genudo-knowledge-tables

GenuDo Knowledge Tables Management

Overview

Knowledge tables provide structured, vector-indexed factual grounding for GenuDo AI agents. Each table contains a defined column schema and row records retrieved via semantic hybrid search at runtime when leads ask factual or product questions.

Key Tools

Task	MCP Tool
View tables, columns, and attached pipelines	list_knowledge_tables
Create a new knowledge table schema	create_knowledge_table
Insert or update knowledge rows (vector points)	upsert_knowledge_points
Test runtime semantic retrieval	search_knowledge_table
Delete rows by stable ID	delete_knowledge_points

---

Step-by-Step Workflow

1. Discovery & Schema Inspection

Call list_knowledge_tables to inspect:

• Existing table names, descriptions, and IDs.
• Column schemas (e.g., product_name, pricing, features, eligibility).
• Which pipelines are linked to each table.

2. Table Creation

When creating a new table (create_knowledge_table):

• name: Descriptive identifier (e.g., summer_camp_pricing_2026).
• description: Clear "when-to-use" prompt guiding the AI router on when to query this table.
• columns: Array of column names that define the row schema.

3. Upserting Knowledge Points

Call upsert_knowledge_points to populate or update data:

• table_id: Target knowledge table ID.
• points: Array of data rows.
• Stable ID (default_id): Always assign a deterministic, stable ID (e.g., prod-pro-annual, faq-refund-policy). If a point with the same default_id exists, it updates in-place; otherwise, it inserts a new row.
• Complete Columns: Supply values for all defined schema columns to avoid incomplete facts.

4. Verifying Retrieval

Always test retrieval before declaring the knowledge base ready:

• Call search_knowledge_table with table_id, query, and optional limit.
• Use realistic customer phrasing (e.g., "how much does the family plan cost?").
• Verify that the expected row is returned in the top matches with high relevance.