Knowledge Base Management

General For Super Administrators Advanced Configuration Last updated: June 20, 2025 Version: 1.0

Knowledge Base Management

Learn how to manage AI assistant knowledge bases, including vector stores, file uploading, deduplication, and synchronization between help articles and AI assistants in the Shifts platform.

Overview

The Knowledge Base Management system in Shifts enables super administrators to manage the information sources that power the platform’s AI assistants. This includes uploading help articles to vector stores, managing database knowledge, and ensuring AI assistants have access to accurate, up-to-date information. This article explains how to effectively manage knowledge bases for optimal AI assistant performance.

Understanding Vector Stores

Vector stores are special databases used by AI assistants to efficiently retrieve relevant information when answering questions. In the Shifts platform:

  1. Help Article Vector Stores: Store help documentation for customer support
  2. Database Knowledge Vector Stores: Store database schema information and explanations
  3. Business-Specific Vector Stores: Store knowledge specific to individual businesses

Each vector store contains files that have been processed into embeddings (numerical representations of text) that allow the AI to quickly find relevant information.

Accessing Knowledge Base Management

To access knowledge base management features:

  1. Log in with super administrator credentials
  2. Navigate to System Admin > Help Articles
  3. Use the Vector Store Files tab to view all uploaded files
  4. Use the Upload to Assistant buttons to manage assistant knowledge

Managing Help Article Knowledge Bases

Uploading Help Articles to Assistants

To upload help articles to an AI assistant:

  1. Navigate to System Admin > Help Articles
  2. You can upload either:
    • All Articles: Click Upload All to Assistant button
    • Specific Article: Click the dropdown menu for an article and select Upload to Assistant
  3. The system will show an upload progress modal with real-time status
  4. Once completed, the articles will be available to the AI assistant

Associating Help Articles with Assistants

To set a default assistant for help articles:

  1. Navigate to System Admin > Help Articles
  2. When uploading, the system will automatically associate the assistant
  3. A banner will display the currently associated assistant
  4. To remove the association, click Remove Association

Monitoring Vector Store Files

To view all files in the vector store:

  1. Navigate to System Admin > Help Articles > Vector Store Files
  2. The page displays:
    • File ID: The OpenAI file identifier
    • Filename: The name of the uploaded file
    • Content Hash: Unique identifier for deduplication
    • Article: Associated help article
    • Vector Store ID: The OpenAI assistant’s vector store identifier
    • Business ID: Which business the file belongs to (or 0 for system)

Managing Database Knowledge Bases

Database knowledge bases store information about your database schema, common queries, and domain-specific knowledge for the Database Assistant.

Accessing Database Knowledge Management

  1. Navigate to System Admin > Database Knowledge Bases
  2. This page lists all database knowledge entries

Creating New Knowledge Entries

To add new database knowledge:

  1. Click Add New Knowledge Entry
  2. Fill in the form:
    • Title: A descriptive name
    • Category: Select from schema, queries, domain, examples, or tips
    • Content: The knowledge content (can include code examples, explanations, etc.)
  3. Click Create Knowledge Entry

Syncing Database Knowledge to Assistants

To sync knowledge with the Database Assistant:

  1. Navigate to System Admin > Database Knowledge Bases
  2. Click Sync All to Assistant
  3. The system will upload all entries to the Database Assistant
  4. A progress indicator will show the status of the upload

File Deduplication

The system automatically prevents duplicate files from accumulating in vector stores through intelligent deduplication:

How Deduplication Works

  1. Content Hashing: Each file’s content is hashed to create a unique identifier
  2. Filename Pattern Recognition: The system recognizes date suffixes in filenames
  3. Automatic Cleanup: Before uploading a new version of a file, older versions are removed
  4. Reference Tracking: The database maintains references between help articles and vector store files

Filename Pattern Examples

The system recognizes various date suffix patterns in filenames:

  • help_article_20250408.md
  • help_article_2025-04-08.md
  • help_article-20250408.md
  • help_article_2025_04_08.md

Multi-Business Knowledge Management

For organizations with multiple businesses:

Business-Specific Knowledge

  1. Isolated Knowledge: Each business has its own isolated knowledge base
  2. System-Wide Knowledge: Super administrators can upload system-wide knowledge (business_id = 0)
  3. Shared vs. Private: Choose which knowledge is shared across all businesses and which is business-specific

Business Knowledge Customization

To customize knowledge for specific businesses:

  1. Navigate to System Admin > Businesses > [select business]
  2. Click on AI Settings > Knowledge Management
  3. Configure which knowledge sources are available to this business
  4. Optionally upload business-specific documents

Managing Assistant Configurations

To configure how assistants use knowledge bases:

  1. Navigate to System Admin > Assistant Configurations
  2. Find the assistant you want to configure
  3. Click Edit
  4. In the Knowledge Sources section:
    • Enable/disable help article knowledge
    • Enable/disable database knowledge
    • Configure custom knowledge sources
    • Set knowledge retrieval parameters (number of results, relevance threshold)

Best Practices

For optimal knowledge base management:

  1. Regular Updates: Keep help articles and knowledge entries up to date
  2. Structured Content: Format help articles with clear headings and sections
  3. Semantic Titles: Use descriptive titles that capture the main topic
  4. Test Regularly: Ask assistants questions to verify knowledge retrieval
  5. Monitor Size: Keep track of vector store size and number of files
  6. Version Control: Maintain article versioning for important updates
  7. Business Relevance: Ensure business-specific knowledge is properly isolated

Troubleshooting

Common Issues

  1. Upload Failures:
    • Check the OpenAI API key is valid
    • Verify the assistant configuration is correct
    • Look for error messages in the upload log
  2. Knowledge Not Retrieved:
    • Verify the file was successfully uploaded (check Vector Store Files)
    • Check if the content is well-structured and clear
    • Test with different phrasings of the question
  3. Duplicate Content:
    • If duplicate content appears, manually check the Vector Store Files
    • Delete unnecessary duplicates using the OpenAI dashboard
    • Re-upload the knowledge with a clear filename

Related Resources

This article should be updated when:

  1. Vector store management features are enhanced
  2. New knowledge source types are added
  3. Deduplication algorithms are updated
  4. Assistant configuration options change
  5. Knowledge synchronization methods are modified