A knowledge base is a curated library of uploaded documents and extracted web content that an AI Agent draws from when answering customer or internal inquiries. Knowledge bases are assigned to AI Agent Profiles via the Access Knowledge Base tool — a single agent can reference multiple knowledge bases, and the same knowledge base can be shared across multiple agents.

Use Cases

• Customer-facing AI bot referencing public support documentation, product manuals, or FAQs
• Internal AI bot for agents and staff to query internal processes, policies, or reference materials
• Department-specific bots drawing from isolated knowledge bases scoped to their audience

Accessing AI Knowledge Bases

AI Knowledge Bases is accessible to accounts with Admin permissions.

1. Navigate to AI Configuration > AI Knowledge Bases
2. All configured knowledge bases are listed in the modal

Creating a Knowledge Base

1. Click "Add Knowledge Base" to add a new knowledge base

   

2. Enter a Knowledge Base Name — be specific about its purpose (e.g., "Public Support Articles", "Appliance Manuals – Microwaves", "Internal HR Policies"). Specificity matters especially when managing multiple libraries

3. Enter a Description — this tells the AI when and why to use this knowledge base. When a profile has access to multiple libraries, the AI uses the description to determine the most relevant source for each inquiry

4. Save changes

Adding Documents and Sources

1. Select the edit icon for the applicable knowledge base

2. Click "Add Document" to add a new source

   

3. Choose the source type:

   • File Upload — Upload a file directly from the local machine (PDF, Word document, etc.)
   • Web Extraction — Extract content from a specific web page using the built-in Web Extractor tool

4. Save changes

Using the Web Extractor Tool

1. While logged into the installed Contact Center application, navigate to the knowledge base, add a new source, select Web Extraction, and then select Open Extractor
2. Enter the desired URL into the address bar and select Load
3. When the desired page is displayed, select Extract Current Page
4. Navigate to additional pages by entering a new URL or by clicking links within the open page, then extract each page as needed
5. Extracted pages populate on the left side of the extractor with their processing status
6. Close the extractor when finished and save changes in the menu

Editing a Knowledge Base Description

To modify a knowledge base description after creation, select the edit description icon inline with the library name, make changes, and select Apply.

Description updates take effect for subsequent AI interactions and do not require re-uploading documents.

Content Best Practices

The AI Agent's responses are grounded entirely in the content of its assigned knowledge bases. The quality and structure of that content directly affects response accuracy.

Document formatting:

• Use headers and subheaders to organize topics
• Keep paragraphs short and scannable
• Use bullet points for key information
• Avoid content that is primarily images — AI relies on text
• Ensure scanned PDFs use selectable text rather than flat image scans

Knowledge base organization:

• A single well-structured knowledge base is preferable to fragmented libraries — the AI navigates one source more efficiently than two with overlapping content
• A second knowledge base is most effective when content is completely separate in audience or subject (e.g., a customer-facing library vs. an internal employee-only library, or two unrelated product lines with no shared process)
• Each knowledge base should have a clear, accurate description — descriptions are the AI's primary tool for choosing which source to consult

Validating uploads:

• Use the Markdown Preview within the UI to check the content, layout, and readability of uploaded documents before deploying the agent
• Use the Test Agent tool within the AI Agent Profile to simulate real customer prompts and verify that the correct knowledge base is selected and that responses are accurate

Troubleshooting inaccurate responses:

1. Check the source document — is the answer clearly stated? If not, revise the document or create a focused article on that topic
2. Check the knowledge base description — if the AI selected the wrong library, update the description to better scope its purpose

For full guidance on testing and optimizing prompts, see the AI Knowledge Base Best Practices documentation on guide.xima.cloud.