143 lines
4.8 KiB
Python
143 lines
4.8 KiB
Python
import os
|
||
import asyncio
|
||
import json
|
||
from pathlib import Path
|
||
from typing import List
|
||
|
||
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
|
||
|
||
# --- Embeddings and vector store (Qdrant + Ollama) ---
|
||
from langchain_ollama import OllamaEmbeddings
|
||
from langchain_qdrant import Qdrant
|
||
from langchain_core.documents import Document
|
||
|
||
# Load OpenRouter key for LLM (required by Ollama embeddings as well)
|
||
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
|
||
|
||
# LLM configuration (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 via Ollama (nomic-embed-text)
|
||
embeddings = OllamaEmbeddings(model="nomic-embed-text")
|
||
|
||
# Qdrant client (in‑memory for demo; replace with host/port for prod)
|
||
qdrant_client = Qdrant(
|
||
collection_name="faq_collection",
|
||
url="http://localhost:6333", # default Qdrant local URL
|
||
embedding_function=embeddings,
|
||
)
|
||
|
||
# --- Data loading and indexing ---
|
||
DATA_DIR = Path("data")
|
||
CHROMA_PERSIST = Path("./qdrant_faq") # not used directly but kept for compatibility
|
||
|
||
|
||
def load_faq_to_qdrant() -> None:
|
||
"""Read all .md files from DATA_DIR, chunk them, embed and store in Qdrant."""
|
||
if not DATA_DIR.exists():
|
||
print("Data directory not found. Create 'data/' with .md files.")
|
||
return
|
||
docs: List[Document] = []
|
||
for md_file in DATA_DIR.glob("*.md"):
|
||
text = md_file.read_text(encoding="utf-8")
|
||
# Simple split by double newlines as a naive chunker
|
||
for i, chunk in enumerate(text.split("\n\n")):
|
||
docs.append(Document(page_content=chunk, metadata={"source": md_file.name, "chunk": i}))
|
||
# Add to Qdrant
|
||
qdrant_client.add_documents(docs)
|
||
print(f"Indexed {len(docs)} chunks into Qdrant.")
|
||
|
||
# --- Tools ---
|
||
@tool
|
||
def search_course_docs(query: str, k: int = 3) -> str:
|
||
"""Search the local FAQ collection for relevant passages."""
|
||
results = qdrant_client.similarity_search(query, k=k)
|
||
if not results:
|
||
return "No relevant information found in the course materials."
|
||
return "\n\n---\n\n".join(r.page_content for r in results)
|
||
|
||
# MCP-style tool: fetch metadata from a static JSON file
|
||
META_FILE = Path("meta.json")
|
||
|
||
@tool
|
||
def fetch_course_meta(query: str) -> str:
|
||
"""Return course metadata that matches the query.
|
||
For demo purposes, we load a static JSON file and perform a simple keyword search.
|
||
"""
|
||
if not META_FILE.exists():
|
||
return "Metadata file not found."
|
||
data = json.loads(META_FILE.read_text(encoding="utf-8"))
|
||
# Very naive matching: return entries where query is a substring of any value
|
||
matches = []
|
||
for key, value in data.items():
|
||
if isinstance(value, str) and query.lower() in value.lower():
|
||
matches.append(f"{key}: {value}")
|
||
if not matches:
|
||
return "No metadata matches your query."
|
||
return "\n".join(matches)
|
||
|
||
# --- 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 a course.\n"
|
||
"When a user asks about course materials, use the search_course_docs tool.\n"
|
||
"When a user asks about schedule, metadata, or other non‑material info, use fetch_course_meta.\n"
|
||
"Do not call both tools unless absolutely necessary.\n"
|
||
"In your final answer, prepend 'source: chroma' if you used search_course_docs,\n"
|
||
"or 'source: mcp_meta' if you used fetch_course_meta."
|
||
),
|
||
)
|
||
|
||
# --- CLI ---
|
||
PRESET_QUESTIONS = [
|
||
"What topics are covered in the first lecture?",
|
||
"How can I access the lecture slides?",
|
||
"What is the schedule for the next week?",
|
||
]
|
||
|
||
async def run_agent(question: str, thread_id: str = "session-1"):
|
||
result = await agent.ainvoke(
|
||
{"messages": [HumanMessage(content=question)]},
|
||
{"configurable": {"thread_id": thread_id}},
|
||
)
|
||
return result["messages"][-1].content
|
||
|
||
async def main():
|
||
# Ensure the vector store is populated
|
||
load_faq_to_qdrant()
|
||
|
||
print("\n--- FAQ Bot Demo ---\n")
|
||
for i, q in enumerate(PRESET_QUESTIONS, 1):
|
||
print(f"Q{i}: {q}")
|
||
answer = await run_agent(q, thread_id=f"demo-{i}")
|
||
print(f"A{i}: {answer}\n")
|
||
|
||
# Interactive mode
|
||
print("Enter your own questions (type 'exit' to quit):")
|
||
while True:
|
||
user_q = input("> ")
|
||
if user_q.lower() in {"exit", "quit"}:
|
||
break
|
||
answer = await run_agent(user_q, thread_id="interactive")
|
||
print(answer)
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main())
|