evo-ai/src/services/agent_builder.py
2025-04-25 17:20:55 -03:00

114 lines
4.9 KiB
Python

from typing import List, Optional, Tuple
from google.adk.agents.llm_agent import LlmAgent
from google.adk.agents import SequentialAgent, ParallelAgent, LoopAgent
from google.adk.models.lite_llm import LiteLlm
from src.utils.logger import setup_logger
from src.core.exceptions import AgentNotFoundError
from src.services.agent_service import get_agent
from src.services.custom_tools import CustomToolBuilder
from src.services.mcp_service import MCPService
from sqlalchemy.orm import Session
from contextlib import AsyncExitStack
logger = setup_logger(__name__)
class AgentBuilder:
def __init__(self, db: Session):
self.db = db
self.custom_tool_builder = CustomToolBuilder()
self.mcp_service = MCPService()
async def _create_llm_agent(self, agent) -> Tuple[LlmAgent, Optional[AsyncExitStack]]:
"""Cria um agente LLM a partir dos dados do agente."""
# Obtém ferramentas personalizadas da configuração
custom_tools = []
if agent.config.get("tools"):
custom_tools = self.custom_tool_builder.build_tools(agent.config["tools"])
# Obtém ferramentas MCP da configuração
mcp_tools = []
mcp_exit_stack = None
if agent.config.get("mcpServers"):
mcp_tools, mcp_exit_stack = await self.mcp_service.build_tools(agent.config)
# Combina todas as ferramentas
all_tools = custom_tools + mcp_tools
return LlmAgent(
name=agent.name,
model=LiteLlm(model=agent.model, api_key=agent.api_key),
instruction=agent.instruction,
description=agent.config.get("description", ""),
tools=all_tools,
), mcp_exit_stack
async def _get_sub_agents(self, sub_agent_ids: List[str]) -> List[Tuple[LlmAgent, Optional[AsyncExitStack]]]:
"""Obtém e cria os sub-agentes LLM."""
sub_agents = []
for sub_agent_id in sub_agent_ids:
agent = get_agent(self.db, sub_agent_id)
if agent is None:
raise AgentNotFoundError(f"Agente com ID {sub_agent_id} não encontrado")
if agent.type != "llm":
raise ValueError(f"Agente {agent.name} (ID: {agent.id}) não é um agente LLM")
sub_agent, exit_stack = await self._create_llm_agent(agent)
sub_agents.append((sub_agent, exit_stack))
return sub_agents
async def build_llm_agent(self, root_agent) -> Tuple[LlmAgent, Optional[AsyncExitStack]]:
"""Constrói um agente LLM com seus sub-agentes."""
logger.info("Criando agente LLM")
sub_agents = []
if root_agent.config.get("sub_agents"):
sub_agents_with_stacks = await self._get_sub_agents(root_agent.config.get("sub_agents"))
sub_agents = [agent for agent, _ in sub_agents_with_stacks]
root_llm_agent, exit_stack = await self._create_llm_agent(root_agent)
if sub_agents:
root_llm_agent.sub_agents = sub_agents
return root_llm_agent, exit_stack
async def build_composite_agent(self, root_agent) -> Tuple[SequentialAgent | ParallelAgent | LoopAgent, Optional[AsyncExitStack]]:
"""Constrói um agente composto (Sequential, Parallel ou Loop) com seus sub-agentes."""
logger.info(f"Processando sub-agentes para agente {root_agent.type}")
sub_agents_with_stacks = await self._get_sub_agents(root_agent.config.get("sub_agents", []))
sub_agents = [agent for agent, _ in sub_agents_with_stacks]
if root_agent.type == "sequential":
logger.info("Criando SequentialAgent")
return SequentialAgent(
name=root_agent.name,
sub_agents=sub_agents,
description=root_agent.config.get("description", ""),
), None
elif root_agent.type == "parallel":
logger.info("Criando ParallelAgent")
return ParallelAgent(
name=root_agent.name,
sub_agents=sub_agents,
description=root_agent.config.get("description", ""),
), None
elif root_agent.type == "loop":
logger.info("Criando LoopAgent")
return LoopAgent(
name=root_agent.name,
sub_agents=sub_agents,
description=root_agent.config.get("description", ""),
max_iterations=root_agent.config.get("max_iterations", 5),
), None
else:
raise ValueError(f"Tipo de agente inválido: {root_agent.type}")
async def build_agent(self, root_agent) -> Tuple[LlmAgent | SequentialAgent | ParallelAgent | LoopAgent, Optional[AsyncExitStack]]:
"""Constrói o agente apropriado baseado no tipo do agente root."""
if root_agent.type == "llm":
return await self.build_llm_agent(root_agent)
else:
return await self.build_composite_agent(root_agent)