Update server/chatStream.js

server/chatStream.js

@@ -1,6 +1,40 @@
 import OpenAI from "openai";
+import { Client } from "@gradio/client";
 import { LIGHTNING_BASE } from "./config.js";
 
+// ── Gradio search client singleton ─────────────────────────────────────────
+// Connect once at module load and reuse across all searches.
+// If the connection drops, gradioClient is reset to null so the next call
+// reconnects automatically rather than retrying a dead client forever.
+let gradioClient = null;
+let gradioConnecting = null; // in-flight connect promise, prevents thundering herd
+
+async function getGradioClient() {
+  if (gradioClient) return gradioClient;
+  if (gradioConnecting) return gradioConnecting;
+  gradioConnecting = Client.connect("incognitolm/Web-Search")
+    .then(c  => { gradioClient = c; gradioConnecting = null; return c; })
+    .catch(e => { gradioConnecting = null; throw e; });
+  return gradioConnecting;
+}
+
+async function gradioSearch(query) {
+  // Try once; if it fails reset the singleton so the next call gets a fresh connection.
+  try {
+    const client = await getGradioClient();
+    const r = await client.predict("/perform_search", { query });
+    // r.data is an array; the search results are in the first element.
+    // Accept a string, an array, or an object — normalise all to a string.
+    const raw = Array.isArray(r.data) ? r.data[0] : r.data;
+    if (!raw) throw new Error("Empty response from search endpoint");
+    return typeof raw === "string" ? raw : JSON.stringify(raw);
+  } catch (err) {
+    // Invalidate the singleton so the next search attempt reconnects.
+    gradioClient = null;
+    throw err;
+  }
+}
+
 const SYSTEM_PROMPT =
   "CRITICAL RULE: Every response MUST use HTML <span data-color=\"{COLOR NAME}\"> tags to color main points and headings. " +
   "COLORS MUST HAVE MEANING AND CONSISTENCY ACROSS THE ENTIRE CONVERSATION. " +
@@ -259,10 +293,7 @@ async function processToolCalls(ws, toolCalls, tools, accessToken, clientId, abo
 
     try {
       if (call.function.name === "ollama_search") {
-        const { Client } = await import("@gradio/client");
-        const client = await Client.connect("incognitolm/Web-Search");
-        const r = await client.predict("/perform_search", { query: args.query });
-        result = JSON.stringify(r.data);
+        result = await gradioSearch(args.query);
       }
 
       else if (call.function.name === "read_web_page") {