Codex on a Mac Mini, Controlled From My Phone

I've started playing around with using the ChatGPT iPhone app as a way to talk to Codex running on my Mac Mini.

The setup is simple enough. Codex is always open on the Mac Mini in my office, and that machine is always on. The funny bit is I barely ever actively sit at it. The screens are connected to the Mac Mini on one input, but most of the time I'm bouncing between my work laptop and personal laptop instead. I rarely bother switching over unless I actually need to (pretty much just updates at this stage, should probably automate that more).

That means the iPhone app is less interesting as a "mobile AI app" thing and more interesting as a remote control for a machine that's already there and already running.

So far, that's been properly useful for small jobs.

Nothing especially exciting yet. Mostly things like updating my personal wiki, syncing repos across devices, and nudging along little bits of admin that I'd otherwise forget about until later. It's the sort of stuff where I don't want to go and sit down at a desk for it, but I do want it done.

That's the part I like most at the minute. It changes the threshold for whether I bother doing something.

If Codex is already open on a machine at home, and I can kick off a quick task from my phone while I'm downstairs or out somewhere, a lot more little maintenance jobs start to feel worth doing. Not in a cursed "AI runs my life now" way. More in a "I may as well get this out of the way now" way.

The app is not flawless though.

I've had a fair few moments where it seems to stall trying to establish the connection back to the Mac Mini. Sometimes even really basic stuff, like asking it to do a git pull, seems to get stuck for a bit. You also see a lot of SSH-looking stuff in the agent's thinking when it's trying to sort itself out, which does not exactly fill you with confidence when all you asked for was something dead simple.

To be fair, it has usually recovered. It can be a bit awkward, then eventually sort itself out, and the notifications help because you can at least see that something is happening rather than it just vanishing into the void.

So I wouldn't describe it as smooth. I would describe it as surprisingly usable, with occasional bits of faff.

That feels about right for this kind of thing at the moment.

When it works, it's good because it makes an always-on Codex setup feel genuinely available. Not theoretically available. Actually available. I don't have to be in front of the machine that Codex lives on. I just need that machine to stay alive and reachable.

I think that's the real point here. The Mac Mini matters more than the phone app.

If Codex only existed on whichever laptop I happened to be using, I don't think I'd care much about doing this from my phone. But when there's a stable machine at home that's always on, always signed in, and always ready to go, the phone app starts making a lot more sense. It's not replacing a computer. It's more like a remote terminal with better manners.

Right now I wouldn't trust it with anything important, or anything fiddly, or honestly anything where I'd be annoyed if it hung halfway through. But for low-stakes tasks, it already feels useful enough that I'll probably keep using it.

That's roughly where I've landed so far. It's a decent way to poke Codex from my phone, provided you're happy with a bit of weirdness and the occasional pause while it remembers how to talk to its own machine.

Early days, but promising.