Eno Reyes

And so if you build an explicit structure into your compression strategy, hey, files are all going to go here.

Decisions are all going to go here.

That where the agent is currently operating is going to go right here.

And you also preserve some of the continuity in the conversation, right?

Maybe the last message or the last couple of messages should actually be preserved.

So the agent is able to pick back up without losing track.

That's huge.

Compression ratio is obviously the wrong strategy as well.

So like, it doesn't matter if you compress more, if it's worse, because you'll end up actually spending more tokens to get back to where you were.

And then you've lost the like 1% delta.

Yeah, totally.

I mean, I think that the, uh, so, so the question is, was that very long session because of the strategy?

Uh, yeah, totally.

Like technically this is, uh, this can go infinitely.

Uh, we have like a strategy at the very end.

If it were to accidentally run over by over compressing, uh, we have a way of like flushing, in which case you might see a minor decline at that very, very, very long end of a, of a potentially multi-day session.

Um,

But but really, this strategy allows you to run for a very long time.

I mean, we have customers that have quite literally run sessions that go well into the 80 million tokens like and so really the thought process here is

If the question is, can it recall the most precise detail about at the very beginning of the session, the second question you asked, very unlikely that it will be able to.