The Knowledge Base section allows AIUNIFY Call Center customers to create collections of business information that AI Agents can reference during telephone conversations.
A Knowledge Base may contain:
After a Knowledge Base is created, it can be assigned to an AI Agent from the AI Agent creation or editing page.
When the AI Agent responds to a caller, AIUNIFY Call Center adds the selected Knowledge Base information to the Agent’s conversation context.
The AI Agent’s system prompt explains its role and behavior.
The Knowledge Base provides the business facts the Agent may need while performing that role.
For example:
The system prompt controls how the Agent behaves.
The Knowledge Base supplies information the Agent can use while speaking.
The system prompt and Knowledge Base serve different purposes.
The system prompt should define:
The Knowledge Base should contain:
Avoid placing all business information inside the system prompt.
Separating instructions from business information makes both areas easier to maintain.
The Knowledge Base section is an advanced Conversational AI feature.
The Customer account must have approved Business KYC before opening the standard Knowledge Base pages.
Without Business KYC approval, the Customer may be redirected to the verification section or prevented from opening the feature.
To resolve access restrictions:
Business KYC requirements are discussed in the Customer Account Settings and KYC chapter.
To open Knowledge Bases:
The page is located at:
The Knowledge Base page displays saved Knowledge Base records and management controls.
The Knowledge Base page may display:
Each Knowledge Base appears as a card.
When the content is longer than the available card area, only a short preview is shown.
Open Edit to review the complete stored content.
When no Knowledge Bases are available, the page displays an empty-state message such as:
The page may also instruct the Customer to create the first Knowledge Base.
To begin:
The Search field checks both:
To search:
Examples of useful searches include:
The search is performed on the records currently loaded into the page.
The current page requests up to 100 Knowledge Base records.
Search is then performed within those loaded records.
When the platform contains more than 100 Knowledge Bases, older or additional records may not appear in the current page or search results.
Businesses should use clear names and avoid creating unnecessary duplicate Knowledge Bases.
To create a Knowledge Base:
The creation page is located at:
The name is required.
It can contain up to 255 characters.
Use a name that clearly identifies the subject and intended AI Agent.
Examples include:
Avoid names such as:
A clear name makes the correct Knowledge Base easier to select from the AI Agent configuration page.
The Create Knowledge Base page provides three content-entry tabs:
Each method produces text that is stored in the Knowledge Base content field.
Select the method based on the source of the information.
Plain Text is the most direct method.
To use it:
Plain Text is recommended when:
The backend allows Knowledge Base content containing up to approximately:
Content exceeding the supported limit will fail validation or be shortened during certain imports.
The fact that the system can store 500,000 characters does not mean the AI Agent uses all 500,000 characters during every conversation.
The current conversation engine normally includes only the first portion of the selected Knowledge Base in the AI context.
The File tab allows the Customer to upload a text-based file.
Supported file types are:
The maximum file size is:
To upload a file:
The server reads the contents of the file and stores the resulting text.
A plain-text file is usually the safest upload option.
It contains little or no formatting and is easy for the system to process.
A Markdown file may contain headings, lists, links, and basic text formatting.
Markdown structure can help organize the content.
A comma-separated file can store structured information such as:
The system stores the CSV text. It does not automatically transform every row into a special searchable database.
An HTML file can be uploaded, but the file-upload workflow stores its raw contents.
HTML tags may therefore remain in the saved Knowledge Base.
For cleaner Agent context, consider converting HTML files to plain text before uploading.
The uploaded file may be as large as 10 MB, but the stored content is limited to approximately 500,000 characters.
When the extracted file content exceeds that limit, the system shortens it before storage.
Important information should appear near the beginning of the file.
Do not rely on information located near the end of a very large document.
The File tab displays the selected file’s text in a preview field and allows the preview to be edited.
However, in the current source version, saving while the File tab remains selected uploads the original selected file.
Changes typed only into the preview field may not be included in the saved content.
