Big players already provide these functionalities.
Even on local, we anyway set OpenAI keys, so processing is also happening on cloud hosted models.
Why to unnecessary consume memory by using this feature?
In short, how the Agentic RAG Differs from ChatGPT or NotebookLM?