How to stop custom GPT displaying Confirm+Deny buttons?

GPT builders
, ,
1

I’m working on a custom GPT, that’s communicate with a remote database via a Flask server.

But each time an endpoint is accessed, GPT pop-ups two buttons that ask the user to Confirm in order to continue. I may understand if it ask for one time in a conversation, but it happens for each message that involved a touch of an endpoint.

How to eliminate those buttons ?

image

1 Like
2

Toggle this to ‘Always allow’ doesn’t solve the problem

image

3

I’ve udpated my question, since I found the solution. May help another mate.

On short, add this line in your Schema, to each Action:
image

On large, all info explained here by chatgpt itself during the conversation.

Understanding the Solution

By using the x-openai-isConsequential field to manage whether actions require explicit user confirmation or can be allowed automatically. Here’s a breakdown of how this works:

  1. x-openai-isConsequential Field:
  • This is a flag that you can include in the schema for actions that a GPT might perform.
  • If this field is set to true, the GPT will prompt the user for confirmation before executing the action. This means the action is considered consequential or critical, so additional user consent is required.
  • If the field is set to false, the GPT will provide an option to the user to allow the action permanently, bypassing the need for repeated confirmation. Essentially, the user can choose to “always allow” the action without further prompts.
  • If the x-openai-isConsequential field is not present, the system defaults to:
    • GET Operations: Automatically treated as false (no confirmation required).
    • Other Operations: Automatically treated as true (confirmation required).

Example Usage

Here’s how you might adjust your schema code based on this solution:

json

Copy code

{
  "actions": [
    {
      "name": "exampleAction",
      "description": "An example action that requires user confirmation",
      "x-openai-isConsequential": true,
      "parameters": {
        // your action parameters here
      }
    }
  ]
}

In this example, exampleAction will always prompt the user for confirmation before execution due to the x-openai-isConsequential field being set to true.

Alternatively, if you want to allow users to bypass repeated confirmations:

json

Copy code

{
  "actions": [
    {
      "name": "anotherAction",
      "description": "Another action where users can choose to always allow",
      "x-openai-isConsequential": false,
      "parameters": {
        // your action parameters here
      }
    }
  ]
}

Implementation Notes

  • Ensure you add the x-openai-isConsequential field to the relevant actions in your schema.
  • Adjust the value based on whether you want the action to always prompt for confirmation (true) or allow a one-time permission (false).
  • Test the behavior of these actions in your environment to ensure they work as expected.

By following this approach, you can fine-tune how your GPT interacts with users and manage the balance between usability and security effectively.

10 Likes
4

Great explanation! I too have come across this problem and now I no longer will. Cheers!

1 Like
5

Happy it helps!
I’ve struggled for days to find a solution.

1 Like
6

Thanks so much for coming back to share your solution!

Sorry nobody chimed in sooner.

But again, appreciate you coming back here to share. That’s the community spirit we want to foster here…Well done…

2 Likes
7

I’m using the x-openai-isConsequential field in our GPT action as follows:

“paths”: {
“/planets”: {
“post”: {
“summary”: “Get planetary positions”,
“description”: “Returns planetary positions based on the provided date, time, and location.”,
“operationId”: “getPlanetaryPositions”,
“x-openai-isConsequential” : false,
“requestBody”: {

but even after I ‘Always Allow’ this web API once, I still am asked everytime the action is run,

1 Like