To make sure edits are preserved:
Always reopen the Knowledge Base after saving and confirm that the expected content was stored.
Before uploading:
A clean, focused file produces better AI responses than an unedited document dump.
The URL tab allows the Customer to retrieve text from a publicly accessible webpage.
To import a webpage:
A valid URL must begin with:
or:
HTTPS is recommended.
When a URL is fetched, AIUNIFY Call Center:
The URL itself is not stored as a reusable content source in the current Knowledge Base record.
Only the extracted content is saved.
The URL importer retrieves one webpage.
It does not automatically crawl:
To import several webpages, the Customer must retrieve and combine the necessary information manually or create separate Knowledge Bases.
Imported webpages often contain unrelated text, including:
Always review the fetched content before saving.
Delete anything the AI Agent should not use.
A URL import may fail when the page:
A page that looks correct in a browser may still block server-based retrieval.
Some modern webpages load their primary content after the page opens through JavaScript.
The current URL importer performs a direct server request. It does not operate a full browser.
As a result, dynamically generated text may be missing.
When this occurs:
Only import content from trusted sources.
A webpage can contain:
AI-generated systems can be affected by instructions embedded inside imported content.
After fetching a webpage:
Do not automatically trust imported webpage content.
After the webpage is fetched, the content appears in an editable text area.
The Customer can:
Edits made to fetched URL content are included when the Knowledge Base is saved.
The system technically allows a Knowledge Base to be created with:
However, an empty Knowledge Base provides no information to the AI Agent.
Create an empty record only when content will be added immediately afterward.
An Agent assigned to an empty Knowledge Base will not receive a useful Knowledge Base context section.
Use clear sections.
A recommended structure is:
Clear headings help administrators maintain the information and help ensure the most important information appears early.
Frequently asked questions work well in a direct Q&A format.
Example:
Use clear and complete answers.
Avoid answers that depend on unstated context.
The current AI conversation engine normally uses only the first approximately 4,000 characters of the selected Knowledge Base.
Place the most important and frequently requested information near the beginning.
Recommended order:
Information near the end of a long Knowledge Base may not be included in the Agent’s active conversation context.
Although the Knowledge Base record can store much more information, the default AI Agent configuration limits the amount inserted into a conversation prompt to approximately 4,000 characters.
This limit helps reduce:
The limit can be changed by the platform administrator, but Customers should design content around the current default.
The current source version does not perform advanced document retrieval, document chunking, embeddings, or semantic search across the entire Knowledge Base.
Instead, it places a limited portion of the selected Knowledge Base directly into the AI Agent’s system context.
This means:
Customers should not treat the current Knowledge Base as an unlimited document-search system.
Customers can create multiple Knowledge Bases for different purposes.
Examples include:
Multiple AI Agents can use the same saved Knowledge Base.
However, each AI Agent can select only one saved Knowledge Base at a time in the current configuration.
An AI Agent has one saved knowledge_base_id field.
The Agent can therefore use:
The current interface does not allow several saved Knowledge Bases to be selected simultaneously for one Agent.
To combine several topics:
Keep the combined record concise enough that important information remains within the active context limit.
To assign a Knowledge Base while creating an AI Agent:
The Knowledge Base dropdown loads available saved records by name.
To change an existing Agent:
The selected Knowledge Base becomes the Agent’s saved Knowledge Base reference.
The AI Agent page supports two methods.
Select a record from the dropdown.
Advantages include:
Select:
Then enter information directly inside the Agent configuration.
Advantages include:
Use a saved Knowledge Base when information must be shared or updated centrally.
When a saved Knowledge Base is selected, the Agent configuration uses the saved record instead of the custom inline Knowledge Base field.
The interface clears or hides the custom-text area while a saved record is selected.
Before replacing custom Agent text with a saved Knowledge Base, copy any custom information that must be preserved.
