6.1 — Generating the database schema

VibeMap designs a complete database schema from your project's features, user stories, and pages. The AI identifies every data entity your application needs, defines columns and types, and maps the relationships between them. The result is a starting-point schema your AI coding agent (or you) can extend in your real database.

Tip — generate pages first. Schema is intentionally the last Plan step, after Pages: page layouts tell the schema generator what data each screen actually displays and edits, which produces noticeably more complete tables. You can generate schema earlier, but pages-first is the recommended order (it's the order the sidebar and journey strip suggest).

How schema generation works

1. Navigate to the Schema tab in the project sidebar.
2. Click Generate Schema.
3. The AI analyzes your features, user stories, and acceptance criteria.
4. It identifies data entities (users, projects, payments, etc.) and determines how they relate.
5. A complete schema is generated, typically in 30–60 seconds.

The AI applies standard database design principles: sensible normalization, consistent naming conventions (snake_case), and automatic audit fields on every table.

What gets generated

Tables and columns

Each table includes:

• Primary key — UUID id field
• Typed columns — text, integer, boolean, timestamp, jsonb, enum, etc., matched to each field's purpose
• Constraints — NOT NULL, UNIQUE, CHECK, and DEFAULT values where appropriate
• Audit fields — created_at and updated_at timestamps on every table

Relationships

The AI maps connections between tables:

• One-to-many — e.g., a user has many projects (foreign key on projects.user_id)
• Many-to-many — e.g., projects and tags linked through a junction table
• One-to-one — e.g., a user has one profile record
• Cascade rules — ON DELETE and ON UPDATE behaviors set per relationship

Exploring your schema

VibeMap provides two primary views for working with the generated schema.

Diagram view

An interactive diagram that visualizes your entire data model.

• Table nodes show the table name and its columns
• Relationship lines connect foreign keys to their referenced tables
• Drag and zoom to organize the layout for large schemas
• Click a table to see full column details in the side panel

This is the best view for understanding how your data entities connect at a glance.

Tables view

A detailed, searchable list of every table and its columns.

• Expand any table to see every column with its type, constraints, and default value
• Search and filter to find specific tables or column names
• Column metadata — nullable status, unique constraints, index presence

Use the tables view when you need to inspect or edit individual fields.

How the schema connects to your project

The schema doesn't exist in isolation — it traces directly back to your features:

• A feature like "Team Management" produces tables for teams, team_members, and invitations.
• User stories under that feature define which columns each table needs (e.g., "As a team admin, I want to set member roles" adds a role column to team_members).
• Pages that display or modify data reference these tables through their data dependencies.

If you regenerate features or add new user stories, regenerating the schema will incorporate those changes.

Regenerating the schema

Once a schema exists, the Regenerate Schema button opens a dialog with two modes:

• Update existing schema (default) — the AI compares your current schema against your features, user stories, and pages, and merges in whatever is missing: new tables, columns, and relationships. Nothing outside the changes it proposes is touched, so manual edits are preserved. This is the right choice after adding or changing features.
• Regenerate from scratch — generates a fresh schema and replaces the current one. The replacement is atomic: your existing schema is only removed after the new one has been generated and parsed successfully, so a failed run never leaves you without a schema.

The dialog also has an optional guidance field where you can steer the run in plain language — for example, "Add tables for the new notification feature" or "Split the settings JSON column into proper columns."

Schema generation and regeneration are available on Starter plans and above. Plans with the manual model selector can also pick which AI model runs the regeneration; other plans use the automatically selected model.

Tips

• Generate features and pages first. The more context the AI has, the more accurate the schema.
• Review relationships in the diagram view before moving forward — missing or incorrect foreign keys are easier to spot visually.
• Edit tables directly — you can add columns, rename fields, or adjust types without regenerating the entire schema.
• Treat it as a starting point. Once your AI coding agent picks up work via the MCP handoff, it can extend the schema in your real database. The generated schema is the v1 starting shape, not the final word.

Next steps

With your schema defined, proceed to the Conversational agent to refine your project, or return to Pages to adjust your UI.