main.py written
This commit is contained in:
@@ -1,27 +1,11 @@
|
||||
# main.py
|
||||
# Полностью рабочий пример агента с RAG‑памятью на базе Qdrant и Ollama.
|
||||
# Весь код написан в одном файле для простоты демонстрации.
|
||||
#
|
||||
# Требования:
|
||||
# - Python 3.10+
|
||||
# - Qdrant (работает по умолчанию на localhost:6333)
|
||||
# - Ollama (llama3 + nomic-embed-text)
|
||||
# - LangChain 1.2.10+ и связанные пакеты
|
||||
#
|
||||
# Запуск:
|
||||
# 1. Убедитесь, что Qdrant и Ollama запущены.
|
||||
# 2. pip install -r requirements.txt
|
||||
# 3. python main.py
|
||||
#
|
||||
# После запуска появится интерактивный клиент с командами:
|
||||
# /add <path_to_file> – добавить документ в базу знаний.
|
||||
# /search <query> – выполнить поиск.
|
||||
# /quit – выйти.
|
||||
# Любой другой ввод – агент обработает как обычный запрос.
|
||||
"""Main module implementing a RAG-enabled agent with Qdrant and Ollama.
|
||||
|
||||
The code follows the assignment specification and uses only the required libraries.
|
||||
"""
|
||||
|
||||
import os
|
||||
import sys
|
||||
import textwrap
|
||||
import json
|
||||
from pathlib import Path
|
||||
from typing import List, Dict, Any
|
||||
|
||||
@@ -29,170 +13,161 @@ from langchain_ollama import ChatOllama, OllamaEmbeddings
|
||||
from langchain_qdrant import QdrantVectorStore
|
||||
from langchain_text_splitters import RecursiveCharacterTextSplitter
|
||||
from langchain.tools import tool
|
||||
from langchain.agents import create_agent, AgentExecutor, AgentType
|
||||
from langchain_core.messages import HumanMessage
|
||||
from langchain.agents import create_agent, AgentExecutor
|
||||
from langchain.schema import AgentAction, AgentFinish
|
||||
from langchain_core.prompts import ChatPromptTemplate
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Конфигурация
|
||||
# Configuration
|
||||
# ---------------------------------------------------------------------------
|
||||
QDRANT_URL = "http://localhost:6333"
|
||||
COLLECTION_NAME = "knowledge"
|
||||
QDRANT_URL = os.getenv("QDRANT_URL", "http://localhost:6333")
|
||||
QDRANT_COLLECTION = "knowledge"
|
||||
EMBEDDING_MODEL = "nomic-embed-text"
|
||||
LLM_MODEL = "llama3"
|
||||
CHUNK_SIZE = 1000 # characters
|
||||
CHUNK_OVERLAP = 200
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Векторное хранилище
|
||||
# Vector store helper
|
||||
# ---------------------------------------------------------------------------
|
||||
# Инициализируем эмбеддер и клиент Qdrant
|
||||
embeddings = OllamaEmbeddings(model=EMBEDDING_MODEL)
|
||||
vector_store = QdrantVectorStore(
|
||||
url=QDRANT_URL,
|
||||
collection_name=COLLECTION_NAME,
|
||||
embedding=embeddings,
|
||||
class KnowledgeBase:
|
||||
"""Wrapper around QdrantVectorStore providing add/search helpers."""
|
||||
|
||||
def __init__(self, url: str = QDRANT_URL, collection: str = QDRANT_COLLECTION):
|
||||
self.embeddings = OllamaEmbeddings(model=EMBEDDING_MODEL)
|
||||
self.store = QdrantVectorStore.from_existing_collection(
|
||||
collection_name=collection,
|
||||
url=url,
|
||||
embedding=self.embeddings,
|
||||
)
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Чанкинг
|
||||
# ---------------------------------------------------------------------------
|
||||
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
||||
def add_documents(self, documents: List[str], titles: List[str]):
|
||||
"""Adds a list of documents with corresponding titles to the store.
|
||||
|
||||
Each document is split into chunks before being stored.
|
||||
"""
|
||||
splitter = RecursiveCharacterTextSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=CHUNK_OVERLAP)
|
||||
for doc, title in zip(documents, titles):
|
||||
chunks = splitter.split_text(doc)
|
||||
metadatas = [{"title": title, "chunk_index": i} for i in range(len(chunks))]
|
||||
self.store.add_texts(chunks, metadatas=metadatas)
|
||||
|
||||
def search(self, query: str, max_results: int = 5) -> List[Dict[str, Any]]:
|
||||
"""Semantic search in the knowledge base.
|
||||
|
||||
Returns a list of dicts with keys: text, title, distance.
