Update app.py

app.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import os, json, time, uuid, asyncio, logging
 from typing import Any, AsyncGenerator
+from contextlib import asynccontextmanager
 
 from dotenv import load_dotenv
 from fastapi import FastAPI, HTTPException, Request, Depends
@@ -71,19 +72,37 @@ async def verify_key(request: Request) -> None:
     if not auth.startswith("Bearer ") or auth[7:] != API_KEY:
         raise HTTPException(status_code=401, detail="Invalid or missing API key")
 
+# ---------------------------------------------------------------------------
+# Lifespan context manager (modern FastAPI pattern)
+# ---------------------------------------------------------------------------
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    # Startup
+    log.info("Starting up - connecting to Gradio client...")
+    await get_client()
+    log.info("Startup complete.")
+    yield
+    # Shutdown (if needed)
+    log.info("Shutting down.")
+
 # ---------------------------------------------------------------------------
 # App
 # ---------------------------------------------------------------------------
 
-app = FastAPI(title="Falcon H1R API", version="3.0.0")
-app.add_middleware(
-    CORSMiddleware, allow_origins=["*"],
-    allow_credentials=True, allow_methods=["*"], allow_headers=["*"],
+app = FastAPI(
+    title="Falcon H1R API",
+    version="3.1.0",
+    lifespan=lifespan,
 )
 
-@app.on_event("startup")
-async def startup():
-    await get_client()
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
 
 # ---------------------------------------------------------------------------
 # Business logic - EXACTLY like the HTML chatbot
@@ -231,6 +250,18 @@ async def _stream_sse(text: str, req: ChatCompletionRequest) -> AsyncGenerator[s
 # Routes
 # ---------------------------------------------------------------------------
 
+@app.get("/")
+async def root():
+    return {
+        "service": "Falcon H1R OpenAI-compatible API",
+        "version": "3.1.0",
+        "endpoints": {
+            "health": "/health",
+            "models": "/v1/models",
+            "chat": "/v1/chat/completions",
+        },
+    }
+
 @app.get("/health")
 async def health():
     return {"status": "ok", "model": MODEL_ID, "space": HF_SPACE_URL}