95 lines
3.7 KiB
Python
95 lines
3.7 KiB
Python
import os
|
||
import asyncio
|
||
from dotenv import load_dotenv
|
||
from langchain_openai import ChatOpenAI
|
||
from langchain_core.prompts import PromptTemplate
|
||
from langchain_core.output_parsers import PydanticOutputParser
|
||
from pydantic import BaseModel, Field
|
||
from langchain_core.messages import HumanMessage
|
||
from langchain.tools import tool
|
||
from deepagents import create_deep_agent
|
||
from deepagents.backends import CompositeBackend, LocalShellBackend, FilesystemBackend
|
||
|
||
# Load environment variables
|
||
load_dotenv()
|
||
|
||
# ---------- LLM INITIALIZATION ----------
|
||
# Using OpenRouter via langchain-openai as required
|
||
llm = ChatOpenAI(
|
||
model_name="gpt-4o-mini", # example model available on OpenRouter
|
||
base_url="https://openrouter.ai/api/v1/chat/completions",
|
||
api_key=os.getenv("OPENROUTER_API_KEY"),
|
||
temperature=0.0,
|
||
)
|
||
|
||
# ---------- Pydantic MODEL ----------
|
||
class TaskCard(BaseModel):
|
||
title: str = Field(description="Краткое название задания")
|
||
subject: str = Field(description="Предмет или тема задания")
|
||
deadline_hint: str = Field(description="Краткая подсказка о сроке сдачи, например 'к пятнице'")
|
||
deliverable_type: str = Field(description="Тип сдаваемого материала: отчёт, код, презентация и т.д.")
|
||
grading_hints: list[str] = Field(description="Список критериев оценки, упомянутых в тексте")
|
||
|
||
# ---------- PROMPT AND PARSER ----------
|
||
parser = PydanticOutputParser(pydantic_object=TaskCard)
|
||
|
||
prompt_template = """You are an assistant that extracts structured information from a short task description.
|
||
|
||
Input: {task_text}
|
||
|
||
Output must be a JSON object with the following fields:
|
||
{format_instructions}
|
||
|
||
Respond ONLY with the JSON object.
|
||
"""
|
||
|
||
prompt = PromptTemplate(
|
||
template=prompt_template,
|
||
input_variables=["task_text"],
|
||
partial_variables={"format_instructions": parser.get_format_instructions()},
|
||
)
|
||
|
||
# ---------- TOOL THAT RUNS THE CHAIN ----------
|
||
@tool
|
||
def parse_task(task_text: str) -> str:
|
||
"""Parse a task description into a structured JSON object."""
|
||
chain = prompt | llm | parser
|
||
result = chain.invoke({"task_text": task_text})
|
||
# Return the validated object as JSON string for the agent to forward
|
||
return result.model_dump_json()
|
||
|
||
# ---------- BACKEND AND AGENT ----------
|
||
backend = CompositeBackend([
|
||
LocalShellBackend(workspace_dir="./workspace"),
|
||
FilesystemBackend(),
|
||
])
|
||
|
||
agent = create_deep_agent(
|
||
model=llm,
|
||
tools=[parse_task],
|
||
backend=backend,
|
||
system_prompt="You are a helpful assistant that parses a single task description into structured data using the provided tool.",
|
||
)
|
||
|
||
# ---------- MAIN EXECUTION ----------
|
||
async def main():
|
||
# Example input – single line, no dialogue
|
||
task_description = "Сдайте к пятнице мини-отчёт по LangChain: 2 страницы, упор на агентов. Оценка: за полноту и за пример кода."
|
||
result = await agent.ainvoke(
|
||
{"messages": [HumanMessage(content=task_description)]},
|
||
{"configurable": {"thread_id": "session-1"}},
|
||
)
|
||
# The agent will return the tool output as the last message
|
||
output_json = result["messages"][-1].content
|
||
print("\n--- Parsed JSON ---")
|
||
print(output_json)
|
||
# Load into Pydantic for pretty printing and summary
|
||
card = TaskCard.parse_raw(output_json)
|
||
print("\n--- Validated Object ---")
|
||
print(card.model_dump())
|
||
print("\n--- Summary ---")
|
||
print(f"Title: {card.title}\nSubject: {card.subject}\nDeadline: {card.deadline_hint}\nDeliverable: {card.deliverable_type}\nGrading: {', '.join(card.grading_hints)}")
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main())
|