PHP AI Agents with Neuron: From Install to Code Review Assistant

Initial Skepticism About PHP for AI

The author previously resisted using PHP for AI tasks, not because PHP is incapable, but because it seemed unnecessary. Calling an OpenAI API from PHP versus Python is essentially the same HTTP request. However, a real project need changed this view: adding a code review assistant to an existing PHP application. Building a separate Python service would introduce cross-language calls, dual environments, and dual dependency maintenance.

Discovering Neuron AI

A community member shared Neuron AI, a PHP agent framework. Installation is a single Composer command requiring PHP 8.1+: composer require neuron-core/neuron-ai Generating an agent skeleton uses the provided CLI tool:

./vendor/bin/neuron make:agent App\Neuron\MyAgent

The generated class extends Agent and requires implementing a provider() method to specify the LLM provider. No complex configuration files or initialization boilerplate.

Minimal Working Example

A basic agent using OpenAI's gpt-4o-mini model:

<?php

namespace App\Neuron;

use NeuronAI\Agent\Agent;
use NeuronAI\Agent\SystemPrompt;
use NeuronAI\Providers\AIProviderInterface;
use NeuronAI\Providers\OpenAI\OpenAI;

class MyAgent extends Agent
{
    protected function provider(): AIProviderInterface
    {
        return new OpenAI(
            key: 'sk-your-api-key',
            model: 'gpt-4o-mini',
        );
    }

    public function instructions(): string
    {
        return (string) new SystemPrompt(
            background: [
                "You are a friendly AI assistant that answers in Chinese."
            ],
        );
    }
}

Invoking the agent:

use App\Neuron\MyAgent;
use NeuronAI\Chat\Messages\UserMessage;

$response = MyAgent::make()
    ->chat(new UserMessage("Hello, who are you?"))
    ->getMessage();

echo $response->getContent();

Output: "Hello! I am an AI assistant built with the Neuron framework..." No Python, no extra service, no cross-language calls.

Building a Code Review Assistant with Tools

The core requirement: read files, analyze code, suggest fixes. In Neuron, this is done by adding "Tools" to the agent. A custom ReadFileTool extends Tool:

<?php

namespace App\Neuron\Tools;

use NeuronAI\Tools\Tool;

class ReadFileTool extends Tool
{
    public function __construct()
    {
        parent::__construct(
            name: 'read_file',
            description: 'Read the content of a specified PHP file',
        );
    }

    public function handle(string $path): string
    {
        if (!file_exists($path)) {
            return "File not found: {$path}";
        }

        return file_get_contents($path);
    }
}

Register the tool in the agent:

public function tools(): array
{
    return [
        new ReadFileTool(),
    ];
}

Now when asked "Check app/Http/Controllers/UserController.php for security issues", the agent autonomously calls read_file, reads the file, analyzes it, and returns suggestions. This autonomous tool use distinguishes an agent from a simple chatbot.

Unexpected Strengths of Neuron

1. Designed for PHP Developers

Neuron's code style is idiomatic PHP: 100% type declarations, clear interfaces, excellent IDE autocompletion. Parameter types are discoverable without guessing.

2. Smooth Laravel Integration

A dedicated neuron-laravel package provides an Artisan command: php artisan neuron:agent MyAgent And a Facade for simple usage:

use NeuronAI\Laravel\Facades\Neuron;
use NeuronAI\Chat\Messages\UserMessage;

$response = Neuron::chat(new UserMessage('Hello'))->getMessage();

3. Transparent Debugging via Inspector

Developing agents often means wondering "Why did it answer that?" Neuron integrates with Inspector. Setting an INSPECTOR_INGESTION_KEY enables a dashboard showing every step: which tool was called, with what parameters, and what the LLM returned. This is crucial for debugging complex agents.

When Should PHP Do AI?

The author's conclusion: it depends on the scenario.

Model training, deep learning, large-scale data processing → Python remains the better choice.

Adding AI features to an existing PHP project (smart customer service, code review, document Q&A, content summarization) → Neuron is more comfortable than calling a separate Python service.

AI logic and business logic live in the same codebase, sharing database, cache, auth, and deployment pipeline. No dual environments, no cross-language serialization or network overhead. Neuron is not a toy; it includes multi-agent orchestration, RAG, tool approval, conversation history management, and a middleware system.

Getting Started Recommendations

Install and run the minimal "Hello" example to get a feel.

Add a tool — even just "read file" or "query database" — and observe autonomous invocation.

Use php artisan neuron:agent in a Laravel project to generate a production agent, configure system prompts, and connect real project data.

For complex workflows, consult the Workflow and RAG documentation.

Final thought: Don't dismiss PHP for AI because of language bias. As Neuron's docs state, "Your next application will be agentic." The language matters less than the ability to build it.