Plan once, build with any AI

Your AI builds it. Scaffolds is what remembers why.

Twenty minutes of good reasoning in one chat ends in a decision, and then the tab closes. Scaffolds holds the structure of a product and the decisions behind it as one model that Claude, ChatGPT, Gemini or your editor connects straight to, so what was settled on Tuesday is still there on Friday, in whichever AI you happen to open next.

  • One document

    Per project, revision checked, so concurrent editing cannot silently clobber a decision

  • 6 pages, 47 blocks

    In the scaffold for this site, which is public and readable without an account

  • 4 journeys

    Mapped across those pages, each with an entry, a goal, steps and an exit

  • Connected or exported

    Your own AI reads the project over one endpoint, or takes the whole thing as Markdown. Both end in the same review

  • Stable identifiers

    Page and block ids survive renames and moves, which is what makes an AI edit a merge

  • Desktop and mobile

    A real mobile projection of the same model, not a shrunken canvas

Home

Plan websites as connected systems

See the system, not separate artefacts

Work with your AI, not ours

Architecture

Intent

  • Evaluate Scaffolds
  • Plan and deliver websites
  • Review and collaborate
  • Understand the AI workflow

The project behind this page, cycling through the lenses it is seen from. Colour is audience intent, not layout role.

One model, four views

See the system, not separate artefacts

Most website planning is spread across a sitemap in one tool, wireframes in another, journeys in a slide and the reasoning in somebody's notes. Scaffolds keeps them as one model, so a change in any view is a change to the same thing.

Architecture

The page tree, with top level pages as separate trees rather than one sprawling diagram. This is the backbone every other view hangs from.

Wireframes

A page is a stack of blocks, and a block is a stack of elements drawn as schematic line work. Enough to agree structure, not so much that it pretends to be design.

User journeys

An intent becomes an explicit route across pages and into external systems, with an entry, a goal, ordered steps and an exit.

Reasoning

Purpose, evidence, open questions and red flags live on the block they belong to. The argument stays attached to the thing it is about.

Product in use

What it looks like on real work

These are the actual product surfaces, not mock-ups. Each one is a different lens over the same document.

Architecture

The Scaffolds canvas showing the page tree for this website, each page a card of coloured blocks.

The whole site on one canvas

Pages as cards, blocks stacked inside them, colour showing which audience each section serves. Pan and zoom rather than scroll, because a sitemap is a space, not a list.

Page detail

A page opened in detail, showing its blocks with the reasoning recorded against each one.

The reasoning sits with the section

Open a page and every block carries its purpose, the evidence behind it and anything still unresolved. The argument is attached to the thing it is about.

User journeys

The user journeys view, showing an intent routed through pages from entry to exit.

Intent becomes an explicit route

Each journey names its entry, its goal, its ordered steps through real pages and its exit. It is a path across the architecture, not a diagram beside it.

Mobile

The same project on a phone, showing the page hierarchy as a list.

A projection, not a shrunken canvas

On a phone the same model becomes a hierarchy and focused detail views, so a stakeholder can review the work properly from a train.

AI

Work with your AI, not ours

There is no model inside Scaffolds, and there is not going to be one. You bring your own: ChatGPT, Claude, Gemini, Cursor, whatever your organisation has already approved, and it reads the plan instead of guessing at it. That answer used to sound like a gap. Now that every one of them can connect to a project directly, it is the reason this works at all.

Connecting takes a minute and does not involve the word “MCP”: pick your AI, follow its own steps, and it reads the project where it lives. If yours cannot connect, the whole project still exports as one Markdown file and comes back the same way. Both roads end in the same review, and neither writes anything on its own.

  • A connected AI reads the goals, the rules and what has already been agreed
  • It asks you about anything nobody has decided, rather than deciding it quietly
  • The most it can send back is a question or a proposal, which waits for you
  • An absent field keeps its value, so a careless edit cannot wipe a red flag

The round trip

  1. 01Connect, or exportPick your AI and follow its own setup, and it reads the project where it is, or take the whole thing as one Markdown file with the rules at the top.
  2. 02ConverseYou work in your own AI. It asks questions before it changes anything.
  3. 03ReturnA proposal arrives, or the file comes back and is parsed deterministically, with no model involved.
  4. 04ReviewAdditions, edits, moves and removals are listed. Removals are counted separately.
  5. 05ApplyOne revision, one undo away.

The wider loop

From planning to implementation context

Planning artefacts usually die at handover, because the reasoning was never in a form anything downstream could read. A structured model does not have that problem.

  1. 01

    Define intent

    Who the site is for and what the organisation needs it to do, before any page exists.

  2. 02

    Shape architecture

    The page tree and the blocks each page is made of, with purpose attached.

  3. 03

    Map journeys

    How each intent moves through the structure and into external systems.

  4. 04

    Attach reasoning

    Evidence, open questions and red flags recorded against the section they concern.

  5. 05

    Hand context forward

    The approved structure and its reasoning travel to delivery as one readable file.

Components, content requirements, search, design tokens, measurement, accessibility and systems are all part of the model now, and a coding agent can read any of them directly. What is not built is the far end of the loop: reading the repository back, and telling you where the thing that got built has drifted from the thing you planned. That is the direction, and this page will not pretend it is here yet. The enterprise page sets out what is available now and what is still ahead.

Enterprise

Built for programmes, not for a single designer

Scaffolds started inside consulting work on large website programmes, and the model reflects it.

Shared project truth

One document per project, revision checked, with per project sharing. A reviewer opens a link and reads the work without an account and without learning a design tool.

Collaboration and review

Comments on blocks, pinned notes anywhere on the canvas, a roster of who is working on it, and version history you can look back through.

No AI provider lock-in

Scaffolds carries the structured context and owns the import rules. Which model reads it is your decision, and it can change without touching the project.

Explore a real scaffold, no account needed

The plan for this website is itself a public Scaffolds project. Open it and you get the same thing a client gets when a link is shared with them: the architecture, the blocks, the journeys and the reasoning behind every section.

Start with one project, free

One project, every planning surface, and your own AI connected to it. Not a trial with a clock on it. What Free does not include is a second project, which is the honest moment to pay. Programmes with procurement and a security review to pass are still a conversation.