import os
import asyncio
from langchain_openai import ChatOpenAI
from langchain_core.messages import HumanMessage
from langchain.tools import tool
from deepagents import create_deep_agent
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
from pydantic import BaseModel, Field
from langchain_core.output_parsers import PydanticOutputParser

# Настройка LLM через OpenRouter
llm = ChatOpenAI(
    model="openai/gpt-oss-20b:free",
    base_url="https://openrouter.ai/api/v1",
    api_key=os.getenv("OPENAI_API_KEY"),
    temperature=0.0,
)

# Бэкенд для хранения файлов и выполнения команд
backend = CompositeBackend(
    [
        LocalShellBackend(workspace_dir="./workspace"),
        FilesystemBackend(),
    ]
)

# Инструмент для проверки корректности вывода
@tool
def validate_output(output: str) -> str:
    """Проверяет, что вывод содержит все необходимые поля."""
    issues = []
    if "goal:" not in output.lower():
        issues.append("Missing goal")
    if "timeline:" not in output.lower():
        issues.append("Missing timeline")
    if "resources:" not in output.lower():
        issues.append("Missing resources")
    if "milestones:" not in output.lower():
        issues.append("Missing milestones")
    if "evaluation:" not in output.lower():
        issues.append("Missing evaluation")
    return "OK" if not issues else f"Issues: {', '.join(issues)}"

# Модель структуры плана
class PlanOutput(BaseModel):
    goal: str = Field(description="Краткое описание цели обучения")
    timeline: str = Field(description="План по времени (месяцы/недели)")
    resources: list[str] = Field(description="Список ресурсов и курсов")
    milestones: list[str] = Field(description="Ключевые контрольные точки")
    evaluation: str = Field(description="Методы оценки прогресса")

parser = PydanticOutputParser(pydantic_object=PlanOutput)

# Создание агента
agent = create_deep_agent(
    model=llm,
    tools=[validate_output],
    backend=backend,
    system_prompt=(
        "You are a helpful agent that creates a structured AI fluency plan. "
        "Return the plan in a format that matches the PlanOutput schema. "
        "After generating the plan, use the validate_output tool to ensure all fields are present."
    ),
)

async def main():
    # Запрос к агенту
    result = await agent.ainvoke(
        {"messages": [HumanMessage(content="Build a personal AI fluency plan for me.")]},
        {"configurable": {"thread_id": "ai-fluency-plan"}},
    )
    # Получаем последний вывод агента
    plan_text = result["messages"][-1].content
    # Парсим в структуру
    try:
        plan = parser.parse(plan_text)
    except Exception as e:
        print("Failed to parse plan:", e)
        print("Raw output:", plan_text)
        return
    # Выводим план
    print("\nPersonal AI Fluency Plan")
    print("------------------------")
    print(f"Goal: {plan.goal}\n")
    print(f"Timeline:\n{plan.timeline}\n")
    print("Resources:")
    for r in plan.resources:
        print(f"- {r}")
    print("\nMilestones:")
    for m in plan.milestones:
        print(f"- {m}")
    print(f"\nEvaluation:\n{plan.evaluation}")

if __name__ == "__main__":
    asyncio.run(main())