Using Database tools in an Agent

Configure agents to query and interact with your database using SQL queries and metadata retrieval.

​

Prerequisites

• Database tools already configured for your database type
• Project with agent building permissions
• Database server accessible via public IP

💡 Setup Required: Ensure you’ve configured your database tools first by following the setup guide for your specific database type.

​

Supported Database Types

• MySQL - Standard metadata + query pattern
• PostgreSQL - Standard metadata + query pattern
• MS SQL Server - Standard metadata + query pattern
• Snowflake - Standard metadata + query pattern
• MongoDB - Standard pattern with Find + Aggregate tools
• Databricks - Simplified query-only pattern

​

Agent Patterns

​

Agent Pattern (MySQL, PostgreSQL, MS SQL, Snowflake, MongoDB)

Most database tools use a two-agent architecture:

1. Metadata Retrieval Agent - Scheduled agent that extracts database schema
2. Query Agent - Interactive agent that answers user questions

Required Components:

• Database-specific SQL Query Tool
• Database-specific Metadata Retrieval Tool
• Memory object (for metadata storage)
• LLM model

​

Simplified Pattern (Databricks)

Databricks uses a single-agent:

1. Query Agent - Direct SQL query execution with LLM processing

Required Components:

• Databricks SQL Query Tool
• LLM model

​

Setting Up Standard Database Agents

​

Step 1: Create Metadata Retrieval Agent

1. Create a new agent in Agent Canvas

2. Add the following components:

   • LLM model
   • [Database] Metadata Retrieval Tool (e.g., “MySQL Metadata Retrieval Tool”)
   • Memory object

3. Configure connections:

   • Input → LLM
   • LLM → Output
   • Memory → Output

4. Add the Metadata Retrieval Tool to the LLM

5. Schedule the agent:

   • Click the gear icon on the Input block
   • Toggle Schedule Execution ON
   • Set repeat interval based on how often your database schema changes

💡 Scheduling Tip: Run metadata retrieval daily for frequently changing databases, or weekly for stable schemas.

​

Step 2: Create Query Agent

1. Create a new agent in Agent Canvas
2. Add the following components:
   • LLM model
   • [Database] SQL Query Tool (e.g., “MySQL SQL Query Tool”)
   • Memory object (same as metadata agent)

For MongoDB: Add both MongoDB Find Tool and MongoDB Aggregate Tool

3. Configure connections:
   • Input → LLM
   • LLM → Output
   • Memory → LLM

4. Add the SQL Query Tool(s) to the LLM

⚠️ Important: Run the Metadata Retrieval Agent at least once before using the Query Agent to populate database schema information.

​

Setting Up Databricks Agents

​

Single Query Agent Setup

1. Create a new agent in Agent Canvas

2. Add the following components:

   • Databricks SQL Query Tool
   • LLM model

3. Configure connections:

   • Input → Databricks SQL Query Tool
   • Tool → LLM
   • LLM → Output

4. Configure the SQL Query Tool:

   • Click on the Databricks SQL Query Tool
   • Set it to receive value from Input Step Result

​

Using Parameterized Queries

For dynamic queries, use Agent Variables in your SQL statements:

1. Click on the [Database] SQL Query Tool (e.g. Databricks SQL Query Tool)
2. Enter your SQL query with variables: SELECT * FROM {{table_name}} WHERE date > '{{start_date}}'
3. Save the configuration

Variables will appear as blue tags and can be set dynamically at runtime.

​

Using Your Database Agents

​

Query Examples

Ask your agents natural language questions about your data:

"Show me all customers from the last month"
"What are the top 10 products by sales?"
"Find all orders with status 'pending'"
"Summarize revenue by region for Q4"

​

MongoDB-Specific Queries

"Find all documents in the users collection where age > 25"
"Aggregate sales data by product category"
"Show me the most recent orders from each customer"

​

Direct SQL Queries

For SQL Tools, you can pass SQL queries directly:

SELECT customer_id, SUM(order_value) 
FROM orders 
WHERE order_date >= '2024-01-01' 
GROUP BY customer_id

​

Best Practices

💡 Tip: Add custom prompts to your LLM with:

• Database schema context
• Common query patterns for your domain
• Output formatting preferences
• Business logic explanations

💡 Memory Management: Use the same Memory object across both agents to ensure consistent metadata access.

⚠️ Connection Requirements: All database servers must be accessible via public IP. Contact support@airia.com for IP whitelisting assistance.