To stop using a saved Knowledge Base:
Custom text that was cleared when a saved Knowledge Base was selected may need to be entered again.
During a conversation, the system builds an Agent prompt containing:
The Knowledge Base appears as context, while the system prompt appears as instructions.
The Agent then uses the combined information to generate a response.
The AI Agent generates a response based on its available context.
A Knowledge Base improves consistency, but it does not guarantee that the Agent will repeat every sentence exactly.
The Agent may:
Customers must test important answers before using an Agent with live callers.
When a rule must be followed exactly, state it clearly.
Weak example:
Better example:
Avoid vague words such as:
Use them only when the business policy is genuinely flexible.
Do not include conflicting statements.
Example of a conflict:
and later:
The AI Agent may choose either statement.
Before saving:
Knowledge Base content does not automatically update when business information changes.
Time-sensitive information may include:
Review and update these details regularly.
Avoid including temporary promotions in a permanent Knowledge Base unless someone is responsible for removing them.
Do not place sensitive credentials in a Knowledge Base.
Never include:
An AI Agent may repeat information placed in its Knowledge Base.
Store only information the Agent is authorized to communicate.
Avoid placing personal customer data or regulated information into a general Knowledge Base.
A shared Knowledge Base should not contain:
Call-specific information should come from approved call context or a properly secured business system rather than a general reusable Knowledge Base.
To edit:
The editing page is located at:
The existing content loads into the Plain Text field.
To edit:
This is the most reliable method for small corrections.
To replace the stored content using a file:
The file’s contents replace the current stored content.
The system does not automatically merge the old and new files.
Keep an external backup before replacing important information.
To replace content using a webpage:
The fetched text replaces the current content when saved.
The system does not retain the source URL for automatic future updates.
A Knowledge Base imported from a URL is a stored copy of the text retrieved at that moment.
It does not automatically update when the webpage changes.
To refresh it:
Customers should establish a schedule for reviewing information imported from websites.
The current Knowledge Base feature does not provide customer-facing:
Before making major changes:
This provides a manual rollback option.
AIUNIFY Call Center caches the assembled AI Agent system prompt for approximately 10 minutes.
The cache is connected to the AI Agent record and the platform’s base instruction file.
Editing only the Knowledge Base may not immediately invalidate the Agent’s cached prompt.
As a result, an AI Agent may continue using the previous Knowledge Base content for up to approximately 10 minutes.
After an important update:
Do not test only within a call that began before the update.
One saved Knowledge Base may be assigned to several AI Agents.
This is useful when Agents share:
Editing a shared Knowledge Base can affect every AI Agent using it.
Before changing a shared record:
A support Agent and a sales Agent may interpret the same information differently because they have different system prompts.
To delete:
The interface warns that deletion cannot be undone.
Delete only when the Knowledge Base is no longer required.
The AI Agent table uses a relationship that is cleared when the selected Knowledge Base is deleted.
This prevents the Agent record itself from being deleted.
However, the affected AI Agent will no longer have that saved Knowledge Base assignment.
The Agent may then operate without useful Knowledge Base context unless custom text is available.
After deletion:
Do not delete a Knowledge Base before reviewing its AI Agent assignments.
The Knowledge Base list does not currently display how many AI Agents use each record.
Before deleting or significantly changing a Knowledge Base, manually review AI Agent configurations.
A future interface improvement could display assigned Agent counts.
The current source version creates Knowledge Base records without a Customer or user ownership field.
The Knowledge Base API also retrieves records without filtering them by the signed-in Customer.
This means saved Knowledge Base records are not fully isolated by Customer account in the current implementation.
Until the platform is updated to enforce Customer ownership:
This is an important current-version security limitation.
The platform developer should update the Knowledge Base feature to include:
user_id or Customer-owner IDAfter that improvement, Customers should see and manage only records owned by their own account.
After assigning a Knowledge Base:
Use questions such as:
Also ask questions with information that should not be available.
