| | """ |
| | Agente de processamento de contexto inicial para sugestão de queries SQL |
| | """ |
| | import logging |
| | import asyncio |
| | from typing import Optional, Dict, Any |
| | from langchain_openai import ChatOpenAI |
| | from langchain_anthropic import ChatAnthropic |
| | from langchain_community.llms import HuggingFaceEndpoint |
| | from langchain.schema import HumanMessage |
| |
|
| | from utils.config import ( |
| | TEMPERATURE, |
| | AVAILABLE_MODELS, |
| | OPENAI_MODELS, |
| | ANTHROPIC_MODELS, |
| | REFINEMENT_MODELS |
| | ) |
| |
|
| |
|
| | class ProcessingAgentManager: |
| | """ |
| | Gerenciador do agente de processamento de contexto inicial |
| | """ |
| | |
| | def __init__(self, model_name: str = "gpt-4o-mini"): |
| | self.model_name = model_name |
| | self.llm = None |
| | self._initialize_llm() |
| | |
| | def _initialize_llm(self): |
| | """Inicializa o modelo LLM baseado no nome fornecido""" |
| | try: |
| | |
| | model_id = AVAILABLE_MODELS.get(self.model_name, self.model_name) |
| | |
| | |
| | if model_id not in AVAILABLE_MODELS.values(): |
| | model_id = REFINEMENT_MODELS.get(self.model_name, model_id) |
| | |
| | |
| | if model_id in OPENAI_MODELS: |
| | |
| | if model_id == "o3-mini": |
| | |
| | self.llm = ChatOpenAI(model=model_id) |
| | else: |
| | |
| | self.llm = ChatOpenAI(model=model_id, temperature=TEMPERATURE) |
| | |
| | elif model_id in ANTHROPIC_MODELS: |
| | |
| | self.llm = ChatAnthropic( |
| | model=model_id, |
| | temperature=TEMPERATURE, |
| | max_tokens=4096, |
| | max_retries=2, |
| | timeout=60.0 |
| | ) |
| | |
| | else: |
| | |
| | self.llm = HuggingFaceEndpoint( |
| | endpoint_url=f"https://api-inference.huggingface.co/models/{model_id}", |
| | temperature=TEMPERATURE, |
| | max_new_tokens=1024, |
| | timeout=120 |
| | ) |
| | |
| | logging.info(f"Processing Agent inicializado com modelo {model_id}") |
| | |
| | except Exception as e: |
| | logging.error(f"Erro ao inicializar Processing Agent: {e}") |
| | |
| | self.llm = ChatOpenAI(model="gpt-4o-mini", temperature=TEMPERATURE) |
| | logging.warning("Usando GPT-4o-mini como fallback") |
| | |
| | def recreate_llm(self, new_model: str): |
| | """ |
| | Recria o LLM com novo modelo |
| | |
| | Args: |
| | new_model: Nome do novo modelo |
| | """ |
| | old_model = self.model_name |
| | self.model_name = new_model |
| | self._initialize_llm() |
| | logging.info(f"[PROCESSING] Modelo alterado de '{old_model}' para '{new_model}'") |
| | |
| | async def process_context(self, context_prompt: str) -> Dict[str, Any]: |
| | """ |
| | Processa o contexto inicial e retorna sugestão de query |
| | |
| | Args: |
| | context_prompt: Prompt com contexto e pergunta do usuário |
| | |
| | Returns: |
| | Resultado do processamento com pergunta e sugestão de query |
| | """ |
| | try: |
| | logging.info(f"[PROCESSING] ===== INICIANDO PROCESSING AGENT =====") |
| | logging.info(f"[PROCESSING] Modelo utilizado: {self.model_name}") |
| | logging.info(f"[PROCESSING] Tamanho do contexto: {len(context_prompt)} caracteres") |
| |
|
| | |
| | if hasattr(self.llm, 'ainvoke'): |
| | |
| | logging.info(f"[PROCESSING] Executando chamada assíncrona para {self.model_name}") |
| | response = await self.llm.ainvoke([HumanMessage(content=context_prompt)]) |
| | output = response.content |
| | else: |
| | |
| | logging.info(f"[PROCESSING] Executando chamada síncrona para {self.model_name}") |
