AI Chat
Text-to-SQL
AI Chat
Text-to-SQL
The Text-to-SQL feature allows users to write natural language queries that are automatically converted into PostgreSQL SQL statements and executed against a PostgreSQL database. This feature leverages AI technology to interpret user intent and generate semantically correct SQL based on the database schema.
Architecture
The Text-to-SQL feature is implemented as a tool within the dba.ai platform with the following components:
- LLM Integration: Uses Anthropic’s Claude model to convert natural language to SQL
- SQL Validation: Parses and validates the generated SQL to ensure it’s safe to execute
- Query Execution: Executes the validated SQL against the specified database instance
- Security: Enforces read-only queries by default to prevent unintended data modifications
Implementation Details
SQL Generation Process
- The feature extracts the database schema (DDL) from the target database
- The natural language query and schema are sent to the LLM using a specialized prompt
- The LLM generates SQL code wrapped in
<sql>tags - The code is extracted, validated, and checked for read-only operations
- The SQL is executed against the database and results are returned to the user
Security Features
- Read-Only Enforcement: By default, only read operations (SELECT statements) are allowed
- SQL Injection Prevention: Generated SQL is parsed and validated before execution
- Permission Controls: Users must have proper authorization for the organization and instance
Usage
To use the Text-to-SQL feature:
- Select a PostgreSQL instance in the dba.ai interface
- Submit a natural language query describing what data you want to retrieve
- The system will convert your query to SQL, execute it, and display the results
Example Inputs
- “Show me all users who signed up last month”
- “What’s the average transaction amount by product category?”
- “Find the top 10 customers by order value”
Limitations
- Requires SQL execution to be enabled for the instance
- Limited to read-only operations by default
- Works best with clear, concise queries that reference existing tables and columns
- Complex queries may require refinement or manual SQL writing
Future Improvements
- Support for controlled write operations with additional safeguards
- Retry policy for invalid SQL generation
- Enhanced schema understanding for more complex queries
- Interactive query refinement
Was this page helpful?