WordPress · performance

Fast is easy.Fast and still working is the job.

Velocity Forge measures what your visitors actually get — and names the control, the file and the plugin behind every slow interaction, not just the number. It defers only what is safe to defer, and with Pro the rule it talked you into is measured for a day and taken back on its own if the site got slower. Every change has an undo, including one that works when the admin will not load.

See what it does See pricing

being forged — launching with the Avakode suite

assets defer

jquery-core rail
wc-checkout rail
stripe rail
slider-revolution deferred
share-buttons deferred

defer

20 handles · 15 URL patterns · never optimisable, by any setting

The elementINP and LCP name the control, the file and the plugin behind them — in the free plugin

17caching and optimisation plugins recognised, with our overlapping work stood down

5 checksthe recovery drill presses the real escape hatches and puts every one of them back

The rails

Making a site faster is easy. Not breaking it is the product.

Nothing is optimised on activation — the plugin measures first. And 20 handles and 15 URL patterns are hard-railed: they cannot be optimised at all, by any setting, filter or database edit.

requestverdict

jQuery jquery-core untouched

WooCommerce checkout wc-checkout untouched

Stripe SDK stripe untouched

Captcha turnstile untouched

Block editor styles wp-block-library untouched

Our own beacon velocityforge-beacon untouched

Slider plugin slider-revolution deferred

Share buttons share-buttons deferred

Chat widget chat-widget deferred

Theme animations theme-animations deferred

4deferred, delayed or dequeued

6refused: these are rails

A rule aimed at jQuery, the block editor, the cart, Stripe, a captcha or our own beacon simply never applies. That is not a default you can change — it is the reason the other rules are safe to switch on at all.

And nothing here is permanent: every rule reverts on its own, safe mode switches the lot off, and a constant in wp-config.php stands the plugin down without reading the database.

Attribute · apply · revert

It names the element, watches the rule, and takes the rule back when the numbers say so.

Attribution down to the control and the plugin behind it, an optimiser with rails you cannot switch off, and a rule that undoes itself when a day of real visitors disagrees with it.

The element, not just the number

LCP, INP, CLS and TTFB from real visits — and for a slow interaction, the control it hit, whether the time went on a busy main thread, on your own handler or on the repaint afterwards, and which plugin owns the script. A 75th percentile per element over 28 days, never one visitor’s anecdote.

The optimiser, as one purchase (Pro)

Defer, async, delay-until-interaction or dequeue per handle; critical CSS inlined under a 14 KB cap; WebP and AVIF sidecars with your originals kept; autoload cleanup with a backup each. jQuery, checkout, payment SDKs and captchas are never optimisable, even by mistake.

Rules that take themselves back (Pro)

A new rule is measured for a day against the week before it. If LCP or INP is worse by more than your threshold, it is reverted through the same revert your own button uses, the reason is written onto the rollback event, and the numbers arrive by e-mail. Never on too few measurements, never in Safe Mode, never on a rule you pinned.

A regression that says what changed

After an update the row does not only say the page got slower. It names the scripts and stylesheets the front end gained, lost or resized — read from a snapshot of the real page taken before the files were replaced.

Autoloaded options with an owner

The heaviest autoloaded options, each with the plugin that wrote it. “wp_options is four megabytes” is not something you can act on until the rows have names beside them.

The way out, drilled (Free)

Five checks press the real escape hatches — the panic file, a loopback on the genuine secret URL, the safe-mode option, the backups table — and restore each of them exactly as found. Safe Mode also sits in the admin bar on the front end, where you notice the breakage.

Critical CSS

9 rules of the 25, chosen by looking at the page.

No headless browser and no guesswork: the extractor reads the HTML your server actually sent and matches selectors against the first screen. What it keeps is inlined; the rest arrives without blocking the render.

60% of the stylesheet kept out of the first paint

the stylesheet 1,467 B

inlined for the first screen 583 B

@font-face always survivesa face that arrives late is a screen of invisible text, which is a worse bug than a slow one.
@keyframes only if something kept uses theman animation nobody references is weight, not motion.
Capped at 14 KBthe first congestion window — past it, inlining costs a round trip instead of saving one.
No match, no changewhen no record fits the request the original stylesheet is served exactly as before.

