README

A Laravel AI Agent framework with first-class multi-tenancy support.

Key Highlights

• First-Class Multi-Tenancy — Team isolation, system vs custom agents, per-team cost attribution
• 10+ LLM Providers — OpenAI, Anthropic, Gemini, Mistral, Ollama, and more via Prism PHP
• 5 Vector Stores — Pinecone, Qdrant, Weaviate, Chroma, PgVector for semantic memory
• Full Workflow Engine — Sequential, parallel (in-process or queued across workers), conditional, loop, and human-in-the-loop patterns
• LLM-as-Judge Evaluation — Built-in evaluation framework with custom assertions and metrics
• Production Ready — Rate limiting, caching, circuit breakers, and comprehensive event system

Installation

composer require agenticorchestrator/agenticorchestrator

Publish the configuration:

php artisan vendor:publish --provider="AgenticOrchestrator\AgenticOrchestratorServiceProvider"

Run migrations:

php artisan migrate

Quick Start

Create Your First Agent

php artisan agent:make CustomerSupportAgent

<?php

namespace App\Agents;

use AgenticOrchestrator\Agents\Agent;
use AgenticOrchestrator\Tools\Attributes\Tool;

class CustomerSupportAgent extends Agent
{
    protected string $name = 'Customer Support';
    protected string $description = 'Handles customer inquiries';
    protected string $model = 'gpt-4o';

    public function instructions(): string
    {
        return "You are a helpful customer support agent for {$this->team->name}.";
    }

    #[Tool('Look up customer order')]
    public function lookupOrder(string $orderId): array
    {
        return $this->team->orders()
            ->where('id', $orderId)
            ->firstOrFail()
            ->toArray();
    }
}

Use Your Agent

use App\Agents\CustomerSupportAgent;

// Simple usage
$response = CustomerSupportAgent::make()
    ->forTeam($team)
    ->respond('What is the status of order #12345?');

echo $response->content;

// With streaming
$stream = CustomerSupportAgent::make()
    ->forTeam($team)
    ->stream('Tell me about your return policy');

foreach ($stream as $chunk) {
    echo $chunk->content;
}

Multi-Tenancy Support

// System agents (platform-wide, read-only)
class SystemHelpAgent extends Agent
{
    protected bool $isSystem = true;
}

// Custom agents (team-owned)
class TeamSalesAgent extends Agent
{
    protected bool $isSystem = false;
}

// Get available agents for a team
$agents = $team->availableAgents(); // System + team's custom agents

Workflow Orchestration

use AgenticOrchestrator\Workflows\Workflow;
use AgenticOrchestrator\Workflows\Patterns\ParallelPattern;
use AgenticOrchestrator\Workflows\Steps\{AgentStep, ConditionalStep};

class OnboardingWorkflow extends Workflow
{
    public function definition(): array
    {
        return [
            new AgentStep(WelcomeAgent::class, output: 'welcome'),

            ParallelPattern::make([
                new AgentStep(AccountSetupAgent::class, output: 'account'),
                new AgentStep(PreferencesAgent::class, output: 'prefs'),
            ]),

            new ConditionalStep(
                condition: fn ($ctx) => $ctx->get('customer.plan') === 'enterprise',
                then: new AgentStep(EnterpriseSetupAgent::class),
            ),

            new AgentStep(SummaryAgent::class, output: 'summary'),
        ];
    }
}

// Execute
$result = OnboardingWorkflow::make()
    ->forTeam($team)
    ->run(['customer' => $customerData]);

Memory Systems

// Configure memory in your agent
protected array $memory = [
    'driver' => 'vector',
    'vector_store' => 'pinecone',
    'namespace' => 'support',
];

// Use memory
$this->getMemory()->store('customer_123', $preferences);
$data = $this->getMemory()->recall('customer_123');
$results = $this->getMemory()->search('shipping policy', limit: 5);

Features

Core Features

• Multi-Tenancy: Team isolation, system vs custom agents, per-team cost tracking
• 10+ LLM Providers: Via Prism PHP (OpenAI, Anthropic, Gemini, Mistral, Ollama, etc.)
• Tool System: Attribute-based tools with parallel execution
• Memory: Session, cache, database, vector (5 stores), and RAG drivers
• Streaming: Real-time token streaming with callbacks

Advanced Features

• Workflows: Sequential, parallel, conditional, loop, human-in-the-loop patterns
• Agent Delegation: Agents can delegate tasks to other agents
• Evaluation: LLM-as-judge with custom assertions and metrics
• Error Handling: Retry strategies, circuit breakers, fallback providers
• Usage Tracking: Tokens, cost, latency tracking per team/agent/user

Production Features

• Rate Limiting: Per-user, per-team, per-agent limits
• Caching: Response, embedding, and tool result caching
• Events: Comprehensive event system for all operations
• Testing: Fakes, mocks, and evaluation framework

Configuration

// config/agent-orchestrator.php
return [
    'default_provider' => 'openai',
    'default_model' => 'gpt-4o',

    'multi_tenancy' => [
        'enabled' => true,
        'team_model' => \App\Models\Team::class,
        'system_agents' => [
            \App\Agents\SystemHelpAgent::class,
        ],
    ],

    'memory' => [
        'default' => 'cache',
        'drivers' => [
            'vector' => [
                'store' => 'pinecone',
                'embedding_provider' => 'openai',
            ],
        ],
    ],

    'rate_limiting' => [
        'per_team' => ['requests' => 1000, 'period' => 60],
    ],
];

Artisan Commands

Testing

use AgenticOrchestrator\Facades\Agent;

// Fake responses
Agent::fake([
    CustomerSupportAgent::class => 'Mocked response',
]);

// Assertions
Agent::assertResponded(CustomerSupportAgent::class);
Agent::assertToolCalled(CustomerSupportAgent::class, 'lookupOrder');

Requirements

• PHP 8.3+
• Laravel 11.0, 12.0, or 13.0

Documentation

Full documentation available in the docs folder.

Contributing

Please see CONTRIBUTING for details.

Security

If you discover any security issues, please email agenticorchestrator@proton.me instead of using the issue tracker.

Credits

• Built with Prism PHP for LLM abstraction
• Inspired by LarAgent, Vizra ADK, and Neuron AI

License

The MIT License (MIT). Please see License File for more information.