feat: Implement OAuth-based auth flow with gr.LoginButton and request context

app.py

@@ -112,7 +112,7 @@ def auto_select_perspectives(query: str, n: int = 4) -> List[str]:
     return selected[:n]
 
 
-def call_perspective(perspective_name: str, query: str) -> str:
+def call_perspective(perspective_name: str, query: str, request: gr.Request = None) -> str:
     """Generate response from a single perspective using HF Inference API."""
     p = get_perspective(perspective_name)
     if not p:
@@ -121,12 +121,20 @@ def call_perspective(perspective_name: str, query: str) -> str:
     if not HAS_LLM:
         return f"[{perspective_name.upper()}] Demo response (LLM unavailable)"
 
+    # Extract auth token from request headers (set by Gradio OAuth)
+    auth_token = None
+    if request and hasattr(request, 'headers') and 'authorization' in request.headers:
+        auth_token = request.headers['authorization'].replace('Bearer ', '').strip()
+
+    # Create client with user's auth token if available
+    inference_client = InferenceClient("meta-llama/Llama-3.1-8B-Instruct", token=auth_token) if auth_token else client
+
     try:
         messages = [
             {"role": "system", "content": p.system_prompt},
             {"role": "user", "content": query}
         ]
-        response = client.chat_completion(
+        response = inference_client.chat_completion(
             messages,
             max_tokens=256,
             temperature=0.7,
@@ -141,9 +149,17 @@ def call_perspective(perspective_name: str, query: str) -> str:
             return f"[{perspective_name}] Error: {str(e)[:80]}"
 
 
-def generate_synthesis(perspectives_responses: Dict[str, str], query: str) -> str:
+def generate_synthesis(perspectives_responses: Dict[str, str], query: str, request: gr.Request = None) -> str:
     """Generate synthesis from all perspective responses."""
 
+    # Extract auth token from request headers
+    auth_token = None
+    if request and hasattr(request, 'headers') and 'authorization' in request.headers:
+        auth_token = request.headers['authorization'].replace('Bearer ', '').strip()
+
+    # Create client with user's auth token if available
+    inference_client = InferenceClient("meta-llama/Llama-3.1-8B-Instruct", token=auth_token) if auth_token else client
+
     # Try LLM synthesis first if available
     if HAS_LLM:
         perspective_text = "\n\n".join(
@@ -162,7 +178,7 @@ Brief unified insight:"""
                 {"role": "system", "content": "Synthesize multi-perspective responses concisely."},
                 {"role": "user", "content": synthesis_prompt}
             ]
-            response = client.chat_completion(
+            response = inference_client.chat_completion(
                 messages,
                 max_tokens=256,
                 temperature=0.7,
@@ -680,7 +696,8 @@ def process_message(
     chat_history: List,
     state: Dict,
     perspective_mode: str,
-    custom_perspectives: List[str]
+    custom_perspectives: List[str],
+    request: gr.Request = None
 ) -> Tuple:
     """Main conversation handler with all visualizations."""
 
@@ -708,7 +725,7 @@ def process_message(
     # ===== STEP 4-6: Generate & Evaluate =====
     perspectives_responses = {}
     for perspective_name in selected_perspectives:
-        response = call_perspective(perspective_name, user_msg)
+        response = call_perspective(perspective_name, user_msg, request)
         perspectives_responses[perspective_name] = response
 
     aegis = state.get("aegis") or AEGIS()
@@ -728,7 +745,7 @@ def process_message(
     mean_tension = np.mean(list(tensions.values())) if tensions else 0.3
 
     # ===== STEP 7: Synthesis =====
-    synthesis = generate_synthesis(perspectives_responses, user_msg)
+    synthesis = generate_synthesis(perspectives_responses, user_msg, request)
     chat_history.append({"role": "assistant", "content": synthesis})
 
     # ===== STEP 8-9: Resonance & Memory =====
@@ -1129,40 +1146,20 @@ def create_interface():
         </div>
         """)
 
-        # Token Input Panel
+        # OAuth Login with HuggingFace
         with gr.Group():
-            gr.Markdown("### 🔐 HuggingFace Token (Optional)")
-            gr.Markdown("_Paste your HF API token to enable full LLM synthesis. Get one at [hf.co/settings/tokens](https://huggingface.co/settings/tokens)_")
+            gr.Markdown("### 🔐 Sign in with HuggingFace")
+            gr.Markdown("_Login with your HF account to use your inference quota for full LLM synthesis._")
             with gr.Row():
-                token_input = gr.Textbox(
-                    label="HF API Token",
-                    type="password",
-                    placeholder="hf_...",
-                    scale=4,
+                login_button = gr.LoginButton(
+                    scale=2,
+                    size="lg",
                 )
-                apply_token_btn = gr.Button("Apply Token", variant="primary", scale=1)
-
-            token_status = gr.Markdown(
-                "✓ Ready (demo mode — perspectives use built-in synthesis)",
+            auth_note = gr.Markdown(
+                "✓ Ready for analysis — LLM features unlocked when logged in",
                 visible=True
             )
 
-            def apply_token(token_str):
-                """Apply HF token to enable inference."""
-                if not token_str or token_str.strip() == "":
-                    return "❌ No token provided"
-                try:
-                    login(token=token_str.strip())
-                    return "✅ Token applied! LLM synthesis enabled."
-                except Exception as e:
-                    return f"⚠️ Token invalid: {str(e)[:60]}"
-
-            apply_token_btn.click(
-                apply_token,
-                token_input,
-                token_status,
-            )
-
         with gr.Tabs():
             # =================== CHAT TAB ===================
             with gr.Tab("Explore", id="chat"):
@@ -1301,14 +1298,14 @@ Created by **Jonathan Harrison** to explore recursive reasoning, multi-perspecti
                 """)
 
         # Event handling
-        def on_submit(msg, history, st, mode, custom):
-            result = process_message(msg, history, st, mode, custom)
+        def on_submit(msg, history, st, mode, custom, request: gr.Request):
+            result = process_message(msg, history, st, mode, custom, request)
             return result
 
         # Wire submit events
         send_btn.click(
             on_submit,
-            [msg_input, chatbot, state, perspective_mode, custom_perspectives],
+            [msg_input, chatbot, state, perspective_mode, custom_perspectives, gr.Request()],
             [chatbot, state, aegis_display, coherence_display, nexus_display,
              psi_display, memory_display, coverage_display, memory_browser,
              spiderweb_plot, coherence_plot, tension_plot, aegis_plot, memory_plot, nexus_plot],
@@ -1317,7 +1314,7 @@ Created by **Jonathan Harrison** to explore recursive reasoning, multi-perspecti
 
         msg_input.submit(
             on_submit,
-            [msg_input, chatbot, state, perspective_mode, custom_perspectives],
+            [msg_input, chatbot, state, perspective_mode, custom_perspectives, gr.Request()],
             [chatbot, state, aegis_display, coherence_display, nexus_display,
              psi_display, memory_display, coverage_display, memory_browser,
              spiderweb_plot, coherence_plot, tension_plot, aegis_plot, memory_plot, nexus_plot],