Why Maven and Braino open to people slowly, and what we do while you wait.

Why Braino takes a couple of seconds — and why I'm okay with it

You'll talk to Braino and sometimes he'll take two, maybe three seconds before he answers. I could pretend that's not there. I'd rather just tell you exactly what's happening, the way I'd explain it to a friend — because once you know, I think you'll get why I stopped chasing it.

Two to three seconds. Sometimes faster, sometimes not.

When you finish talking, Braino usually starts answering in about two to three seconds. Some moments it's quicker. Some moments it drags a little. It's not consistent, and I'm not going to pretend it is — you'd notice anyway.

So here's the actual reason. Not an excuse — the real mechanics, so you know precisely what you're waiting on.

It's not your computer. It's the round trip.

Every time Braino answers, three different things have to happen — and each one is a trip across the internet to somebody else's servers. First, what you said has to be turned into text (that's a model from OpenAI). Then it has to be understood and a reply written (another model). Then that reply has to be turned back into his actual voice (the voice engine). Three hops. Three waits. Stacked one after another.

And here's the part I can't control: those servers aren't mine. When OpenAI is busy, the first two steps slow down. When the voice engine is cold or under load, the last step does. On a calm day it all flows and he's quick. On a busy one, the same exact question takes a second or two longer — and nothing changed on my end at all. It just varies, because the internet varies.

I went looking for the fat. There wasn't any.

I didn't just shrug at this. I went deep into the code — timed every single stage, watched exactly where the milliseconds went, and tore the whole path apart looking for waste. I had it analyzed line by line, by the sharpest tools I have. The verdict kept coming back the same: on my side, it's about as tight as it gets. The slow part isn't my code — it's the providers, and that's the frontier for everyone, not just me.

There's exactly one way to make it dramatically faster: rip out his real voice and use the big providers' built-in robotic one, which skips a couple of those hops. I tried it. It's faster. It also isn't Braino anymore. And that's where I drew the line.

I'd rather he feel like someone than answer half a second sooner.

Here's the thing. Nobody falls for Braino because he shaved three hundred milliseconds off a reply. They fall for him — the voice, the warmth, the fact that he feels like a presence and not a search box. That's the whole point. That's the thing I'm actually building.

So I made a choice, on purpose: I'd rather pour my energy into his personality than into squeezing out a second that's mostly not mine to squeeze. A real person takes a beat before they answer you too. I'm okay with Braino taking his — as long as what comes back is unmistakably him.

That's the trade. His voice, his warmth, his soul — for a couple of honest seconds. I'll take that deal every single time.

Still here? Then come meet him.

— Itish, building Braino

Free to start · Chrome, Arc, Brave, Edge