Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.laragent.ai/llms.txt

Use this file to discover all available pages before exploring further.

LarAgent doesn’t contain any built-in Implementations of RAG, instead it gives you a solid foundation and a frame to inject your prefered RAG approach elegantly
Since there are multiple approaches and multiple implementations per each (including in PHP), we decided to not reinvent the wheel. For more info check RAG section in Guides.

​
Prompt method of an Agent

The best place to inject your RAG solution is the prompt method of Agent class, which initially looks like this: 
public function prompt($message)
{
    return $message;
}
When using Agent, the message from respond method is going through prompt method at first, so here you can mutate/enhance the user message as well as query the database for extra context. Let’s imagine we have (any type of) RAG implemented as RetrivalService with Search method. 
use App\Services\RetrivalService;

// ...

public function prompt($message)
{
    $results = RetrivalService::search($message);

    return $message;
}
Additionally, we can format it as well structured context using blade template: 
use App\Services\RetrivalService;

// ...

public function prompt($message)
{
    $results = RetrivalService::search($message);
    $context = view('prompts.rag_context', ["result" => $results]);

    return $message;
}
The most comfortable way to add given context in chat sequence, is the “Developer” message role, since it is allowed to be added in any point at chat history sequence as well as doesn’t requires any further maintenance 
use App\Services\RetrivalService;
use LarAgent\Messages\DeveloperMessage;

// ...

public function prompt($message)
{
    $results = RetrivalService::search($message);
    $context = view('prompts.rag_context', ["result" => $results])->render();
    $devMsg = new DeveloperMessage(view('prompts.support_agent_context', [
        'documents' => $docs,
    ])->render());

    return $message;
}
And the only thing we are still missing is actual passing the $devMsg to the chat history, which is pretty simple: 
use App\Services\RetrivalService;
use LarAgent\Messages\DeveloperMessage;

// ...

public function prompt($message)
{
    $results = RetrivalService::search($message);
    $context = view('prompts.rag_context', ["result" => $results])->render();
    $devMsg = new DeveloperMessage(view('prompts.support_agent_context', [
        'documents' => $docs,
    ])->render());
    $this->chatHistory()->addMessage($devMsg);

    return $message;
}
That’s it! Your agent now has needed context to answer the user’s question! For more specific types of RAG implementations please check the Guides.