fix(app): restaura suporte a tool calling sem SyntaxError

app.py

@@ -787,6 +787,9 @@ def chat_completions(req: ChatRequest):
                 "max_tokens": max_tokens,
                 "temperature": temperature if temperature else 0.0,
             }
+            if req.tools:
+                payload["tools"] = [t.model_dump(exclude_none=True) for t in req.tools]
+                payload["tool_choice"] = req.tool_choice or "auto"
 
             @async_retry_with_backoff(max_retries=3, initial_delay=1, backoff_factor=2)
             async def call_openrouter_api():
@@ -815,7 +818,9 @@ def chat_completions(req: ChatRequest):
             if "choices" not in data or not data["choices"]:
                 raise APIError(f"Resposta inválida da OpenRouter API: {data}")
 
-            text = data["choices"][0]["message"]["content"]
+            message_obj = data["choices"][0]["message"]
+            text = message_obj.get("content", "") or ""
+            tool_calls = message_obj.get("tool_calls")
             usage = data.get("usage", {})
             prompt_tokens = usage.get("prompt_tokens", 0)
             completion_tokens = usage.get("completion_tokens", 0)
@@ -836,6 +841,9 @@ def chat_completions(req: ChatRequest):
                 "max_tokens": max_tokens,
                 "temperature": temperature if temperature else 0.0,
             }
+            if req.tools:
+                payload["tools"] = [t.model_dump(exclude_none=True) for t in req.tools]
+                payload["tool_choice"] = req.tool_choice or "auto"
 
             payload_bytes = len(json.dumps(payload))
             logger.info(f"[GROQ DEBUG] Payload size: {payload_bytes} bytes, Messages: {len(chat)}")
@@ -867,7 +875,9 @@ def chat_completions(req: ChatRequest):
             if "choices" not in data or not data["choices"]:
                 raise APIError(f"Resposta inválida da Groq API: {data}")
 
-            text = data["choices"][0]["message"]["content"]
+            message_obj = data["choices"][0]["message"]
+            text = message_obj.get("content", "") or ""
+            tool_calls = message_obj.get("tool_calls")
             usage = data.get("usage", {})
             prompt_tokens = usage.get("prompt_tokens", 0)
             completion_tokens = usage.get("completion_tokens", 0)
@@ -888,6 +898,9 @@ def chat_completions(req: ChatRequest):
                 "max_tokens": max_tokens,
                 "temperature": temperature if temperature else 0.0,
             }
+            if req.tools:
+                payload["tools"] = [t.model_dump(exclude_none=True) for t in req.tools]
+                payload["tool_choice"] = req.tool_choice or "auto"
 
             @async_retry_with_backoff(max_retries=3, initial_delay=1, backoff_factor=2)
             async def call_mistral_api():
@@ -916,7 +929,9 @@ def chat_completions(req: ChatRequest):
             if "choices" not in data or not data["choices"]:
                 raise APIError(f"Resposta inválida da Mistral API: {data}")
 
-            text = data["choices"][0]["message"]["content"]
+            message_obj = data["choices"][0]["message"]
+            text = message_obj.get("content", "") or ""
+            tool_calls = message_obj.get("tool_calls")
             usage = data.get("usage", {})
             prompt_tokens = usage.get("prompt_tokens", 0)
             completion_tokens = usage.get("completion_tokens", 0)
@@ -935,6 +950,9 @@ def chat_completions(req: ChatRequest):
                 "temperature": temperature if temperature else 0.0,
             }
 
+            if req.tools:
+                gemini_payload["tools"] = convert_tools_to_gemini_format(req.tools)
+
             if model_id.startswith("gemini"):
                 gemini_payload["generationConfig"]["thinkingConfig"] = {"thinkingBudget": 0}
 
@@ -984,6 +1002,25 @@ def chat_completions(req: ChatRequest):
         elapsed = time.time() - t0
         logger.info(f"Resposta gerada em {elapsed:.1f}s ({backend}, web={req.web})")
 
+        message = {"role": "assistant", "content": text}
+        openai_tool_calls = None
+        if 'tool_calls' in locals() and tool_calls:
+            openai_tool_calls = tool_calls
+        elif 'function_calls' in locals() and function_calls:
+            openai_tool_calls = [
+                {
+                    "id": f"call_{uuid.uuid4().hex[:8]}",
+                    "type": "function",
+                    "function": {
+                        "name": fc["name"],
+                        "arguments": json.dumps(fc.get("args", {})),
+                    },
+                }
+                for fc in function_calls
+            ]
+        if openai_tool_calls:
+            message["tool_calls"] = openai_tool_calls
+
         return {
             "id": f"chatcmpl-{uuid.uuid4().hex[:8]}",
             "object": "chat.completion",
@@ -992,8 +1029,8 @@ def chat_completions(req: ChatRequest):
             "mode": req.mode,
             "choices": [{
                 "index": 0,
-                "message": {"role": "assistant", "content": text},
-                "finish_reason": "stop"
+                "message": message,
+                "finish_reason": "tool_calls" if openai_tool_calls else (finish_reason or "stop"),
             }],
             "usage": {
                 "prompt_tokens": prompt_tokens,
@@ -1310,6 +1347,9 @@ async def chat_completions_async(req: ChatRequest):
                 "max_tokens": max_tokens,
                 "temperature": temperature if temperature else 0.0,
             }
+            if req.tools:
+                payload["tools"] = [t.model_dump(exclude_none=True) for t in req.tools]
+                payload["tool_choice"] = req.tool_choice or "auto"
 
             async with httpx.AsyncClient() as client:
                 resp = await client.post(
@@ -1334,7 +1374,9 @@ async def chat_completions_async(req: ChatRequest):
             if "choices" not in data or not data["choices"]:
                 raise APIError(f"Resposta inválida da OpenRouter API: {data}")
 
