Download the PHP package asignua/filament-redirects without Composer

Filament Redirects

A redirect manager and a 404 log for Filament panels, with one button that turns a dead URL from the log into a redirect.

The best-known Filament 5 redirect plugin has a few thousand downloads and none of the existing ones combines a 404 log with a one-click redirect. Redirect plugins stop at the table of rules. This one also watches what your visitors (and Googlebot) actually hit, aggregates it by path with a referrer sample and the kind of client, drops the scanner noise, and lets you answer a row with a redirect in one click. And the redirects themselves are the kind an SEO audit will not flag: they only fire on a real 404, they are always one hop (chains are collapsed when you save, loops are refused), a slashed /old/ reaches the target of /old in a single hop, and HEAD is served like GET.

• Screenshots
• Requirements
• Installation
• Quick start
• Putting the middleware on your site — read this one
• How paths are stored
• The 404 log
• Multilingual sites
• Creating redirects from code
• Hooks of the registry
• Authorization
• Commands and schedule
• Configuration
• Gotchas
• Translations
• AI agents
• Testing

Screenshots

The redirects list - one hop, any status code, a hit counter per redirect:

The 404 log - what visitors and Googlebot actually hit, with a referrer sample and the kind of client:

Create redirect on a row opens a modal prefilled by Redirects::suggestUsing(); the row leaves the log and the old URL redirects at once:

Requirements

• PHP 8.3+
• Laravel 12 or 13
• Filament 5

Installation

Optionally publish the config:

The package ships no CSS or JavaScript (it is built from Filament's own components), so there is nothing to filament:assets.

Register the plugin in your panel provider:

[!WARNING] Restrict who can manage redirects. Without authorize() (or a redirects.manage gate) every user who can enter the panel can create a redirect from any path of your site to any external URL - an author or editor account is then enough to turn your domain into a phishing redirector. See Authorization.

Then put the middleware on your site - the next sections explain why the package cannot do that for you.

Quick start

An ordinary Laravel app: append the middleware globally in bootstrap/app.php.

Add a redirect in the Redirects resource - or just visit a dead URL on your site, open 404 log and click Create redirect on its row.

Putting the middleware on your site

Redirects::middleware() returns [RedirectTrailingSlash::class, RedirectFallbackMiddleware::class].

• RedirectFallbackMiddleware is an after middleware: it looks nothing up until the request has already produced a 404, so a live page is never touched.
• RedirectTrailingSlash acts before the route, and only on a slashed GET/HEAD address (/old/): it looks /old up in the cached map (and, when a row exists, asks Redirects::resolvesUsing() whether a live page sits there) and answers with the target in one hop, short-circuiting the route. Unslashed requests pass through untouched.

Both send a relative Location for an internal target, so a forged Host / X-Forwarded-Host header never ends up in a cached permanent redirect.

For a site that serves everything through one catch-all route (a CMS front), put it on that route:

The trap that is the reason the package registers nothing by itself: a package cannot push a global middleware (or one into the web group) from its own boot(). The HTTP kernel is built in public/index.php before the service providers boot, so afterResolving(Kernel::class)->pushMiddleware(...) and pushMiddlewareToGroup(...) silently do nothing in a real request - while appearing to work in tests, which resolve the kernel later. Route middleware is resolved at dispatch, and bootstrap/app.php is read before the kernel exists; those two are reliable.

You do not need a locale middleware in front of them: both parse the language from the path themselves.

How paths are stored

old_path and to_path hold the shape of a request path after the language prefix is split off:

Everything is normalised on save, in one place (RedirectRepository), so pasting https://site.test/en/shop/item/ into a field of an en row stores shop/item: our own domain (app.url, plus anything you add with Redirects::ownUrlsUsing(), each with its www. twin and port) and the row's own language prefix are stripped, a foreign domain is kept. Paths are stored percent-decoded (привіт, not %D0%BF...) and matched against the decoded request path, so a redirect typed in any alphabet fires for the encoded URL a browser sends. The leading slash exists only in the form - the field shows https://site.test + /shop as an affix, so a redirect to the home page is typed as / (an empty target means Gone, not home).

The form refuses spaces, a query string (?) or fragment (#) in a source (only the path is matched), a foreign host as a source, the home page as a source (it never 404s), a duplicate source in the same language, a source that starts with another language's prefix (it could never match), a target equal to the source or leading back to it, and a broken URL as a target.

The 404 log

The fallback middleware writes every unmatched GET that has no redirect into the log: one row per (path, language) with a hit counter, first/last seen, a referrer sample and the User-Agent family (Googlebot, Chrome, curl... - the full string is deliberately not stored).

• Create redirect (row action) opens a modal for the target; Redirects::suggestUsing() can prefill it. The row leaves the log and the redirect counts the hits from then on. A loop is refused on the target field.
• Create redirects (bulk action) sends many paths to one target; rows that would loop stay in the log and are reported. The target is typed relative to one language, so the selected rows must be of one language. "Gone" needs no target.
• Recheck closes rows that now have a redirect - and, with Redirects::resolvesUsing(), rows whose path now shows a page. The panel button checks the 2000 most recently seen rows (one web request); redirects:recheck walks the whole log. Prune old entries removes rows by age.
• Filters: language, bots / no bots, with / without a referrer. A Gone redirect and a matched redirect are deliberate decisions and are never logged.

