Update hf_demo.py

hf_demo.py

@@ -1171,24 +1171,30 @@ if __name__ == "__main__":
     # Mount FastAPI on the Gradio app if available
     if FASTAPI_AVAILABLE:
         print("🔄 Mounting FastAPI endpoints on Gradio app...")
+        # This returns a FastAPI app that includes both FastAPI routes and Gradio
         app = mount_gradio_app(api_app, demo, path="/")
         print("✅ FastAPI endpoints mounted successfully")
         print("   Access API at: https://petter2025-agentic-reliability-framework.hf.space/api/v1/...")
+        
+        # For FastAPI app, we need to use uvicorn to run it
+        import uvicorn
+        print("🚀 Starting combined FastAPI+Gradio server on port 7860...")
+        uvicorn.run(app, host="0.0.0.0", port=7860, log_level="info")
     else:
-        app = demo
-    
-    # Launch the combined app
-    app.launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        share=False,
-        debug=False,
-        prevent_thread_lock=True,
-        show_error=True,
-        quiet=False
-    )
+        # Fallback to Gradio-only mode
+        print("⚠️ FastAPI not available, running Gradio only")
+        demo.launch(
+            server_name="0.0.0.0",
+            server_port=7860,
+            share=False,
+            debug=False,
+            prevent_thread_lock=True,
+            show_error=True,
+            quiet=False
+        )
     
-    # Keep the app alive
-    import time
-    while True:
-        time.sleep(1)
+    # Keep the app alive (only needed for Gradio fallback)
+    if not FASTAPI_AVAILABLE:
+        import time
+        while True:
+            time.sleep(1)