Update app.py

app.py

@@ -12,9 +12,8 @@ from mistralai import Mistral
 import google.generativeai as genai
 from huggingface_hub import snapshot_download 
 
-# --- SEGURANÇA: RATE LIMITER ANTI-SPAM ---
-# Permite 10 requisições por minuto por IP (ajustado para seus testes)
-MAX_REQUESTS_PER_MINUTE = 10
+# --- SEGURANÇA: RATE LIMITER ---
+MAX_REQUESTS_PER_MINUTE = 15
 BLOCK_TIME_SECONDS = 60
 ip_access_log = defaultdict(list)
 
@@ -23,11 +22,9 @@ def verify_rate_limit(request: gr.Request):
     client_ip = request.client.host
     current_time = time.time()
     ip_access_log[client_ip] = [t for t in ip_access_log[client_ip] if current_time - t < BLOCK_TIME_SECONDS]
-    
     if len(ip_access_log[client_ip]) >= MAX_REQUESTS_PER_MINUTE:
-        print(f"⛔ BLOQUEIO: IP {client_ip} barrado por spam.")
+        print(f"⛔ BLOQUEIO: IP {client_ip} barrado.")
         return False
-    
     ip_access_log[client_ip].append(current_time)
     return True
 
@@ -55,7 +52,7 @@ def download_local_model():
     try: snapshot_download(repo_id=LOCAL_MODEL_ID)
     except Exception as e: print(f"⚠️ Aviso: {e}")
 
-# --- BACKENDS DE EXECUÇÃO ---
+# --- BACKENDS ---
 
 @spaces.GPU(duration=120) 
 def run_local_h200(messages):
@@ -75,11 +72,10 @@ def run_groq(messages, model_id):
     for m in messages:
         if isinstance(m['content'], list): return "⚠️ Groq não lê imagens. Use Gemini/Pixtral."
     if not groq_client: return "❌ Erro: API Key Groq ausente."
-    
     clean_msgs = [{"role": m['role'], "content": m['content']} for m in messages]
     try:
         completion = groq_client.chat.completions.create(
-            model=model_id, messages=clean_msgs, temperature=0.7, max_tokens=8192 # Aumentei tokens para os GPT OSS
+            model=model_id, messages=clean_msgs, temperature=0.7, max_tokens=8192
         )
         return completion.choices[0].message.content
     except Exception as e: return f"❌ Groq Error: {e}"
@@ -123,11 +119,11 @@ def run_gemini(messages, model_id):
                         if os.path.exists(path): parts.append(Image.open(path))
             if parts: chat_history.append({"role": role, "parts": parts})
         
-        last_msg_content = messages[-1]['content']
+        last_msg = messages[-1]['content']
         current_parts = []
-        if isinstance(last_msg_content, str): current_parts.append(last_msg_content)
-        elif isinstance(last_msg_content, list):
-            for item in last_msg_content:
+        if isinstance(last_msg, str): current_parts.append(last_msg)
+        elif isinstance(last_msg, list):
+            for item in last_msg:
                 if item.get('type') == 'text': current_parts.append(item['text'])
                 elif item.get('type') == 'image_url':
                     path = item['image_url']['url']
@@ -138,22 +134,22 @@ def run_gemini(messages, model_id):
         return response.text
     except Exception as e: return f"❌ Gemini Error ({model_id}): {e}"
 
-# --- ROTEADOR INTELIGENTE ---
+# --- ROTEADOR ---
 def router(message, history, model_selector, request: gr.Request):
     if not verify_rate_limit(request):
         return f"⛔ LIMITADO: Aguarde para enviar mais mensagens."
 
-    # Normalização de Histórico (Blindagem)
     formatted_history = []
-    for turn in history:
-        if isinstance(turn, dict): formatted_history.append(turn)
-        elif isinstance(turn, (list, tuple)) and len(turn) >= 2:
-            u = turn[0]['text'] if isinstance(turn[0], dict) and 'text' in turn[0] else str(turn[0])
-            b = str(turn[1]) if turn[1] else ""
-            formatted_history.append({"role": "user", "content": u})
-            if b: formatted_history.append({"role": "assistant", "content": b})
+    # Proteção contra history=None ou formatos estranhos
+    if history:
+        for turn in history:
+            if isinstance(turn, dict): formatted_history.append(turn)
+            elif isinstance(turn, (list, tuple)) and len(turn) >= 2:
+                u = turn[0]['text'] if isinstance(turn[0], dict) and 'text' in turn[0] else str(turn[0])
+                b = str(turn[1]) if turn[1] else ""
+                formatted_history.append({"role": "user", "content": u})
+                if b: formatted_history.append({"role": "assistant", "content": b})
     
-    # Payload Atual
     current_content = []
     text = message.get("text", "")
     files = message.get("files", [])
@@ -163,8 +159,7 @@ def router(message, history, model_selector, request: gr.Request):
     if not files: formatted_history.append({"role": "user", "content": text})
     else: formatted_history.append({"role": "user", "content": current_content})
 
