Agent

Easily implement LLM interactions extending the basic Agent class.

Neuron provides you with the Agent class you can extend to inherit the main features of the framework and create fully functional agents. This class automatically manages some advanced mechanisms for you such as memory, tools and function calls, up to the RAG systems. We will go into more detail about these aspects in the following sections.

This ensure the portability of your agent because all the moving parts are encapsulated into a single entity that you can run wherever you want in your application, or release as stand alone composer packages.

Let's start creating an AI Agent summarizong YouTube videos. We start creating the YouTubeAgent class extending NeuronAI\Agent:

use NeuronAI\Agent;
use NeuronAI\Providers\AIProviderInterface;

class YouTubeAgent extends Agent
{
    protected function provider(): AIProviderInterface
    {
        ...
    }
}

Minimal implementation

Here is an example of the minimum implementation of an agent, in the following section we will add other features and capabilities:

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

class YouTubeAgent extends Agent
{
    protected function provider(): AIProviderInterface
    {
        // return an AI provider (Anthropic, OpenAI, Mistral, etc.)
        return new Anthropic(
            key: 'ANTHROPIC_API_KEY',
            model: 'ANTHROPIC_MODEL',
        );
    }
}

The only required method to implement is provider(). It's where we assign the agent the connection to the LLM we want to use. Let's assume it's Claude by Anthropic.

You can also use other providers like OpenAI, or Ollama if you want to run the model locally. Check out the supported providers.

System instructions

The second important building block is the system instructions. System instructions provide directions for making the AI ​​act according to the task we want to achieve. They are fixed instructions that will be sent to the LLM on every interaction.

That’s why they are defined by an internal method, and stay encapsulated into the agent entity. Let's implement the instructions() method:

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

class YouTubeAgent extends Agent
{
    protected function provider(): AIProviderInterface
    {
        // return an AI provider instance (Anthropic, OpenAI, Mistral, etc.)
        return new Anthropic(
            key: 'ANTHROPIC_API_KEY',
            model: 'ANTHROPIC_MODEL',
        );
    }
    
    public function instructions(): string
    {
        return new SystemPrompt(
            background: ["You are an AI Agent specialized in writing YouTube video summaries."],
            steps: [
                "Get the url of a YouTube video, or ask the user to provide one.",
                "Use the tools you have available to retrieve the transcription of the video.",
                "Write the summary.",
            ],
            output: [
                "Write a summary in a paragraph without using lists. Use just fluent text.",
                "After the summary add a list of three sentences as the three most important take away from the video.",
            ]
        );
    }
}

The SystemPrompt class is designed to take your base instructions and build a consistent prompt for the underlying model reducing the effort for prompt engineering. The properties has the following meaning:

• background: Write about the role of the Agent. Think about the macro tasks it's intended to accomplish.

• steps: Define the way you expect the Agent to behave. Multiple steps help the Agent to act consistently.

• output: Define how you want the agent to respond. Be explicit on the format you expect.

We highly recommend to use the SystemPrompt class to increase the quality of the results, in alternative you can just return a simple string:

class YouTubeAgent extend Agent
{
    ...
    
    public function instructions(): string
    {
        return "You are an AI Agent specialized in writing YouTube video summaries.";
    }
}

Talk to the YouTubeAgent

We are ready to test how the agent respond to our message based on the new instructions.

use NeuronAI\Chat\Messages\UserMessage;

$response = YouTubeAgent::make()
    ->chat(
        new UserMessage("Who are you?")
    );
    
echo $response->getContent();
// Hi, I'm a frindly AI agent specialized in summarizing YouTube videos!
// Can you give me the URL of a YouTube video you want a quick summary of?

Message

The agent always accept input as Message class, and return Message instances.

As you saw in the example above we sent a UserMessage instance to the agent and it responded with an AssistantMessage instance. A list of assistant message and user message creates a chat.

We will learn more about ChatHistory later, but it's important to know that the unified interface for the agent input and response is the Message object.