AI Designer: Proposing New Tables From a Feature Description

Where AI Engineer finds and fixes problems in a schema that already exists, AI Designer works the other direction: describe a feature or module in plain language, and Gemini proposes whatever new tables, columns, and relationships would support it. The two sit side by side as the app’s two AI-powered tabs, grouped together in their own pill in the mode switcher — AI Engineer patches what’s broken, AI Designer builds what’s missing. This works the same way whether you’re adding onto a schema you’ve already built or starting completely blank — see the “Designing a schema from scratch” section below.

Starting a conversation

Opening AI Designer swaps the right-hand panel for a persistent chat thread — the Diagram canvas itself stays the main pane, since a proposal is meant to be seen on the real diagram rather than in a separate preview. Type a request like “add a module for tracking production costs per project” and send it (Enter, or Shift+Enter for a new line) to call the /api/design endpoint, which needs the same GEMINI_API_KEY as the other AI features.

Designing a schema from scratch

AI Designer doesn’t require a schema to already exist. On a brand-new schema, before any DDL has been pasted or analyzed, DDL mode’s empty “Import DDL” state shows a callout with a Use AI Designer button that jumps straight to this tab — no SQL to write first. The chat and canvas work exactly the same either way; the only difference is what you’re asked to describe. With no existing schema, Gemini designs a complete, sensible set of tables, columns, and relationships from a blank slate instead of modifying anything — every table in the resulting proposal renders as new. See Starting From Scratch: Designing a Schema with AI Designer for a full walkthrough of this path.

Seeing the proposal

A proposal renders directly on the canvas: new tables get the same blue highlight used elsewhere in the app for unsaved additions, so you can see exactly what’s being suggested in context — no separate diff modal to cross-reference against the diagram. Your real, saved schema is completely untouched until you explicitly confirm; switching to plain Diagram mode and back always shows the unmodified diagram, proposal included.

Refining a proposal

A proposal isn’t final the moment it appears — send a follow-up like “also add a currency column” and it compounds on what’s already proposed rather than starting over, since each turn is sent along with the prior turns’ context. The thread stays alive if you switch to another mode and back; it only clears once you act on it, or once you load a different DDL baseline (running AI Engineer’s analysis, loading a sample, a saved schema, or a historical version).

Finishing up

Once you’re happy with a proposal, three actions are available:

• Confirm — folds the proposal into your real working schema through the same re-parse path used elsewhere in the app, preserving layout and notes for tables that weren’t touched, and clears the thread
• Edit DDL — hands the proposal’s DDL to the ordinary DDL-mode editor for manual hand-tuning, which you then apply yourself via the normal Save & Update flow
• Discard — drops the proposal and the thread; your schema is untouched

As with accepting an AI Engineer fix, confirming or manually editing a proposal doesn’t save it to your account by itself — the top-bar Save button is still the actual persistence step.

Same usage budget as the other AI features

/api/design shares the same identity check, monthly token cap, and usage recording as AI Schema Analysis, Applying Fixes, and Change Request generation — see Free vs. Pro: Plans and Limits for the actual numbers. There’s no separate budget or gate specific to AI Designer.