rag_agent.py

import os
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, ServiceContext
from llama_index.core.llms import OpenAI
from llama_index.core.tools import QueryEngineTool, ToolMetadata

# Build Service Context with preferred model
llm = OpenAI(model="gpt-3.5-turbo", temperature=0)
service_context = ServiceContext.from_defaults(llm=llm)

# Load and index documents
documents = SimpleDirectoryReader("kb").load_data()
index = VectorStoreIndex.from_documents(documents, service_context=service_context)
query_engine = index.as_query_engine()

# Tool wrapper to integrate with a multi-tool agent if needed
rag_tool = QueryEngineTool(
    query_engine=query_engine,
    metadata=ToolMetadata(name="RAGSearch", description="Answers questions using a local knowledge base")
)

# Agent class (Hugging Face-compatible)
class BasicAgent:
    def __init__(self):
        self.tool = rag_tool

    def __call__(self, question: str) -> str:
        print(f"🧠 RAG Agent received: {question}")
        response = self.tool.query_engine.query(question)
        return str(response)