probably has to do with the format of what is being returned.

I'm not an expert at this, but the Accept header in an HTTP request is telling the request what format of response you will accept. You currently have "/" which is any kind of MIME_type.

https://www.geeksforgeeks.org/http-headers-accept/ 

So whatever that means, you might know better than me. 

Then you are putting the response directly into a teams message. You appear to be selecting the first item value by doing all of the [0] stuff on the body.

I would put a Compose before the teams message and insert the same thing you are using in the teams message body. Check and see what the output looks like.

If the Compose is blank, then you are getting no values. If the Compose has text in the response then you can look and see if there is something about the format you need to adjust.

[tregan]

You guys are the f***ing best. Thanks for all the excellent help and suggestions! 

I ended up using a combo of things -  @wskinnermctc 's suggestion to fiddle around with the output formatting put me on the right track, as it was indeed generating a proper response in the http response body, so just had to get SOMETHING into Teams. First I got it to just return the entire JSON response to make sure it was actually sending something, which worked. Then, I used the "Parse Json" action in Power Automate after the HTTP response was returned, and then that enabled me to use dynamic content to directly map the content value from the response to the body of the msg flowbot returns in teams, and voilla - ChatGPT inside of MS Teams 🙂

Only tricky thing was I had to click "generate from sample" to get the schema populated (can't save without a schema value). I just pasted in the Json response from the Power Automate Successful Run Log, clicked "Done", and then it started running perfectly.

Thanks again!!! Case Closed, and interestingly...no security blocks in place to prevent this...yet lol.