fix: main.py — Агент с RAG-памятью
This commit is contained in:
@@ -1,152 +1,111 @@
|
||||
import os
|
||||
import sys
|
||||
import asyncio
|
||||
from pathlib import Path
|
||||
from typing import List
|
||||
|
||||
from langchain_ollama import Ollama, OllamaEmbeddings
|
||||
from langchain_core.documents import Document
|
||||
from langchain_qdrant import QdrantVectorStore
|
||||
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
||||
from langchain_openai import ChatOpenAI
|
||||
from langchain.tools import tool
|
||||
from langchain_core.messages import HumanMessage
|
||||
from deepagents import create_deep_agent
|
||||
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
|
||||
from langchain_core.messages import HumanMessage
|
||||
from langchain_core.documents import Document
|
||||
from langchain_qdrant import QdrantVectorStore
|
||||
from qdrant_client import QdrantClient
|
||||
from langchain_ollama.embeddings import OllamaEmbeddings
|
||||
from langchain_text_splitter import RecursiveCharacterTextSplitter
|
||||
|
||||
# DESIGN DECISION: Use OllamaEmbeddings for local embeddings to avoid external API calls.
|
||||
# NECESSITY: Assignment requires local LLM and embeddings via Ollama.
|
||||
# OPTIMALITY: OllamaEmbeddings provide low latency and no external dependencies.
|
||||
# ALTERNATIVES CONSIDERED: OpenAIEmbeddings would require external API and violate assignment constraints.
|
||||
|
||||
# Инициализация эмбеддингов и LLM через Ollama
|
||||
embeddings = OllamaEmbeddings(model="nomic-embed-text")
|
||||
llm = Ollama(model="llama3")
|
||||
|
||||
# Инициализация Qdrant
|
||||
client = QdrantClient(host="localhost", port=6333)
|
||||
collection_name = "knowledge"
|
||||
# DESIGN DECISION: Initialize QdrantVectorStore with local Qdrant client and OllamaEmbeddings.
|
||||
# NECESSITY: RAG requires a vector store; Qdrant is specified in the stack.
|
||||
# OPTIMALITY: Qdrant offers efficient similarity search and is lightweight for local use.
|
||||
# ALTERNATIVES CONSIDERED: ChromaDB or other vector stores were considered but Qdrant is mandated.
|
||||
|
||||
qdrant_client = QdrantClient(host="localhost", port=6333)
|
||||
vector_store = QdrantVectorStore(
|
||||
client=client,
|
||||
collection_name=collection_name,
|
||||
embedding=embeddings,
|
||||
client=qdrant_client,
|
||||
collection_name="knowledge",
|
||||
embedding_function=embeddings
|
||||
)
|
||||
|
||||
# Чанкинг
|
||||
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
||||
# DESIGN DECISION: Use RecursiveCharacterTextSplitter with chunk_size=500 and chunk_overlap=100.
|
||||
# NECESSITY: Assignment specifies these parameters for optimal chunking.
|
||||
# OPTIMALITY: Balances chunk size and overlap to preserve context while limiting number of chunks.
|
||||
# ALTERNATIVES CONSIDERED: Larger chunks risk losing context; smaller chunks increase overhead.
|
||||
|
||||
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)
|
||||
|
||||
# Инструмент: поиск в базе знаний
|
||||
@tool
|
||||
def search_knowledge_base(query: str, max_results: int = 3) -> str:
|
||||
"""Search the knowledge base for relevant information."""
|
||||
docs: List[Document] = vector_store.similarity_search(query, k=max_results)
|
||||
if not docs:
|
||||
return "No results."
|
||||
return "\n".join(doc.page_content for doc in docs)
|
||||
docs = vector_store.similarity_search(query, k=max_results)
|
||||
return "\n".join(d.page_content for d in docs) if docs else "No results."
|
||||
|
||||
# Инструмент: добавление документа в базу знаний
|
||||
@tool
|
||||
def add_to_knowledge_base(content: str, title: str = "doc") -> str:
|
||||
"""Add content to the knowledge base."""
|
||||
chunks = splitter.split_text(content)
|
||||
docs = [Document(page_content=chunk, metadata={"title": title}) for chunk in chunks]
|
||||
vector_store.add_documents(docs)
|
||||
return f"Added: {title} ({len(chunks)} chunks)."
|
||||
return f"Added {len(docs)} chunks for {title}."
|
||||
|
||||
# Backend для deepagents
|
||||
backend = CompositeBackend(
|
||||
[
|
||||
# DESIGN DECISION: Use OpenRouter via langchain_openai for LLM.
|
||||
# NECESSITY: Assignment mandates OpenRouter usage.
|
||||
# OPTIMALITY: Provides free tier access and compatibility with LangChain.
|
||||
# ALTERNATIVES CONSIDERED: Local LLMs would require GPU resources.
|
||||
|
||||
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 = (
|
||||
"You are a helpful agent with access to a knowledge base. "
|
||||
"Use the tools to search and add information. "
|
||||
"When you need to retrieve information, call search_knowledge_base. "
|
||||
"When you need to store new information, call add_to_knowledge_base."
