Conversation Summary Memory

Conversation Summary memory generates a summary of the entire conversation, including responses from the user, agent, and tool calls. This is a direct implementation of Langchain's conversation summary memory.

Getting Started

To implement conversation summary memory, follow these steps:

1. Add the HasMemory trait and implement the resolveMemory method in your agent.
2. Include the ManagesMemory trait in your agent.
3. In your Blade prompt view, use the @include('synapse::Parts.MemoryAsMessages') snippet to display the memory as messages.
4. Set the memory type to ConversationSummaryMemory in the resolveMemory method.

IMPORTANT

To keep things simple with memory, the handle method input should always have an input key. This is what is stored as the user's content. For example:
$agent->handle(['input' => 'How Are you?']);

Example:

<?php

use UseTheFork\Synapse\Agent;
use UseTheFork\Synapse\Integrations\OpenAIIntegration;
use UseTheFork\Synapse\Memory\ConversationSummaryMemory;
use UseTheFork\Synapse\Contracts\Agent\HasMemory;
use UseTheFork\Synapse\Traits\Agent\ManagesMemory;

class SimpleAgent extends Agent implements HasMemory
{
    use ManagesMemory;  

    protected string $promptView = 'synapse::Prompts.SimplePrompt';

    public function resolveIntegration(): Integration
    {
        return new OpenAIIntegration();
    }

    public function resolveMemory(): Memory
    {
        return new ConversationSummaryMemory();
    }
}