Philosophy
Why there is no notification in TermRover?
It's deliberate. Agent notifications glue you to your phone and hand it your attention. Set and forget is the better workflow: you decide when to check in, and you stay in control.
You might ask why TermRover doesn’t buzz your phone when an agent finishes, or stops to ask a question, or hits an error. Other tools race to add exactly that. For TermRover it’s a deliberate choice, and I want to explain it, because underneath it’s a choice about who is in charge of your attention.
I built TermRover so you can work on the move, from wherever you happen to be. But working on the move is the opposite of being glued to a screen all day. The whole idea is that you decide when you want to work. Always.
Notifications invert the relationship
A notification is an interruption you didn’t ask for, scheduled by something other than you. The moment an agent can ping you, it has a claim on your attention. It decides when you look at your phone. You stop being the person directing the work and become the person on call for it.
That’s backwards. The agent works for me. I should decide when I check in, on my terms, when it suits whatever I’m doing. Not the other way around.
And the interruptions don’t even buy you much. An agent that pings you the second it has a question trains you to sit and wait for the ping, which is just babysitting with extra steps. You end up watching a progress bar through a notification shade.
Set and forget is a better way to work
The workflow I want is the one you’d use for anything that runs on its own. You start it, you walk away, you come back when you’re ready. The dryer doesn’t text you a question halfway through, and you don’t stand in front of it for the whole cycle. You start it and you get on with your life.
Long-running agent work is the same. Kick it off, put the phone in your pocket, do something else. When there’s a natural gap, you check in: read what happened, answer anything that’s waiting, point it at the next thing. The work batches into the moments you choose, instead of scattering itself across your whole day.
This only works if checking in is genuinely easy. That’s the other half of the design: tmux keeps every session alive while you’re away, and the session picker gets you back into any of them in a tap. Set and forget needs a frictionless way to come back, and that’s the part TermRover does invest in.
A small example
Picture a dinner out. The phone is face down on the table, and the person you’re with steps away for a minute. In that minute you flip the phone over, open TermRover, and nudge the agent: answer its question, give it the next instruction, lock the phone, put it back down. Ten seconds, on your initiative, in a gap you chose.
Now picture the same evening with notifications on. The agent buzzes during dinner. It buzzes again. You feel the phone go and you’re already half out of the conversation wondering what it wants. The work didn’t get any faster. You just got pulled out of your evening by a script.
The first version keeps you in control. The second hands control to your own tools. TermRover is built for the first one.
In control, by default
So there’s no agent-notification feature, and that’s intentional, not a gap on a roadmap. You decide when to look. Between those moments you’re free to be wherever you are and actually be there.
This is the same principle behind the rest of how I work with agents: reach anything from anywhere, but only when you choose to. The reaching is for your convenience. It was never meant to become a leash.