Create an Assistant
Overview
Admins can create AI Assistants in the Admin Portal and prepare them for testing and deployment to users. Newly created Assistants will be disabled by default until an Admin enables it.
In the AI Assistant creation window, the Admin will complete 3 sections:
Setting up basic information about the AI Assistant
Customizing the AI Assistant’s expected behavior
Connecting the AI Assistant to a knowledge base
Configure Assistant
In this section, enter basic information about the AI Assistant, such as the name, description of use case, and assign a visual icon for the AI Assistant. These basic fields will be publicly displayed in the User Portal with the description of each AI Assistant displayed as a tooltip on hover.
This section is also where the Admin will select the type of Assistant to create:
Standard Type
Image Generator Type
Depending on the Assistant type selected, the Customize Assistant section will display different fields for Admin configuration.
Customize Assistant
In this section, you can customize your assistant interactions by providing specific instructions and guidelines for the connected knowledge base. When you edit assistant settings, they'll take effect in all new chats you create with the assistant. Existing chats won't be updated.
Select a Large Language Model (LLM) that will be associated with the AI Assistant. This can be the MangoApps Managed LLM or any other third-party LLM that has an active connection to the domain. If no third-party LLM has been configured in the LLMs tab of the module, only the MangoApps Managed LLM will be available as an option.
Create Instructions
Admins will then have the option to edit custom Instructions for the AI Assistant. Any provided instructions will be passed to the AI Assistant as an initial prompt and serves as baseline context to the AI Assistant when interacting with users.
AI Assistants will come pre-loaded with a default instruction set that can be changed by the Admin. Instructions have a maximum of 3000 characters and support Markdown formatting.
Best Practices for Creating Instructions
Creating effective instructions for an AI assistant involves clarity, precision, and structure. The goal is to ensure the assistant understands the task, its parameters, and any constraints, while maintaining flexibility to handle diverse inputs.
Here are some best practices:
Be Clear and Specific: Avoid ambiguity; use clear, unambiguous language to avoid confusion. For example, instead of saying "Tell me about the weather," specify “What’s the current weather in New York City?”
Provide Context: Offer relevant context or background details, especially for specific tasks or inquiries, such as “Summarize this news article from a business perspective.”
Set Clear Expectations: Be specific about the format or style of the expected output. For example, "Provide a one-paragraph summary" or "Create a bulleted list." If there's a limitation or a focus, mention it. Example: “Find news articles from the last 24 hours” or “Use only reliable academic sources.”
Use Natural Language, but Be Consistent: Avoid idioms and slang; use straightforward language without overly complex or colloquial expressions that might confuse the AI.
Anticipate Edge Cases: Provide instructions for handling exceptions or ambiguous cases. For example, “If no data is available, return ‘No data found.’" When possible, specify default behaviors. Example: “If I don't specify a city, assume I mean New York.”
Be Mindful of Length and Complexity: Avoid overloading with too many details; provide only the necessary amount of information. Long or overly complex instructions can lead to errors in interpretation.
Allow for Iteration: Provide space for follow-ups; expect that your initial instruction may require refining. Encourage rephrasing or correction when the first output isn’t what you wanted. Give feedback loops; if the assistant doesn’t fully meet your needs, provide feedback to clarify or expand on your original instructions.
Utilize Constraints and Flexibility: Scope the task by clearly defining what should be included and excluded in the response. Example: “Write a 100-word summary about climate change, but focus only on economic impacts.”
By adhering to these best practices, you can enhance the clarity and effectiveness of your AI assistant instructions.
Greeting and Prompts
Next, customize a Greeting Message that the AI Assistant will always display when the AI Assistant is opened in the User Portal. Select from various Substitution Tags to use as text placeholders:
FIRST NAME: Displays the logged-in user’s first name
DISPLAY NAME: Displays the logged-in user’s first and last name
NETWORK NAME: Displays the name of the current MangoApps network domain
TIME GREETING: Displays a dynamic greeting depending on the logged-in user’s time zone
Default Prompts are text prompts that will always appear upon selecting the AI Assistant in the User Portal. Default Prompts are an easy way to help users understand what types of questions they can ask the AI Assistant and can capture common employee use cases.
Prompts will appear in the AI Assistant's user interface as ordered and can be reordered. Admins can create up to 4 different prompts using the following fields:
Prompt Title: Name of the default prompt. Title is visible to the user, title content is not sent to the AI Assistant. Max 100 characters.
Prompt Text: Text content of the default prompt. Text is visible to the user, prompt content is sent to the AI Assistant for a response. Max 500 characters.
Prompt Instructions (optional): Additional context sent alongside the prompt text only when the specified default prompt is selected. Content is not visible to the user. Max 2000 characters.
Image Generator Assistant Type
Default Prompts cannot be created for Image Generator Assistant types. Instead, Image Styles can be created as default art styles for users to select when generating images. Image Styles will appear in the order they are shown, and can be reordered.
A maximum of 10 Image Styles can be created. When selected by a user, the Image Style title will be sent along with the user’s provided prompt as context to help generate the prompted image.
Knowledge Base Setup
The last step of creating a new AI Assistant is configuring the AI Assistant’s knowledge base. The knowledge base is the foundation for the AI Assistant’s training and defines what the AI Assistant knows and can answer questions about.
If Model Knowledge Base is selected, the Admin can configure the AI Assistant’s Browsing Settings, which determines whether the AI Assistant is able to search the internet for additional information if its internal knowledge base is insufficient to generate a response to a prompt.
When Web Browsing Mode is selected, the AI Assistant will be able to search the internet for up-to-date information relevant to the given prompt and access external URLs for summary or analysis.
When Local Browsing Mode is selected, the AI Assistant will exclusively use its internal knowledge base to answer questions. It will be unable to access internet links and will respond to user queries without pulling any real-time information from the web.
The Knowledge Base Setup section is not applicable for Image Generator Assistant types.
Role-Based Access Control
Finally, select the user segment(s) that will have access to the AI Assistant. Only the user segment(s) selected in the Enable Access For field will be able to view and interact with the AI Assistant in the User Portal.
If one or more user segments have been selected, the Everyone option will be disabled. To enable the AI Assistant for Everyone, ensure that no other user segment has already been selected.
Last updated
