Update app.py

app.py

@@ -6,6 +6,7 @@ from huggingface_hub import InferenceClient
 # Helper to read a secret (fallback is useful when you run locally)
 # ----------------------------------------------------------------------
 def _secret(key: str, fallback: str = "") -> str:
+    """Return the value of a secret or the supplied fallback."""
     return os.getenv(key, fallback)
 
 
@@ -18,26 +19,32 @@ def respond(
     max_tokens: int,
     temperature: float,
     top_p: float,
-    hf_token: gr.OAuthToken,
 ):
     """
     Generate a response using the HuggingFace Inference API.
 
-    The system prompt is taken from the secret **prec_chat**.
-    Users cannot edit it from the UI.
+    * System prompt = secret `prec_chat`
+    * HF inference token = secret `HF_TOKEN`
     """
     # 1️⃣  Load the system prompt (fallback = generic assistant)
     system_message = _secret("prec_chat", "You are a helpful assistant.")
 
-    # 2️⃣  Initialise the HF inference client
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
+    # 2️⃣  Load the HF inference token
+    hf_token = _secret("HF_TOKEN")
+    if not hf_token:
+        raise RuntimeError(
+            "HF_TOKEN not found in secrets. Add it to secrets.toml (or via the Space UI)."
+        )
+
+    # 3️⃣  Initialise the HF inference client
+    client = InferenceClient(token=hf_token, model="openai/gpt-oss-20b")
 
-    # 3️⃣  Build the message list for the chat‑completion endpoint
+    # 4️⃣  Build the message list for the chat‑completion endpoint
     messages = [{"role": "system", "content": system_message}]
-    messages.extend(history)                     # previous turns
-    messages.append({"role": "user", "content": message})  # current query
+    messages.extend(history)                     # previous conversation turns
+    messages.append({"role": "user", "content": message})  # current user query
 
-    # 4️⃣  Stream the response back to the UI
+    # 5️⃣  Stream the response back to the UI
     response = ""
     for chunk in client.chat_completion(
         messages,
@@ -85,7 +92,9 @@ with gr.Blocks() as demo:
 # Launch – protect the UI with the credentials from secrets.
 # ----------------------------------------------------------------------
 if __name__ == "__main__":
-    # Pull the allowed credentials from secrets (fail fast if missing)
+    # ------------------------------------------------------------------
+    # 1️⃣  Pull the allowed credentials from secrets (fail fast if missing)
+    # ------------------------------------------------------------------
     allowed_user = _secret("CHAT_USER")
     allowed_pass = _secret("CHAT_PASS")
 
@@ -95,12 +104,15 @@ if __name__ == "__main__":
             "Add CHAT_USER and CHAT_PASS to secrets.toml (or via the HF Spaces UI)."
         )
 
+    # ------------------------------------------------------------------
+    # 2️⃣  Launch
+    # ------------------------------------------------------------------
     demo.launch(
         auth=(allowed_user, allowed_pass),   # <-- Gradio's built‑in basic auth
-        # Turn off server‑side rendering to avoid the i18n locale error
-        ssr_mode=False,
-        # In a remote environment (HF Spaces, Docker, cloud VM) you need a shareable link:
-        share=True,                         # <-- remove if you run locally and can reach http://127.0.0.1:7860
-        # Optional – listen on all interfaces (useful inside containers)
-        server_name="0.0.0.0",
+        ssr_mode=False,                     # <-- avoids the i18n locale error
+        # In a Space we **must not** set share=True (Spaces already give a public URL)
+        # If you run locally and want a shareable link, add share=True here.
+        server_name="0.0.0.0",              # listen on all interfaces (needed in containers)
+        # Optional: give the app a nice title
+        # title="Secure Chatbot",
     )