How Optimevra works
Optimevra crawls a site you own, looks at each page through eight lenses, and returns findings you can act on. Everything below is generated from the same configuration the product runs on.
The eight lenses
Every analyzed page is judged against all eight. They are separated because the fixes are different jobs — a contrast failure and a slow image are not the same work, and lumping them into one score tells you nothing about who should pick it up.
| Lens | What it looks at |
|---|---|
| UX | Whether the path to the thing a visitor came for is obvious — navigation that matches how people actually look, and pages that do not make someone hunt. |
| UI | Visual consistency and hierarchy: spacing, alignment, type scale and whether the important thing on a page looks important. |
| Accessibility | Whether the page works for people using a screen reader or a keyboard — labels, contrast, focus order, and images that describe themselves. |
| Performance | What makes the page slow to become usable: oversized images, render-blocking resources, and work done before anything appears. |
| Mobile | How the page behaves on a phone — tap targets, text that stays readable without zooming, and layouts that do not overflow. |
| Conversion | Whether the page earns the click: whether the call to action is findable, and whether the copy around it does any work. |
| Content | Whether the writing is clear, current and does not contradict itself from one page to the next. |
| Technical | The machine-readable layer: titles, meta descriptions, headings, structured data, canonicals and the things search engines read. |
What counts against your allowance
One number is the real limit: AI-analyzed pages per month. Everything else in the table is scope.
A page counts only if it actually reached the model. Near-duplicates, pages carried forward from a previous scan and paths you have excluded spend nothing and do not decrement it — so re-scanning a site that has not changed is free. That is deliberate: you should be able to re-check a fix without paying to be told nothing else moved.
Pages per crawl is a ceiling on how much of a site one crawl covers, not a cost control. It was mistaken for one internally, which is how an unlimited-scan tier read as a single small number when it was really that many pages, times the number of sites, times as many scans as you like.
What each plan includes
| Free | Starter | Pro | Agency | |
|---|---|---|---|---|
| Sites | 1 | 1 | 2 | 5 |
| Pages per crawl | 1 | 60 | 120 | 250 |
| AI-analyzed pages per month | 1 | 60 | 135 | 500 |
| Scans | 1 | Unlimited | Unlimited | Unlimited |
| History kept | — | 90 days | 365 days | — |
| Copilot messages per month | 20 | 100 | 300 | 1,000 |
| Copilot rate | Enough to see what the assistant does, at a gentle rate. | Comfortable for one person working through a site. | Sized for a team working in parallel without queuing behind each other. | No rate ceiling — the monthly budget is the only bound. |
One Evra subscription covers Optimevra alongside every other product — see pricing.
Copilot
Copilot answers questions about a finding and can propose the code change that fixes it. Two limits apply and they do different jobs: the monthly message count is the budget, and the rate tier above caps how fast you may spend it — not how much. A team on a high tier and a solo user on a standard one have the same monthly allowance if their plan says so; the tier only decides whether they queue behind each other.
What we do not do
Optimevra does not change your site. It reads pages and reports findings, and Copilot can write a suggested patch — but nothing is applied for you, and there is no agent with credentials to your host. Only crawl sites you own or are authorised to test.
Something here wrong or missing? This page is generated from the same configuration the product runs on, so if it disagrees with what you see in the app, that is a bug worth hearing about.