Simple view

LLMs by themselves can only provide you with the information they have been trained on
LLMs with Tools can fetch information, perform computations, and more on fixed piece of information/data
LLMs with Environments can work alongside you in sync, fetching latest information and performing actions. Ex: Copilot of Code, Calender, Image editing etc.
Simple view

Advanced workflow

Tools are first class citizens
Tools are defined as classes to enable defining complex logic with memory.
Tools act on Environments
Inspired from Reinforcement Learning, provide snapshots of Environments to the LLM to work on.
Reactive framework
Bi-directional access between the tools and the chat instance.
Streaming made intuitive
Simple but powerful streaming API to build great user experiences.