ryderasking/laravel-ai-trace

LangSmith-style waterfall tracing for Laravel AI SDK workloads.

1.0.6 2026-04-09 10:42 UTC

Requires

• php: ^8.3
• illuminate/auth: ^12.0 || ^13.0
• illuminate/console: ^12.0 || ^13.0
• illuminate/database: ^12.0 || ^13.0
• illuminate/http: ^12.0 || ^13.0
• illuminate/routing: ^12.0 || ^13.0
• illuminate/support: ^12.0 || ^13.0
• illuminate/view: ^12.0 || ^13.0
• livewire/livewire: ^3.6 || ^4.0

Requires (Dev)

• laravel/pint: ^1.18
• orchestra/testbench: ^10.0
• phpunit/phpunit: ^11.0

MIT

• RyderAsKing

README

Laravel AI Trace is a tracing and debugging package for AI workflows built with Laravel AI SDK (laravel/ai).

It captures full trace -> span -> event lifecycle data with SDK-native hooks so you can inspect execution order, retries, tool calls, streaming milestones, failures, and timing with waterfall-style visibility.

The dashboard provides period-based trend cards, a filterable trace explorer, and quick drill-in links for incident investigation and performance triage.

Compatibility

• Laravel: 12+ (currently tested on 12.x and 13.x)
• PHP: 8.3+
• Livewire: 3.6+ (including 4.x)
• Laravel AI SDK (laravel/ai): currently validated against v0.3.2

How It Works

Laravel AI Trace subscribes to Laravel AI SDK lifecycle events (for example agent start/end, tool start/end, stream completion, and failover events), then:

1. Correlates each callback to a trace and active span context.
2. Persists traces, spans, and events in relational tables.
3. Deduplicates repeated callbacks inside a configurable TTL window.
4. Applies privacy mode/redaction before storing sensitive content.
5. Serves dashboard/query data through package services.

This package is SDK-native by design. It does not use HTTP fallback instrumentation.

Dashboard Stack

The dashboard is package-owned UI and uses:

• Livewire components for card rendering, polling, filters, and URL-backed state
• Blade components for layout, cards, tables, and inspector primitives
• Alpine.js + Chart.js for chart interactions and dataset updates
• Tailwind CSS for styling (with compiled package assets)

The design is inspired by Laravel Pulse, but the dashboard does not require a runtime dependency on laravel/pulse.

Installation

1. Install Laravel AI SDK (if not already installed):

composer require laravel/ai

1. Install Laravel AI Trace:

composer require ryderasking/laravel-ai-trace

1. Publish package configuration and migrations:

php artisan vendor:publish --tag=ai-trace-config
php artisan vendor:publish --tag=ai-trace-migrations

1. Run migrations:

php artisan migrate

1. Optional smoke test:

php artisan ai-trace:smoke

Dashboard Authorization Setup

Dashboard authorization is controlled by the host application via gate.

By default, Laravel AI Trace checks gate viewAiTrace (ai-trace.dashboard.authorization_gate). Define it in your AppServiceProvider:

<?php

namespace App\Providers;

use App\Models\User;
use Illuminate\Support\Facades\Gate;
use Illuminate\Support\ServiceProvider;

class AppServiceProvider extends ServiceProvider
{
    public function boot(): void
    {
        Gate::define('viewAiTrace', function (User $user): bool {
            return $user->is_admin;
        });
    }
}

Behavior details:

• If the configured gate exists, access is decided by that gate.
• If no gate is defined, dashboard access falls back to local environment only.
• If ai-trace.dashboard.enabled is false, the dashboard routes return 404.

Configuration

Published config file: config/ai-trace.php

Important keys:

• ai-trace.enabled: master package toggle
• ai-trace.track_ai_sdk: subscribe/unsubscribe SDK event tracing
• ai-trace.record_content_mode: none, hash, redacted, full
• ai-trace.sample_rate: trace sampling (0..1)
• ai-trace.dedup_ttl_seconds: event deduplication window
• ai-trace.stream.max_events_per_invocation: cap stored stream timeline events
• ai-trace.retention_days: retention horizon for stored traces
• ai-trace.dashboard.*: domain/path/middleware/gate settings

Routes

Dashboard routes are registered by the package service provider:

• ai-trace.dashboard: /{dashboard-path} (default path: /ai-trace)
• ai-trace.dashboard.trace: /{dashboard-path}/traces/{traceId}

Data Model

Core persistence tables:

• ai_traces: one row per end-to-end workflow
• ai_spans: one row per operation (agent/tool/operation step)
• ai_span_events: timeline markers for span-level events

This structure enables parent-child waterfall reconstruction and detailed drilldown inspection.

Production Notes

• Keep record_content_mode=redacted (or stricter) for sensitive environments.
• Use gate-based authorization for all non-local dashboard access.
• Tune sample_rate and stream event caps based on traffic profile.
• Plan retention cleanup strategy based on retention_days and compliance needs.