174 lines
6.6 KiB
Python
174 lines
6.6 KiB
Python
"""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 json
|
|
from pathlib import Path
|
|
from typing import List, Dict, Any
|
|
|
|
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
|
|
from langchain.schema import AgentAction, AgentFinish
|
|
from langchain_core.prompts import ChatPromptTemplate
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Configuration
|
|
# ---------------------------------------------------------------------------
|
|
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
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Vector store helper
|
|
# ---------------------------------------------------------------------------
|
|
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,
|
|
)
|
|
|
|
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
|
|
]
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Global knowledge base instance
|
|
# ---------------------------------------------------------------------------
|
|
kb = KnowledgeBase()
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Tools
|
|
# ---------------------------------------------------------------------------
|
|
@tool("search_knowledge_base")
|
|
def search_knowledge_base(query: str, max_results: int = 5) -> str:
|
|
"""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."
|
|
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")
|
|
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."
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Agent definition
|
|
# ---------------------------------------------------------------------------
|
|
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."
|
|
)
|
|
|
|
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])
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# 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 interactive_loop():
|
|
print("RAG Agent CLI. Commands: /add <title> <content>, /search <query>, /load <dir>, /quit")
|
|
while True:
|
|
try:
|
|
user_input = input("> ")
|
|
except (KeyboardInterrupt, EOFError):
|
|
print("\nExiting.")
|
|
break
|
|
if not user_input:
|
|
continue
|
|
if user_input.startswith("/quit"):
|
|
print("Goodbye!")
|
|
break
|
|
elif user_input.startswith("/add "):
|
|
try:
|
|
_, 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", ""))
|
|
|
|
if __name__ == "__main__":
|
|
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")
|