Files
task-6a1d75c5fd30e81cf3126ae7/main.py
T

144 lines
5.1 KiB
Python

import os
import asyncio
import json
import httpx
from pathlib import Path
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
from langchain_chroma import Chroma
from langchain_core.documents import Document
from langchain.tools import tool
from deepagents import create_deep_agent
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
from langchain_core.messages import HumanMessage
# ---------------------------
# Configuration
# ---------------------------
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
if not OPENAI_API_KEY:
raise RuntimeError("OPENAI_API_KEY not set in environment")
# LLM via OpenRouter
llm = ChatOpenAI(
model="openai/gpt-oss-20b:free",
base_url="https://openrouter.ai/api/v1",
api_key=OPENAI_API_KEY,
temperature=0.0,
)
# Embeddings for Chroma
embeddings = OpenAIEmbeddings(
model="text-embedding-3-small",
base_url="https://openrouter.ai/api/v1",
api_key=OPENAI_API_KEY,
)
# ---------------------------
# Chroma DB utilities
# ---------------------------
CHROMA_DIR = Path("./chroma_faq")
CHROMA_DIR.mkdir(parents=True, exist_ok=True)
vector_store = Chroma(
collection_name="faq_collection",
persist_directory=str(CHROMA_DIR),
embedding_function=embeddings,
)
def load_faq_to_chroma(md_dir: str = "data"):
"""Load all .md files from md_dir into ChromaDB.
Each file is split into chunks and added to the vector store.
"""
md_path = Path(md_dir)
if not md_path.exists():
raise FileNotFoundError(f"Markdown directory {md_dir} not found")
for md_file in md_path.glob("*.md"):
text = md_file.read_text(encoding="utf-8")
# Simple chunking: split by double newlines
chunks = [c.strip() for c in text.split("\n\n") if c.strip()]
docs = [Document(page_content=c, metadata={"source": md_file.name}) for c in chunks]
vector_store.add_documents(docs)
vector_store.persist()
# ---------------------------
# Tools
# ---------------------------
@tool
def search_course_docs(query: str, k: int = 3) -> str:
"""Search the local FAQ ChromaDB for relevant passages."""
docs = vector_store.similarity_search(query, k=k)
if not docs:
return "No relevant information found in the course materials."
return "\n\n---\n\n".join(f"{d.metadata.get('source', 'unknown')}\n{d.page_content}" for d in docs)
@tool
def fetch_course_meta(query: str) -> str:
"""Mock MCP-style tool that fetches course metadata from a local JSON file.
In production this would be an HTTP call to an MCP server.
"""
# For simplicity, we use a local JSON file. In a real scenario, replace with httpx.get.
meta_path = Path("course_meta.json")
if not meta_path.exists():
return "Course metadata not available."
data = json.loads(meta_path.read_text(encoding="utf-8"))
# Very naive search: return any entry where query is a substring of title or description
results = [f"{item['title']}: {item['description']}" for item in data if query.lower() in item.get('title', '').lower() or query.lower() in item.get('description', '').lower()]
return "\n".join(results) if results else "No matching metadata found."
# ---------------------------
# Agent setup
# ---------------------------
backend = CompositeBackend([
LocalShellBackend(workspace_dir="./workspace"),
FilesystemBackend(),
])
agent = create_deep_agent(
model=llm,
tools=[search_course_docs, fetch_course_meta],
backend=backend,
system_prompt=(
"You are a helpful FAQ bot for the course.\n"
"When a user asks about course content, use the search_course_docs tool.\n"
"When a user asks about schedule, metadata, or other non-content info, use fetch_course_meta.\n"
"Do not call both tools unless absolutely necessary.\n"
"In your final answer, prepend 'source: chroma' or 'source: mcp_meta' to indicate the origin."
),
)
# ---------------------------
# CLI
# ---------------------------
PRESET_QUESTIONS = [
"What is the deadline for the final project?", # likely in metadata
"Explain the concept of tokenization in NLP.", # content
"How many lectures are there in the first module?", # content
]
async def run_agent(question: str, thread_id: str = "session-1"):
result = await agent.ainvoke(
{"messages": [HumanMessage(content=question)]},
{"configurable": {"thread_id": thread_id}},
)
# The last message is the agent's reply
reply = result["messages"][-1].content
print(f"\nQ: {question}\nA: {reply}\n")
async def main():
# Load data into Chroma if not already persisted
if not CHROMA_DIR.exists() or not any(CHROMA_DIR.iterdir()):
print("Loading FAQ data into ChromaDB...")
load_faq_to_chroma()
print("\n--- Predefined questions ---")
for q in PRESET_QUESTIONS:
await run_agent(q)
print("\n--- Interactive mode (type 'exit' to quit) ---")
while True:
user_input = input("You: ")
if user_input.lower() in {"exit", "quit"}:
break
await run_agent(user_input)
if __name__ == "__main__":
asyncio.run(main())