164 lines
6.0 KiB
Python
164 lines
6.0 KiB
Python
import os
|
||
import asyncio
|
||
import json
|
||
from pathlib import Path
|
||
from typing import List
|
||
|
||
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
|
||
from langchain_chroma import Chroma
|
||
from langchain_core.documents import Document
|
||
from langchain_core.messages import HumanMessage
|
||
from langchain.tools import tool
|
||
from deepagents import create_deep_agent
|
||
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 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 via OpenRouter
|
||
embeddings = OpenAIEmbeddings(
|
||
model="text-embedding-3-small",
|
||
base_url="https://openrouter.ai/api/v1",
|
||
api_key=OPENAI_API_KEY,
|
||
)
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Chroma persistence
|
||
# ---------------------------------------------------------------------------
|
||
CHROMA_DIR = Path("./chroma_faq")
|
||
CHROMA_DIR.mkdir(parents=True, exist_ok=True)
|
||
vector_store = Chroma(
|
||
collection_name="faq_collection",
|
||
embedding_function=embeddings,
|
||
persist_directory=str(CHROMA_DIR),
|
||
)
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Utility: load markdown files into Chroma
|
||
# ---------------------------------------------------------------------------
|
||
@tool
|
||
def load_faq_to_chroma() -> str:
|
||
"""Load all .md files from data/ into the Chroma vector store.
|
||
This tool is idempotent – it will overwrite existing collection.
|
||
"""
|
||
data_dir = Path("data")
|
||
if not data_dir.exists():
|
||
return "Data directory not found."
|
||
docs: List[Document] = []
|
||
for md_file in data_dir.glob("*.md"):
|
||
text = md_file.read_text(encoding="utf-8")
|
||
docs.append(Document(page_content=text, metadata={"source": md_file.name}))
|
||
if docs:
|
||
vector_store.delete_collection()
|
||
vector_store.add_documents(docs)
|
||
vector_store.persist()
|
||
return f"Loaded {len(docs)} documents into Chroma."
|
||
return "No markdown files found."
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Tool: search knowledge base
|
||
# ---------------------------------------------------------------------------
|
||
@tool
|
||
def search_course_docs(query: str, k: int = 3) -> str:
|
||
"""Search the FAQ collection for relevant passages.
|
||
Returns a string with the top k passages and a source tag.
|
||
"""
|
||
docs = vector_store.similarity_search(query, k=k)
|
||
if not docs:
|
||
return "No relevant information found in the FAQ."
|
||
results = "\n\n".join(f"{i+1}. {doc.page_content}" for i, doc in enumerate(docs))
|
||
return f"source: chroma\n{results}"
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# MCP-style tool: fetch course metadata
|
||
# ---------------------------------------------------------------------------
|
||
# For simplicity we use a static JSON file in the repo. In production this
|
||
# would be an HTTP call to an MCP server.
|
||
METADATA_FILE = Path("course_meta.json")
|
||
|
||
@tool
|
||
def fetch_course_meta(query: str) -> str:
|
||
"""Return metadata that matches the query.
|
||
The function performs a simple keyword search in the static JSON.
|
||
"""
|
||
if not METADATA_FILE.exists():
|
||
return "Metadata file not found."
|
||
data = json.loads(METADATA_FILE.read_text(encoding="utf-8"))
|
||
matches = [item for item in data if query.lower() in item.get("title", "").lower()]
|
||
if not matches:
|
||
return "No metadata matches the query."
|
||
return f"source: mcp_meta\n" + json.dumps(matches, indent=2)
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Backend setup
|
||
# ---------------------------------------------------------------------------
|
||
backend = CompositeBackend([
|
||
LocalShellBackend(workspace_dir="./workspace"),
|
||
FilesystemBackend(),
|
||
])
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Agent definition
|
||
# ---------------------------------------------------------------------------
|
||
agent = create_deep_agent(
|
||
model=llm,
|
||
tools=[load_faq_to_chroma, search_course_docs, fetch_course_meta],
|
||
backend=backend,
|
||
system_prompt=(
|
||
"You are a helpful FAQ bot for a course. "
|
||
"Use the search_course_docs tool for questions about lecture materials. "
|
||
"Use fetch_course_meta for questions about schedule or metadata. "
|
||
"Do not call both tools unless necessary. "
|
||
"Always prefix your answer with the source tag (chroma or mcp_meta)."
|
||
),
|
||
)
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# CLI helpers
|
||
# ---------------------------------------------------------------------------
|
||
PRESET_QUESTIONS = [
|
||
"What is the deadline for the final project?", # should hit metadata
|
||
"Explain the concept of polymorphism in OOP.", # should hit FAQ
|
||
"How many lectures are there in the first module?", # metadata
|
||
]
|
||
|
||
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 FAQ is loaded once
|
||
await agent.ainvoke(
|
||
{"messages": [HumanMessage(content="load_faq_to_chroma()")]},
|
||
{"configurable": {"thread_id": "init"}},
|
||
)
|
||
print("\n--- Preset questions ---")
|
||
for q in PRESET_QUESTIONS:
|
||
ans = await run_agent(q)
|
||
print(f"Q: {q}\nA: {ans}\n")
|
||
print("\n--- Interactive mode (type 'exit' to quit) ---")
|
||
while True:
|
||
user_input = input("> ")
|
||
if user_input.lower() in {"exit", "quit"}:
|
||
break
|
||
ans = await run_agent(user_input)
|
||
print(ans)
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main())
|