@aeriondyseti/vector-memory-mcp 2.3.0-rc.6 → 2.4.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aeriondyseti/vector-memory-mcp",
-  "version": "2.3.0-rc.6",
+  "version": "2.4.0",
   "description": "A zero-configuration RAG memory server for MCP clients",
   "type": "module",
   "main": "server/index.ts",
@@ -30,6 +30,7 @@
     "test:quick": "bun test",
     "test:coverage": "bun test --preload ./tests/preload.ts --coverage",
     "benchmark": "bun test tests/benchmark.test.ts --preload ./tests/preload.ts",
+    "benchmark:update": "bun run scripts/update-benchmarks.ts",
     "test:preload": "bun run tests/preload.ts",
     "smoke": "bun run scripts/smoke-test.ts",
     "warmup": "bun run scripts/warmup.ts",

package/server/core/embeddings.service.ts

@@ -23,6 +23,14 @@ export class EmbeddingsService {
     return this._dimension;
   }
 
+  get isReady(): boolean {
+    return this.session !== null;
+  }
+
+  async warmup(): Promise<void> {
+    await this.initialize();
+  }
+
   private async initialize(): Promise<void> {
     if (this.session) return;
     if (!this.initPromise) {

package/server/transports/http/server.ts

@@ -111,10 +111,22 @@ export function createHttpApp(memoryService: MemoryService, config: Config): Hon
         embeddingDimension: config.embeddingDimension,
         historyEnabled: config.conversationHistory.enabled,
         pluginMode: config.pluginMode,
+        embeddingReady: memoryService.getEmbeddings().isReady,
       },
     });
   });
 
+  // Warmup endpoint — triggers ONNX model load if not already cached
+  app.post("/warmup", async (c) => {
+    const embeddings = memoryService.getEmbeddings();
+    if (embeddings.isReady) {
+      return c.json({ status: "already_warm" });
+    }
+    const start = Date.now();
+    await embeddings.warmup();
+    return c.json({ status: "warmed", elapsed: Date.now() - start });
+  });
+
   // Search endpoint
   app.post("/search", async (c) => {
     try {