Edit Mode

The agents edit tab (AI->Agents) contains options for creating and editing agents for HyperSkill.

The agents window when editing

By default, the agent settings will be locked until you create or select an agent.

You can edit the following properties of agents:

Property	Description
Name	The name of the agent.
When to run	Determines when the agent should generate a response. The options include "Run after user utterance", "Run through state machine", and "Run after character".
Use History	Determines if the agent should look at the last speech action or the entire conversational history. Generally the agent should look at the entire history.
Language Model	Determines which LLM to use. Azure refers to GPT4, Google refers to Google's Gemini Pro.
Instructions	The instructions the agent is intended to follow.

The options for "When to run" refer to the following:

Option	Description
Run after user utterance	The agent will run after the user says or types something into HyperSkill every time. This is useful for evaluating the user at every dialogue turn.
Run through state machine	The agent will run only when the state action run agent is used. This is useful to run the agent only when it is needed.
Run after character	The agent will run after the chit-chat roleplaying character finishes speaking. This is useful if you want to include the character response in an analysis.

When you click on the "+Add" button under "Describe Your Agent's Output", a new box will be added where you can describe what the agent should update in HyperSkill.

An empty output box after clicking on the "+Add" button.

We feed names and descriptions to a powerful language model, which then uses them to create new content that matches those descriptions. The created content can be automatically stored in a specific area (attribute) for you to use easily. You can create as much outputs as you want, but access in HyperSkill might become restricted depending on your subscription plan and usage. You can edit the following fields:

Property	Description
Name of Output	A descriptive name for the LLM to use.
Description of Output	This explains what kind of content the LLM should create. It could be something specific like "a score from 0-100" or more general like "feedback for the learner."
Output Destination	This tells the system where to put the generated content. It's like a specific label on a box where you want the information stored. These dropdowns point to a particular piece of data (attribute) for a virtual object.