rag-v1

/**
 * @module MemoryService
 * Manages the long-term, persistent memory of the AI Agent.
 * This service simulates interaction with a vector database (e.g., Pinecone, ChromaDB).
 */

// Define the structure for a stored memory chunk
export interface MemoryChunk {
    id: string;
    timestamp: number;
    topic: string; // Key topic derived from the conversation
    summary: string; // The summarized content of the session
    source_context: string[]; // List of key documents/sources used in this session
}

/**
 * A simulated service for interacting with a persistent memory store.
 */
export class MemoryService {
    private static storage: Map<string, MemoryChunk[]> = new Map();

    /**
     * Saves the current conversation summary and context to long-term memory.
     * @param topic The primary subject of the conversation (e.g., "Q3 Budgeting").
     * @param summary A detailed summary generated by the LLM describing what was accomplished.
     * @param sources List of documents/sources used in this session.
     */
    public static async saveMemory(topic: string, summary: string, sources: FileHandle[]): Promise<void> {
        const newChunk: MemoryChunk = {
            id: Date.now().toString(), // Simple unique ID for simulation
            timestamp: Date.now(),
            topic: topic,
            summary: summary,
            source_context: Array.from(sources).map(f => f.name),
        };

        if (!MemoryService.storage.has(topic)) {
            MemoryService.storage.set(topic, []);
        }
        MemoryService.storage.get(topic)!.push(newChunk);

        console.log(`\n[MEMORY] Successfully saved memory chunk for topic: "${topic}".`);
    }

    /**
     * Retrieves relevant past memories based on a new query or focus topic.
     * @param query The current user query.
     * @param focusTopic The determined conversational focus.
     * @returns A string containing synthesized context from past sessions.
     */
    public static async retrieveMemory(query: string, focusTopic: string): Promise<string> {
        console.log(`\n[MEMORY] Searching long-term memory for topics related to "${focusTopic}" and query: "${query}"...`);

        // --- SIMULATION START ---
        // In a real system, this would be an API call to the vector database (e.g., cosine similarity search).
        // We simulate finding relevant memories based on keywords.

        const allTopics = Array.from(MemoryService.storage.keys());
        let retrievedContext = "";

        if (allTopics.length === 0) {
            return "No long-term memory found.";
        }

        // Simple simulation: If the query contains a keyword that matches any stored topic, we retrieve it.
        for (const topic of allTopics) {
            if (query.toLowerCase().includes(topic.toLowerCase()) || focusTopic.toLowerCase().includes(topic.toLowerCase())) {
                const chunks = MemoryService.storage.get(topic)!;
                retrievedContext += `\n--- RECALL: Previous Session on "${topic}" ---\n`;
                // Just grab the summary of the most recent session for that topic
                const latestChunk = chunks[chunks.length - 1];
                retrievedContext += `Summary: ${latestChunk.summary}\n`;
            }
        }

        if (retrievedContext) {
            return `[CONTEXT FROM MEMORY]: ${retrievedContext}`;
        } else {
            return "No relevant memories found in past sessions.";
        }
        // --- SIMULATION END ---
    }
}