Noise never reaches the table (filament-redirects.not_found):

and for anything the config cannot say: Redirects::ignoreUsing(fn (Request $request, string $language, string $path): bool => ...). Paths longer than the column (191) are skipped, not truncated - truncated ones would merge different URLs. A failed log write never breaks the response.

Multilingual sites

A single-language site needs no setup. Otherwise tell the registry how your URLs map to languages:

• The row has a language column and unique(old_path, language); the form shows the language select only when there is more than one language.
• A row's own prefix is stripped from what you type; another language's prefix is kept in a target (en/shop is a valid target of a uk row) and refused in a source of the unprefixed language.
• A target in another language. On a row of a prefixed language the first segment of a target that is ANOTHER prefixed language is served in that language (de/aktion on an en row -> /de/aktion, not /en/de/aktion). To reach a page of the unprefixed language from a prefixed row, paste its whole address (https://site.test/about on an en row): it is kept whole and served from the site root. Chains that cross languages are not compacted (compaction works inside one language).
• For a CMS that keeps languages in its own config, pass a closure - it is resolved at the moment of use: Redirects::useLocaleUrls(fn (): LocaleUrls => new PrefixedLocaleUrls(...)).
• For any other URL scheme implement Asignua\FilamentRedirects\Contracts\LocaleUrls (parse, url, prefix, default, all, unprefixed) - for example, a CMS that resolves URLs through its own link table.

Creating redirects from code

The Redirects::create() API is the door for anything that is not the panel - a listener that redirects the old slug when a record is renamed, a seeder, a migration:

It normalises, rejects a loop, a duplicate source or a source with ?/# (ValidationException), collapses chains in one transaction and flushes the cache. Never write the model directly: it has $guarded = ['*'], and a field that the repository does not assign is silently not saved. If you must update the table with raw SQL, call Redirects::flushCache() afterwards.

The optional entity morph (entity_type, entity_id) records which record a redirect leads to; give it explicitly, or let Redirects::entityUsing(fn (string $language, string $path): ?Model => ...) find it from the target.

Hooks of the registry

Everything that is a closure lives on Asignua\FilamentRedirects\Redirects (set it in AppServiceProvider); each hook has a neutral default.

resolvesUsing must answer for an anonymous visitor: if it shows drafts to a logged-in admin, a draft counts as found and its 404 row disappears while visitors still get a 404 (the recheck also runs from the panel).

Authorization

Without authorize() the redirects.manage gate decides when it is defined; otherwise everyone who can enter the panel can use the resources - including pointing any 404 path at an external URL. Set one of the two unless every panel user is trusted with that. The 404 log item follows the redirects item in the navigation.

Commands and schedule

filament-redirects.schedule.enabled (off by default, needs the Laravel scheduler) runs prune at 03:10 and recheck at 03:20; both times are configurable.

Configuration

Gotchas

• The middleware must be route middleware or appended in bootstrap/app.php - see above. Nothing will tell you when it is not running; hits staying at 0 will.
• One hop only. Chains are collapsed when a redirect is saved (saving B -> C re-points the existing A -> B to A -> C). Do not add hop-following to the middleware: it would cost a query per hop on every 404, scanners included. Only permanent rows (301/308) take part in that: a temporary B -> C (302/307) never re-points other rows, and a new redirect is not flattened through it, so removing a campaign 302 leaves the permanent rows as they were. A redirect into a "Gone" source keeps its target (the Gone row answers). Loops are refused through any active redirect.
• A redirect never wins over a real page. It is a fallback for a 404: a row whose source is a live URL is inert. /old/ follows the same rule when resolvesUsing() is set.
• Browsers cache a 301 (and 308) forever, so a later edit (a new target, "Gone") never reaches a visitor who already followed it. The response carries Cache-Control: max-age=<cache_ttl>, must-revalidate so changes arrive within an hour. A 301 that is already cached in a browser cannot be recalled; clear that cache once.
• Livewire and Cache-Control. Livewire pushes DisableBackButtonCacheMiddleware onto the kernel's global stack, and once any Livewire component rendered on a 404 page it overwrites the response's Cache-Control with no-store. The redirect response resets that flag, so the bounded max-age survives.
• Gone. Stored with code 404, served with redirects.gone_status (default 410; set 404 for plain not-found) on the site's own error page. Unlike an unmatched path it is not logged - it is a decision.
• A scanned URL arrives collapsed. A scanner that requests https://site.com/x as a path leaves https:/site.com/x in your log after nginx merges the slashes; the path field recognises that form too and strips your own domain.
• Bots in the log. ignore_bots is off on purpose. Turn it on only if you want humans' broken links alone.
• Raw SQL bypasses the cache invalidation. Call Redirects::flushCache() after it.

Translations

English, Ukrainian, German, Spanish, French, Italian, Dutch, Polish, Brazilian Portuguese and Turkish. Publish to override:

AI agents

The package ships Laravel Boost guidelines (resources/boost/guidelines/core.blade.php) covering the registry, the middleware trap, the path canon and the write path.

Testing

The tests run on Orchestra Testbench with an in-memory SQLite database and a workbench panel (admin).

Changelog

See the CHANGELOG.

License

MIT - see the LICENSE.