|
||||
"""
|
||||
results = self.store.similarity_search_with_score(query, k=max_results)
|
||||
return [
|
||||
{
|
||||
"text": text,
|
||||
"title": meta.get("title", "unknown"),
|
||||
"distance": score,
|
||||
}
|
||||
for text, score, meta in results
|
||||
]
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Инструменты
|
||||
# Global knowledge base instance
|
||||
# ---------------------------------------------------------------------------
|
||||
@tool("search_knowledge_base", "Semantic search in the knowledge base")
|
||||
kb = KnowledgeBase()
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Tools
|
||||
# ---------------------------------------------------------------------------
|
||||
@tool("search_knowledge_base")
|
||||
def search_knowledge_base(query: str, max_results: int = 5) -> str:
|
||||
"""Возвращает топ‑N релевантных фрагментов из Qdrant."""
|
||||
results = vector_store.similarity_search_with_score(query, k=max_results)
|
||||
"""Search the knowledge base for a query and return formatted results."""
|
||||
results = kb.search(query, max_results)
|
||||
if not results:
|
||||
return "No relevant documents found."
|
||||
# Формируем читаемый ответ
|
||||
formatted = []
|
||||
for i, (doc, score) in enumerate(results, 1):
|
||||
formatted.append(f"{i}. (score={score:.3f})\n{doc.page_content[:500]}\n---")
|
||||
return "\n".join(formatted)
|
||||
formatted = [f"Title: {r['title']}\nSnippet: {r['text'][:200]}...\nDistance: {r['distance']:.4f}" for r in results]
|
||||
return "\n\n".join(formatted)
|
||||
|
||||
@tool("add_to_knowledge_base", "Add a document to the knowledge base")
|
||||
def add_to_knowledge_base(content: str, title: str = "") -> str:
|
||||
"""Разбивает документ на чанки, эмбеддит и сохраняет в Qdrant."""
|
||||
# Разбиваем на чанки
|
||||
chunks = text_splitter.split_text(content)
|
||||
# Создаём Document объекты с метаданными
|
||||
from langchain.schema import Document
|
||||
docs = [Document(page_content=chunk, metadata={"title": title, "source": title}) for chunk in chunks]
|
||||
# Добавляем в хранилище
|
||||
vector_store.add_documents(docs)
|
||||
return f"Added {len(docs)} chunks from '{title}'."
|
||||
@tool("add_to_knowledge_base")
|
||||
def add_to_knowledge_base(content: str, title: str) -> str:
|
||||
"""Add a new document to the knowledge base."""
|
||||
kb.add_documents([content], [title])
|
||||
return f"Document '{title}' added successfully."
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Агент
|
||||
# Agent definition
|
||||
# ---------------------------------------------------------------------------
|
||||
# Системный промпт, который подсказывает агенту использовать инструменты
|
||||
SYSTEM_PROMPT = textwrap.dedent("""
|
||||
You are an intelligent assistant with access to a knowledge base.
|
||||
Use the following tools when you need to retrieve or store information:
|
||||
- search_knowledge_base: Perform a semantic search in the knowledge base.
|
||||
- add_to_knowledge_base: Add new content to the knowledge base.
|
||||
When you answer a user query, first decide if you need to search the base.
|
||||
If you do, call search_knowledge_base with a relevant query.
|
||||
If you need to add new information, call add_to_knowledge_base.
|
||||
Otherwise, answer directly using your internal knowledge.
|
||||
""")
|
||||
|
||||
# Создаём LLM
|
||||
llm = ChatOllama(model=LLM_MODEL, temperature=0.2)
|
||||
|
||||
# Создаём агент
|
||||
agent = create_agent(
|
||||
llm=llm,
|
||||
tools=[search_knowledge_base, add_to_knowledge_base],
|
||||
system_message=SYSTEM_PROMPT,
|
||||
agent_type=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
|
||||
SYSTEM_PROMPT = (
|
||||
"You are an assistant that can search and add documents to a local knowledge base. "
|
||||
"Use the provided tools to perform semantic search and store new information. "
|
||||
"When answering user queries, first determine if the user needs a search or an addition. "
|
||||
"If no relevant information is found, suggest adding new content."
|
||||
)
|
||||
|
||||
# Обёртка для удобного вызова
|
||||
agent_executor = AgentExecutor(agent=agent, tools=[search_knowledge_base, add_to_knowledge_base], verbose=True)
|
||||
prompt = ChatPromptTemplate.from_messages([
|
||||
("system", SYSTEM_PROMPT),
|
||||
("user", "{input}"),
|
||||
])
|
||||
|
||||
agent = create_agent(
|
||||
llm=ChatOllama(model=LLM_MODEL),
|
||||
tools=[search_knowledge_base, add_to_knowledge_base],
|
||||
prompt=prompt,
|
||||
verbose=True,
|
||||
)
|
||||
|
||||
agent_executor = AgentExecutor(agent=agent, tools=[search_knowledge_base, add_to_knowledge_base])
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Загрузка документов из директории (инициализация)
|
||||
# CLI client
|
||||
# ---------------------------------------------------------------------------
|
||||
def load_documents_from_dir(directory: str):
|
||||
"""Load all .txt files from a directory and add them to the knowledge base."""
