edulazaro / toxicfilter-sdk
The official PHP client for the ToxicFilter content moderation API.
Requires
- php: ^8.1
- ext-curl: *
- ext-json: *
Requires (Dev)
- phpunit/phpunit: ^10.5|^11.0|^12.0
Suggests
None
Provides
None
Conflicts
None
Replaces
None
README
ToxicFilter PHP SDK
The official PHP client for ToxicFilter.
composer require edulazaro/toxicfilter-sdk
use ToxicFilter\Client; $tf = new Client(getenv('TOXICFILTER_KEY')); $verdict = $tf->text('Check this message', [ 'locales' => ['en'], 'surface' => 'comment', 'reference' => 'comment_9931', ]); if ($verdict->blocked()) { return refuse($verdict->reason()); // the first reason; reasons() has them all } if ($verdict->needsReview()) { return hold($verdict->id(), $verdict->reasons()); } publish();
Using Laravel?
Laratox wraps this client for Laravel: a validation rule, a facade that reads as a sentence and a fake for your tests.
composer require edulazaro/laratox
ToxicFilter::text($comment->body)->check()->blocked(); $request->validate(['body' => ['required', Moderated::text()]]);
Reading a verdict
$verdict->decision(); // 'allow', 'review' or 'block' $verdict->allowed(); // and needsReview(), blocked() $verdict->reason(); // the first reason, or null $verdict->reasons(); // every reason, in words you can show the author $verdict->flagged(); // the categories that crossed a line: ['spam'] $verdict->scores(); // every category that scored, 0 to 1 $verdict->score('spam'); // one of them, 0.0 when it did not score $verdict->signals(); // every finding, with its evidence $verdict->topics(); // what it is ABOUT, per subject: ['gambling' => 0.82] $verdict->topic('crypto'); $verdict->id(); // mod_..., the name of this decision $verdict->reference(); // your own id, as you sent it $verdict->project(); // the project it was filed under $verdict->effort(); // 'low', 'medium' or 'high': the effort applied $verdict->modelRead(); // whether the model read it $verdict->modelWhy(); // why it did not although it could, or null $verdict->cached(); // answered from a verdict already reached $verdict->charged(); // credits this call cost $verdict->creditsRemaining(); $verdict->policy(); // the rules it was judged under: slug, version, overridden $verdict->tookMs();
Three decisions, not two. review is where the uncertainty is allowed to live: forced to
choose between publishing and deleting, a threshold set safely deletes real posts and one
set kindly publishes the abuse. There is no isToxic() here for the same reason: fifteen
categories collapsed into one boolean is somebody else's policy in your code.
How hard it looks
effort says how far a check may go, on every call, on a batch and on each of its items:
effort |
What happens | Cost |
|---|---|---|
low |
The free checks only | 1 credit |
medium |
The model reads only what the free checks left in doubt | 1 credit, plus the model's tokens when it reads |
high |
The model reads everything not already refused on hard evidence | 1 credit, plus the model's tokens |
Left out, it is your policy's default if it has one, then the kind's own: medium for text
and conversations, high for images and prompts, low for names, emails, signups and links.
Names, emails and links only take low: anything else is a validation error with the code
effort_unavailable. An image asked for at medium is read at high, and the answer says
so.
$tf->text($comment, ['effort' => 'low']); // never the model $tf->text($listing, ['effort' => 'high']); // the model reads it unless it was refused outright $verdict->effort(); // the level applied $verdict->modelRead(); // whether the model read it $verdict->modelWhy(); // 'settled', 'conversation_sampling', 'test_key' or 'unavailable', or null
What it does for you
Retries the right failures and never the wrong one. A 429 or a 5xx is asked again with
a growing wait; a QuotaExhausted is not, ever. Those two are constantly confused and the
correct behaviour is opposite: retry a rate limit, stop dead on a quota.
Makes those retries safe. Every call carries an Idempotency-Key, generated per call,
so a request that timed out and is asked again is judged once and billed once.
Waits as long as the service asked, and no longer. A 429 carries retry_after, and
honouring it beats guessing: the service knows when its own window turns over. It is bounded
by maxWait (30 seconds by default) all the same, because a number on the wire should not
decide how long your own request hangs.
Never reads anything but an answer as allow. A redirect (a base URL on http://,
which this client deliberately does not follow), an empty body, or a proxy's HTML page is a
ServerError, retried like one, and never a verdict. A verdict with no decision, or one the
client does not know, throws from decision(), allowed(), needsReview() and blocked()
rather than defaulting to anything.
