143 lines
4.9 KiB
Python
143 lines
4.9 KiB
Python
import os
|
||
import asyncio
|
||
from pathlib import Path
|
||
from langchain_community.embeddings import OllamaEmbeddings
|
||
from langchain_chroma import Chroma
|
||
from langchain_core.documents import Document
|
||
from langchain_openai import ChatOpenAI
|
||
from langchain.tools import tool
|
||
from deepagents import create_deep_agent
|
||
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
|
||
from langchain_core.messages import HumanMessage
|
||
|
||
# ---------------------------
|
||
# 1. Embeddings & Chroma setup
|
||
# ---------------------------
|
||
# Using OllamaEmbeddings with nomic-embed-text as required by the "Исправить" section.
|
||
# The embeddings are used for both loading the FAQ and for the search tool.
|
||
embeddings = OllamaEmbeddings(model="nomic-embed-text")
|
||
|
||
# Persistent Chroma collection for the FAQ knowledge base.
|
||
vector_store = Chroma(
|
||
collection_name="faq_collection",
|
||
embedding_function=embeddings,
|
||
persist_directory="./chroma_faq"
|
||
)
|
||
|
||
# ---------------------------
|
||
# 2. Load FAQ markdown files into Chroma
|
||
# ---------------------------
|
||
|
||
def load_faq_to_chroma(data_dir: str = "data"):
|
||
"""Load all .md files from *data_dir* into the persistent Chroma collection.
|
||
Each file is split into documents with a simple line‑based splitter.
|
||
"""
|
||
data_path = Path(data_dir)
|
||
if not data_path.exists():
|
||
raise FileNotFoundError(f"Data directory {data_dir} not found")
|
||
docs = []
|
||
for md_file in data_path.glob("*.md"):
|
||
text = md_file.read_text(encoding="utf-8")
|
||
# Simple split by double newlines to create chunks
|
||
for i, chunk in enumerate(text.split("\n\n")):
|
||
docs.append(Document(page_content=chunk, metadata={"source": md_file.name, "chunk": i}))
|
||
vector_store.add_documents(docs)
|
||
vector_store.persist()
|
||
|
||
# ---------------------------
|
||
# 3. Tools
|
||
# ---------------------------
|
||
@tool
|
||
def search_course_docs(query: str, k: int = 3) -> str:
|
||
"""Search the FAQ knowledge base for relevant information."""
|
||
docs = vector_store.similarity_search(query, k=k)
|
||
if not docs:
|
||
return "No relevant information found in the FAQ."
|
||
return "\n\n---\n\n".join(f"**{doc.metadata.get('source')}** (chunk {doc.metadata.get('chunk')}):\n{doc.page_content}" for doc in docs)
|
||
|
||
@tool
|
||
def fetch_course_meta(query: str) -> str:
|
||
"""Mock MCP‑style tool that returns course metadata.
|
||
In production this would perform an HTTP GET to an MCP server.
|
||
Here we simply return a static JSON string based on the query.
|
||
"""
|
||
# Simple static mapping for demo purposes
|
||
meta = {
|
||
"schedule": "Monday 10:00-12:00, Wednesday 14:00-16:00",
|
||
"instructor": "Dr. Ivanov",
|
||
"credits": "3"
|
||
}
|
||
key = query.lower().strip()
|
||
return meta.get(key, f"No metadata found for '{query}'.")
|
||
|
||
# ---------------------------
|
||
# 4. Agent setup with deepagents
|
||
# ---------------------------
|
||
# LLM via OpenRouter as per course requirement
|
||
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,
|
||
)
|
||
|
||
backend = CompositeBackend([
|
||
LocalShellBackend(workspace_dir="./workspace"),
|
||
FilesystemBackend(),
|
||
])
|
||
|
||
# System prompt instructs the agent to choose the appropriate tool and to label the source.
|
||
system_prompt = (
|
||
"You are a helpful FAQ assistant.\n"
|
||
"When a user asks a question about course materials, use the tool `search_course_docs`.\n"
|
||
"When a user asks about schedule, instructor, or credits, use the tool `fetch_course_meta`.\n"
|
||
"Do not call both tools unless necessary.\n"
|
||
"In your final answer, prepend the source label: `source: chroma` or `source: mcp_meta`."
|
||
)
|
||
|
||
agent = create_deep_agent(
|
||
model=llm,
|
||
tools=[search_course_docs, fetch_course_meta],
|
||
backend=backend,
|
||
system_prompt=system_prompt,
|
||
)
|
||
|
||
# ---------------------------
|
||
# 5. CLI
|
||
# ---------------------------
|
||
PRESET_QUESTIONS = [
|
||
"What topics are covered in the first lecture?",
|
||
"Who is the instructor for this course?",
|
||
"When is the next class?"
|
||
]
|
||
|
||
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 response
|
||
return result["messages"][-1].content
|
||
|
||
async def main():
|
||
# Ensure FAQ is loaded
|
||
load_faq_to_chroma()
|
||
|
||
print("--- 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_input = input("> ")
|
||
if user_input.lower() in {"exit", "quit"}:
|
||
break
|
||
answer = await run_agent(user_input, thread_id="interactive")
|
||
print(answer)
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main())
|