Database Node

Overview

Nadoo AI provides two database node types for accessing structured data within workflows:

Database Node (database-node) — Execute SQL queries directly against connected relational databases.
Database Semantic RAG Node (database-semantic-rag-node) — Translate natural language questions into SQL queries using LLM-powered semantic understanding.

Together, these nodes let you build workflows that read, query, and reason over structured data alongside unstructured document retrieval.

The Database Node executes parameterized SQL queries against a connected database and returns the results as structured data in the workflow context.

Configuration

{
  "type": "database-node",
  "config": {
    "connection_id": "conn_01HX7A9B2C3D4E5F",
    "query": "SELECT name, email, plan FROM users WHERE status = :status ORDER BY created_at DESC LIMIT :limit",
    "parameters": {
      "status": "{{user_status}}",
      "limit": 10
    },
    "output_variable": "query_results",
    "timeout_seconds": 30
  }
}

Parameter	Type	Default	Description
`connection_id`	string	—	ID of the database connection configured in the platform (required)
`query`	string	—	SQL query with named parameters using `:param` syntax
`parameters`	object	`{}`	Parameter values, supports workflow variable references
`output_variable`	string	`"query_results"`	Context variable to store the results
`timeout_seconds`	number	`30`	Maximum execution time before the query is cancelled

Output

{
  "query_results": {
    "columns": ["name", "email", "plan"],
    "rows": [
      {"name": "Alice Kim", "email": "alice@example.com", "plan": "enterprise"},
      {"name": "Bob Park", "email": "bob@example.com", "plan": "pro"}
    ],
    "row_count": 2,
    "execution_time_ms": 45
  }
}

Parameterized Queries

Always use parameterized queries (:param syntax) instead of string interpolation. Parameterized queries prevent SQL injection and ensure proper type handling.

Parameters can reference workflow context variables:

{
  "query": "SELECT * FROM orders WHERE customer_id = :customer_id AND status IN (:statuses)",
  "parameters": {
    "customer_id": "{{user_id}}",
    "statuses": ["pending", "processing"]
  }
}

Supported Operations

The Database Node supports read-only queries by default for safety:

Operation	Supported	Notes
`SELECT`	Yes	Full query support with joins, aggregations, subqueries
`INSERT`	Configurable	Must be explicitly enabled in connection settings
`UPDATE`	Configurable	Must be explicitly enabled in connection settings
`DELETE`	Configurable	Must be explicitly enabled in connection settings
`DDL`	No	Schema modifications are never permitted through workflow nodes

Database Semantic RAG Node

The Database Semantic RAG Node translates natural language questions into SQL queries using an LLM, executes the generated SQL, and returns the results. This enables non-technical users to query databases using plain English.

How It Works

Schema Retrieval

The node retrieves the database schema (tables, columns, types, relationships) for the connected database.

Natural Language to SQL

The user’s question and the database schema are sent to an LLM, which generates a SQL query.

Query Validation

The generated SQL is validated for safety (no destructive operations) and syntactic correctness.

Execution

The validated query is executed against the database, and results are returned.

Configuration

{
  "type": "database-semantic-rag-node",
  "config": {
    "connection_id": "conn_01HX7A9B2C3D4E5F",
    "question": "{{user_message}}",
    "model": "gpt-4o",
    "schema_filter": {
      "include_tables": ["users", "orders", "products"],
      "exclude_columns": ["password_hash", "api_key"]
    },
    "max_rows": 100,
    "output_variable": "semantic_results"
  }
}

Parameter	Type	Default	Description
`connection_id`	string	—	Database connection ID (required)
`question`	string	—	Natural language question to translate to SQL
`model`	string	`"gpt-4o"`	LLM model used for SQL generation
`schema_filter.include_tables`	string[]	all tables	Restrict which tables the LLM can query
`schema_filter.exclude_columns`	string[]	`[]`	Hide sensitive columns from the LLM
`max_rows`	number	`100`	Maximum number of rows to return
`output_variable`	string	`"semantic_results"`	Context variable for the results

Output

{
  "semantic_results": {
    "question": "How many orders were placed last month?",
    "generated_sql": "SELECT COUNT(*) as order_count FROM orders WHERE created_at >= '2026-02-01' AND created_at < '2026-03-01'",
    "columns": ["order_count"],
    "rows": [{"order_count": 1247}],
    "row_count": 1,
    "execution_time_ms": 120
  }
}

The output includes the generated_sql so downstream nodes (or the user) can inspect exactly what query was executed.

Schema Filtering

Use schema_filter to control what the LLM sees:

include_tables — Only expose specific tables. This improves SQL generation accuracy by reducing noise and prevents the LLM from accessing tables with sensitive data.
exclude_columns — Hide sensitive columns (passwords, API keys, PII) from the schema. The LLM will not reference these columns in generated queries.

{
  "schema_filter": {
    "include_tables": ["products", "categories", "inventory"],
    "exclude_columns": ["cost_price", "supplier_contract_id"]
  }
}

Example: Data Analysis Workflow

A workflow that lets users ask questions about business data in natural language:

The user asks: “What were our top 5 products by revenue last quarter?”
Semantic RAG Node generates and executes the SQL query.
Condition Node checks if any results were returned.
AI Agent Node summarizes the tabular results into a natural language response with insights.

Database Connections

Database connections are configured at the platform level and referenced by connection_id in workflow nodes. Supported databases include:

Database	Connection String Format
PostgreSQL	`postgresql://user:pass@host:5432/dbname`
MySQL	`mysql://user:pass@host:3306/dbname`
SQLite	`sqlite:///path/to/database.db`

Database connections are managed in Settings > Database Connections in the platform UI. Connection credentials are encrypted at rest.

Best Practices

Always use parameterized queries

Never concatenate user input directly into SQL strings. Use the :param syntax and the parameters object to prevent SQL injection.

Filter schemas for Semantic RAG

The fewer tables and columns the LLM sees, the more accurate the generated SQL will be. Always use schema_filter to expose only the tables relevant to the workflow’s purpose.

Set reasonable timeouts and row limits

Prevent runaway queries with timeout_seconds and max_rows. A query that returns 10,000 rows will overwhelm the workflow context and downstream LLM processing.

Validate Semantic RAG output

The generated SQL may not always be correct. For critical workflows, add a review step where the user can confirm the generated query before execution, or use a Condition Node to verify result reasonableness.

Next Steps

Search Knowledge

Combine structured and unstructured data retrieval

Python Code Node

Post-process query results with custom logic

Condition Node

Branch based on query results

AI Agent Node

Summarize and analyze database results with LLM

Workflow Engine

Node Types

AI Agent Strategies

Database Node

Overview

Database Node

Configuration

Output

Parameterized Queries

Supported Operations

Database Semantic RAG Node

How It Works

Configuration

Output

Schema Filtering

Example: Data Analysis Workflow

Database Connections

Best Practices

Next Steps

Search Knowledge

Python Code Node

Condition Node

AI Agent Node

Workflow Engine

Node Types

AI Agent Strategies

​Overview

​Database Node

​Configuration

​Output

​Parameterized Queries

​Supported Operations

​Database Semantic RAG Node

​How It Works

​Configuration

​Output

​Schema Filtering

​Example: Data Analysis Workflow

​Database Connections

​Best Practices

​Next Steps

Search Knowledge

Python Code Node

Condition Node

AI Agent Node

Overview

Database Node

Configuration

Output

Parameterized Queries

Supported Operations

Database Semantic RAG Node

How It Works

Configuration

Output

Schema Filtering

Example: Data Analysis Workflow

Database Connections

Best Practices

Next Steps