Introduction to Model Builder in Salesforce Agentforce

As of now, we have seen Agent Builder, where we can build a new AI agent, and we also learned about Prompt Builder, where we can build various types of prompt templates.

Now, AI agents in Agentforce can be customized to fulfill business needs, such as predicting customer actions or generating correct replies. In short, to train the AI agents using our own data, the third builder we have is a model builder, which allows us to customize AI models, such as predictive and generative models.

In this tutorial, we will learn about the introduction to model builder in Salesforce Agentforce. In that, I will explain how to configure the standard model, how to create the custom model, and different advanced settings in Einstein Studio in Salesforce.

Table of Contents

Model Builder in Salesforce Agentforce

A model builder in Agentforce is a feature of Einstein Studio that allows you to connect your data with your preferred large language model (LLM) and allows us to customize AI models, such as predictive and generative models.

The model builder allows Salesforce professionals to create custom AI models customized to specific business needs without requiring extensive data science expertise or integration. It is a no-code tool for connecting external models or creating a model from scratch.

In the Prompt Builder, we generated content using various LLM models. However, with the advent of Gen II, an organization can configure its own models through the model builder if they do not want to use Salesforce-provided LM models.

Many of the organizations started building their own LM models, which are hosted on various platforms such as Azure or Open AI. A few of the organizations have also built their in-house models, which are trained on their business-specific data, and these customers would want to use their own models as well.

So, in Model Builder, with minimal configurations, the customer can bring or connect to the models hosted outside of Salesforce and use them in Prompt Builder, etc. The Model Builder platform provides a user-friendly interface and guided workflow to facilitate the model-building process.

Model Bilder in Salesforce Agentforce

Now, let’s understand when it is appropriate to use the model builder and how we can connect out-of-the-box models, custom models, or our own models through the builder to configure them.

Now, let’s navigate to our Salesforce org to see the model builder. The model builder can be accessed through the Einstein Studio, which is part of Data Cloud. So, you can search for the Einstein Studio or Data Cloud app in the App Launcher.

In the data cloud app, we also have Einstein Studio. Here, you will see the generative models and the model library. These are the generative models we have already seen in our prompt builder. So, in the Prompt Builder, when we were selecting the models, we had some of Salesforce’s models, which were the same models.

These are the LLM models, which are Salesforce-shared models. For example, OpenAI models are Salesforce-shared models, and Salesforce-hosted models are also available.

Navigate to the prompt builder and open any prompt template. As part of our prompt builder, we created the prompt template and selected a model against which it would be executed.

If you click on the Model Type drop-down, you will see all the standard models(image 1), which are the same as those in the generative model.

But if you go to the custom model (image 2) here, since there is no custom model in this org, it is showing blank. But the moment we add a custom model, it will start showing up here and in the model library.

Different Models in Salesforce Agentforce

Configure the Standard Model in Salesforce Agentforce

When you prompt the AI agent, the response generated is according to the settings defined by Salesforce. Those settings or configurations are done in the Einstein Studio, where we can build a custom model or configure a standard model in Salesforce.

Let’s edit the OpenAI GPT 4 model. To do so, from this drop-down menu, click on Edit.

Configure Models in Salesforce Agentforce

In edit mode, you will see the model playground where we can configure the standard and custom models.

Introduction to Model Builder in Salesforce Agentforce

In the Advanced Settings, you will see all the configurations that Salesforce has defined for you. For example, the temperature is 0.5, the frequency is 0, and the presence penalty is 0.

If you want to change these parameters, you should customize this model according to your needs and adjust the temperature.

What is the Temperature of a Model?

If you increase a model’s temperature, it will become more creative, which means that it can assume things or generate creative content. But if you keep this value at a lower end, it will be very predictable. You can predict the response because whatever you ask it, it will show that it will not add any creativity while generating the content.

Currently, the value is 0.5, which means that it is at the lower end. Let’s say that we want to be a little more creative, so we can change this value to 1.5.

What is the Frequency Penalty of a Model?

The frequency penalty is how often this model will repeat a word. A higher number indicates that the word will be less likely to be repeated. So, for example, if you are generating a pitch or content where you do not want the same word to be repeated, then you can increase this value.

Right now, it is at a stable position of 0.0. As we increase this to 1, this model will repeat fewer words in the content.

What is the Presence Penalty of a Model?

The presence penalty, which determines how many unique words will be present in the response, is also currently set to zero. So, let’s increase that, and now you have configured OpenAI’s GPT four model according to your needs.

Models Advanced Settings in Salesforce Agentforce

After that, you can click on the Update button. However, since we cannot change the standard model, you need to create a new one. To do so, you need to click the Save as New button after clicking on update.

Then, provide the Name and Description for your model and click the Create Model button.

Now, let’s go back to the models again. Here in the generative, you will see that we have a GPT 4 Custom Model, which is one of the custom models that we have created right now by customizing some of the parameters of the standard GPT model that Salesforce gave us.

If we return to our prompt builder, let’s refresh it. Now we can see the Custom Model that we have created, the GPT 4 Custom Model, available here.

Similarly, you can change the configuration of the out-of-the-box models or bring your own model, which will start showing up here, and use it for various activities related to Salesforce AI.

One more thing about Model Builder is that it also gives us a playground

where we can test the new model.

Conclusion

I hope you have got an idea about the introduction to model builder in Salesforce Agentforce. In that, I have explained how to configure the standard model, how to create the custom model, and different advanced settings in Einstein Studio in Salesforce.