How PromptForge works

Context detection

When you type a brief, PromptForge reads it live and extracts structured context. It recognises:

• The product name and a one-line description, usually from the first sentence.
• Labelled lines such as Audience:, Goals:, Screens:, Stack: and Rules:.
• Bullet points as features, hex codes as brand colours, and known framework names as stack.

It also detects the project type — web app, mobile app, website, e-commerce store, game, extension or campaign — and adapts every export: game packs get HUD and menu screens, engine conventions and play-testing in the definition of done; web apps get access-control and state-handling rules.

Anything you type into the structured fields overrides what was detected. Attached context kits fill whatever is still empty.

Generators

Each export target has its own generator that knows the tool's conventions — where the file lives, which sections it reads, how long prompts should be and in what order to send them. Coding exports also include stack-specific rules for what you use (for example Next.js server/client boundaries, Supabase row-level security, verified payment webhooks for Paystack or Stripe, typed GDScript and signals for Godot) and suggested verification commands. Generators are deterministic: the same context always produces the same pack.

The PromptForge AI layer

Signed-in users can add PromptForge AI on top of the standard engine. It can structure a messy brief, suggest questions worth answering, and rewrite exports to be more specific to your product — always using only facts from your brief.

Workspaces and storage

Saved packs and kits live in a workspace. Every account has a personal workspace; Team plan users can create shared workspaces with roles. Data is protected with row-level security so only members of a workspace can read it. See Privacy & security.