The first week

Measure, change one thing, and let it check its own work.

Measure, then read the names

Activate and change nothing. Within days Monitoring names the ten controls that cost the most interaction time and the five elements that delay the first paint, with the file and the plugin behind each.

Apply one rule

Work the asset list one handle at a time. The rails stop you breaking checkout — and with Pro the rule you just made is watched for a day and undone on its own if a real day of visitors disagrees.

Read what the update did

Every plugin update is compared against your baseline, and the row says which assets the front end gained or lost. It reaches you by e-mail, or with Pro through Telegram, Slack or your own webhook.

The way back

Break the site on purpose. Then take any way out you like.

The measure of a performance plugin is not how fast it makes a good day. It is what you can do on a bad one — and whether the escape hatch still opens when the database is the thing that went.

your site velocity forge · active

the site is up database: fine

three ways back Nothing is broken yet. Press «Break it» below, then take any of these three.

Three ways out, and the last of them reads nothing at all.

And if you delete the plugin outright: uninstall restores every autoload value it changed and has not reverted — opt-in or not, and whether or not velocityforge_safe_mode was ever switched on. A performance plugin must not leave its white screen behind.

Shares the road with

Knows your stack

WP Rocket

LiteSpeed Cache

W3 Total Cache

Autoptimize

NitroPack

Perfmatters

WooCommerce

Elementor

Comparison

Velocity Forge vs the caching plugin you already have

What you are comparing	Velocity Forge	a page cache
What it works on	Assets, critical CSS and images — what the browser has to do	The HTML your server has to build
How it knows it worked	Real-user field data, p75 by device over 28 days — and per element, so a number has a name	Usually nothing; you re-run a synthetic test
When a rule turns out to be wrong	With Pro it is watched for a day and reverted on its own, with the reason and the numbers	It stays on until somebody notices
What a slow release tells you	Which scripts and stylesheets the update added, dropped or grew on the front end	Not covered
Undoing one change	Every rule reverts on its own, Safe Mode sits in the admin bar, a wp-config constant disables the plugin without touching the database	Deactivate, all of it, and hope
Being told a page regressed	E-mail free; Telegram, Slack or your own webhook with Pro	Not covered
The first minute after installing	Nothing changes — it measures first and asks you to apply rules one at a time	Switch it on and the page is usually faster straight away their win
Page caching	Not its job — keep the one you have	Exactly its job, and it is good at it their win

Velocity Forge is not a page cache and does not want to be one. Run both: the cache saves your server, this saves the browser.

Pricing

One licence for the optimiser, the watch and the alert that wakes you.

No upsells, no surprises.

Personal

$49 /year

1 site

Get it on avakode.com

What runs where in Velocity Forge

Which work happens inside your own install, what leaves it, what needs a paid plan, and what a lapsed licence leaves behind.

What the free tier does inside your own install

INP and LCP named down to the plugin
The regression names the assets an update changed
Recovery drill — a button that presses the escape hatches on a day when nothing is wrong: the panic constant, a probe write to the panic file's directory, a real loopback request to the secret panic route with the real secret, the safe-mode option's round trip, and the autoload backups table. Everything it touches is put back exactly as it was found, including the on-disk copy of the secret

What leaves your site when a feature is on, and where it goes

Usage telemetry is off until you switch it on: every send reads the site option `forge_telemetry_consent` first, and stops when it is not "1".
With telemetry on, batches of events POST to https://api.avakode.com/telemetry/batch.
The in-admin feedback form POSTs your rating and comment to https://api.avakode.com/feedback when you press Send.
The Forge Suite dashboard health card GETs https://api.avakode.com/health.

What needs Pro, credits or an outside service

Pro is a separate plugin. With no Pro add-on loaded the plan is free, whatever is stored in the database.
Licence activation and every AI request go to https://api.avakode.com.
AI actions spend credits from your Avakode balance; the price per action is decided on the server (1 credit for a translation, 3 for a whole post).
A request over 10,000 tokens costs a multiple of the base action price.

