Microsoft 365 Community Power Platform Connectors, 8 Jun 2023

Previous Meeting

Next Meeting

⏯

youtube image

►

From YouTube: Power Platform Custom Connector for ChatGPT

Description

Create a connector and use it in a Power App in less than 15 minutes. In this 12-minute developer-focused demo, Samir Daoudi delivers a two-phase demo. Phase 1 - Create and configure a custom Power Platform connector (open template, set authentication, add OpenAPI key, define an action and test) and in Phase 2 - use connector in a Power App. Create a simple canvas app, add connector, add call to access OpenAI via connector. Template fields and terminology explained throughout demo.

This PnP Community demo is taken from the Microsoft 365 & Power Platform Development Community call recorded on June 8, 2023.

Demo Presenter
• Samir Daoudi (LogiSam)

Community Call Conversation
• aka.ms/June8-Demo1

Supporting materials
• Article - Power Platform – ChatGPT Custom Connector | https://daoudisamir.com/power-platform-chatgpt-custom-connector/

Learn more
• Microsoft 365 Unified Sample gallery - https://aka.ms/community/samples
• Microsoft 365 Platform Community in YouTube - https://aka.ms/community/videos
• Microsoft 365 Platform Community - http://aka.ms/community/home

A

B

But it's going to be mainly a demo uh basically and today we're going to talk about how we can create an open AI to rgbt, connector and and consume that in Power Platform, so the demo will be in two phases: the first phase, we're going to create the connector, configure it and do everything with the API key and so on. The second part will be to use this connector, either in powerapps or power automate a little bit about myself.

B

My name is Samuel Dodie I'm, Paul platform, uh architect I've been working with Microsoft Technologies since forever. Since I remember I blog frequently, you can hit me on Twitter LinkedIn or you can follow my my blog I'm sure you're, not here for this. So let me switch to more important stuffs as I said, um we'll be doing uh two things today. So the first thing is to create the custom connector. In order to do that, we will go to make.powerapp.com. We are all familiar with that.

B

If you don't see the custom connectors here, you can go to discover more and it should be somewhere here so custom connector. Let me pin this here and we get started so at the moment in this sentence. I don't have any custom. Connector I'm gonna get started with a new custom connector from blank I'm, going to call this demo Dot PT next right. So um when it comes to creating the custom connector, you will be going through a wizard, so I'm gonna quickly go through this the so you can select an icon for your connector.

B

The most important part is to uh to define the API endpoint. The way you can authenticate with this connector and some details about the request, body structure and so on. So the first thing uh here is: is the host the the API endpoint, and this is um API dot. Openai I have blog posts about um uh how to create this custom connector with screenshots step by step instructions. So I'm gonna try to go quickly through uh the steps to create this connector.

B

Next, um it's the way you're going to authenticate with the with this open, AI API and the preferred way is to use an API key. The parameter will be and we call it parameter label. Let me call it API key and the parameter name, which is quite important. You can type that it's authorization. We will be passing the authorization in the header, so I'm, going to leave that as it is uh now next is defining the different steps, uh the different actions for my connector.

B

So let me zoom out a little bit so I'm going to have just one action for uh for today's demo and I'm gonna call this um yeah ask chart GPT description, more user-friendly way. uh Ask anything, and this is the operation I did. This is the the action that would be uh calling in powerapps or power automate, and let's call this ask okay.

B

Now, the way our request will be structured, um I'm gonna import from sample, so this will be a post request in terms of URL. um This keeps changing to be honest with you, so you can go back to um the open, API reference uh site where you can look today, we'll be looking only at the completion. So basically the completion uh endpoints will be this one I'm just gonna copy it.

B

As I said, I have blog posts where I'm going slowly step by step, with screenshots and so on, but uh for today, I'm just gonna those steps, so the header structure, it's just depending on the content type we'll be talking uh with the Json formats and um in terms of the the request body, um open, AI uh API- expect at least those five parameters. So basically what we have here, we have the model, which is the language that we'll be interacting with. uh In this case, it's text DaVinci zero, zero, three uh prompt.

B

This is the question or the uh what should be asking in charge equity in this case I just predefined this for test Purpose with uh what's SharePoint metadata, you have three other important parameters that you need to set max tokens. As the number of tokens you will be you'll be using.

B

um You can create a free account with the open AI you get I think 18 credits per month that you can use for um all the instructions that you'll be making with openai uh temperature is a parameter which goes from zero to two, where zero is very precise. Answer two can be very vague and the N is the number of responses that you are expecting from uh from chargpt.

