Skip to content

SQL Connector

Key Concepts

The SQL Connector block allows seamless integration between your SQL database and an LLM-powered workflow in INTELLITHING. It is designed to be:

  • Robust: Capable of handling large-scale and complex database systems.
  • Scalable: Supports high-volume query handling.
  • Secure: Includes built-in mechanisms to detect and prevent prompt injection attacks.
  • Reliable: Features multiple evaluation and correction layers for accurate data retrieval.

Key Definitions

Term Definition
SQL Connector A block in INTELLITHING that enables secure, intelligent access to SQL databases.
LLM Large Language Model that generates and interprets natural language queries.
Database Configuration The set of credentials and metadata used to establish a connection to a SQL database.
Workflow Node A stage in the workflow pipeline that performs a specific action, such as initializing or executing queries.

Setup Guide: Configuring a New SQL Connection

Follow the steps below to connect a SQL database to INTELLITHING:

1. DB Type

  • Choose the database type (e.g., MySQL, PostgreSQL) from the dropdown list.
  • This selection determines the database driver and connection method.

2. DB Host

  • Enter the host address of your database.
    Example: 127.0.0.1 or database.example.com

3. DB Port

  • Provide the port used by your database:
  • MySQL: 3306
  • PostgreSQL: 5432

4. DB Name

  • Specify the name of the target database.
  • Ensure the database exists and contains the relevant tables.

5. DB User

  • Input the username with permission to access the necessary tables.

6. DB Password

  • Enter the corresponding password for the DB user.
  • Use the eye icon to toggle visibility if needed.

7. DB Table

  • Enter the name(s) of the table(s) you want to access.
  • Click Add to DB Tables to include the table in your configuration.
  • This helps scope access to only necessary data—important for large databases.

8. Description

  • Provide a description for the router engine to contextualize this connection.
    Example: Database of sales data. Used to answer queries related to sales.

  • Optional in custom workflows but recommended for maintainability.

9. Save

  • Click Save to store your configuration and return to the main editor pane.

Workflow Nodes

The SQL Connector block includes a sequence of nodes for processing SQL queries. Each node contributes to secure and structured interaction with your database.

Node Descriptions

initialize_workflow (First Node)

  • Initializes the environment, resources, and configurations required to begin workflow execution.

index_all_tables

  • Scans and indexes all defined tables in the connected database.
  • Organizes data for efficient retrieval and future operations.

prepare_table_retrieval

  • Analyzes the database schema to prepare for targeted table retrieval.

get_vector_index_dict

  • Generates a vectorized index dictionary from the tables for similarity-based or semantic query matching.

retrieve_table_context

  • Retrieves relevant context from the selected table(s) based on the user query or workflow need.

generate_sql_prompt

  • Constructs a database-aware prompt to guide the LLM in generating valid SQL queries.

parse_sql_from_response

  • Parses and validates the SQL query returned by the LLM to ensure safe and executable syntax.

setup_sql_retriever

  • Configures the connection for executing the SQL query on the specified database.

execute_sql_and_generate_response (Last Node)

  • Executes the SQL query and processes the results.
  • Outputs a structured response for downstream blocks or for direct return to the user.

Best Practices for Successful SQL Integration

  • Verify credentials: Double-check database hostname, port, user, and password.
  • Test before use: After saving, test the connection to ensure proper configuration.
  • Use descriptions: Helps differentiate between multiple connections, especially with the default router.
  • Firewall considerations: If your database is behind a firewall, you can activate a proxy for your database under security tab.

Example Use Case

To integrate a customer data warehouse for analytics:

  1. Configure connection to customer_db with table user_activity.
  2. Use generate_sql_prompt to interpret natural language questions like "Show top users last month."
  3. Parse and execute the query with parse_sql_from_response and execute_sql_and_generate_response.
  4. Return results to the user via an LLM or connected application interface.