The Agent should not invent unsupported answers.
Callers may ask the same question in different ways.
Test variations such as:
The Knowledge Base should provide enough clear information for the Agent to answer reasonable variations.
After testing:
The spoken response may sound acceptable while still containing a factual error.
Always review the transcript.
The Knowledge Base can explain when a caller should be transferred, but actual call-transfer behavior is controlled through the AI Agent settings.
Use the system prompt and transfer settings for instructions such as:
Use the Knowledge Base for supporting facts such as:
Do not rely on Knowledge Base text alone to activate the technical transfer function.
AI Agents assigned to SMS-capable numbers may also use their configured instructions and business context when generating text responses.
Keep Knowledge Base responses suitable for both voice and SMS when the same Agent handles both channels.
Avoid:
Create a separate Agent and Knowledge Base when voice and SMS require substantially different behavior.
Confirm:
Agents may not receive Knowledge Base management access unless specifically authorized.
Confirm:
Review the validation message displayed by the page.
Confirm that the file extension is:
Also confirm:
PDF, DOCX, XLSX, image, audio, and video files are not supported by the current Knowledge Base uploader.
The current File tab uploads the selected source file.
Edits made only inside the preview may not replace the file contents during submission.
Correct the original file or save the corrected text through the Plain Text tab.
Reopen the record afterward to verify the result.
Confirm:
When retrieval continues to fail, copy the page’s approved text manually.
The page may load its content dynamically through JavaScript.
The importer does not run a full web browser.
Copy the required information from the webpage and use Plain Text instead.
URL extraction removes HTML but may still produce:
Edit the fetched content before saving.
Use headings and Q&A formatting to improve clarity.
Confirm:
Sign out and sign back in when necessary.
Possible causes include:
Wait approximately 10 minutes, resave the Agent, and begin a new test call.
The stored Knowledge Base may exceed the active conversation-context limit.
Move the required information closer to the beginning.
Shorten or divide the content.
Remove duplicate and irrelevant sections.
Review:
State prices explicitly.
Example:
Avoid presenting several unrelated numbers without clear labels.
Add an instruction to the system prompt such as:
Also:
A Knowledge Base reduces unsupported answers but cannot eliminate them completely.
Deletion removes the Knowledge Base record and clears its relationship with assigned AI Agents.
Create or select a replacement record and reassign it to the affected Agents.
Deleted content cannot be restored through the current customer interface.
The current source does not fully isolate Knowledge Bases by Customer ownership.
Do not edit or delete an unfamiliar record.
Record its name and report it to the platform administrator.
Do not place sensitive content into the Knowledge Base feature until Customer-level data isolation has been corrected and verified.
Review active Knowledge Bases:
When prices, promotions, availability, or schedules change frequently.
For standard business information and frequently asked questions.
For stable policies, service descriptions, and general company information.
When any of the following changes:
Record the date of each review in an external change log.
Before assigning a Knowledge Base to a live AI Agent, confirm:
Knowledge Bases provide business information that AI Agents can reference during voice and messaging conversations.
The Knowledge Base feature requires approved Business KYC.
Customers can create content using:
Supported files include TXT, Markdown, CSV, and HTML, with a maximum file size of 10 MB.
The system can store approximately 500,000 characters, but the current AI Agent conversation engine normally places only the first approximately 4,000 characters into the Agent’s active context.
Important information should therefore be concise, clearly organized, and positioned near the beginning.
A saved Knowledge Base may be shared by multiple AI Agents, but each Agent can select only one saved Knowledge Base at a time.
Customers can edit, replace, search, and delete Knowledge Bases. Deleting a Knowledge Base clears its assignment from connected AI Agents.
The current version does not provide automatic website refresh, version history, semantic document retrieval, or complete Customer-level ownership isolation.
Customers should avoid confidential content, review all imported information, maintain an external backup, test every assigned AI Agent, and monitor call transcripts for accuracy.