Every failure is a type you can branch on:
| Status | Exception | Retried |
|---|---|---|
| 401 | AuthenticationError |
no |
| 402 | QuotaExhausted, with remaining(), required(), renewsAt() |
never |
| 404 | NotFound |
no |
| 422 | InvalidRequest, with fields() |
no |
| 429 | RateLimited, with retryAfter() |
yes |
409 idempotency_in_flight, 5xx, network, not an answer |
ServerError |
yes |
any other 409 (appeal_filed, no_restriction, ...) |
ApiError |
no |
All of them extend ToxicFilter\Exception\ApiError, which carries status, errorCode and
the decoded payload.
use ToxicFilter\Exception\QuotaExhausted; use ToxicFilter\Exception\RateLimited; try { $verdict = $tf->text($comment); } catch (QuotaExhausted $e) { // Stop calling. No amount of retrying produces credits. alert("out of credits, {$e->remaining()} left, renews {$e->renewsAt()}"); } catch (RateLimited $e) { // Already retried, and still too many. Back off properly. }
When the model is down
A verdict reached without the model because the provider was failing comes back with
degraded set, and is billed as the cheap call. It is a separate statement from
modelRead() being false: modelWhy() saying settled means the cheap detectors were
enough, degraded means nobody could read it, and only the first is reassuring. Hold or
queue what matters to you when you see it.
In a conversation the model is sometimes allowed and deliberately not run (it reads a message
when the free detectors found something, when a lead type is half there, or every few
messages). That is a third statement, and modelWhy() says it:
$verdict->modelWhy(); // 'conversation_sampling', or null when the model read it
Projects
An organization can moderate several sites, one project each. Name the project and the verdict is filed there, with its own activity, review queue and webhooks; leave it out and it goes to your default project. The keys and the credits are the organization's.
$verdict = $tf->text($comment, ['project' => 'forum']); $verdict->project(); // 'forum' $tf->batch($items, ['project' => 'forum']); // the whole batch, on the envelope $tf->records(['project' => 'forum']); // one project's queue $tf->batches(['project' => 'forum']); // its recent batches
A project that does not exist is refused with an InvalidRequest (unknown_project).
Rules without a policy
Send the line you care about and nothing else is acted on. No stored policy is looked up, and no default of ours is laid underneath.
$verdict = $tf->text($comment, [ 'rules' => ['thresholds' => ['sexual' => ['block' => 0.7]]], ]);
A category you did not mention still scores and still appears in signals; it just does not
decide anything. A name that is not a real category, subject or lead type is a 422: a line
that acts on nothing looks exactly like a line that works.
Send rules together with a policy and they are laid over it instead: the call wins for
what it names, the policy keeps everything else, and words are added to its lists.
$verdict->policy() then has 'overridden' => true.
$verdict = $tf->text($comment, [ 'policy' => 'comments', 'rules' => ['thresholds' => ['spam' => ['block' => 0.6]]], ]);
Domain rules travel the same way: lists of domains or endings (shop.example covers its
subdomains, ru everything under it), a domain registered recently, and what Cloudflare's
filtering resolvers block. They apply to the links in a text, to url() and to the domain of
an email address.
$verdict = $tf->text($comment, [ 'rules' => ['domains' => [ 'block' => ['ru'], 'young' => ['days' => 30, 'action' => 'review'], 'malware' => 'block', ]], ]);
The rest of the answer
$verdict->redacted(); // the content with the personal data masked, when you asked $verdict->context(); // repeats, near-duplicates, the actor's record and what it moved $verdict->shadow(); // what a policy you are trialling would have said. Never what happened $verdict->facts(); // noticed, not a finding: a language, an age signal, a fingerprint $verdict->degraded(); // part of the pipeline could not run
redacted is usually worth more than a refusal: throwing a whole comment away because it
carried one phone number throws away everything else the person wrote.
$verdict = $tf->text($comment, ['redact' => true, 'actor' => 'user_8812']); publish($verdict->redacted() ?? $comment);
Who is writing
Every verdict also carries leads: what kind of lead wrote it, scored per type. Neither a
harm nor a subject, but who is on the other side and what they want.
$verdict = $tf->conversation($messages, [ 'rules' => ['leads' => ['free_work_for_equity' => ['block' => 0.6]]], ]); $verdict->leads(); // ['free_work_for_equity' => 0.9, 'no_budget' => 0.7] $verdict->lead('sales_pitch'); // 0.0 when nothing of that type showed $verdict->blocked(); // true: your rule discarded it
Types: free_work_for_equity, no_budget, unrealistic_expectations, free_consulting,
scope_creep, no_show, sales_pitch, partnership_offer, job_seeker,
student_or_survey, support_request. Send the conversation rather than one message and
everything that person said counts.
