Deploy SaulGPT backend with full ChromaDB, Supabase, and Gradio API server

app.py

@@ -61,26 +61,32 @@ demo = gr.Interface(
     description="Backend engine for SaulGPT. Integrated with Supabase PostgreSQL, ChromaDB, and ZeroGPU acceleration. Multi-device web client available at https://legalgpt-india.web.app"
 )
 
-# 3. Attach all FastAPI backend routes to Gradio's internal FastAPI app
-demo.app.include_router(fastapi_app.router)
-
-# 4. Attach CORS middleware so the Firebase frontend can communicate freely
-demo.app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
-)
-
-# 5. Attach BYOK middleware
-demo.app.middleware("http")(byok_middleware)
-
-# 6. Initialize database during server launch
+# Initialize Supabase database
 try:
     init_db()
 except Exception as _e:
     print(f"init_db check: {_e}")
 
 if __name__ == "__main__":
-    demo.launch()
+    # Launch Gradio server in non-blocking mode to attach FastAPI routes
+    server_app, local_url, share_url = demo.launch(prevent_thread_lock=True)
+    
+    # Mount all FastAPI routes onto the live Gradio server
+    server_app.include_router(fastapi_app.router)
+    
+    # Add CORS middleware for client web apps
+    server_app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+    
+    # Add BYOK header middleware
+    server_app.middleware("http")(byok_middleware)
+    
+    print(f"SaulGPT API & Gradio UI live on {local_url} with full routes attached.")
+    
+    # Block main thread to keep server alive
+    demo.block_thread()