Update app.py

app.py

@@ -1,4 +1,4 @@
-import os, re, json
+import os, re, json, threading, traceback
 from functools import lru_cache
 
 import gradio as gr
@@ -302,7 +302,14 @@ def clarityops_reply(user_msg, history, tz, uploaded_files_paths):
 
         return history + [(user_msg, safe_out)]
     except Exception as e:
-        return history + [(user_msg, f"Error: {e}")]
+        # Surface the error in-chat so the websocket doesn’t die silently
+        err = f"Error: {e}"
+        try:
+            # Optionally log stack for ops
+            traceback.print_exc()
+        except Exception:
+            pass
+        return history + [(user_msg, err)]
 
 # ---------- Theme & CSS ----------
 theme = gr.themes.Soft(primary_hue="teal", neutral_hue="slate", radius_size=gr.themes.sizes.radius_lg)
@@ -340,6 +347,18 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
         js="() => Intl.DateTimeFormat().resolvedOptions().timeZone",
     )
 
+    # --- Background warmup so first message doesn't time out ---
+    def _warmup():
+        def _bg():
+            try:
+                # Preload local fallback model quietly (download may be large)
+                load_local_model()
+            except Exception:
+                pass
+        threading.Thread(target=_bg, daemon=True).start()
+        return ""
+    demo.load(_warmup, None, None)  # fire-and-forget
+
     gr.Markdown("# ClarityOps Augmented Decision AI")
 
     # Main chat (tuple-format history)
@@ -375,12 +394,21 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
 
     uploads.change(fn=_store_uploads, inputs=[uploads, state_uploaded], outputs=state_uploaded)
 
-    # Send / Enter handlers (set concurrency_limit here per Gradio deprecation notice)
+    # Send / Enter handlers (defensive wrapper)
     def _on_send(user_msg, history, tz, up_paths):
-        if not user_msg or not user_msg.strip():
-            return history, "", history
-        new_history = clarityops_reply(user_msg.strip(), history or [], tz, up_paths or [])
-        return new_history, "", new_history
+        try:
+            if not user_msg or not user_msg.strip():
+                return history, "", history
+            new_history = clarityops_reply(user_msg.strip(), history or [], tz, up_paths or [])
+            return new_history, "", new_history
+        except Exception as e:
+            # Return the error in the chat so UI doesn't show a websocket failure
+            err = f"Error: {e}"
+            try:
+                traceback.print_exc()
+            except Exception:
+                pass
+            return (history or []) + [(user_msg or "", err)], "", (history or []) + [(user_msg or "", err)]
 
     send.click(
         fn=_on_send,
@@ -402,6 +430,8 @@ with gr.Blocks(theme=theme, css=custom_css) as demo:
 
 if __name__ == "__main__":
     port = int(os.environ.get("PORT", "7860"))
+    # max_threads gives the worker pool; leave as-is unless you need more parallelism
     demo.launch(server_name="0.0.0.0", server_port=port, show_api=False, max_threads=8)
 
 
+