What happens when the licence lapses

Free features keep working. The wp.org base ships no licensing SDK: without the Pro add-on answering "pro" through the `forge_pro_plan` filter the plan reads free.
When the paid ladder answers "free" in so many words, the stored grace marker `forge_license_last_pro_at` is cleared and Pro gates close on the next admin request.
Pro keeps working for 6 hours of silence from the licence server — that window covers an outage, not a revoked licence.
A lapse removes no data. Tables and options go away only when you uninstall the plugin.

How long data is kept, and how you take it out

Field beacons are kept 30 days by default; the setting accepts 1–365.
What `beacon_retention_days` counts changed in 1.1.0, and a one-time migration rewrote the stored value.
Pro exports field data as a download.

The same five questions for every tool →

Switching

Your cache plugin stays, and it is recognised by name.

Nothing to migrate

No settings to port. Install it, let it measure for a week, and read the elements it names before you change anything.

It stands down where your cache plugin already works

Seventeen caching and optimisation plugins are recognised, and the notice says which of our features stepped back for yours — so the two never fight over the same handle.

Every change is reversible, and you can prove it

Rules revert one at a time, the drill presses all three escape hatches on a good day, and uninstall restores any autoload backup still outstanding whether or not you opted in.

FAQ

Asked before the first rule goes live

How is this different from WP Rocket?

It starts from your field data instead of a checklist, and it goes further than the metric: the slow interaction is named down to the control, the file and the plugin that shipped it. It will not let you optimise the handles that break stores, and every change is reversible — including from a constant in wp-config.php when the admin is unreachable. Deleting the plugin restores what it changed.

Can I run it alongside my cache plugin?

Yes. Seventeen caching and optimisation plugins are recognised, and our overlapping work switches itself off while one of them is active. The notice names every feature we stood down — not only the one that overlaps — because a plugin that quietly stops doing four of the five things you installed it for is worse than one that says so.

An update slowed the site down. What does it actually tell me?

Which assets changed. Before the files are replaced the plugin takes a snapshot of the real front page’s script and stylesheet queue, and the regression row reads the difference: what was added, what disappeared, what grew. “LCP is up 400 ms” is a symptom; “this update added three render-blocking stylesheets” is something you can take to its author.

Will it change my site on its own?

Only with Pro, only for a rule you just created, and only after refusing six ways first: no licence, no beacon, fewer measurements than your sample floor, Safe Mode on, a hard-railed handle, or a rule you pinned. When it does act it reverts one rule through the ordinary revert path, writes the reason onto the rollback event and e-mails you the numbers and how to put it back.

What happens if I break something?

Revert the single rule, hit Safe Mode from the admin bar on whichever screen you noticed it, use the panic URL, or add one line to wp-config.php. The last one is checked before any database read, because the day you need it the database is often the problem — and the recovery drill lets you prove all of that works on a day when nothing is wrong.

Does it send my data anywhere?

Measurement never leaves your site — no field reading of any kind is sent anywhere by either tier. The free plugin does make requests of its own, and every one of them goes to your site or to an asset your own pages already load: the audit fetches each URL you scan, the pre-update snapshot reads your front page so a regression can say which assets an update changed, each asset gets a HEAD to size it, and the recovery drill calls your own secret panic route to prove the escape hatch works. The one part of the free download that talks to us is the shared Forge admin layer every product carries: a service health check when you open the Forge dashboard, and the feedback box if you choose to answer it — neither carries a measurement. Pro talks to three parties, each for a reason you can see in Settings: our Forge API — the licence check, account connect, the credits balance and prices, plugin updates, the AI advisor when you press it (it refuses to answer with no field data to reason about), and usage telemetry while it is switched on; Google PageSpeed Insights, on your own key and quota for the daily lab audit, never proxied through us; and the alert channels you configure — Telegram, Slack or your own webhook — when a regression fires. Critical CSS reads your own pages, nothing else.

Measure first