Many of the latest LLMs now have the ability to use tools - functions you can write that the LLM can call if it needs more data, say from web search, or to execute a task, like sending an email. When combined with the ability to look up data from a vector database using unstructured inputs, that gives you inputs, outputs, memory, and a brain - everything I need to build a smart butler to make my life easier!

I'll go through each of these systems to explain how they work, and then I'll put them together using the POLM stack (that's Python, OpenAI, LlamaIndex, & MongoDB).

If I have time, I'll also talk about some of the other things LLMs can do (like writing and executing code) that could be useful for an AI agent.

If everything goes to plan, I'll show you a system that can schedule tasks, tell me interesting things that are happening,
and generally make me look like someone who know's what they're doing on a daily basis. Let's hope I don't accidentally unleash an unrestrained digital robot upon the Internet.