-            text = data["choices"][0]["message"]["content"]
+            message_obj = data["choices"][0]["message"]
+            text = message_obj.get("content", "") or ""
+            tool_calls = message_obj.get("tool_calls")
             usage = data.get("usage", {})
             prompt_tokens = usage.get("prompt_tokens", 0)
             completion_tokens = usage.get("completion_tokens", 0)
@@ -1355,6 +1397,9 @@ async def chat_completions_async(req: ChatRequest):
                 "max_tokens": max_tokens,
                 "temperature": temperature if temperature else 0.0,
             }
+            if req.tools:
+                payload["tools"] = [t.model_dump(exclude_none=True) for t in req.tools]
+                payload["tool_choice"] = req.tool_choice or "auto"
 
             payload_bytes = len(json.dumps(payload))
             logger.info(f"[GROQ DEBUG] Payload size: {payload_bytes} bytes, Messages: {len(chat)}")
@@ -1382,7 +1427,9 @@ async def chat_completions_async(req: ChatRequest):
             if "choices" not in data or not data["choices"]:
                 raise APIError(f"Resposta inválida da Groq API: {data}")
 
-            text = data["choices"][0]["message"]["content"]
+            message_obj = data["choices"][0]["message"]
+            text = message_obj.get("content", "") or ""
+            tool_calls = message_obj.get("tool_calls")
             usage = data.get("usage", {})
             prompt_tokens = usage.get("prompt_tokens", 0)
             completion_tokens = usage.get("completion_tokens", 0)
@@ -1403,6 +1450,9 @@ async def chat_completions_async(req: ChatRequest):
                 "max_tokens": max_tokens,
                 "temperature": temperature if temperature else 0.0,
             }
+            if req.tools:
+                payload["tools"] = [t.model_dump(exclude_none=True) for t in req.tools]
+                payload["tool_choice"] = req.tool_choice or "auto"
 
             async with httpx.AsyncClient() as client:
                 resp = await client.post(
@@ -1446,6 +1496,9 @@ async def chat_completions_async(req: ChatRequest):
                 "temperature": temperature if temperature else 0.0,
             }
 
+            if req.tools:
+                gemini_payload["tools"] = convert_tools_to_gemini_format(req.tools)
+
             if model_id.startswith("gemini"):
                 gemini_payload["generationConfig"]["thinkingConfig"] = {"thinkingBudget": 0}
 
@@ -1476,7 +1529,10 @@ async def chat_completions_async(req: ChatRequest):
 
             logger.info(f"[GEMINI DEBUG] finishReason={data['candidates'][0].get('finishReason')}, content={data['candidates'][0].get('content', {})}")
 
-            text = data["candidates"][0]["content"]["parts"][0]["text"]
+            parts = data["candidates"][0]["content"]["parts"]
+            text_parts = [p.get("text", "") for p in parts if "text" in p]
+            text = "".join(text_parts)
+            function_calls = [p["functionCall"] for p in parts if "functionCall" in p]
             usage = data.get("usageMetadata", {})
             prompt_tokens = usage.get("promptTokenCount", 0)
             completion_tokens = usage.get("candidatesTokenCount", 0)
@@ -1491,6 +1547,25 @@ async def chat_completions_async(req: ChatRequest):
         elapsed = time.time() - t0
         logger.info(f"Resposta gerada em {elapsed:.1f}s ({backend}, web={req.web})")
 
+        message = {"role": "assistant", "content": text}
+        openai_tool_calls = None
+        if 'tool_calls' in locals() and tool_calls:
+            openai_tool_calls = tool_calls
+        elif 'function_calls' in locals() and function_calls:
+            openai_tool_calls = [
+                {
+                    "id": f"call_{uuid.uuid4().hex[:8]}",
+                    "type": "function",
+                    "function": {
+                        "name": fc["name"],
+                        "arguments": json.dumps(fc.get("args", {})),
+                    },
+                }
+                for fc in function_calls
+            ]
+        if openai_tool_calls:
+            message["tool_calls"] = openai_tool_calls
+
         return {
             "id": f"chatcmpl-{uuid.uuid4().hex[:8]}",
             "object": "chat.completion",
@@ -1499,8 +1574,8 @@ async def chat_completions_async(req: ChatRequest):
             "mode": req.mode,
             "choices": [{
                 "index": 0,
-                "message": {"role": "assistant", "content": text},
-                "finish_reason": "stop"
+                "message": message,
+                "finish_reason": "tool_calls" if openai_tool_calls else (finish_reason or "stop"),
             }],
             "usage": {
                 "prompt_tokens": prompt_tokens,