136 lines
4.3 KiB
Python
136 lines
4.3 KiB
Python
import os
|
|
import asyncio
|
|
from pathlib import Path
|
|
|
|
from dotenv import load_dotenv
|
|
from langchain_openai import ChatOpenAI, OpenAIEmbeddings
|
|
from langchain_chroma import Chroma
|
|
from langchain_core.documents import Document
|
|
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
|
from langchain.tools import tool
|
|
from langchain_community.tools.tavily_search import TavilySearchResults
|
|
|
|
from deepagents import create_deep_agent
|
|
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
|
|
from langchain_core.messages import HumanMessage
|
|
|
|
load_dotenv()
|
|
|
|
# ---------- LLM ----------
|
|
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,
|
|
)
|
|
|
|
# ---------- Vector Store ----------
|
|
embeddings = OpenAIEmbeddings(
|
|
model="text-embedding-3-small",
|
|
base_url="https://openrouter.ai/api/v1",
|
|
api_key=os.getenv("OPENAI_API_KEY"),
|
|
)
|
|
|
|
def create_vectorstore(persist_directory: str = "./chroma_db") -> Chroma:
|
|
return Chroma(
|
|
collection_name="knowledge",
|
|
embedding_function=embeddings,
|
|
persist_directory=persist_directory,
|
|
)
|
|
|
|
vectorstore = create_vectorstore()
|
|
|
|
def load_documents(directory: str, vectorstore: Chroma) -> None:
|
|
path = Path(directory)
|
|
if not path.is_dir():
|
|
return
|
|
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
|
docs = []
|
|
for file_path in path.rglob("*"):
|
|
if file_path.suffix.lower() not in {".txt", ".md"}:
|
|
continue
|
|
text = file_path.read_text(encoding="utf-8")
|
|
chunks = splitter.split_text(text)
|
|
for i, chunk in enumerate(chunks):
|
|
metadata = {"source": str(file_path), "chunk_index": i}
|
|
docs.append(Document(page_content=chunk, metadata=metadata))
|
|
if docs:
|
|
vectorstore.add_documents(docs)
|
|
|
|
# Load initial documents (run once or each start)
|
|
load_documents("./documents", vectorstore)
|
|
|
|
# ---------- Tools ----------
|
|
@tool
|
|
def search_local_kb(query: str, top_k: int = 3) -> str:
|
|
"""
|
|
Perform a semantic search in the local Chroma knowledge base.
|
|
Returns the concatenated contents of the most relevant documents.
|
|
"""
|
|
docs = vectorstore.similarity_search(query, k=top_k)
|
|
if not docs:
|
|
return "No relevant information found in the local knowledge base."
|
|
return "\n---\n".join(doc.page_content for doc in docs)
|
|
|
|
tavily_tool = TavilySearchResults(max_results=5)
|
|
|
|
@tool
|
|
def web_search(query: str) -> str:
|
|
"""
|
|
Search the web using Tavily and return a short summary of the top results.
|
|
"""
|
|
results = tavily_tool.run(query)
|
|
if not results:
|
|
return "No web results found."
|
|
# results is a list of dicts; extract title and url
|
|
lines = []
|
|
for r in results:
|
|
title = r.get("title", "No title")
|
|
url = r.get("url", "")
|
|
lines.append(f"{title}: {url}")
|
|
return "\n".join(lines)
|
|
|
|
# ---------- Backend ----------
|
|
backend = CompositeBackend(
|
|
[
|
|
LocalShellBackend(workspace_dir="./workspace"),
|
|
FilesystemBackend(),
|
|
]
|
|
)
|
|
|
|
# ---------- Agent ----------
|
|
system_prompt = (
|
|
"You are an AI assistant that can answer questions using two sources: "
|
|
"a local knowledge base (accessed via the tool `search_local_kb`) and the web (accessed via the tool `web_search`). "
|
|
"Decide which tool to use based on the user query. "
|
|
"When you use a tool, include the source name in your final answer: "
|
|
"`Source: chromadb` for local knowledge, `Source: tavily` for web results. "
|
|
"If both sources are needed, combine them and list both sources."
|
|
)
|
|
|
|
agent = create_deep_agent(
|
|
model=llm,
|
|
tools=[search_local_kb, web_search],
|
|
backend=backend,
|
|
system_prompt=system_prompt,
|
|
)
|
|
|
|
# ---------- CLI ----------
|
|
async def chat_loop() -> None:
|
|
print("AI Assistant (type 'exit' to quit)")
|
|
thread_id = "session-1"
|
|
while True:
|
|
user_input = input("\nYou: ").strip()
|
|
if user_input.lower() in {"exit", "quit"}:
|
|
print("Goodbye!")
|
|
break
|
|
|
|
result = await agent.ainvoke(
|
|
{"messages": [HumanMessage(content=user_input)]},
|
|
{"configurable": {"thread_id": thread_id}},
|
|
)
|
|
answer = result["messages"][-1].content
|
|
print(f"\nAssistant: {answer}")
|
|
|
|
if __name__ == "__main__":
|
|
asyncio.run(chat_loop()) |