| | response = await asyncio.get_event_loop().run_in_executor( |
| | None, |
| | lambda: self.llm.invoke([HumanMessage(content=context_prompt)]) |
| | ) |
| | output = response.content if hasattr(response, 'content') else str(response) |
| |
|
| | logging.info(f"[PROCESSING] Resposta recebida do modelo ({len(output)} caracteres)") |
| |
|
| | |
| | processed_result = self._parse_processing_response(output) |
| |
|
| | result = { |
| | "success": True, |
| | "output": output, |
| | "processed_question": processed_result.get("question", ""), |
| | "suggested_query": processed_result.get("query", ""), |
| | "query_observations": processed_result.get("observations", ""), |
| | "model_used": self.model_name |
| | } |
| |
|
| | |
| | if result['suggested_query']: |
| | logging.info(f"[PROCESSING] ✅ Query SQL extraída com sucesso") |
| | else: |
| | logging.warning(f"[PROCESSING] ❌ Nenhuma query SQL foi extraída") |
| |
|
| | logging.info(f"[PROCESSING] ===== PROCESSING AGENT CONCLUÍDO =====") |
| | return result |
| | |
| | except Exception as e: |
| | error_msg = f"Erro no Processing Agent: {e}" |
| | logging.error(error_msg) |
| | |
| | return { |
| | "success": False, |
| | "output": error_msg, |
| | "processed_question": "", |
| | "suggested_query": "", |
| | "model_used": self.model_name |
| | } |
| | |
| | def _parse_processing_response(self, response: str) -> Dict[str, str]: |
| | """ |
| | Extrai query SQL e observações da resposta |
| | |
| | Args: |
| | response: Resposta do modelo |
| | |
| | Returns: |
| | Dicionário com query e observações extraídas |
| | """ |
| | try: |
| | import re |
| |
|
| | query = "" |
| | observations = "" |
| |
|
| | |
| | obs_match = re.search(r'Observações:\s*(.*?)(?:\n|$)', response, re.IGNORECASE) |
| | if obs_match: |
| | observations = obs_match.group(1).strip() |
| |
|
| | |
| | sql_patterns = [ |
| | |
| | r'```sql\s*(.*?)\s*```', |
| | |
| | r'```\s*(WITH.*?)\s*```', |
| | r'```\s*(SELECT.*?)\s*```', |
| | |
| | r'Opção de querySQL:\s*(WITH.*?)(?=Observações:|$)', |
| | r'Opção de querySQL:\s*(SELECT.*?)(?=Observações:|$)', |
| | |
| | r'(WITH\s+.*?;)', |
| | r'(SELECT\s+.*?;)' |
| | ] |
| |
|
| | for pattern in sql_patterns: |
| | match = re.search(pattern, response, re.DOTALL | re.IGNORECASE) |
| | if match: |
| | query = match.group(1).strip() |
| | break |
| |
|
| | |
| | if query: |
| | |
| | query = query.replace('```', '').replace('sql', '').strip() |
| |
|
| | |
| | query = query.strip('\n').strip() |
| |
|
| | |
| | if not observations: |
| | obs_patterns = [ |
| | r'Observações:\s*(.*)', |
| | r'Observacoes:\s*(.*)', |
| | ] |
| | for pattern in obs_patterns: |
| | match = re.search(pattern, response, re.IGNORECASE | re.DOTALL) |
| | if match: |
| | observations = match.group(1).strip() |
| | break |
| |
|
| | return { |
| | "question": "", |
| | "query": query, |
| | "observations": observations |
| | } |
| |
|
| | except Exception as e: |
| | logging.error(f"Erro ao extrair query e observações: {e}") |
| | return { |
| | "question": "", |
| | "query": "", |
| | "observations": "" |
| | } |
| |
|
| |
|
| | def get_default_processing_agent(model_name: str = "gpt-4o-mini") -> ProcessingAgentManager: |
| | """ |
| | Cria um Processing Agent com configurações padrão |
| | |
| | Args: |
| | model_name: Nome do modelo a usar |
| | |
| | Returns: |
| | ProcessingAgentManager configurado |
| | """ |
| | return ProcessingAgentManager(model_name) |
| |
|