Plan mode is the safe zone. It produces a Blueprint — a typed plan you can edit, share with a teammate, and feed into Build. The chat rail is the primary way to evolve it, but every field is also editable inline.
App goal
A one-line summary of what the project is. The AI uses this every time it plans: it grounds intent. Keep it short. "Track 200 students, fees, and balances for a community school" is much more useful than a paragraph.
Audience
Who will use the tool. Audience drives copy choices and role suggestions. "Bursar and headmaster, no students" is more useful than "Schools".
Pages
Planned screens. Each page has a name and a purpose sentence. Pages are not the same as routes; they are conceptual destinations the operator will spend time on.
Tables
Planned tables (logical, not yet physical). Each table has a name and a purpose. Build from plan is what turns these into real schema.
Roles
Who has access to what. Today this is descriptive — there is no role-based auth at the launch tier. The Blueprint roles set the right context for the AI when it generates copy and access copy in scenes.
Integrations
External services this project depends on — Stripe, Email, WhatsApp, Sheets, Storage, AI. Each integration carries a status (planned / connected / not_needed) and a notes field. We never store raw API keys here.
Tasks
A backlog the AI generates as it plans. You can add or edit tasks manually. Tasks have status draft / active / ready / done. They are guidance for the operator, not gates on the AI.
When to use Plan mode
- At the start of a project, before the AI has touched any tables.
- After Build, when the schema does not match the operation you actually run.
- Before a sales call, to make sure the Blueprint reads cleanly to a non-developer.
- Whenever you want to pause and reflect on structure without risking the database.