Optimize connection lifecycle to load incrementally, and fix NVIDIA model in research route

index.js

@@ -56,46 +56,51 @@ async function loadAndConnectServers() {
     }
   }
 
-  // Close existing active clients
+  // Close and remove any clients that are no longer in the registry
   for (const [name, entry] of activeClients.entries()) {
-    try {
-      await entry.client.close();
-    } catch (e) {}
+    if (!registry[name]) {
+      try {
+        await entry.client.close();
+      } catch (e) {}
+      activeClients.delete(name);
+      console.log(`[Registry] Closed and removed offline sub-server: ${name}`);
+    }
   }
-  activeClients.clear();
 
-  // Connect to registered servers in parallel
-  const connectionPromises = Object.entries(registry).map(async ([name, serverConfig]) => {
-    try {
-      console.log(`[Registry] Connecting to sub-server: ${name}...`);
-      let transport;
-      if (serverConfig.type === "sse") {
-        transport = new SSEClientTransport(new URL(serverConfig.data.url));
-      } else if (serverConfig.type === "stdio") {
-        transport = new StdioClientTransport({
-          command: serverConfig.data.command,
-          args: serverConfig.data.args || [],
-          env: { ...process.env, ...serverConfig.data.env }
-        });
-      } else {
-        return;
+  // Connect to registered servers in parallel (only those not already connected!)
+  const connectionPromises = Object.entries(registry)
+    .filter(([name]) => !activeClients.has(name))
+    .map(async ([name, serverConfig]) => {
+      try {
+        console.log(`[Registry] Connecting to sub-server: ${name}...`);
+        let transport;
+        if (serverConfig.type === "sse") {
+          transport = new SSEClientTransport(new URL(serverConfig.data.url));
+        } else if (serverConfig.type === "stdio") {
+          transport = new StdioClientTransport({
+            command: serverConfig.data.command,
+            args: serverConfig.data.args || [],
+            env: { ...process.env, ...serverConfig.data.env }
+          });
+        } else {
+          return;
+        }
+
+        const client = new Client(
+          { name: `proxy-${name}`, version: "1.0.0" },
+          { capabilities: {} }
+        );
+
+        await client.connect(transport);
+        const toolsResult = await client.listTools();
+        const tools = toolsResult.tools || [];
+        
+        activeClients.set(name, { client, tools });
+        console.log(`[Registry] Successfully registered '${name}' with ${tools.length} tools.`);
+      } catch (e) {
+        console.error(`[Registry Error] Failed to connect to '${name}':`, e.message);
       }
-
-      const client = new Client(
-        { name: `proxy-${name}`, version: "1.0.0" },
-        { capabilities: {} }
-      );
-
-      await client.connect(transport);
-      const toolsResult = await client.listTools();
-      const tools = toolsResult.tools || [];
-      
-      activeClients.set(name, { client, tools });
-      console.log(`[Registry] Successfully registered '${name}' with ${tools.length} tools.`);
-    } catch (e) {
-      console.error(`[Registry Error] Failed to connect to '${name}':`, e.message);
-    }
-  });
+    });
 
   await Promise.allSettled(connectionPromises);
   console.log(`[Registry] Parallel load complete. Active servers: ${activeClients.size}`);
@@ -293,7 +298,7 @@ app.post("/api/research", express.json(), async (req, res) => {
           "Content-Type": "application/json"
         },
         body: JSON.stringify({
-          model: "stepfun-ai/step-3.7-flash",
+          model: "nvidia/llama-3.1-nemotron-70b-instruct",
           messages: [
             { role: "system", content: "You are a research summarizer. Compress the raw text into a strict 500-word markdown brief detailing novel formula structures, chemical symbols, or algorithms." },
             { role: "user", content: rawText.substring(0, 30000) }