113 lines
3.6 KiB
Python
113 lines
3.6 KiB
Python
"""
|
|
Agent with RAG memory using Qdrant and Ollama.
|
|
"""
|
|
|
|
import os
|
|
import sys
|
|
from pathlib import Path
|
|
from typing import List, Dict, Any
|
|
|
|
from langchain_ollama import OllamaEmbeddings, Ollama
|
|
from langchain_qdrant import QdrantVectorStore
|
|
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
|
from langchain.tools import tool
|
|
from langchain.agents import create_agent, AgentExecutor, AgentType
|
|
from langchain.schema import AgentAction, AgentFinish
|
|
from langchain.callbacks import get_openai_callback
|
|
|
|
# Configuration
|
|
QDRANT_URL = os.getenv("QDRANT_URL", "http://localhost:6333")
|
|
QDRANT_COLLECTION = os.getenv("QDRANT_COLLECTION", "rag_agent")
|
|
Ollama_EMBED_MODEL = os.getenv("Ollama_EMBED_MODEL", "nomic-embed-text")
|
|
Ollama_MODEL = os.getenv("Ollama_MODEL", "llama3")
|
|
|
|
# Initialize embeddings and vector store
|
|
embeddings = OllamaEmbeddings(model=Ollama_EMBED_MODEL)
|
|
vector_store = QdrantVectorStore(
|
|
url=QDRANT_URL,
|
|
collection_name=QDRANT_COLLECTION,
|
|
embeddings=embeddings,
|
|
)
|
|
|
|
# Text splitter
|
|
splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
|
|
|
|
# Tools
|
|
@tool
|
|
def search_knowledge_base(query: str, max_results: int = 5) -> str:
|
|
"""Semantic search in the knowledge base."""
|
|
docs = vector_store.similarity_search(query, k=max_results)
|
|
if not docs:
|
|
return "No relevant documents found."
|
|
return "\n\n".join([f"{i+1}. {doc.page_content[:200]}..." for i, doc in enumerate(docs)])
|
|
|
|
@tool
|
|
def add_to_knowledge_base(content: str, title: str) -> str:
|
|
"""Add a document to the knowledge base."""
|
|
# Split content into chunks
|
|
chunks = splitter.split_text(content)
|
|
# Create documents with metadata
|
|
docs = []
|
|
for i, chunk in enumerate(chunks):
|
|
docs.append(
|
|
{
|
|
"page_content": chunk,
|
|
"metadata": {"title": title, "chunk_index": i},
|
|
}
|
|
)
|
|
# Add to vector store
|
|
vector_store.add_documents(docs)
|
|
return f"Document '{title}' added with {len(chunks)} chunks."
|
|
|
|
# Agent setup
|
|
SYSTEM_PROMPT = (
|
|
"You are an AI assistant with access to a local knowledge base. "
|
|
"Use the provided tools to search and add information. "
|
|
"When answering, rely on the knowledge base and the LLM."
|
|
)
|
|
|
|
# Create agent with tools
|
|
agent = create_agent(
|
|
llm=Ollama(model=Ollama_MODEL),
|
|
tools=[search_knowledge_base, add_to_knowledge_base],
|
|
system_prompt=SYSTEM_PROMPT,
|
|
agent_type=AgentType.ZERO_SHOT_REACT_DESCRIPTION,
|
|
)
|
|
|
|
executor = AgentExecutor(agent=agent, tools=[search_knowledge_base, add_to_knowledge_base], verbose=True)
|
|
|
|
# CLI
|
|
def main():
|
|
print("RAG Agent CLI. Commands: /add <title> <content>, /search <query>, /quit")
|
|
while True:
|
|
try:
|
|
inp = input("> ")
|
|
except EOFError:
|
|
break
|
|
if not inp:
|
|
continue
|
|
if inp.startswith("/quit"):
|
|
print("Bye!")
|
|
break
|
|
if inp.startswith("/add"):
|
|
parts = inp.split(maxsplit=2)
|
|
if len(parts) < 3:
|
|
print("Usage: /add <title> <content>")
|
|
continue
|
|
title, content = parts[1], parts[2]
|
|
print(add_to_knowledge_base(content, title))
|
|
continue
|
|
if inp.startswith("/search"):
|
|
query = inp[len("/search"):].strip()
|
|
if not query:
|
|
print("Usage: /search <query>")
|
|
continue
|
|
print(search_knowledge_base(query))
|
|
continue
|
|
# Treat as normal user query
|
|
response = executor.invoke({"input": inp})
|
|
print(response.get("output", ""))
|
|
|
|
if __name__ == "__main__":
|
|
main()
|