🆕Understanding Your Knowledge Base
Overview
Each AI Assistant is trained on a specific set of knowledge, which is then accessed by the AI Assistant to respond to user-generated queries. These sets of knowledge are known as knowledge bases. Within MangoApps, there are multiple knowledge base types that each correspond to a different scope of training data.
Model Knowledge Base
When Model Knowledge Base is selected, the AI Assistant relies on the internal knowledge base of the associated AI model to answer user queries. By default, this option is enabled when creating a new Assistant.
Admins can also configure the AI Assistant’s browsing capabilities:
Web Browsing Mode: The Assistant can search the internet for up-to-date information relevant to the prompt and access external URLs for summaries or analysis.
Local Browsing Mode: The Assistant will only use its internal knowledge base, without accessing the internet, to provide responses.
In addition, the Assistant can analyze text and images in PDF files (up to 50 pages and 100 MB in size) to extract answers. JSON files can also be uploaded, enabling users to query the data they contain directly.
The Model Knowledge Base is best suited for general use cases with publicly available large language models such as OpenAI ChatGPT or Google Gemini.
Company Knowledge Base
When Company Knowledge Base is selected, the AI Assistant will utilize company-wide content to answer user queries. Admins can select from the following module content to include in an AI Assistant’s training:
Company Libraries
Company Pages
Company Posts
Network Drive Files
People Directory (Available for Colleague Finder Only)
Admins can add up to 250 files not managed by the MangoApps auto-governance engine to the company knowledge bases. After that number is reached, additional files must be enabled for auto-governance. Any links, documents, or online content will respect their original access permissions, meaning users without permission to a specific item will still be unable to access that item even if it is included in an AI Assistant’s response.
Team Knowledge Base
When Team Knowledge Base is selected, the AI Assistant will utilize a specified team’s content to answer user queries. First, type to search and select an existing MangoApps team (e.g. workspace, group, department, etc.) Then, admins can select from the following team content to include in an AI Assistant’s training:
Pages
Posts
Wikis
Files
Admins can add up to 250 files not managed by the MangoApps auto-governance engine to the team knowledge base. After that number is reached, additional files must be enabled for auto-governance.
If the selected team is public, then access permission can be granted to any user segment. If the selected team is private or unlisted, then access permission will only be granted to members of the team.
Any links, documents, or online content will respect their original access permissions, meaning users without permission to a specific item will still be unable to access that item even if it is included in an AI Assistant’s response.
Bulk File Selection
To optimize your AI Assistant’s knowledge, you can add multiple files at once. You can select up to 250 files, with a maximum size of 50MB per file. Supported file types include: .pdf, .txt, .docx, .pptx, .odt, .odp, and .html.
Adding Files
Select files from the left panel: Click on the files you want to add to your AI Assistant’s training knowledge.
Add multiple files using the Shift key:
Click the first file in the folder.
Hold down the Shift key and click the last file. All files in between will be selected.
The Add button will display the total number of selected files. Click Add to include them.
Select an entire folder:
Click the folder title to select all files within it.
The Add button will update with the total count of selected files. Click Add to include them.
Removing Files in Bulk
Use the same selection method (Shift key or folder click) on files already added to your Assistant.
Click Remove to delete the selected files from your Assistant’s knowledge base.
Tracker Knowledge Base
When Tracker Knowledge Base is selected, the AI Assistant will utilize a specific tracker’s content to answer user queries. In addition to using a generative AI model to create natural language responses, the AI Assistant can also search a Tracker Knowledge Base to find and return the most relevant, fixed, question-and-answer entry based on the user's query.
First, select an existing Tracker to be used as the AI Assistant’s data source. The Tracker must contain at least 2 required fields. One of the required fields will contain all of the Admin’s predefined questions. The AI Assistant will find the closest match between the user’s query and one of these predefined questions. The other required field will contain all of the Admin’s predefined answers, paired to its corresponding question field.
When a match between the user query and the predefined Tracker question has been found, the AI Assistant will respond to the user with the content in the corresponding predefined answer field.
This knowledge base also offers the option to choose the Response Method. Choose whether the assistant will return a fixed response from a pre-defined answer column or if it will generate a dynamic response using all the tracker data as a whole.
The Tracker Knowledge Base is ideal for meeting FAQ or Q&A-style use cases where Admins would want to provide a response to common or frequently asked employee questions.
MangoApps Guides Knowledge Base
The Guides Knowledge Base is only available for the Admin Help Expert, which comes as an out-of-the-box AI Assistant in the AI Studio module.
When Guides Knowledge Base is selected, the AI Assistant will utilize the MangoApps Help Guides to answer questions about MangoApps modules, features, system requirements, and more.
AI Assistants Knowledge Base
The AI Assistants Knowledge Base is only available for Find an AI Expert, which comes as an out-of-the-box AI Assistant in the AI Studio module.
When AI Assistants Knowledge Base is selected, the AI Assistant will utilize the names, descriptions, and icons of all of the currently active AI Assistants in the domain to answer employee questions about which AI Assistant they should direct their query.
Last updated