Creating and Working with AI Assistants in AmplifyAI

This document provides a step-by-step guide on building and deploying assistants in AmplifyAI, managing group access, and best practices for designing educational assistants.

Kirstie Knighton
19 steps
17 minutes

Cr8

Building Your Assistant

Assistants are custom AI helpers (similar to personal custom GPTs or Gems) configured for a particular purpose. They can be created by individuals for personal use or built to live inside a collaborative group. You can create an assistant for a repeated role or workflow. Additionally, once you have built an assistant, you can deploy it to specific student groups so that only your enrolled students can access it. Give it a name and instructions and add reference materials it should use. Test it before sharing with others.

Open the Assistant's Panel by clicking the sparkles tab.

Click "Assistant."

Give the assistant a clear name. Add a short description (which will be visible to students if you share it with them).

Give your assistant instructions (your prompt). The instructions tells the AI its role, knowledge scope, tone, and constraints. See our hints for writing prompts for help.

Click the "Assistant disclaimer message" field. A response disclaimer is highly suggested when sharing assistants with students to remind users (students) that their conversations can be monitored by the teacher. It can also be used to remind users to verify the AI answers or to explain the assistant's limits.

You can enforce the model that the assistant will use. Choose the model that best fits the assistant's purpose. See our guide for choosing models if you need help with this.

You can also enforce the thinking level of the model. Off or Low should be used for almost all everyday work. Medium and High use additional computing resources and may cause users to reach their chat limits faster.

If the assistant needs to reference materials, then you can upload data sources. However, upload only current, authoritative, accessible sources that you are permitted to use. Once files are initially uploaded, users chat with the assistant without repeatedly burning token upload limits for those documents.

You can also attach website data sources for the assistant to reference.

Click "Save"

## Attach the assistant to a chat **1.** Start a New Chat and select the @ icon below the prompt input box. **2.** Choose your assistant. Upload any documents specific to the task and wait for the green check mark. **3.** Type the current request. The assistant supplies the reusable instructions; your message supplies the task for this conversation.

Allow several minutes after saving for source changes to take effect. Test ordinary questions, ambiguous questions, missing information, adversarial prompts, and requests outside the assistant's scope.

Sharing Assistants

**1.** Select the share control for the item, or open Sharing from the profile menu. **2.** Add the intended recipients under People and enter the required Note explaining what you are sharing. **3.** Select the intended items. Assistants appear under Prompts in the sharing dialog. Check that no unintended items are selected. **4.** Select Share. Ask a recipient to confirm that they can open the material. For an assistant, tell recipients its purpose, which documents they should provide, its limitations, and whom to contact about errors. Keep its reference materials current.

Layered Assistants

Layered Assistants are an advanced builder feature. A parent assistant acts as a router, while child assistants perform specialized work. The builder includes a description, decision guide, AI-generated routing support, child assistants, sub-parents, and a publishing path. Builders should test incorrect routing, missing categories, ambiguous requests, permissions, and escalation to a human. Layered Assistants are available for personal use or for use in groups.

Before building a layered assistant, it helps to understand how the components interact: **The Parent Assistant** (router) acts as the primary entry point. It reads incoming user queries, analyzes context, and decides which specialized child assistant is best equipped to handle the task. **Child Assistants** are specialized, standalone assistants (or domain-specific tools) containing tailored instructions, data sources, or models designed to execute specific sub-tasks.

Layered Assistants are for **ADVANCED USE ONLY.** Before creating a group or Layered Assistant, confirm ownership, membership, permissions, source maintenance, sharing rules, support responsibility, and what happens when the builder leaves Mercer.

Open the main navigation menu and select the "Layered Assistants" tab (situated alongside your standard Assistants and Prompt Templates) on the left side of your screen.

This screen will pop up. Give your Layered Assistant a clear name and description.

**Define the Parent Router:** Give your router a clear name and description. Write custom routing instructions if necessary, explaining to the parent model how it should evaluate user prompts and choose between child options.

You can also click on the "AI Generate Layered Assistant" to let the wizard create different levels and the assistants that belong to the different levels.

**Assign Child Assistants:** Select the specific child assistants you want to bundle under this parent router by clicking on their name (or the + sign) from the right side of the menu.

Once you've added all of the child assistants, click "save" to establish the hierarchy. **Updates and Maintenance:** You can return to the Layered Assistants configuration menu at any time to add new child assistants, adjust routing logic, or prune outdated child nodes as your workflow evolves.

Once saved, your layered router will appear in your left-hand navigation list alongside your regular assistants. **Prompt Execution:** Type your request naturally. The parent router will dynamically process your input, determine the correct underlying child agent context, and pull the appropriate structured response.