-    # --- SELEÇÃO DE MODELOS (ATUALIZADA) ---
-    
+    # SELEÇÃO (IDs CORRIGIDOS DO SEU PRINT)
     if "Gemini" in model_selector:
         tid = "gemini-1.5-flash"
         if "3.0" in model_selector: tid = "gemini-3.0-pro-preview" 
@@ -176,13 +171,12 @@ def router(message, history, model_selector, request: gr.Request):
     elif "Mistral" in model_selector:
         tid = "mistral-large-latest"
         if "Pixtral" in model_selector: tid = "pixtral-large-latest"
-        elif "2509" in model_selector: tid = "magistral-medium-2509" # <--- Seu pedido!
+        elif "2509" in model_selector: tid = "magistral-medium-2509" # Pedido aceito
         elif "2512" in model_selector: tid = "mistral-large-2512"
         elif "Codestral" in model_selector: tid = "codestral-2508"
         return run_mistral(formatted_history, tid)
 
     elif "Groq" in model_selector:
-        # Mapeamento do seu Print
         if "120B" in model_selector: tid = "openai/gpt-oss-120b"
         elif "20B" in model_selector: tid = "openai/gpt-oss-20b"
         else: tid = "llama-3.3-70b-versatile"
@@ -195,35 +189,48 @@ def router(message, history, model_selector, request: gr.Request):
 
 # --- INTERFACE ---
 with gr.Blocks() as demo:
-    gr.Markdown("# 🔀 APIDOST v5 (Full Arsenal & Secured)")
+    gr.Markdown("# 🔀 APIDOST v6 (Endpoint Fixed)")
     
+    # Lista de Modelos Atualizada
+    models_list = [
+        "✨ Google: Gemini 3.0 Pro (Experimental)",
+        "✨ Google: Gemini 2.5 Pro",
+        "✨ Google: Gemini 2.5 Flash",
+        "✨ Google: Gemini 2.0 Flash",
+        "☁️ Groq: GPT OSS 120B (OpenAI) 🆕",
+        "☁️ Groq: GPT OSS 20B (OpenAI) 🆕",
+        "☁️ Groq: Llama 3.3 70B",
+        "🇫🇷 Mistral: Magistral Medium 2509 🆕",
+        "🇫🇷 Mistral: Pixtral Large (Vision) 🖼️",
+        "🇫🇷 Mistral: Large 2512 (Dez/25)",
+        "🇫🇷 Mistral: Codestral 2508",
+        "🔥 Local H200: Qwen 2.5 Coder 32B"
+    ]
+
     with gr.Row():
-        model_dropdown = gr.Dropdown(
-            choices=[
-                "✨ Google: Gemini 3.0 Pro (Experimental)",
-                "✨ Google: Gemini 2.5 Pro",
-                "✨ Google: Gemini 2.5 Flash",
-                "✨ Google: Gemini 2.0 Flash",
-                "☁️ Groq: GPT OSS 120B (OpenAI) 🆕",
-                "☁️ Groq: GPT OSS 20B (OpenAI) 🆕",
-                "☁️ Groq: Llama 3.3 70B",
-                "🇫🇷 Mistral: Magistral Medium 2509 🆕",
-                "🇫🇷 Mistral: Pixtral Large (Vision) 🖼️",
-                "🇫🇷 Mistral: Large 2512 (Dez/25)",
-                "🇫🇷 Mistral: Codestral 2508",
-                "🔥 Local H200: Qwen 2.5 Coder 32B"
-            ],
-            value="🔥 Local H200: Qwen 2.5 Coder 32B",
-            label="Cérebro Escolhido",
-            interactive=True
-        )
+        model_dropdown = gr.Dropdown(choices=models_list, value=models_list[-1], label="Cérebro", interactive=True)
 
+    # 1. Interface de Chat VISUAL (para você testar no HuggingFace)
     chat = gr.ChatInterface(
         fn=router,
         additional_inputs=[model_dropdown],
         multimodal=True, 
     )
 
+    # 2. PONTE DE API INVISÍVEL (A SOLUÇÃO DO SEU PROBLEMA)
+    # Isso cria explicitamente o endpoint "/chat" que o seu JavaScript está procurando.
+    # Ele aceita 'message' (multimodal), 'history' (estado) e 'model_selector' (dropdown).
+    api_bridge = gr.Interface(
+        fn=router,
+        inputs=[
+            gr.MultimodalTextbox(label="message"), # O JS manda {text:..., files:...}
+            gr.State(value=[], label="history"),   # O JS pode mandar lista vazia []
+            gr.Dropdown(choices=models_list, label="model_selector") # O JS manda a string do modelo
+        ],
+        outputs=[gr.Textbox(label="response")],
+        api_name="chat" # <--- AQUI! Isso garante que activeClient.predict("/chat") funcione.
+    )
+
 if __name__ == "__main__":
     download_local_model()
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+    demo.queue(api_open=True).launch(server_name="0.0.0.0", server_port=7860)