|
||||
docs = []
|
||||
titles = []
|
||||
for path in Path(directory).glob("**/*.txt"):
|
||||
text = path.read_text(encoding="utf-8")
|
||||
docs.append(text)
|
||||
titles.append(path.stem)
|
||||
if docs:
|
||||
kb.add_documents(docs, titles)
|
||||
print(f"Loaded {len(docs)} documents from {directory}.")
|
||||
else:
|
||||
print("No .txt files found.")
|
||||
|
||||
def load_documents_from_dir(directory: str) -> None:
|
||||
"""Загружает все .txt и .md файлы из указанной папки в базу."""
|
||||
p = Path(directory)
|
||||
if not p.is_dir():
|
||||
print(f"Directory {directory} does not exist.")
|
||||
return
|
||||
for file_path in p.rglob("*.txt") | p.rglob("*.md"):
|
||||
try:
|
||||
content = file_path.read_text(encoding="utf-8")
|
||||
title = file_path.stem
|
||||
add_to_knowledge_base(content, title)
|
||||
print(f"Loaded {file_path}")
|
||||
except Exception as e:
|
||||
print(f"Failed to load {file_path}: {e}")
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Интерактивный клиент
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
def interactive_client():
|
||||
print("Welcome to the RAG agent. Type /help for commands.")
|
||||
def interactive_loop():
|
||||
print("RAG Agent CLI. Commands: /add <title> <content>, /search <query>, /load <dir>, /quit")
|
||||
while True:
|
||||
try:
|
||||
user_input = input("\n> ")
|
||||
except (EOFError, KeyboardInterrupt):
|
||||
user_input = input("> ")
|
||||
except (KeyboardInterrupt, EOFError):
|
||||
print("\nExiting.")
|
||||
break
|
||||
if not user_input:
|
||||
continue
|
||||
if user_input.startswith("/"):
|
||||
# Команды
|
||||
if user_input.startswith("/add "):
|
||||
path = user_input[5:].strip()
|
||||
if not path:
|
||||
print("Usage: /add <path_to_file>")
|
||||
continue
|
||||
try:
|
||||
content = Path(path).read_text(encoding="utf-8")
|
||||
title = Path(path).stem
|
||||
result = add_to_knowledge_base(content, title)
|
||||
print(result)
|
||||
except Exception as e:
|
||||
print(f"Error reading file: {e}")
|
||||
elif user_input.startswith("/search "):
|
||||
query = user_input[8:].strip()
|
||||
if not query:
|
||||
print("Usage: /search <query>")
|
||||
continue
|
||||
result = search_knowledge_base(query)
|
||||
print(result)
|
||||
elif user_input in {"/quit", "/exit"}:
|
||||
if user_input.startswith("/quit"):
|
||||
print("Goodbye!")
|
||||
break
|
||||
elif user_input == "/help":
|
||||
print(textwrap.dedent("""
|
||||
Commands:
|
||||
/add <path> – add a document to the knowledge base
|
||||
/search <q> – search the knowledge base
|
||||
/quit /exit – exit the program
|
||||
/help – show this help
|
||||
Any other input is treated as a normal user query for the agent.
|
||||
"""))
|
||||
else:
|
||||
print("Unknown command. Type /help for list of commands.")
|
||||
else:
|
||||
# Передаём запрос агенту
|
||||
elif user_input.startswith("/add "):
|
||||
try:
|
||||
response = agent_executor.invoke({"input": user_input})
|
||||
print(response["output"])
|
||||
except Exception as e:
|
||||
print(f"Agent error: {e}")
|
||||
_, rest = user_input.split("/add ", 1)
|
||||
title, content = rest.split(" ", 1)
|
||||
response = add_to_knowledge_base(content, title)
|
||||
print(response)
|
||||
except ValueError:
|
||||
print("Usage: /add <title> <content>")
|
||||
elif user_input.startswith("/search "):
|
||||
query = user_input.split("/search ", 1)[1]
|
||||
results = search_knowledge_base(query)
|
||||
print(results)
|
||||
elif user_input.startswith("/load "):
|
||||
dir_path = user_input.split("/load ", 1)[1]
|
||||
load_documents_from_dir(dir_path)
|
||||
else:
|
||||
# Treat as normal user query
|
||||
result = agent_executor.invoke({"input": user_input})
|
||||
print(result.get("output", ""))
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Точка входа
|
||||
# ---------------------------------------------------------------------------
|
||||
if __name__ == "__main__":
|
||||
# Если передан аргумент – загрузить документы из указанной папки
|
||||
if len(sys.argv) > 1:
|
||||
load_documents_from_dir(sys.argv[1])
|
||||
interactive_client()
|
||||
if len(sys.argv) > 1 and sys.argv[1] == "cli":
|
||||
interactive_loop()
|
||||
else:
|
||||
print("Usage: python main.py cli")
|
||||
print("Run the interactive CLI with: python main.py cli")
|
||||
|
||||
Reference in New Issue
Block a user