Phone
Mid-range Android, slow 4G connection, processor four times slower than a laptop.
- 95Performance
- 100Accessibility
- 100Best practices
- 100SEO
We measure Sairo the way browsers do. Here are the numbers, what they mean for the people who visit you, and how to check them yourself.
The homepage on 7 September 2026, measured with Lighthouse 12.8. Ninety and above is green in every category. The phone run is the demanding one: it emulates a mid-range device on a slow connection.
Mid-range Android, slow 4G connection, processor four times slower than a laptop.
Laptop-class device on a fast connection.
Lighthouse scores move a few points between runs and with the distance from our servers. These are single runs, not the best of several.
Scores summarise; these are the measurements that produce them, from the same phone run.
The same discipline applies to the script you install. It is small, it waits for your page, and it stays out of your layout.
No tricks aimed at the test. Each of these makes the page lighter for a real visitor, and the score follows.
Every public page is generated as static HTML when we release, not when you visit. The server sends a finished page instead of assembling one per request.
The typeface is self-hosted, there are no analytics tags or tracking pixels, and the content security policy only permits sairo.app. Your visit is a conversation with one server.
The site draws its icons from the exact path data it needs. That replaced a 204 KB icon library with 29 KB of shapes on every page.
Phones get a still image of the assistant, captured from the scene itself. Desktops load the WebGL model only after the page is ready and the browser is idle.
Versioned assets are cached for a year and images for a month. Responses use brotli over HTTP/2 and HTTP/3, so repeat visits cost almost nothing.
A 100 on Lighthouse accessibility is the floor. Keyboard navigation, visible focus, screen reader labels and reduced-motion support are part of every component. Our accessibility approach.
We publish these numbers because we did the work in the open. A performance pass that day moved the homepage on a phone from an unpredictable score to a consistently green one.
| Measure | Before | After |
|---|---|---|
| Lighthouse performance | 62 to 92, depending on the run | 95 |
| Total Blocking Time | 110 ms to 2,650 ms | 10 ms |
| Speed Index | 5.3 s | 3.7 s |
| Bytes over the wire | 645 KB in 34 requests | 344 KB in 28 requests |
| JavaScript | 340 KB | 143 KB |
| HTML | 48 KB, built per request | 23 KB, built ahead of time |
| Lighthouse SEO | 92 | 100 |
| Back/forward cache | Not eligible | Eligible |
Nothing here needs our word for it. Any of these gives you an independent reading in about a minute.
Google's hosted Lighthouse, from Google's own network. Open the report for sairo.app (opens in a new tab)
The same engine on your own machine, with the report opened in your browser when it finishes.
npx lighthouse https://sairo.app/ --viewOpen the Lighthouse panel, choose Mobile, and analyse the page. Your device, your connection and your distance from our servers all move the result a little, which is normal.
Lighthouse simulates a slow connection and a slower processor, then measures real timing on the machine running it. Background work, network conditions and distance from our servers each move the result by a few points. We publish one dated run rather than the best of several.
It adds one asynchronous script of about 15 KB compressed that loads after your page. Its launcher is fixed in a corner, so it causes no layout shift, and its styles stay inside Shadow DOM. Measure your own pages before and after installation with the same tool; our guide on measuring chatbot performance explains how.
Lab data. The scores come from Lighthouse runs on the date shown, not from aggregated visitor measurements. We measure again after significant changes and update the date on this page.
Try a sample, review the result, and decide what comes next.