Deploy

Shared from $4.99/mo · Virtual from $6.99/mo · Bare metal from $35.99/mo

DeployClient Login

SaaS · Website auditing

MideTuWeb

A free SEO and performance auditor that loads a site in a real browser and grades it, sized to stay cheap and still survive a viral afternoon.

  • MideTuWebFree site audit
  • Technical SEO reportTitles, headings, indexation
  • Performance reportSpeed, resources, caching
  • Mobile-first checkLayout, tap targets, readability
Visit MideTuWeb

Architecture published with the organisation’s permission.

The problem

  • Hard constraints
  • A scan must finish or fail cleanly, never hang, however slow the site being audited is.
  • A spike may make the queue longer. It may not take the site down or lose a job.
  • Every report stays retrievable, so a returning visitor can see whether their site improved.
  • The whole platform has to cost a predictable monthly figure, because the product earns nothing per scan.
  • Running an untrusted page in a real browser is a security boundary, not a detail.

MideTuWeb scans a website in seconds and hands back a prioritised list of what is wrong with it: technical SEO, performance, mobile behaviour, on-page structure. It is free, it asks for no registration, and it is aimed at people building real sites rather than at agencies buying seats.

Which makes the workload unusual for a SaaS. Almost none of the cost is serving the page a visitor sees; nearly all of it is the audit itself — a headless browser fetching a stranger’s site, waiting out its slowest resource, laying it out at two viewport sizes and measuring. That is minutes of somebody else’s latency, held open, per scan.

And the traffic is spiky in the worst way: quiet for days, then a link lands somewhere and a thousand people all want the same thing at once. On a per-request cloud that is a bill nobody budgeted; on one small box it is a queue that never drains.

Architecture

  1. Edge

    Public app and API

    1 ×Volt 2

  2. Application

    Audit workers

    2 ×Apex 2

  3. Application

    Queue and scheduler

    1 ×Volt 3

  4. Storage

    Reports and database

    1 ×Forge 13

One audit, end to end. The public tier only enqueues and reads back; the workers reach the internet and nothing else; the store is the only tier that keeps anything.

What it runs on

  • Edge

    Public app and API

    1 ×Volt 2

    4 vCore · 8 GB · 75 GB NVME

    One virtual server carrying the site, the report viewer and the API that starts a scan, on 8 GB. Deliberately the smallest thing in the deployment: the page a visitor loads does almost nothing except hand the work to the queue and then read a result back out.

    From $12.99 per month

  • Application

    Audit workers

    2 ×Apex 2

    AMD Ryzen 9 5900X · 32 GB DDR4 · 2× 512 GB NVMe

    Two Ryzen 9 5900X machines at 4.8 GHz running headless Chromium in locked-down containers. An audit is a browser waiting on somebody else’s server, so what decides throughput is clock and how many browsers fit beside each other — 12 cores at that clock is the shape that answers both, and the second machine is the spike.

    From $134.99 per month, $127.00 one-time setup

  • Application

    Queue and scheduler

    1 ×Volt 3

    6 vCore · 12 GB · 100 GB NVME

    The queue, the rate limiter and the scheduler for repeat scans, on a machine of their own. Separated from the workers because a queue that dies with the worker it lives on loses exactly the jobs nobody has run yet.

    From $18.99 per month

  • Storage

    Reports and database

    1 ×Forge 13

    2 x Xeon E5-2660v2 · 128 GB DDR3 · 1× 480 GB SSD + 1× 8 TB SATA

    Every report ever generated, kept and addressable: a 20-core machine with 128 GB, an SSD for the live database and 8 TB behind it for the reports themselves. A free tool that forgets last month’s audit cannot show anybody whether they improved, which is the one thing that brings them back.

    From $124.99 per month

Techniques worth naming

  1. The browser is the sandbox boundary

    Every audit runs in a container with no credentials, no network route to the queue or the database, and a hard wall-clock limit. The page being audited is hostile by assumption, because sooner or later one of them will be.

  2. A queue that outlives its workers

    Jobs are durable and acknowledged only on completion. A worker can be killed mid-audit — by the time limit, by a deploy, by a hostile page — and the job is picked up again rather than lost with the process that held it.

  3. Degrade the wait, never the answer

    Under load the queue lengthens and the page says so honestly. What does not happen is a shortened audit: a report that skipped the mobile pass to keep up would be a worse product than a report that took four minutes.

  4. Reports are addressable and kept

    Every scan has a stable URL and stays retrievable. It costs disk, which is the cheapest thing in the deployment, and it turns a one-off tool into something a person can come back to after they have made the fixes.

Support

Extended cover, around the clock

Agreed with this account

MideTuWeb

The afternoon a link lands. A free tool has no sales pipeline to warn anybody, so the first sign of a spike is the queue depth — and the answer to it is capacity, at whatever hour it starts.

  1. Watched at the queue, not the door

    The alert that matters here is not the site being down, it is the backlog growing. We are reachable around the clock for that, which is a quieter emergency than most and just as time-critical.

  2. Capacity on the same night

    An extra worker is a machine we confirm and hand over rather than a procurement conversation. What we will not do is promise a delivery time for physical hardware we have not confirmed is in stock.

  3. The same people, not a second queue

    Extended cover changes when we are reachable and how fast a fault is escalated. It does not hand the account to a different team, because there is not one.

Extended cover is agreed per account and written down before it applies. It changes when we are reachable and how quickly a fault is escalated — it does not create a second queue, a premium tier or a response time we have not signed. Everything on the support page is still true here.

What changed

  • A spike lengthens a queue that drains instead of taking the product off the air, and the bill for it is the same figure as a quiet week.
  • A hostile or endlessly slow page costs one worker one time limit rather than the platform.
  • Somebody who ran a scan in March can open it again in September and see what they fixed.
  • The platform costs a fixed monthly figure, which is what makes a genuinely free tool survivable.

No performance percentages appear on this page. We publish figures we have measured under a named method, and these deployments are described by their architecture instead.

Ready when you are

Bring us a workload, not a shopping list

Tell us what the thing actually does and we will tell you which machines fit — including when the answer is a cheaper one than you asked for.

Deploy a serveror compare every plan

Shared from $4.99/moVirtual from $6.99/moBare metal from $35.99/mo