I noticed the new feature of llama-index: the event-driven workflow[1], which allows for building agents based on LLMs.

I think the concept of an event-driven workflow is cool, but I don't find the way llama-index implements it to be intuitive. Therefore, I wrote a simple prototype to demonstrate what I believe is the correct approach. It supports the following features with just a few lines of code:

- Multi-Recv: it supports an event to listen on a group of other events(or self), and triggered only when the group of events produced the results

- Parallel Execution: using python event loop

- Dynamic dispatch: it's a cool feature I think. You can use signals like goto or abort to change the event's return behaviors.

[1]:https://www.llamaindex.ai/blog/introducing-workflows-beta-a-...