Creating Agentforce Custom Actions with Heroku

This tutorial will guide you through configuring an Agentforce Action deployed to Heroku within your Salesforce org. This allows Agentforce agents to access powerful custom-coded actions written in Python and other languages, leveraging Heroku's fully managed and elastic compute service. By the end, you will be able to generate your own badge as shown below!

💡 Heroku Integration Pilot:
These steps utilize generally availble features in Heroku and Salesforce. If you have joined the Heroku Integration pilot program please refer to these alternative instructions.

Step 1 - Action API URL

💡 Salesforce Org Requirements: Ensure you have access to Agentforce in your org by searching for Agents in the Setup menu. If not, you can create a temporary org by logging into Trailhead and creating a Playground per the instructions at the top of this module (you do not need to complete the module). Once you have access to the org, locate Einstein Setup under the Setup menu and Enable Einstein, which will also enable Agentforce.

This tutorial provides examples in Java and Python for coding a basic Agentforce action. You can deploy these to your own Heroku account or proceed for now with the pre-deployed versions we have created for you.

Language	Pre-deployed API URL	Pre-deployed API Test Page	Or Deploy your Own to Heroku
Python	Action API URL	Action API Test Page	Go to this GitHub Repository
Java	Action API URL	Action API Test Page	Go to this GitHub Repository

With the above information, you have two paths to take:

• If you are using the pre-deployed versions, right-click the Action API URL link from above and copy the URL to your clipboard when requested below. Optionally, click the Action API Test URL to try the API out from your browser; this will be the Action API Agentforce will eventually call out to when invoking your action.
• If you want to deploy your own action before proceeding, follow the instructions in the respective repository above and return here with the deployed URL.

Step 2 - Creating a Named Credential

If you have not yet gained access to an Agentforce-enabled environment, please refer to the note above. The following steps assume that you are logged in and have administrator access to the Salesforce Setup menu.

1. Search for Named Credentials under Setup, select the External Credentials tab, and click New, completing the dialog as shown.

   

2. Scroll down the next page to the Principals section and click New, completing the dialog as shown.

   

   The default password for this tutorial is agent.

3. Search for Named Credentials under Setup, select the Named Credentials tab, and click New, completing the dialog as shown.

   

   To obtain the URL value, right-click the desired Pre-deployed API URL above or use the one given to you after completing your own Heroku deployment. The one shown in the screenshot above is the Java pre-deployment.

Step 3 - Registering the Action API

1. Search for External Services under Setup and click Add an External Service. Then, select From API Specification and click Next, completing the dialog as shown, and click Save & Next.

   

   If you have not modified the action code yet or are using the pre-deployments above, you can click here (Java) or here (Python) to obtain the schema and copy-paste it into the Schema field. Log in when prompted with user heroku and password agent.

   Alternatively, to obtain the dynamically generated schema for your own action deployment, refer to the instructions in the respective GitHub repository in the table above to open the API Test (Swagger) page. Then click the link just below the API title in the top left corner of the page, as shown in the examples below. Then copy and paste the entire contents into the Schema field above.

   

   For more information on the API Test (Swagger) page and authentication, see the respective sample code GitHub repositories listed in the table in step 1.

2. Finally, on the next page, select the Operation shown below and click Next.

   

3. Confirm you can see the input and output parameters as shown below, and click Finish.

   

Step 4 - Assigning Permissions

1. Search for Permission Sets under Setup, click New, complete the page as shown, and click Save.

   

2. Locate the External Credential Principal Access section and edit it so that it appears as shown.

   

3. Finally, assign the permission set to your current user using the Manage Assignments button.

Step 5 - Creating an Agentforce Action

1. Search for Flows under Setup and click New Flow. Click Start from Scratch, then Next, and select Autolaunched Flow (No Trigger). Click Create to open the Flow Builder.

2. From the Flow Builder, open the Toolbox by clicking the sidebar icon near the top left of the screen. Click New Resource and create the following variable resources.

   • Resource Type: Variable

     • API Name: AgentActionRequestName
     • Data Type: Text
     • Apex Class: Not Applicable
     • Availability Outside the Flow: Available for Input

   • Resource Type: Variable

     • API Name: AgentActionResponseMessage
     • Data Type: Text
     • Apex Class: Not Applicable
     • Availability Outside the Flow: Available for Output

   • Resource Type: Variable

     • API Name: AgentActionRequest
     • Data Type: Apex Defined
     • Apex Class: ExternalService__HerokuAgentAction_AgentRequest
     • Availability Outside the Flow: None

   Your Flow should look like this so far:

   

3. Click the + icon in the middle of the canvas and select Action. In the Search Actions sidebar, search for Heroku, select the Post Process action, and complete the fields as shown.

   

4. Click the + icon before the Call Heroku element on the canvas, select Assignment, and complete as shown.

   

5. Click the + icon after the Call Heroku element on the canvas, select Assignment, and complete as shown.

   

   Tip: To complete the Value field click into the field and use the popup to select the Outputs from the Call Heroku Action resource, then click _200 and finally message.
   

6. Finally, your Flow should look like this:

   

7. Click Save, enter Heroku Agent Action, and click Save, followed by clicking Activate.

8. Search for Agent Actions under the Setup menu, and click New Agent Action, selecting Flow as the Reference Action Type. In the Reference Action search field, search for Heroku so that you see the following:

   

9. Complete defining the action as follows and click Finish.

   

   Field	Text
   Agent Action Instructions	Use this action in response to requests for a Heroku badge with a name on it.
   AgentActionRequestName Instructions	Name to be placed on the badge.
   AgentActionResponseMessage Instructions	This is the Heroku badge the user requested.

Step 6 - Using Agent Builder

1. Search for Agents under Setup and be sure to toggle to enable Einstein Copilot for Salesforce.

2. Click the Einstein Copilot link at the bottom of the page to find the Open in Agent Builder button.

3. Click the Deactivate button in the top right of the screen if present.

4. Click Topics and the General CRM topic. If the New Version button is shown, click it.

5. Click the Topic Configuration tab and append "Also including requests for badges." to the current contents of the Classification Description field.

6. Click the This Topic's Actions tab and select Add from Asset Library from the New button dropdown.

7. Complete the dialog as shown and click Finish.

   

8. Finally, click the Activate button in the top right of the screen.

Step 7 - Testing your Action

1. Open the Agent Builder for Einstein Copilot.

2. In the Conversation Preview, enter Generate my badge with the name Master Builder on it!

   You can, of course, change the name to your own and share your badge with your friends!

   

Summary

Congratulations on deploying your first Heroku-powered Agentforce Action!

We are working to improve and simplify this process, so please follow this repository to stay informed on updates.

Need some inspiration on what to build?

The examples in this tutorial are intentionally simple, designed to give you a clear starting point for creating your own actions. We have also included fully developed actions below.

Example	Demo Video		Related Content
The Archive Agent action demonstrates using Heroku to access and perform complex computations over unstructured archived data. It uses AI to interpret natural language requests from users to perform a wider array of dynamic calculations on the data.	Link		Blog Code
With the Coral Cloud Collage Agent, we explore how Coral Cloud Resort invites its guests to browse and book unique experiences throughout their stay using Agentforce. With Heroku, we extend the agent's functionality to generate a personalized collage of each guest's adventures, showcasing how custom code deployed on Heroku can create dynamic digital media delivered directly within the Agentforce experience.	Link		Code