diff --git a/package.json b/package.json
index 88ea709..27560ef 100644
--- a/package.json
+++ b/package.json
@@ -1,12 +1,12 @@
 {
   "name": "@four-bytes/four-opencode-brain",
-  "version": "1.7.1",
+  "version": "1.7.2",
   "description": "Unified brain plugin — single SQLite DB for RAG search, memory, and knowledge base",
   "license": "Apache-2.0",
   "type": "module",
   "scripts": {
     "build": "NODE_ENV=production bun run scripts/build.ts",
-    "postinstall": "ln -sf node_modules/node-llama-cpp/llama llama && rm -rf node_modules/node-llama-cpp/bins/linux-x64 && cp -r node_modules/@node-llama-cpp/linux-x64/bins/linux-x64 node_modules/node-llama-cpp/bins/linux-x64",
+    "postinstall": "if [ \"$BRAIN_EMBED_ENABLE\" = \"true\" ] || [ \"$BRAIN_EMBED_ENABLE\" = \"1\" ]; then ln -sf node_modules/node-llama-cpp/llama llama && rm -rf node_modules/node-llama-cpp/bins/linux-x64 && cp -r node_modules/@node-llama-cpp/linux-x64/bins/linux-x64 node_modules/node-llama-cpp/bins/linux-x64; fi",
     "test": "bun test"
   },
   "keywords": [
diff --git a/scripts/build.ts b/scripts/build.ts
index b0d295a..014a0f1 100644
--- a/scripts/build.ts
+++ b/scripts/build.ts
@@ -4,11 +4,15 @@ import { rmSync, renameSync } from "node:fs";
 rmSync("dist", { recursive: true, force: true });
 
 // 2. Server build: Bun bundler on clean dist (FIRST — no stale artifacts)
+const externals = ["@opencode-ai/*"];
+if (process.env.BRAIN_EMBED_ENABLE === "true" || process.env.BRAIN_EMBED_ENABLE === "1") {
+  externals.push("@node-llama-cpp/*");
+}
 const server = await Bun.build({
   entrypoints: ["src/four-opencode-brain.ts"],
   outdir: "dist",
   target: "bun",
-  external: ["@opencode-ai/*", "@node-llama-cpp/*"],
+  external: externals,
   minify: process.env.NODE_ENV === "production",
 });
 
diff --git a/src/embed/embeddingService.ts b/src/embed/embeddingService.ts
index 36472a1..3376d91 100644
--- a/src/embed/embeddingService.ts
+++ b/src/embed/embeddingService.ts
@@ -105,8 +105,14 @@ export class EmbeddingService {
     if (this.initialized) return;
     if (this.initPromise) return this.initPromise;
 
+    log("debug", "embedding-service", "initialize() called", {
+      brainEmbedEnable: process.env.BRAIN_EMBED_ENABLE ?? "(unset)",
+      caller: new Error().stack?.split("\n")[2]?.trim().slice(0, 120),
+    });
+
     this.initPromise = (async () => {
       if (process.env.BRAIN_EMBED_ENABLE !== "true" && process.env.BRAIN_EMBED_ENABLE !== "1") {
+        log("info", "embedding-service", "BRAIN_EMBED_ENABLE not set — skipping real embedding model, using pseudo-embeddings");
         this.initialized = true;
         this._available = false;
         return;
@@ -127,6 +133,7 @@ export class EmbeddingService {
         this.model = await llama.loadModel({ modelPath: resolvedModelPath });
         this.ctx = await this.model.createEmbeddingContext();
         this._available = true;
+        log("info", "embedding-service", "BRAIN_EMBED_ENABLE=1 — real embedding model initialized successfully", { dimensions: this.dimensions || "lazy" });
         log("info", "embedding-service", "Real embedding model loaded successfully");
       } catch (err) {
         this._available = false;