KuzuMem-MCP

// External dependencies import { generateObject } from 'ai'; // Internal services and utilities import { BaseMemoryAgent } from '../base/base-memory-agent'; import { MemoryContextBuilder } from '../context-builder'; import { PromptManager } from '../prompt-manager'; // Type imports import type { ToolHandlerContext } from '../../../mcp/types/sdk-custom'; import type { AnalysisResult } from '../../../schemas/optimization/types'; import type { OptimizationStrategy } from '../prompt-manager'; // Schema imports import { AnalysisResultSchema } from '../../../schemas/optimization/types'; /** * Service responsible for memory graph analysis using LLM intelligence * Analyzes memory graphs to identify optimization opportunities */ export class MemoryAnalysisService extends BaseMemoryAgent { private contextBuilder: MemoryContextBuilder; private promptManager: PromptManager; constructor( memoryService: any, config: any, contextBuilder: MemoryContextBuilder, promptManager: PromptManager, ) { super(memoryService, config); this.contextBuilder = contextBuilder; this.promptManager = promptManager; } /** * Analyze memory graph and identify optimization opportunities */ async analyzeMemory( mcpContext: ToolHandlerContext, clientProjectRoot: string, repository: string, branch: string = 'main', strategy: OptimizationStrategy = 'conservative', ): Promise<AnalysisResult> { const analysisLogger = this.createOperationLogger('analyzeMemory', { repository, branch, strategy, }); try { analysisLogger.info('Starting memory analysis'); // Build comprehensive memory context const memoryContext = await this.contextBuilder.buildMemoryContext( mcpContext, clientProjectRoot, repository, branch, ); // Get additional context for analysis const staleEntityCandidates = await this.contextBuilder.getStaleEntityCandidates( mcpContext, clientProjectRoot, repository, branch, this.getStaleDaysThreshold(strategy), ); const relationshipSummary = await this.contextBuilder.getRelationshipSummary( mcpContext, clientProjectRoot, repository, branch, ); // Build prompts for LLM analysis (with optional MCP sampling) const systemPrompt = this.config.enableMCPSampling ? await this.promptManager.buildContextAwareSystemPrompt( 'analyzer', strategy, mcpContext, clientProjectRoot, repository, branch, true, this.config.samplingStrategy || 'representative', ) : await this.promptManager.buildSystemPrompt('analyzer', strategy); const userPrompt = await this.promptManager.buildUserPrompt( 'analysis', memoryContext, strategy, JSON.stringify({ staleEntityCandidates, relationshipSummary }), ); analysisLogger.debug('Sending analysis request to LLM', { totalEntities: memoryContext.totalEntities, staleCandidates: staleEntityCandidates.length, }); // Generate analysis using LLM with reasoning configuration const result = await generateObject({ model: this.llmClient, system: systemPrompt, prompt: userPrompt, schema: AnalysisResultSchema, temperature: 0.1, // Low temperature for consistent analysis ...this.getReasoningConfig(), }); const analysisResult = result.object as AnalysisResult; analysisLogger.info('Memory analysis completed', { staleEntitiesFound: analysisResult.staleEntities.length, redundancyGroupsFound: analysisResult.redundancies.length, optimizationOpportunities: analysisResult.optimizationOpportunities.length, overallHealthScore: analysisResult.summary.overallHealthScore, }); return analysisResult; } catch (error) { analysisLogger.error('Memory analysis failed:', error); throw new Error(`Memory analysis failed: ${error}`); } } }