AI models, credits & limits

Clyro runs on frontier AI models. This page covers how to get the best out of them, how credits work, and the limitations worth knowing upfront - so you spend credits on results, not surprises.

Choosing a model

Auto, the default

Leave the picker on Auto and Clyro routes each message to the right model for the job - fast models for small tweaks, stronger ones for real builds. For most people, Auto is the right answer all the time.

When to pick manually

Paid plans from Starter up unlock the full lineup. The mental model:

Choice	Best for
Auto	Everything - routes per task, best value
Fast models (Haiku, Gemini Flash)	Color changes, text updates, small fixes
Sonnet	Building sections, multi-file edits, most real work
Opus	Full-page builds, complex interactions, debugging the hard stuff

Tip: Start with Auto. Escalate to Opus only when a task keeps missing - the quality jump is real, and so is the credit cost.

Plan availability

• Free - guided demo builds (no credit allowance); bring your own AI over MCP instead.
• Starter - the full lineup, Opus included, plus image generation.
• Grandfathered plans (Lite, Studio, Agency) keep the model access they signed up with.

How credits work

• Every AI message costs credits based on the work actually done - a text tweak costs little, a multi-section build costs more.
• Your plan's credits reset on renewal. Top-up credits stack on top and keep for 12 months.
• Plan credits are spent first, top-ups after.
• Buying, boosting, and topping up all live in your billing settings.

Connecting an outside agent over MCP consumes no credits - your agent's own subscription powers the thinking, and only image generation is billed.

What the AI can't do

The AI is powerful, but it has real limitations. Knowing them saves you time and credits.

It can't access your live store

The AI works on a copy of your theme files. It can't see your real products, collections, prices, or customers - the preview uses mock data. Once you upload the theme to Shopify, it pulls your real store content.

It can't install apps or third-party scripts

Clyro edits Liquid, JSON, CSS, and JS. It can't install Shopify apps, add tracking pixels, or inject anything that needs app-level permissions. Do that from your Shopify admin after uploading.

It can't build backends

Clyro is your theme - the frontend. No APIs, databases, custom checkout flows, or server-side logic outside Liquid.

It sometimes gets Liquid wrong

The AI is good at Shopify's templating language, not perfect. Clyro's validation layer catches most mistakes before they reach you, and when something does break, paste the error into the chat - the AI is good at fixing its own mistakes.

Complex JavaScript is hit or miss

Simple animations (fades, slide-ups, hover effects) work well. Custom sliders, drag-and-drop, canvas work - expect to iterate more.

It doesn't remember across themes

Each theme is its own conversation. If you want the same style across several themes, describe it each time.

When the output is messy

Working code can still be sloppy - inline styles everywhere, over-complex logic, hardcoded values that should be settings. Three ways to handle it:

Prevent it with constraints

"Add a hero section" gives the AI too much freedom. "Add a hero with a full-width background image, overlay text, and a CTA - make all text and the image editable in the customizer" gives it rails.

Ask for cleanup explicitly

"The hero section has too many inline styles. Refactor them into clean CSS classes."

"This section has hardcoded colors. Make them all customizer settings."

Rewind and retry

If the output is really off, rewind to before the change and go again with a more specific prompt or a stronger model. You can also open the code tab and fix a few lines yourself - sometimes that's faster than explaining.