Knowledge Management

This workflow covers adding, organizing, and maintaining the RAG knowledge base that AI Core uses to provide informed, context-aware responses.

Overview

AI Core includes a knowledge system that supplements the LLM’s general capabilities with company-specific information. When a user asks a question, the AI searches the knowledge base for relevant content and includes it in its reasoning context. This enables the AI to answer questions about your specific processes, policies, and procedures that a general-purpose LLM would not know.

Required permissions

  • AI Manager to create, edit, and delete knowledge sources

  • AI User (read-only access to knowledge during conversations)

Knowledge source types

Each knowledge source has a type that helps organize and prioritize search results:

Type

Description

Model Metadata

Auto-generated descriptions of whitelisted model schemas (fields, types, relationships). Refreshed via the Refresh All Metadata button.

Documentation

Product manuals, system documentation, or help articles.

Procedure

Step-by-step operational procedures (e.g., “How to process a return”).

FAQ

Frequently asked questions with their answers.

Custom

Any other reference material that does not fit the above categories.

Add a knowledge source

  1. Navigate to AI Core ‣ Configuration ‣ Knowledge Sources.

  2. Click New.

  3. Fill in:

    • Name – A descriptive title (e.g., “Return Policy Procedure”)

    • Source Type – Select the appropriate type

    • Content – The full text content. Write in clear, structured language. The AI searches this content when looking for relevant information.

    • Summary (optional) – A brief overview for quick reference

    • Category (optional) – A grouping label (e.g., “sales”, “hr”, “inventory”)

    • Related Model (optional) – Link to a specific Progrid model if the knowledge is model-specific

    • Keywords (optional) – Comma-separated search terms to improve matching

    • Priority – Lower number = higher priority in search results (default: 10)

  4. Click Save.

Tip

Write knowledge content in a structured format with clear headings and bullet points. The AI performs keyword-based search, so include the terms users are likely to ask about.

Generate model metadata

Model metadata knowledge sources are auto-generated from your whitelisted models. They describe each model’s fields, types, and relationships so the AI understands your data structure.

To regenerate all model metadata:

  1. Navigate to AI Core ‣ Configuration ‣ Knowledge Sources.

  2. Filter by Model Metadata type.

  3. Click Refresh All Metadata on any model metadata record.

  4. The system regenerates metadata for all whitelisted models.

Note

Run this after adding new models to the whitelist or after significant schema changes (new fields, renamed models) to keep the AI’s understanding current.

Organize knowledge sources

Use the Category field to group related knowledge sources. Common categories include:

  • sales – Sales processes, pricing rules, discount policies

  • hr – HR procedures, leave policies, onboarding guides

  • inventory – Warehouse procedures, shipping rules

  • finance – Invoicing procedures, payment terms, tax rules

  • general – Company-wide policies, IT procedures

The Related Model field links knowledge to a specific model. When a user asks about that model, the linked knowledge is prioritized in search results.

Archive outdated knowledge

To remove outdated knowledge without deleting it:

  1. Open the knowledge source.

  2. Click Archive (or uncheck Active).

  3. The source is excluded from AI searches but preserved for reference.

Archived sources can be restored by checking Active again.

Next steps