neuron-agent-builder

Neuron AI Agent Builder

This skill helps you create and configure Neuron AI agents for building agentic applications in PHP.

Core Agent Structure

A Neuron agent extends the Agent class and implements key methods:

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

class MyAgent extends Agent
{
    protected function provider(): AIProviderInterface
    {
        return new Anthropic(
            key: 'ANTHROPIC_API_KEY',
            model: 'ANTHROPIC_MODEL',
        );
    }

    protected function instructions(): string
    {
        return (string) new SystemPrompt(
            background: [
                "You are a helpful AI assistant."
            ]
        );
    }
}

Agent Execution Methods

Chat Mode (Synchronous)

For standard back-and-forth conversations:

$agent = MyAgent::make();

$response = $agent->chat(
    new UserMessage("Hello!")
)->getMessage();

echo $response->getContent();

Stream Mode (Real-time)

For streaming responses as chunks arrive. Chunks represents pieces of context that are generated by the model. The TextChunk class is used to represent a piece of text, but there are other chunk types available:

• NeuronAI\Chat\Messages\Chunks\TextChunk
• NeuronAI\Chat\Messages\Chunks\ImageChunk
• NeuronAI\Chat\Messages\Chunks\AudioChunk
• NeuronAI\Chat\Messages\Chunks\ToolCallChunk
• NeuronAI\Chat\Messages\Chunks\ToolResultChunk

$handler = $agent->stream(new UserMessage("Hello"));

foreach ($handler->events() as $event) {
    if ($event instanceof TextChunk) {
        echo $event->content;
    }
}

Streaming Adapters for UI Integration

When connecting a frontend UI to an agent, use streaming adapters to format the response for specific protocols. This enables seamless integration with popular AI UI libraries.

When to use:

• Building a chat interface with React/Vue/Next.js
• Using Vercel AI SDK's useChat hook
• Implementing AG-UI protocol for agent-frontend communication
• Any SSE-based real-time UI updates

Available adapters:

• VercelAIAdapter - Compatible with Vercel AI SDK (useChat, useCompletion)
• AGUIAdapter - AG-UI protocol for agent interactions

Laravel example:

use NeuronAI\Chat\Messages\Stream\Adapters\VercelAIAdapter;

Route::post('/chat', function (Request $request) {
    $handler = MyAgent::make()->stream(
        new UserMessage($request->input('message'))
    );

    $stream = $handler->events(new VercelAIAdapter());

    return response()->stream(
        function () use ($stream) {
            foreach ($stream as $line) {
                echo $line;
                ob_flush();
                flush();
            }
        },
        200,
        $adapter->getHeaders()
    );
});

Structured Output Mode

For extracting structured data from a natural language:

class Person
{
    #[SchemaProperty(description: 'The user name', required: true)]
    public string $name;

    #[SchemaProperty(description: 'What the user loves to eat')]
    public string $preference;
}

$person = $agent->structured(
    new UserMessage("I'm John and I like pizza!"),
    Person::class
);

Providers Configuration

Anthropic

new Anthropic(
    key: $_ENV['ANTHROPIC_API_KEY'],
    model: 'claude-3-5-sonnet-20241022',
)

OpenAI

new OpenAI(
    key: $_ENV['OPENAI_API_KEY'],
    model: 'gpt-4',
)

Ollama (Local)

new Ollama(
    baseUrl: 'http://localhost:11434',
    model: 'llama3',
)

Other Providers

• Gemini - Google AI models
• VertexAI - Google Vertex AI platform
• Mistral - Mistral AI models
• HuggingFace - Open models via HuggingFace
• Deepseek - DeepSeek models
• Grok - XAI models
• AWSBedrockRuntime - AWS Bedrock inference platform
• Cohere - Cohere models
• AzureOpenAI - Use OpenAI models on the Azure platform
• ZAI - ZAI for GLM models

Tools Integration

Adding Built-in Toolkits

use NeuronAI\Tools\Toolkits\MySQL\MySQLToolkit;
use NeuronAI\Tools\Toolkits\Calculator\CalculatorToolkit;

protected function tools(): array
{
    return [
        MySQLToolkit::make(\DB::connection()->getPdo()),
        CalculatorToolkit::make(),
    ];
}

