Data Types

How you wire up your Agents

When you're building your AI agents, you'll find that there're places where you can plug in data to get certain results.

For example, let's image you've got the simplest AI Agent that's just a Prompt Step:

Let's say this prompt has one internal Variable, @animal.

This @animal variable can be replaced with dynamic data when it's actually run by your users (or you while testing).

There are 6 main types of data that can be selected to fill in this variable here.

• Output

  • When you select this data type you can then select any of the previous steps that produce output to fill in this variable.

    • In this case, we could have connect an API Step for fetching the user's favorite animal, or a Prompt Step that generates a demo favorite animal

• Text

  • This is used when you just want to use static, non-changing text. This data type is most useful in Decision Steps

• Variable

  • In the Start Node you can specify data that your users will send directly to Easybeam from your app. So if you wanted to send your user's favorite animal directly from the app, you could select it after selecting "variable" here.

• Messages

  • Her you can select,

    • Most Recent Message

      • This will be the text of the user's most recently sent message. Most useful for pre-processing data used in lookups in the Knowledge or Data steps.

    • Most Recent Image URL

      • If the user has included an image in the most recent message this will be the URL to that image, otherwise it will be empty.

• Conversational

  • This data type is available branches of an Action Step that have defined required conversational data. For example, if you're build an Agent, and you want it to fetch the current weather of a city when requested by a user, you can create an Action Step and define User Location as a piece of conversational data, which could then be accessed by the API Step for the current weather. In other words, conversation data is pieces of information you're asking the AI to try and extract from the current conversation with your user.

• User Secrets

  • User Secrets are sent by your user to Easybeam at the time of request. These are essentially identical to variables, except they're scrubbed from any logs for privacy requirements. Especially useful when you want your Agent to take actions on behalf of a user, which would then require their user token in your system to be included (i.e. update their profile, create a post, etc). This data is only available to the API Step.