Build a GPT Chatbot in Power Apps with Encodian Flowr

2. Encodian Action: AI – Run Prompt (Text)

Example configuration for using the AI – Run Prompt (Text) action:

• Model: GPT_4_1_mini
• Prompt & Conversation: the dynamic ‘prompt’ and ‘conversationHistory’ values from Power Apps (V2) action respectively
• Context: “Please respond only using HTML for <strong>, <br> and bullet lists. You can include emojis.”

Requesting HTML output overrides this model’s default Markdown output, which Canvas Apps have no built-in way to display. However, it can display HTML text using the ‘HTML text’ control. So this ensures the response is ready to be rendered in Power Apps without any formatting conversions. The request to include emojis is simply to make the chat more fun and human-like.

3. Action: Respond to a Power App or flow

The last step in the flow is to respond to the Power App. Return both:

• response: AI – Run Prompt (Text) action’s dynamic ‘Message’ value
• conversationHistory: AI – Run Prompt (Text) action’s dynamic ‘Conversation’ value

This allows the app to persist and rebuild the chat thread, allowing the user to have a conversation with the GPT model and see previous responses from the app.

The app has been designed to look like a chatbot conversation, so the user can enter their message/prompt in the text input box at the bottom of the screen and press the send button to trigger the backend Power Automate flow.

I set the conversation history response from the Power Automate in a collection called colConversation. We will cover the formula later, but let’s start with the collection:

2. Send Button (Power Automate Integration)

The Send button triggers the entire chat cycle.

The OnSelect property of the send button has multiple formulas:

Convert conversation to JSON

We must set a variable called jsonConversation. The value is colConversation converted to JSON format. We need to convert this collection so we can send the current conversation history to Power Automate. If it is the first message, this value will be blank.

Set(jsonConversation,  JSON(colConversation,JSONFormat.Compact))

Set(
    flowResponse,
    AIRunPrompt.Run(promptTxt.Value,{text_1:(jsonConversation)})
)

Reset(promptTxt)

ClearCollect(
    colConversation,
    ForAll(
        Table(ParseJSON(flowResponse.conversationhistory)),
        {
            messageType: Value.messageType,
            message: Text(Value.message)
        }
    )
)

Reset the gallery to refresh the display

Lastly, we must reset the gallery to ensure the data has been updated:

Reset(messageGal)