Everything else
$tf->email('someone@mailinator.com'); $tf->name('asdkjhasd'); $tf->signup(['name' => 'Ana', 'email' => 'ana@example.com', 'bio' => '...']); $tf->image('https://cdn.example.com/photo.jpg'); $tf->image($base64OrDataUri); // or the bytes, if you have not published it $tf->imageData($rawFileContents); // the same thing, said explicitly $tf->url('https://bit.ly/3xYz'); // one link, judged as a link $tf->prompt($whatTheUserTypedIntoYourChatbot); // prompt injection, plus everything else // A message with what came before it. A pile-on is thirty people each writing one // ordinary rude sentence, and no classifier reading one of them can see it. $tf->conversation([ ['author' => 'u1', 'content' => '...'], ['author' => 'u2', 'content' => '...'], ['author' => 'u5', 'content' => 'the one being judged'], ], ['locales' => ['es']]); // Many things in one call. One bad item is an item, not a batch. $batch = $tf->batch([ ['kind' => 'text', 'content' => '...', 'reference' => 'c_1'], ['kind' => 'image', 'url' => '...', 'reference' => 'p_2'], ], ['effort' => 'low']); foreach ($batch->verdicts() as $index => $verdict) { /* ... */ } foreach ($batch->failures() as $index => $error) { /* ... */ } // negative keys: a failure no item owns // A backfill: up to 1,000 items per call, queued, answered immediately, polled or webhooked. $queued = $tf->batchAsync($aThousandComments); $status = $tf->batchStatus($queued->id()); $status->finished(); // and status(), count(), processed(), failed(), creditsCharged() // A backfill read back a page at a time. A cursor, not an offset: rows appear as workers // finish them, so an offset skips whatever was inserted behind it. $page = $tf->batchStatus($queued->id()); while ($page->hasMore()) { $page = $tf->batchStatus($queued->id(), ['after' => $page->nextAfter()]); } // The account's recent batches, newest first, for the id you did not keep. foreach ($tf->batches(['limit' => 20]) as $batch) { $batch->id(); $batch->status(); } // The review queue. $queue = $tf->records(['state' => 'open']); $tf->resolve($id, 'approved', 'ana@example.com'); $tf->feedback($id, 'false_positive'); // free, and the only honest measure we have // And what a held verdict has had done to it. Only record(), resolve() and feedback() // carry this: the rows of records() do not, so read one in full to see its state. $verdict = $tf->record($id); $verdict->reviewState(); // open, approved or rejected $verdict->resolvedBy(); // your own name for whoever decided $verdict->feedback(); // what you already told us, or null $verdict->content(); // only when your policy keeps it, and only until it expires $tf->keys(); // prefixes, modes, last use. Never a secret. $tf->revokeKey($id); // including the one you are calling with. There is no create. $tf->usage(); // credits, windows, prices. Works at zero credits. $tf->ping();
Statements of reasons and appeals
Switch statements on for a project in the panel (Settings, then Statements), give your policy rules a message, and every block comes back with a statement of reasons: what was done, why, whether it was automated, which of your rules it broke and how to contest it. It helps you produce what articles 17, 20 and 24(5) of the Digital Services Act ask for; it is not legal advice.
$verdict = $tf->text($comment, ['project' => 'forum']); $verdict->statement(); // restrictions, territories, facts, automated, ground, redress, text $verdict->statementText(); // the same in plain words, ready to send to the author // Say what your site does with it, when it is not what the rule or the project says. $tf->text($comment, ['restriction' => ['removal', 'account_suspended']]); // Later, from the record, in another language. A verdict that restricts nothing is a 409 // `no_restriction`, raised as an ApiError and never retried. $statement = $tf->statement($verdict->id(), 'es'); // The author contests it. It waits in the review queue under Appeals. $tf->appeal($verdict->id(), 'It was a recipe, not an insult.'); // A person decides, with reasons. appealDecision() is the text to send back. $decided = $tf->resolveAppeal($verdict->id(), 'reversed', 'ana', 'A recipe after all.', 'es'); $decided->appealDecision(); $decided->appeal(); // state, filed_at, reason, resolved_at, resolved_by, explanation $tf->record($id)->transparency(); // uuid and submitted_at, once filed with the Commission // A period for the Commission's Transparency Database, up to 31 days, 100 a page. $page = $tf->transparency('2026-10-01', '2026-10-31'); $next = $tf->transparency('2026-10-01', '2026-10-31', after: $page['next']);
Appealing twice, appealing after the six-month window or resolving an appeal that is not
open is a 409 (appeal_filed, appeal_window_closed, no_open_appeal). An
appeal.resolved webhook tells your site the outcome, which is what puts reversed content
back.
Webhooks
Your endpoint URL is public. Verify before you act:
use ToxicFilter\Webhooks; $event = Webhooks::event( file_get_contents('php://input'), // the RAW body $_SERVER['HTTP_X_TOXICFILTER_SIGNATURE'] ?? '', // absent is a failed check, not a crash getenv('TOXICFILTER_WEBHOOK_SECRET'), ); if ($event === null) { http_response_code(400); exit; }
Configuration
$tf = new Client( key: getenv('TOXICFILTER_KEY'), // tf_live_… or tf_test_… (never charged) baseUrl: 'https://toxicfilter.com', // https: redirects are never followed retries: 2, // for 429 and 5xx; a 402 is never retried timeout: 10, // seconds per attempt maxWait: 30, // ceiling on the wait between attempts );
Notes
Requires PHP 8.1 and ext-curl. No other dependencies: a client for one small API is not
worth a dependency tree, and pinning an HTTP library is how a client gets removed from a
project. Bring your own by implementing ToxicFilter\Transport.
How this is tested
composer install
composer test
No network and no framework: a stub transport answers like the API does. It covers what a caller cannot see and would otherwise discover from an invoice: what is retried, what never is, that a retry reuses its idempotency key while two calls do not, that anything but an answer is refused, and that bytes go out as bytes.
Author
Created by Edu Lazaro for ToxicFilter, the moderation API this client speaks to.
License
The ToxicFilter PHP SDK is open-sourced software licensed under the MIT license.
