Overview

Custom actions are one of the most powerful ways to make the Fibi AI Agent more useful for your customers. You can create completely custom workflows with external APIs to save time on your support loads.

Here are some cool workflows you could create with custom actions:

• Auto-extend trials – Effortlessly manage customer trials without manual intervention.

• Offer discounts – Automatically offer certain discounts to bypass budget issues.

• Account insights – Quickly provide detailed customer account summaries.

• Order tracking – Instantly retrieve order statuses and shipping details.

• Billing & payments – Automate the retrieval of invoices or outstanding balances.

Creating your first custom action

To get started with Custom actions, head over to Settings → Support → Fibi AI Agent → Actions. From there, click on + Create action to create your first custom action.

The creation consists of 4 parts:

1. General settings – Short overview that describes the action to the AI agent

2. API request – Configure the API request the agent should call for this action

3. Test the API response – Make sure the configured API request works as expected, and provide extra instructions for the AI for result parsing

4. Data access – Manage whether the AI agent should get full access to the data retrieved from the API

1. General settings

First, define clearly when your AI agent should invoke the custom action.

Action Name: Keep it concise, descriptive, and under 50 characters. Examples:

• "Extend free trial"

• "Retrieve account balance"

• "Check order status"

When to use: Provide clear, specific scenarios when the agent should trigger this action. Keep instructions concise (ideally under 400 characters). Examples:

• "Use when a customer requests more trial time or needs their trial period extended."

• "Use to quickly fetch the user's latest account balance information."

• "Invoke when customers inquire about their recent order statuses."

2. API request

Next up, you can configure the API request that the agent will call. For this, you have the following options:

2.1. Collect data inputs from the user

List any information the AI Agent needs from the user to perform the action.

Built-in variables (email, userId, userName, userType) are automatically available if the user is authenticated in the Messenger.

2.2. Add API request details

Configure your API endpoint and request method. The URL of the endpoint also supports inputting custom variables you have available from the previous step.

For custom variables, make sure to input the name of an input from the “Collect data inputs from user” section inside two brackets e.g. {{userName}}

Full URL example with custom variable: https://api.example.com/endpoint/{{userId}}

2.3. Modify the request with data

Finally, you can pass any custom data inside the request via either query parameters, headers, or the request body.

The same custom variable syntax option is supported for all of these options.

3. Test the API response

Now it’s time to test the API response to make sure it works as expected.

If you used custom variables in the previous step, you’re expected to provide real example values for the variables used to simulate that the request works.

Optional: Custom Instructions for AI

After receiving the response, you can also give extra instructions for the AI agent of how it should format it back to the user.

4. Data Access

This step can usually be skipped if you are already limiting data access from your own API.

However, if you are using a more generic API, you can optionally control what parts of the response are passed to the AI for processing.

That's it! 🎉 If you have any ideas about how we can improve custom Fibi actions, please post on our feedback board.