diff --git a/api/server.py b/api/server.py
index 6fcd1b5..85f99ea 100755
--- a/api/server.py
+++ b/api/server.py
@@ -65,7 +65,7 @@
 # When set, all requests must include: Authorization: Bearer <key>
 # Health and root endpoints are exempt.
 GATHM_API_KEY = os.environ.get("GATHM_API_KEY", "")
-PUBLIC_PATHS = {"", "/", "/api", "/api/v1", "/api/v1/health"}
+PUBLIC_PATHS = {"", "/", "/api", "/api/v1", "/api/v1/health", "/api/v1/ping"}
 # GUI static files are also public (any path not starting with /api/)
 
 import hashlib
@@ -288,7 +288,13 @@ def do_GET(self):
             else:
                 self._send_json({"error": f"Tool '{tool_name}' not found"}, 404)
 
-        # GET /api/v1/health
+        # GET /api/v1/ping  — lightweight liveness probe used by the GUI.
+        # Unlike /health it does NOT shell out to the orchestrator, so it
+        # returns instantly and never makes the UI look "offline".
+        elif path == "/api/v1/ping":
+            self._send_json({"status": "ok", "service": "gathm-api"})
+
+        # GET /api/v1/health  (full system health — checks every tool, slow)
         elif path == "/api/v1/health":
             result = run_agent_command("health", "all")
             self._send_json(result)
@@ -434,7 +440,10 @@ def main():
         else:
             i += 1
 
-    server = http.server.HTTPServer((host, port), GathmAPIHandler)
+    # ThreadingHTTPServer so a slow request (e.g. the full /health sweep
+    # across all tools) never blocks the GUI from loading or other
+    # requests from being served concurrently.
+    server = http.server.ThreadingHTTPServer((host, port), GathmAPIHandler)
     print(f"""
 ╔══════════════════════════════════════════════════╗
 ║           Gathm Enterprise API Server            ║
diff --git a/gui/app.js b/gui/app.js
index 6dd8bda..29a8454 100644
--- a/gui/app.js
+++ b/gui/app.js
@@ -24,8 +24,9 @@ let isOnline = false;
 
 async function checkConnectivity() {
     try {
-        const res = await fetch(`${API_BASE}/api/v1/health`, {
-            signal: AbortSignal.timeout(5000),
+        // /ping is instant — /health sweeps every tool and would time out.
+        const res = await fetch(`${API_BASE}/api/v1/ping`, {
+            signal: AbortSignal.timeout(4000),
         });
         isOnline = res.ok;
     } catch {
@@ -96,6 +97,34 @@ function hideTyping() {
     typingEl = null;
 }
 
+// ── Render agent reply ─────────────────────────────────────────
+// The /agent/ask endpoint is a tool ROUTER, not a chat LLM. It returns
+// structured JSON; translate it into something human-readable instead of
+// dumping raw JSON into the chat.
+function formatAgentReply(data) {
+    // A tool was matched to the query
+    if (data.matched_tool && data.matched_tool !== 'null') {
+        const desc   = data.description ? `\n\n${data.description}` : '';
+        const action = data.action ? `\n\n→ ${data.action}` : `\n\n→ Run: gathm run ${data.matched_tool}`;
+        return `I can help with that using the “${data.matched_tool}” tool.${desc}${action}`;
+    }
+
+    // No tool matched — give a friendly, useful nudge instead of an error blob
+    if (data.error && /no matching tool/i.test(data.error)) {
+        return "I'm a tool-running assistant, so I work best with task requests. " +
+               "Try things like:\n" +
+               "  • weather in Tokyo\n" +
+               "  • dns records for github.com\n" +
+               "  • ip info 8.8.8.8\n" +
+               "  • define serendipity\n" +
+               "  • crypto price bitcoin";
+    }
+
+    // Any other shape: prefer real output, fall back to the error text
+    return data.raw_output || data.output || data.result || data.error
+        || JSON.stringify(data, null, 2);
+}
+
 // ── Send via API ───────────────────────────────────────────────
 let isSending = false;
 
@@ -125,9 +154,7 @@ async function sendMessage() {
         }
 
         const data = await res.json();
-        const reply = data.raw_output || data.output || data.result
-            || JSON.stringify(data, null, 2);
-        addMessage(reply, 'bot');
+        addMessage(formatAgentReply(data), 'bot');
 
     } catch (err) {
         hideTyping();
diff --git a/gui/styles.css b/gui/styles.css
index 1a312c8..8a01d4b 100644
--- a/gui/styles.css
+++ b/gui/styles.css
@@ -381,6 +381,8 @@ body {
 .message p {
     font-size: 14px;
     line-height: 1.55;
+    white-space: pre-wrap;   /* preserve newlines / bullet lists in replies */
+    word-break: break-word;
 }
 
 .bot-text {
diff --git a/install b/install
index 4ba772e..89d0076 100755
--- a/install
+++ b/install
@@ -1529,7 +1529,9 @@ main() {
     install_ollama "$platform"
     install_llmfit_and_select_model
 
-    install_engineer_deps
+    # TODO(engineer): temporarily disabled — re-enable when resuming the
+    # engineer agent. Uncomment the line below to restore engineer setup.
+    # install_engineer_deps
     install_pilot_deps
     install_playwright_browser
     setup_shortcuts "$platform"