Available Toolkits

• MySQLToolkit - Database queries via MySQL
• PostgreSQLToolkit - Database queries via PostgreSQL
• CalculatorToolkit - Math operations (sum, mean, std, etc.)
• TavilyToolkit - Web search with Tavily
• SESToolkit - Email sending via AWS SES
• JinaToolkit - Reranking with Jina

Creating Custom Tools

use the neuron-tool-creator skills for more complex tool creation:

use NeuronAI\Tools\Tool;
use NeuronAI\Tools\ToolProperty;

class WeatherTool extends Tool
{
    public function __construct()
    {
        parent::__construct(
            name: 'get_weather',
            description: 'Get the current weather for a location',
        );
    }

    /**
     * @return ToolProperty[]
     */
    protected function properties(): array
    {
       return [
            new ToolProperty(
                name: 'location',
                type: ToolPropertyType::String,
                description: 'The city name',
                required: true,
            ),
        ];
    }

    public function __invoke(string $location): mixed
    {
        // Call weather API and return result
        return "The weather in {$location} is sunny, 72°F";
    }
}

System Prompt Engineering

Use SystemPrompt for structured agent instructions:

new SystemPrompt(
    background: [
        "You are a data analyst expert in creating reports.",
    ],
    steps: [
        "Analyze the user's request",
        "Query the database",
        "Generate a summary",
    ],
    output: [
        "Always cite your sources",
        "Never make up data",
    ]
)

Chat History

Agents automatically maintain conversation history. For custom components:

use NeuronAI\History\FileChatHistory;

// In agent class
protected function chatHistory(): ChatHistoryInterface
{
    return new FileChatHistory('/path/to/memory.json');
}

Memory Types

• InMemoryChatHistory - Default, session-based
• FileChatHistory - Persist to file
• SQLChatHistory - Database-backed
• EloquentChatHistory - Laravel Eloquent integration

Content Blocks (Multi-modal)

Agents support multiple content types:

use NeuronAI\Chat\Messages\ContentBlocks\TextContent;
use NeuronAI\Chat\Messages\ContentBlocks\ImageContent;
use NeuronAI\Chat\Enums\SourceType;

$message = new UserMessage([
    new TextContent('Analyze this image:'),
    new ImageContent(
        content: 'https://example.com/image.jpg',
        sourceType: SourceType::URL,
        mediaType: 'image/jpeg'
    ),
]);

CLI Generation

Use the Neuron CLI to generate an agent boilerplate:

vendor/bin/neuron make:agent MyCustomAgent

Common Patterns

Tool Approval Middleware

For human oversight of tool execution:

use NeuronAI\Agent\Middleware\ToolApproval;
use NeuronAI\Agent\Nodes\ChatNode;
use NeuronAI\Agent\Nodes\StreamNode;
use NeuronAI\Agent\Nodes\StructuredOutputNode;

// In agent constructor
$this->middleware([
    ChatNode::class,
    StreamNode::class,
    StructuredOutputNode::class
], new ToolApproval());

Observability with Inspector

Monitor agent execution:

# Set environment variable
INSPECTOR_INGESTION_KEY=your_key_here

Parallel Tool Calls

Execute tools in parallel (requires pcntl):

$agent->parallelToolCalls(true);

Key Decisions

When helping users build agents:

1. Choose execution mode based on requirements:

   • chat() for standard conversations
   • stream() for real-time streaming
   • structured() for data extraction

2. Add tools when the agent needs to:

   • Access external systems (databases, APIs)
   • Perform calculations
   • Search the web
   • Send emails

3. Configure chat history when:

   • Long-running conversations need persistence
   • Multiple sessions should share history
   • Conversation context needs to be shared across agents

4. Use middleware for:

   • Context summarization
   • Tool approval workflows
   • Custom pre/post nodes processing

Project Structure Considerations

For Laravel projects:

namespace App\Neuron;

class MyAgent extends Agent { ... }

For Symfony projects:

• Use dependency injection for providers
• Configure a service in services.yaml