refactor: optimize core engine and retrieval logic for v2.80.43

src/config.ts

@@ -45,6 +45,19 @@ export interface IAgentConfig {
     maxAutoContinuations: number;
     /** 모델이 내부 사고만 출력하고 답변이 없으면 "최종 답변만" 지시로 1회 재생성. */
     finalOnlyRetryOnThoughtLeak: boolean;
+    // ─── Hybrid Semantic Search ───
+    /**
+     * Embedding model name as registered in LM Studio / Ollama. Empty disables
+     * semantic search and the retriever falls back to TF-IDF only. The user
+     * must load this model in the engine before enabling it here.
+     */
+    embeddingModel: string;
+    /**
+     * Blend between TF-IDF (sparse) and embedding cosine (dense) scoring.
+     * 0 = TF-IDF only (status quo), 1 = embedding only.
+     * Default 0.5 = equal weight, a reasonable starting point.
+     */
+    embeddingBlendAlpha: number;
 }
 
 // ─── 경로 정규화 유틸리티 ───
@@ -125,7 +138,9 @@ export function getConfig(): IAgentConfig {
         smallModelContextCap: Math.max(0, cfg.get<number>('smallModelContextCap', 0)),
         autoContinueOnOutputLimit: cfg.get<boolean>('autoContinueOnOutputLimit', true),
         maxAutoContinuations: Math.max(0, Math.min(10, cfg.get<number>('maxAutoContinuations', 4))),
-        finalOnlyRetryOnThoughtLeak: cfg.get<boolean>('finalOnlyRetryOnThoughtLeak', true)
+        finalOnlyRetryOnThoughtLeak: cfg.get<boolean>('finalOnlyRetryOnThoughtLeak', true),
+        embeddingModel: (cfg.get<string>('embeddingModel', '') || '').trim(),
+        embeddingBlendAlpha: Math.max(0, Math.min(1, cfg.get<number>('embeddingBlendAlpha', 0.5))),
     };
 }