Restyle

Restyle edits your Shopify theme from a written description. It writes real theme files in a disposable copy of your theme, checks the page still renders after every write, and hands the result back as a draft you publish yourself.

That last clause is the whole design. Everything below is what it costs to keep that promise honest.

Nothing it does touches your live theme

Restyle never writes to the theme your shoppers see. It duplicates it — about sixteen seconds — and every write for the rest of the session lands in that copy, named so you can find it in your theme list:

[Dynoweb Restyle] Homepage redesign - 2026-08-22

When you save, the changed files are copied into a draft theme — your draft, in your admin, sitting next to every other draft you have. Restyle stops there. Publishing is a button you press in Shopify, not one we press for you.

The editor is your real page, next to the chat

Restyle opens fullscreen inside your Shopify admin: your pages down the left, your live preview in the middle, the chat on the right. The preview is the actual rendered page, not a mockup of it — and you can drive the whole thing by pointing rather than describing.

• Click a section to target it. The picker is element-precise and works on any theme, so "this one" replaces a paragraph explaining which block you meant. Stack several targets at once if a change spans them.
• Paste a screenshot straight from the clipboard as a reference for what you want the section to look like.
• Search your store's pages to switch context without leaving the editor.
• Plan mode makes it write the plan and stop, before any file is touched.
• Desktop, tablet and mobile widths, and a before/after toggle on the preview.
• The Workshop panel puts your store's own CRO findings beside the editor, so the next change worth making is already on screen.

A theme is a workspace — five per shop — and a workspace holds as many chats as you like. Files belong to the theme, not to the chat, so you can close a conversation, open a new one tomorrow, and pick the work up where it stood.

Every write is checked before it counts

A model that edits Liquid will eventually produce Liquid that renders a blank page. So each write is followed by a render check that compares the page before and after on three things: byte size, how many sections it has, and which sections they are. If the page collapsed, the write is reverted and you are told it was.

The interesting part is the exception. "Cut four of these six sections" and "the page exploded" look identical to a naive size check, so the collapse floor scales with how much removal you actually asked for. Without that, every deliberate simplification gets undone and blamed on the model.

Some guardrails are structural rather than statistical, and are evaluated against the finished result rather than each step: a 25-section cap per page, max_blocks per section, and a refusal to remove the last section on a page. A multi-part request — "add a slideshow with three slides" — is folded into one write and one check rather than four round trips.

Some edits are refused before anything is written at all, on the text of the change itself: third-party scripts, iframes to other domains, tracking beacons, shopper data in a URL, eval, cookie reads. The same pre-write pass parses the Liquid and the JSON, so a broken tag comes back as a line-numbered refusal in under a second rather than as a write, a failed render and a revert a minute later. Scripts your theme already had are left alone.

It reaches for the smallest tool first

The same request can be answered four ways, and which one it picks is the difference between a change you can still undo from the theme editor and one that needs a developer. Restyle works down a fixed ladder and stops at the first rung that does the job:

Behind that sit six written notes on how Shopify themes actually behave: schema rules, block limits, where a setting lives versus where it renders. It searches those before it edits rather than guessing from the file in front of it. And it can read your page views, scroll depth and element clicks while it works, so "move the reviews above the fold" is argued from your own traffic rather than from taste.

Long tasks are planned in the open

"Make the homepage feel less cluttered" is not one edit. Restyle plans it into steps and shows you the plan before it touches a file. Each step then runs as its own bounded piece of work against the page as it actually is at that moment, not as the plan imagined it would be. You can pause a run mid-way; resuming re-plans against the page as it stands, so a pause is not a lost afternoon.

A run ends at review, never at a silent "done". You get one row per step, the files it touched, and the change itself — and you can keep all of it or undo individual changes from there.

Undo is linear: rewinding to a change undoes it and everything after it. There is no branching, and there is one deliberate refusal — if a later change touched the same file, Restyle will not restore the older snapshot over it. It names the change that is in the way instead, because silently deleting newer work and reporting success is worse than refusing.

What it does not do

The highest-value thing on this page.

• It does not publish. Ever. You publish, in Shopify admin. If you never press that button, nothing you did here reaches a shopper.
• It cannot place a generated image into a section. It can generate one, but wiring that into a theme's image picker needs file-write scopes that would re-prompt every merchant for consent. It will tell you this rather than pretend.
• It has deliberately fewer tools than DynoAgent — roughly a third. It can read four analytics tools and your catalogue, and write to themes. It cannot create discounts, edit products, or touch anything else in your admin.
• Themes do not merge. Each workspace is independent. Work done in one is not portable to another.
• Rewinding does not un-save a draft. Once files are in your draft theme they are yours; undoing the working copy and silently rewriting your draft are different promises.
• Measurement needs a real page and real traffic. A change is matched to a page URL exactly. If that page does not have enough traffic in the window, the result says "not enough traffic" — not a number we made up to fill the slot.

That last one matters more than it reads. When a published change is measurable, Impact reports it as improved, held steady, or went down, and only attaches money to the first. A tool that edits your storefront and then grades its own homework is worth less than one that will tell you the change did nothing.

What it costs

Restyle is metered in dollars against your monthly AI budget, not in "tasks" or credits. A real task typically costs between two and twenty-six cents.

Your first task is free and, more importantly, it finishes — it is funded to completion, so the first thing you ever ask for cannot die halfway with your theme half-changed. Every task after that is checked against your remaining budget before anything is spent, so the limit arrives as a question up front rather than an interruption mid-edit. See pricing for the monthly budget on each plan.

Where it fits

Restyle runs on the same pipeline as DynoAgent with a different, narrower profile — so a layout argument can be made from your store's own numbers rather than from taste alone.

Most of the time you should not be starting from a blank prompt at all. The changes worth making usually arrive from AI Suggestions, which finds the friction first; Restyle is what applies the fix, and Impact is what tells you whether it was worth applying.