B

In our case, we expect to send uh to last question get one response, so it's going to be always set to one so I'm gonna import my request structure at this stage. I can just go ahead and create the connector so I'm going to save the chains.

B

It may take probably minutes or two to save and we are ready to start testing.

B

Perfect so I I switched to the test step in order to test this connector um you're gonna have to create a new connection, um and this is where you will have to go back to open AI, create your account. um Let me find the right URL for that.

B

As I said, you can have a free account once you, you create your account. You go to API keys and you create a new new secret key, so I'm, basically I'm going to say M365 demo- and this is my key- so I will copy the the key from here. The the format of the API key has to be the following, so you're going to prefix it with better space and then the key that you have just generated right so now, I have created the connector. I will go back to my uh the connection.

B

Sorry I will go back to the connector and go back to test here. Is my connection. um I have two options: I can either fill those parameters or I can use the same request body that I use it to generate the model, which is this one so again, I'm gonna say the model. Is this prompt? This isn't my question and I'm gonna test this connector? If everything goes one, it should return response. 200, and here we go so SharePoint metadata is information blah blah blah, that's fine!

B

Now what we need to do is to copy the um the response body. I will go back to definition. The reason why I'm doing this is because I know I will always be using those parameters. uh I don't need the users to keep feeding me those those details, so um what I will do is I'm going to set some default values here. So, in my request, um oh sorry in the response section, I'm gonna import from uh from sample.

B

um Basically I'm saying this is the the response that you are expecting, so it says: okay, these are the properties that I expect from uh under the response. Again, I will go back to the request and I'm going to set some default values, so content type uh as I said: it's application Json. So if I can copy that.

B

This is required and it's internal. It doesn't mean that the user will not be able to to change it again. I'm going to go to the request body here, I'm going to set some default values, the max tokens. We said it's a thousand, it is a required parameter and it's internal uh model is text. Davinci.

B

Again, it's Acquired, and it's internal so same here and has one and the temperature repeat: zero.

B

Okay, perfect! So visit our connector, it's ready to be used. We can either use it in power, automate powerapps, but for today's demo, I'm gonna use powerapps um I kind of cheated a little bit and I prepared the powerapps canvas application. Here there is nothing it's just the layout, so here we have one screen with the two part cell. You have one text box where the user can ask questions, you have the image of the logged in user. A nice icon here for the to the bot and the response will be will be added here.

B

There is no code yet in this in this application. So the first thing we need to do. Let me just refresh this.

B

Would be to add my custom connector in this uh in this power up so I'll go to data.

B

Just remember my connected demo, the motor to PT, okay, that is my connector all right now, when it comes to uh to using this this connector. uh As I said, we have a question that will be coming from this text. Query and the response will be set to text response. uh I said: I set the default value for the response to be coming from this variable. So basically the button will set the value of this variable.

B

So I'm going to say about a set of R response to my connector name, don't ask ask, is the um if you remember is the operation ID that we said, and then you have the default value. So, um as you can see it's given me what what the the connector is expecting? You can go with the with the default values.

B

So this is the top Max tokens thousand temperature zero and one and the most important is the prompt, which is a question that will be asking uh chat directly. So this will be coming from text. Query failed right. I will quickly uh back to to my connector. Just show you to show you how the response is structured.

B

B

Okay, what's going on here,.

B

um Sorry, I, don't know what's going on here with this connector, but um so, basically, even even though we said uh we want one response with the unparameter is going to send us. uh It's gonna return an array of one item so basically I'm gonna take the First Response um that will be receiving dot choices. So it's coming from uh from the.

B

Yeah, so the response will contain different parameters and we are only interested in the text value um that we got. What I've done here is I've set to this variable. The first response that I'm getting from this custom connector call.

B

um They said we have the choices array which will give us all the the responses and from that response, I'm taking the text, uh the text value right. Okay, let's see if this, if this goes well this time so I'm going to test the application and I will try to ask a question: um what is power?

B

B

There we go so basically a call has been made using my API to open AI uh via the custom. Connector I got an array of responses which contains just one response: I stripped everything out of that response and only displaying the text. So you can do a lot with this. I have another demo where I showed where you can keep the history of conversation by using powerapps collections. But the fact that you have this custom connector. It opened the door to to any options. So you can use it in power virtual agent.

B

You can use it in power to make you can do some translation. You can ask additional details, but um yeah that's pretty much it. Thank you so much over to you. David.

A

Awesome Samir very, very cool, so in less than 15 minutes you have created a connector and used in a PowerApp that is slick, slick slick. So uh lots of questions in the uh in the channel that you kind of look at those and I posted the uh blog post for him as well, so definitely check that out. Thank you. So much foreign.