146 lines
4.3 KiB
Python
146 lines
4.3 KiB
Python
"""
|
|
Arcádia MiroFlow Service — Agentes científicos via MiroFlow + Ollama local
|
|
FastAPI microserviço porta 8006
|
|
"""
|
|
import os
|
|
import sys
|
|
import time
|
|
import uuid
|
|
from typing import Optional
|
|
|
|
from fastapi import FastAPI, HTTPException
|
|
from fastapi.middleware.cors import CORSMiddleware
|
|
from pydantic import BaseModel
|
|
|
|
# Adicionar o submodule MiroFlow ao path
|
|
MIROFLOW_MODULE_PATH = os.path.join(
|
|
os.path.dirname(__file__), "..", "modules", "miroflow"
|
|
)
|
|
sys.path.insert(0, os.path.abspath(MIROFLOW_MODULE_PATH))
|
|
|
|
from miroflow.agents.factory import build_agent
|
|
from miroflow.agents.context import AgentContext
|
|
|
|
OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434")
|
|
RESEARCHER_MODEL = os.getenv("MIROFLOW_RESEARCHER_MODEL", "llama3.1:8b")
|
|
|
|
AGENT_MODELS = {
|
|
"statistician": "llama3.2:3b",
|
|
"fiscal_auditor": "llama3.2:3b",
|
|
"researcher": RESEARCHER_MODEL,
|
|
}
|
|
|
|
|
|
def make_agent_cfg(agent_type: str) -> dict:
|
|
model = AGENT_MODELS.get(agent_type, "deepseek-r1:14b")
|
|
return {
|
|
"type": "IterativeAgentWithToolAndRollback",
|
|
"name": f"arcadia_{agent_type}",
|
|
"max_turns": 10,
|
|
"llm": {
|
|
"provider_class": "UnifiedOpenAIClient",
|
|
"model_name": model,
|
|
"api_key": "ollama",
|
|
"base_url": f"{OLLAMA_BASE_URL}/v1",
|
|
"max_tokens": 8192,
|
|
"temperature": 0.7,
|
|
"top_p": 1.0,
|
|
"min_p": 0.0,
|
|
"top_k": -1,
|
|
"reasoning_effort": None,
|
|
"repetition_penalty": 1.0,
|
|
"max_context_length": -1,
|
|
"async_client": True,
|
|
"disable_cache_control": True,
|
|
"keep_tool_result": -1,
|
|
"use_tool_calls": False,
|
|
"oai_tool_thinking": False,
|
|
},
|
|
}
|
|
|
|
|
|
async def run_agent(agent_type: str, task: str) -> tuple[str, str]:
|
|
"""Retorna (result_text, model_name)"""
|
|
if agent_type not in AGENT_MODELS:
|
|
raise ValueError(f"Agente desconhecido: {agent_type}. Use: {list(AGENT_MODELS.keys())}")
|
|
cfg = make_agent_cfg(agent_type)
|
|
agent = build_agent(cfg)
|
|
ctx = AgentContext(
|
|
task_description=task,
|
|
system_prompt=(
|
|
"Você é um agente científico da Arcádia Suite especializado em análise de dados "
|
|
"empresariais. Responda em português de forma clara e objetiva. "
|
|
"Quando precisar de dados, explique o que analisaria e quais conclusões seriam esperadas."
|
|
),
|
|
initial_user_message=task,
|
|
)
|
|
result = await agent.run(ctx)
|
|
if isinstance(result, dict):
|
|
text = (
|
|
result.get("summary")
|
|
or result.get("final_boxed_answer")
|
|
or result.get("exceed_max_turn_summary")
|
|
or next(
|
|
(m["content"] for m in reversed(result.get("message_history") or []) if m.get("role") == "assistant"),
|
|
None,
|
|
)
|
|
or str(result)
|
|
)
|
|
else:
|
|
text = str(result)
|
|
return text, cfg["llm"]["model_name"]
|
|
|
|
|
|
app = FastAPI(title="Arcádia MiroFlow Service", version="1.0.0")
|
|
app.add_middleware(
|
|
CORSMiddleware,
|
|
allow_origins=["*"],
|
|
allow_credentials=True,
|
|
allow_methods=["*"],
|
|
allow_headers=["*"],
|
|
)
|
|
|
|
|
|
class AnalyzeRequest(BaseModel):
|
|
agent: str
|
|
task: str
|
|
context: dict = {}
|
|
tenant_id: Optional[int] = None
|
|
|
|
|
|
class AnalyzeResponse(BaseModel):
|
|
agent: str
|
|
model: str
|
|
result: str
|
|
execution_id: str
|
|
duration_ms: int
|
|
|
|
|
|
@app.get("/health")
|
|
async def health_check():
|
|
return {"status": "ok", "service": "miroflow", "agents": list(AGENT_MODELS.keys())}
|
|
|
|
|
|
@app.post("/analyze", response_model=AnalyzeResponse)
|
|
async def analyze(req: AnalyzeRequest):
|
|
if req.agent not in AGENT_MODELS:
|
|
raise HTTPException(status_code=422, detail=f"Agente inválido: {req.agent}")
|
|
start = time.time()
|
|
try:
|
|
result_text, model_name = await run_agent(req.agent, req.task)
|
|
except Exception as e:
|
|
raise HTTPException(status_code=500, detail=str(e))
|
|
return AnalyzeResponse(
|
|
agent=req.agent,
|
|
model=model_name,
|
|
result=result_text,
|
|
execution_id=str(uuid.uuid4()),
|
|
duration_ms=int((time.time() - start) * 1000),
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import uvicorn
|
|
port = int(os.environ.get("MIROFLOW_PORT", 8006))
|
|
uvicorn.run(app, host="0.0.0.0", port=port)
|