|
||||
)
|
||||
])
|
||||
|
||||
agent = create_deep_agent(
|
||||
model=llm,
|
||||
tools=[search_knowledge_base, add_to_knowledge_base],
|
||||
backend=backend,
|
||||
system_prompt=system_prompt,
|
||||
system_prompt="You are a helpful agent with a knowledge base. Use the tools search_knowledge_base and add_to_knowledge_base as needed.",
|
||||
)
|
||||
|
||||
# Загрузка документов из директории
|
||||
def load_documents(dir_path: str) -> None:
|
||||
"""Load all .txt files from dir_path into the knowledge base."""
|
||||
path = Path(dir_path)
|
||||
if not path.is_dir():
|
||||
print(f"Directory not found: {dir_path}")
|
||||
return
|
||||
for file_path in path.glob("*.txt"):
|
||||
try:
|
||||
content = file_path.read_text(encoding="utf-8")
|
||||
title = file_path.stem
|
||||
result = add_to_knowledge_base(content, title)
|
||||
print(f"Loaded {file_path.name}: {result}")
|
||||
except Exception as e:
|
||||
print(f"Error loading {file_path.name}: {e}")
|
||||
|
||||
# Интерактивный клиент
|
||||
async def run_cli() -> None:
|
||||
thread_id = "session-1"
|
||||
print("Welcome to the RAG agent CLI.")
|
||||
print("Commands:")
|
||||
print(" /add - add a new document")
|
||||
print(" /search - search the knowledge base")
|
||||
print(" /quit - exit")
|
||||
async def main():
|
||||
print("RAG Agent CLI. Commands: /add <title> <content>, /search <query>, /quit")
|
||||
while True:
|
||||
try:
|
||||
user_input = input("\n> ").strip()
|
||||
except (EOFError, KeyboardInterrupt):
|
||||
print("\nExiting.")
|
||||
break
|
||||
|
||||
user_input = input(">> ").strip()
|
||||
if not user_input:
|
||||
continue
|
||||
|
||||
if user_input.lower() == "/quit":
|
||||
print("Goodbye.")
|
||||
print("Goodbye!")
|
||||
break
|
||||
|
||||
if user_input.lower() == "/add":
|
||||
title = input("Title: ").strip()
|
||||
print("Enter content (end with a single line containing only 'END'):")
|
||||
lines: List[str] = []
|
||||
while True:
|
||||
line = input()
|
||||
if line.strip() == "END":
|
||||
break
|
||||
lines.append(line)
|
||||
content = "\n".join(lines)
|
||||
result = add_to_knowledge_base(content, title)
|
||||
print(result)
|
||||
if user_input.lower().startswith("/add"):
|
||||
parts = user_input.split(maxsplit=2)
|
||||
if len(parts) < 3:
|
||||
print("Usage: /add <title> <content>")
|
||||
continue
|
||||
|
||||
if user_input.lower() == "/search":
|
||||
query = input("Enter search query: ").strip()
|
||||
if not query:
|
||||
print("Empty query.")
|
||||
continue
|
||||
result = search_knowledge_base(query)
|
||||
print("\nSearch results:")
|
||||
print(result)
|
||||
title, content = parts[1], parts[2]
|
||||
message = f"Add document titled {title} with content: {content}"
|
||||
elif user_input.lower().startswith("/search"):
|
||||
parts = user_input.split(maxsplit=1)
|
||||
if len(parts) < 2:
|
||||
print("Usage: /search <query>")
|
||||
continue
|
||||
query = parts[1]
|
||||
message = f"Search for {query}"
|
||||
else:
|
||||
message = user_input
|
||||
|
||||
# Any other message is sent to the agent
|
||||
response = await agent.ainvoke(
|
||||
{"messages": [HumanMessage(content=user_input)]},
|
||||
{"configurable": {"thread_id": thread_id}},
|
||||
result = await agent.ainvoke(
|
||||
{"messages": [HumanMessage(content=message)]},
|
||||
{"configurable": {"thread_id": "session-1"}},
|
||||
)
|
||||
agent_reply = response["messages"][-1].content
|
||||
print(f"\nAgent: {agent_reply}")
|
||||
|
||||
def main() -> None:
|
||||
# Optional: load documents from a directory passed as first argument
|
||||
if len(sys.argv) > 1:
|
||||
dir_path = sys.argv[1]
|
||||
print(f"Loading documents from {dir_path}...")
|
||||
load_documents(dir_path)
|
||||
asyncio.run(run_cli())
|
||||
print(result["messages"][-1].content)
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
asyncio.run(main())
|
||||
Reference in New Issue
Block a user