Function Calling

Estimated Time: 4 minutes

Function calling lets you connect your AI models to external APIs.

There are three primary reasons you would configure function calling:

1. Data Collection (e.g lead generation)
2. Triggering an action (e.g sending an email)
3. Information Retrieval (e.g fetching order status)

This link helps you understand more about the use cases for function calling.

Configuring function calling

Prerequisites

1. Configured your function in OpenAI - Configuring Function Calling in OpenAI

2. Implemented you API endpoints and have the API method (GET or POST) & URL.

3. (optional) If your endpoint needs authentication, then the bearer token string. (Note: If you need a auth mechanism other than bearer token, please contact us)

Sign In

• Sign in to Predictable Dialogs.

Go to the AI Resources page

• Hover on the profile picture. The drop down menu will show up. Click AI Resources.

Select AI Resource

• Click the AI Resource which needs to be configured. You will see a screen like below

• Click on the "Functions" button. This will pull the functions configured on the OpenAI Assistant into the PD app. You will be able to see the list of configured functions as shown in the image below.

• Now, click on the gear/setting icon against the function. This will open a modal as shown below:

Choosing the HTTP Method

• GET: When selected, the function arguments are passed as query parameters. Query parameters are the most common type of parameters. They appear at the end of the request URL after a question mark (?), with different name=value pairs separated by ampersands (&). Currently, we only support query parameters for GET requests. Support for path parameters (variables included directly in the URL path) is planned for a future release. For more details, see this explanation of query vs path parameters.

  Example: If you're using the Open Meteo weather API, a full GET request might look like this:

  https://api.open-meteo.com/v1/forecast?latitude=52.52&longitude=13.41&current=temperature_2m,wind_speed_10m&hourly=temperature_2m,relative_humidity_2m,wind_speed_10m

  In this case, latitude and longitude are dynamic parameters that will be generated by the LLM. When configuring this endpoint in Predictable Dialogs, you would skip the latitude and longitude and enter:

  https://api.open-meteo.com/v1/forecast?current=temperature_2m,wind_speed_10m&hourly=temperature_2m,relative_humidity_2m,wind_speed_10m

  During a chat session, the LLM function call would generate the additional parameters latitude=52.52&longitude=13.41 and automatically append them to this URL.

  • POST: When selected, the function arguments are passed in the body of the request.

• Enter your endpoint url (For a GET request, include the static query parameters.)
• If your endpoint needs authentication, click the checkbox Add authentication and add your bearer token.
• Click Save, to save the configuration.

Testing

You can now start a session on your chatbot and enter the prompt which would trigger the function. You would notice that your function gets called and the chatbot assistant replies accordingly.

Check Details Of Function Call

• Go to the "Sessions" tab of your agent which lists all the chat conversations.
• Select the conversation which triggered the function call.
• Scroll down to the chat message which triggered the function call and you would be able to see
  • The function name which was triggered along with the arguments
  • The endpoint method and url which was used.
  • The result returned from the endpoint.
  • The time duration in ms, it took to make the request.

Understanding Authentication For Endpoints

When configuring your endpoints, you may need to ensure secure access by using authentication. Predictable Dialogs supports a simple and secure authentication method called Bearer Token Authentication.

What is Bearer Token Authentication?

Bearer Token Authentication is a widely-used method for securing API calls. Think of it like using a key to unlock a door:

• Your service (API) provides you with an API Key (like a secret password).
• Each time Predictable Dialogs calls your endpoint, it uses this key as a "Bearer Token" to identify itself and gain access.

How to Set It Up:

1. Obtain Your API Key:

   • Typically, you get this key from your developer or external service provider. Look under their settings, API access, or authentication sections.

2. Enter the API Key:

   • In Predictable Dialogs, check the "Add Authentication" checkbox.
   • Paste your API Key into the provided field.

Security and Privacy:

• Predictable Dialogs securely stores your API keys using encryption.

When is Authentication Needed?

• Authentication is required when your external endpoint expects secure requests.
• If your endpoint does not require authentication, simply leave the "Add Authentication" box unchecked.

If you're unsure whether your endpoint requires authentication, consult your developer or external service's documentation or support team.