Sub Agents provide a straightforward yet powerful way to organize user engagement—simply define the scenarios and provide Aissistant with instructions on how to handle them. When a scenario matches, Aissistant will follow the sub-agent instructions to process the message and tag the conversation with the sub-agent's name. For instance, you can create a "order_tracking" sub-agent with the scenario "user asks where their order is" and the instruction "Check the order status; if shipped, send the tracking number; if not shipped, ask the user to wait patiently."
These versatile tools can be used for a variety of purposes, including:
Classifying, organizing, and analyzing your messages.
Filtering out important messages and allowing Aissistant to respond only to those queries (with the option to "Respond only to messages recognized by listed sub-agents").
Disregarding certain messages, such as advertisements, by using the "no response" option.
Adding additional context or instructions to fine-tune the AI’s behavior in specific scenarios.
How to create a sub agent?
To create a tag, you can just input the necessary details directly into the sub agent creation interface.
Detection Trigger
The detection trigger specifies when sub-agent detection will occur. It has three options: Inbound, Outbound, and Both.
Inbound means sub-agent detection will occur based on the user's request. In this direction, you can use sub-agents to categorize what the user wants and give contextual instructions on how to respond.
Outbound means sub-agent detection happens when the AI generates a response. In this direction, you can use sub-agents to detect what the AI says, such as using the "human_transfer" sub-agent for scenarios where the AI mentions transferring the user to a human.
Both means sub-agent detection will occur in both directions (user request and AI response).
Note: When detecting scenario matches for sub-agents in both directions, AI will review the entire conversation history, not just the current message.
Scenario
A scenario is the trigger that defines the situation in which the corresponding sub-agent should be activated. When describing the scenario, use clear and simple language. For example, you could describe scenarios as "user wants to make a return" or "user wants to cancel an order."
Instruction (Optional but highly recommended)
The instructions for sub-agents are referred to as contextual instructions. Unlike global instructions, which apply to every user message, contextual instructions are provided to AI only when the corresponding sub-agent is identified. These instructions help AI handle messages with greater accuracy and efficiency, reducing the likelihood of hallucinations. For example, for an "order_return" sub-agent, you can provide instructions to ask for the order number, inquire about the reason for the return, share the return policy, and notify the user that the return request will be submitted to the team for approval.
Context, Task, Summary (Optional)
The context refers to the additional information provided to AI when the corresponding sub-agent is detected. For example, if the sub agent relates to a user wanting to make a return, you might include the return policy in this section.
The task outlines what you want AI to achieve with the sub-agent. For instance, for an eSIM sales sub-agent, the task could involve collecting travel details, including the destination, travel days, and number of travelers.
The summary is the information AI generates when the corresponding sub-agent is identified. For example, for the eSIM sales sub-agent, the summary could be:
{
"destination": "the destination of user travel",
"number_of_days": "the number of days the user plans to travel",
"number_of_esim": "the number of eSIMs the user wants"
}
The generated summary will appear in the agent platform as a note or comment, to be reviewed by the human team. It serves as an effective tool for communication between AI and the human team.
Session Controls
Stay Open and Unassigned
When this option is selected, the sub agent transforms into a system sub agent by prefixing 'sys_' to its name. If a ticket or conversation is tagged by a system sub agent, it remains open. Aissistant will disengage after responding, allowing a human agent to take over and continue the conversation with customers.
No Response for Current Interaction
Aissistant will disregard the messages and will neither respond nor draft a reply for any ticket or conversation with this sub agent. It is recommended to enable this for no-reply emails, out-of-office responses, advertisements, spam, and similar messages.
No Response After Current Interaction
This feature allows Aissistant to respond to the immediate message but prevents it from addressing subsequent messages. It's typically used to transition the conversation from Aissistant to the human team.
Close
Once enabled, Aissistant will close the ticket or conversation on your agent platform if it is tagged by this sub agent.
“Respond only to messages recognized by listed sub-agents”, when enabled, Aissistant will only try to respond to incoming messages that have at least one of the tags. It applies to both “Auto-Reply” and “Auto-Draft”.
Assets, Actions, Follow-ups
Sub agents can be linked to assets, actions, and follow-ups.
For instance, you can associate an 'order_return' sub agent with a return policy asset.
For an 'order_tracking' sub agent, you can link it to a 'getShopifyOrder' action, allowing Aissistant to call the action and retrieve tracking details when the sub agent is triggered.
Regarding follow-ups, you can schedule one for two days later if a customer inquires about a product. Aissistant will check in after two days to see if the customer has any additional questions.
How does it work with Agent Platform?
When Sub agent tags a ticket, the tag will be added to the Agent Platform.
Once tag is created, most of Agent Platform provides mechanism to carry on further actions including workflow, automation, notification, etc.
