I am almost done.
This is a month of work, from old proof of concept I did last year.
It is well, kinda calendar agent, besides that it cannot read calendar. And cannot reschedule, or delete events, or do anything at all.
As a control freak, I am thoroughly freaked out if LLM gonna move my things around.
What, LLM, so called AI is useless? By no means. EventFlow uses LLM and cannot NOT use it, by design.
But, power of LLM/AI is used safely, for very specific purpose only, without any ability to change anything. Just observe, and propose neat calendar events from observed. Nothing else, no chats, no agents, no chatbots, none of this stuff.
Few words about development. I will thank few more, but for now.. Without GPT I wouldn’t prevail.. this fast. I mean, i would eventually get here, but.. I am scared to think, how long would it take.
I already made post about my #cogomp technique, which transfers a lot of daunting work to LLM without transfering almost any responsibility to LLM. Ie development of EventFlow used same principles as EventFlow itself.
To think of it, though.. Two prompts from EventFlow are together 3 A4 pages at max.
EventFlow cogomp seed prompt takes atm about 17 pages, and turns GPT into a very useful assistant.
So no, this is not unmaintainable vibecode, chaotic and terrible. This is well structured, well factored and carefully built project, I use myself.
In a fact, I used proof of concept BEFORE it turned to be public project, for simple reason – I never found exactly same functionality. Perhaps Hermes or Openclaw with restrictive settings could emulate EventFlow correctly, but.. Trajectories for Hermes and Openclaw would be longer, much longer than trajectory of EventFlow which takes only one step, and has no tools, besides access to realtime.
Today is day 10 of 14 day closing setting, and I am curious how well it will fare, and how much people will also find it useful, as I do myself.
Leave a Reply