Table of Contents
Layered Assistants allow you to build a hierarchy of AI assistants that work together to handle user requests. A Layered Assistant uses a parent assistant to evaluate incoming questions and automatically route them to the most appropriate child assistant for a response.
This approach helps organize complex workflows, improve response quality, and create specialized experiences for different types of requests. Instead of relying on a single assistant to perform every task, Layered Assistants can direct users to assistants that are configured for specific topics, departments, or functions.
- Intelligent Request Routing: Automatically direct users to the most appropriate assistant based on their request.
- Parent and Child Assistant Structure: Use a parent assistant to make routing decisions while child assistants handle responses.
- Custom Decision Logic: Define routing behavior using natural language instructions and decision guides.
- Specialized Assistants: Create dedicated assistants for different topics, services, or departments.
- Scalable Organization: Manage multiple assistants within a single routing framework.
- Improved User Experience: Ensure users receive responses from the assistant best suited to their needs.
- Flexible Use Cases: Ideal for support desks, academic services, student resources, departmental knowledge bases, and multi-purpose AI solutions.
- Click on the "Assistant Gallery" on the Assistant Tab.
- The Layered Assistants Tab is visible on the left of the top bar.
- Go to the Assistant Tab on the top-left sidebar on the sandbox, and click the Layered Assistants button.
- Give your Layered Assistant a Name at first then write a small description.
- For example: Support Desk Layered Assistant, with the support requests to the right specialist assistant based on issue type, urgency, and context.
- Find the Parent Settings below and give your 1st Parent Assistant a name and add a Description for its role.
- Also, add a Decision Guide for its routing logic. See the example below.
- Add child assistants to the parent assistant.
Note: Add at least 2 child assistants to the parent to save it properly Once 2 child assistants are added and configured, hit the "Save" button at the bottom-right of the screen to save the Layered Assistant.
Decision Guide:
''' Decision Guide Use this assistant as the top-level router for all support-related requests. Route the user to the most relevant sub-parent or leaf assistant based on the issue category:
Billing / payments → choose the Billing Assistant (route to the Billing Assistant)
Account access / login / password → choose the Account Access Assistant (route to the Account Access Assistant)
When deciding, prefer the branch that best matches the user’s main problem:
If they mention charges, invoices, refunds, subscriptions, or payment methods, route to Billing Assistant
If they mention signing in, passwords, MFA, or account recovery, route to Account Access Assistant
If the request is ambiguous:
Ask yourself which branch would most likely resolve the issue fastest
Prefer the branch that handles the user’s immediate pain point
If the issue spans multiple areas, route to the primary root cause first
As the Support Desk Router, you cannot answer general questions too! thats against your guardrails you can only rout to the children assistants which will answer the questions.'''
Child Assistant 1 - Billing Assistant
Child Assistant 2 - Account Access Assistant
- Go to the Assistants Tab > Assistant Gallery > Layered Assistants.
- Click on the Recycle bin icon as annotated below.
- The Created Layered Assistant i.e. Prompts "Support Desk Layered Assistant" should appear on the left sidebar under the Layered Assistants tab and inside.
Tip: You can also create multiple sub-parents & have multiple children/sub-parents inside those sub-parents as well.
Disclaimer: The Publish at Path feature is currently under internal testing and should not be used at this time. The AI Generate feature for DESCRIPTION and DECISION GUIDES is not working, and the CometAI team is working internally to resolve the issue.