Classification

Classification asks a fixed set of questions about content and returns typed answers with probabilities for routing, moderation, and scoring, where free-form text will not do. TypeSafe is the default backend; OpenAI, OpenRouter, and OpenAI-Compatible classify too.

WARNING

Classification is experimental, the API may change in future minor releases of the plugin.

Asking questions

Each question gets a key; the response returns each answer under its key. Boolean measures true-probability, Choice picks one option with per-option probabilities, Score positions the answer on ordered levels:

use Crustum\Ai\Classification;
use Crustum\Ai\Classification\Boolean;
use Crustum\Ai\Classification\Choice;
use Crustum\Ai\Classification\Score;

$result = Classification::of('My invoice is wrong and I need help today.')
    ->questions([
        'urgent' => new Boolean('Does this request need an immediate response?'),
        'department' => new Choice('Which team should handle this request?', [
            'billing' => 'Payments, invoices, and refunds',
            'technical' => 'Bugs, outages, and integrations',
            'sales' => 'Pricing, plans, and upgrades',
        ]),
        'frustration' => new Score('How frustrated is the customer?', [
            'Calm',
            'Frustrated',
            'Very angry',
        ]),
    ])
    ->classify();
echo ($result['urgent']->isTrue() ? 'yes' : 'no') . PHP_EOL;
echo $result['department']->choice . PHP_EOL;
echo $result['frustration']->level() . PHP_EOL;

yes
technical
1

isTrue(threshold: 0.8) gates a boolean at any cutoff (default 0.5); probabilityOf('technical') and probabilities expose the full distribution; confidence reads the spread, null when unmeasurable. Score adds score (probability-weighted), label(), and normalized(). Iterate, count, or pull one answer with answer('urgent'); usage and meta->provider ride along like everywhere else.

Single yes-or-no calls skip the ceremony: Text::of($message)->decide('Is this spam?') returns a boolean, with criteria and threshold shaping the question. Choosing one item from a list works the same way through Collections::of($items)->decide($question, $text, by: 'name'). Keys win over values when keyed, and threshold rejects low-confidence picks with null.

See also

• Providers — capability matrix
• TypeSafe — the classification backend
• OpenRouter — regrouping classification models