Advanced

Nothing here is required to run the package. Reach for it when the defaults are not enough.

Runtime configuration

Config files are frozen once cached at deploy time. If your settings live in a database, or an administrator picks the model, register a resolver — it is asked on every read, so a change takes effect immediately.

use Aaix\LaravelErrorAudit\Facades\ErrorAudit;

public function boot(): void
{
    ErrorAudit::recipientsUsing(fn () => Setting::get('audit_recipients'));
    ErrorAudit::modelUsing(fn () => Setting::get('audit_model'));
}

Named resolvers exist for the common keys (recipientsUsing, senderUsing, modelUsing, providerUsing, periodUsing, minimumLevelUsing, contextLevelUsing, redactUsing, notifiableUsing, channelsUsing). Anything else is reachable by name, and a resolver returning null falls back to config:

ErrorAudit::resolveUsing('ai.max_issues_per_run', fn () => $team->auditQuota());

Delivery

The report is delivered as a notification, so routing is yours. Hand it any notifiable — a user, a team, a collection of models:

ErrorAudit::notifiableUsing(fn () => User::where('receives_audit', true)->get());
ErrorAudit::channelsUsing(fn () => ['mail', 'slack']);

The notification's toArray() carries the whole report — counts and every issue with its urgency, cause and suggestion — so a custom channel can render it any way it likes. To adjust the mail just before it goes out:

ErrorAudit::sending(function ($mail, $report) {
    $mail->cc('archive@example.com');
});

Filtering issues

Drop issues before they reach the analysis, and therefore before they cost anything:

ErrorAudit::filterIssues(
    fn ($issue) => $issue->exceptionClass !== NotFoundHttpException::class
);

Log channels

Channels come from your application's own logging config:

• single and daily → resolved to their files
• stack → expanded; a file used by several channels is read once
• write-nowhere drivers (slack, syslog, null) → skipped

'channels' => [
    'include' => ['*'],
    'exclude' => ['audit'],
],

Both lists accept wildcards; exclusion wins.

Failed jobs

A job that exhausts its retries never reaches a log file — its exception lands in the queue's failed-job store. Failed jobs from the audited period are therefore included in the report like any other issue, under the failed-jobs channel, with the job class carried in the message.

'failed_jobs' => [
    'enabled' => true,
],

Strictly best effort: an application without a failed-job table, or with a failed driver other than database / database-uuids, simply contributes nothing — nothing fails, nothing warns.

Queue

'queue' => [
    'connection' => null,
    'queue' => null,
    'timeout' => 1800,
],

Left null, generation runs on the application's default connection. Name one only to keep the audit off a busy queue.

Assessment cache

'cache' => [
    'store' => null,
],

Assessments live in the framework cache, keyed by fingerprint — no schema, no migration. Name a store to keep the assessments out of a cache you flush regularly. Losing the entries is harmless: the affected issues are analysed once more, nothing else changes.

The cache only saves requests — it never changes what the report says. A report is a pure function of its time window: "new" and every delta compare against the immediately preceding window of the same length, read fresh from the logs, so the same period always yields the same report.

Preview route

Off by default. The one deliberate environment switch in the package, so it reads from .env, like APP_DEBUG:

ERROR_AUDIT_PREVIEW=true

'preview' => [
    'enabled' => env('ERROR_AUDIT_PREVIEW', false),
    'path' => 'error-audit/preview',
    'middleware' => ['web'],
],

• /error-audit/preview renders the report exactly as it would be mailed.
• /error-audit/preview?prompts=1 shows every prompt sent to the AI provider, with no API call made.

Both accept ?since=24+hours and ?until=. Keep it off in production, or guard it with your own middleware.