AI Database Builder

Turn project requirements into a database proposal you can review.

A Java command-line tool that uses Gemini to propose a MySQL schema from a natural-language project brief. It validates the design and prints the SQL before any database operation. Creation requires both --apply and an explicit confirmation.

Try it without an API key or database

Requirements: Java 17 and Maven 3.9+.

mvn --batch-mode verify
java -jar target/ai-database-builder-0.1.0-SNAPSHOT.jar --example

The bundled fictional example proposes a projects table with an ID, title and creation timestamp. Preview mode makes no database connection.

Generate a proposal

Export GEMINI_API_KEY in your shell, then run the JAR without arguments. GEMINI_MODEL selects the model; its default is gemini-3.8-flash. Check Google's model documentation for availability in your account. Requests send the project description to Google's Gemini API and may consume quota.

Configuration names are documented in .env.example. The Java application reads environment variables; it does not load .env automatically. Never put credentials in source files.

Review and create

You can also review a saved design:

java -jar target/ai-database-builder-0.1.0-SNAPSHOT.jar --design design.json
java -jar target/ai-database-builder-0.1.0-SNAPSHOT.jar --design design.json --apply

For creation, export MYSQL_HOST, MYSQL_PORT, MYSQL_USER and MYSQL_PASSWORD. Inspect the SQL and type yes when prompted. Use a disposable database server first. The tool creates a new database and stops on the first error; it does not overwrite an existing database. MySQL DDL is not transactional, so a failed run can leave partially created objects for inspection.

Design boundaries

• Names use ASCII letters, digits and underscores, up to 64 characters.
• Types are allowlisted: common numeric/text/date types, VARCHAR(n) and DECIMAL(p,s).
• A proposal supports up to 50 tables and 100 columns per table.
• Relationships, indexes, keys, constraints and migration planning require further design work; this is a schema prototyping tool.
• A valid response is not necessarily a good data model. Review normalization and business rules yourself.

Verification

Unit tests cover request escaping, response parsing, SQL validation, approval cancellation and failure handling. CI additionally creates and removes an isolated test database on a disposable MySQL service. No live Gemini request is required for tests.

License · Moayad Alshehry