I'll stick with tmux, thanks. I'd take ISC and slow over corporate and fast any day.
Dev who gave HashiCorp its name returns with a faster terminal multiplexer
HashiCorp cofounder Mitchell Hashimoto has launched a new startup to build a server-side terminal multiplexer that allows programmers to switch between multiple persistent work sessions. Superlogical promises its terminal multiplexer will be faster and keep better track of user jobs. And it could be instrumental to building …
COMMENTS
-
-
Saturday 1st August 2026 14:14 GMT Kurgan
NOO!!! You don't understand!!!
This needs a GPU (which is super expensive), is made for agentic AI (which is expensive and useless), and has this thing that state is maintaned on the server and somehow sent to the client, like, you know, IBM 5250 terminals from 1977.
We are well over the "reinventing the wheel" phase. We are reinventing a square wheel made of stone, to be used in a boat instead of on a road. Expensive, proprietary, utterly complex, and totally not needed because cheap, free solutions already exist, which are screen, tmux, mosh, and tssh.
But we need it for agentic AI. We absolutely need it. (not)
-
-
Friday 31st July 2026 22:11 GMT Jamie Jones
As well as agreeing with the above comments, I don't want a terminal emulator to render characters locally that haven't been processed by the server. I remember battling with old dumb terminals that did that, it was a nightmare.
Also, I recently hit the 100 concurrent session limit in "screen". It was still fast and responsive.
And latency to servers is better than ever these days!
-
Wednesday 5th August 2026 14:51 GMT CrazyOldCatMan
I remember battling with old dumb terminals that did that, it was a nightmare
Two modes:
Local terminal just echoes the local input. Press enter, string gets sent over X.25 (or other slow, old serial comms protocol) to far end. Sometimes it survives intact, sometimes doesn't. Forget anything interactive. (Been there, done that)
Local terminal send everything to the other end, waits for acknowledgement and echo. Single character keypress takes ages to perform. Mistakes get propagated, slowly and corrections take even longer because, by this time, you've got so frustrated with the whole thing that you are not waiting for the echo and carry on mashing keys, disrupting the local keyboard buffer.
Been there, done that. Hated it even more than mode one. At least with mode one you could (with most terminals anyway) make sure that the entry is correct at local-edit speed *before* you send it.
-
-
Friday 31st July 2026 22:15 GMT DS999
I used screen years ago
Running a terminal emulator on a souped up Mac Plus, with the sessions running on 50 MHz PA-RISC CPUs, over a 9600 bps modem.
That was pretty snappy, so I'm not sure what problem they are trying to solve in a reality 30+ years hence where my watch absolutely blows the doors off that PA-RISC workstation in every metric, let alone the Mac Plus!
I submitted a few fixes/improvements for screen back in the day, so when I saw this article title all I could think was "what's wrong with screen" lol
-
-
Saturday 1st August 2026 02:26 GMT BoHu
Re: Hashimoto Implements Microsoft Recall for Terminal Sessions
Yeah! Can't wait to see this dishwasher multiplexer precariously balancing its stochastic heft one full three-legged stool level above the kitchen sink ... it'll right bring the surprise and excitment of 3-ring circus clowning straight back into IT, where it's always belonged! (or not?) The kids will love it ... ;)
-
-
Saturday 1st August 2026 08:37 GMT TJ1
"GPU acceleration"
I had to dig through the documentation to figure this out for sure. There are several implicit prerequisites that are not-mentioned:
1. It is for local-only sessions (not remote sessions over SSH)
2. Must be using a graphical windowing system of some form
3. Must have a GPU
So for a typical use-case of tmux or screen where one SSH's to a remote (often headless - no GPU) host where the multiplexer is executing this won't help - won't even work.
-
Saturday 1st August 2026 08:51 GMT Rich 2
I’m confused
“It does require that every connecting client be a very smart, high-functioning, compliant client,” such as one built on libghostty, Hashimoto said. But this approach sounds better than dispatching streams of raw ASCII hither and yon
So, replacing a (relatively) simple ASCII terminal client (like ssh) with something that is “very smart [and] high functioning” is a good (“sounds better”) thing?
Maybe it’s an editorial mistake - maybe the sentence is the wrong way round?
As others have said above, I’ll stick with tmux thanks
-
Saturday 1st August 2026 12:19 GMT Annoyed Grunt
Yeh im in the same boat, dont get the use case for it.
If its the “must be agent aware” thing. then tmux can do this easily enough and if not then herdr is probably a better choice.
Otherwise, i’ve never heard anyone moan about tmux or screen nuking cpu or being slow.
Mind you I dont get ghostty either. Iterm2 is fine and doesnt need server side profiles to be installed to display text properly.
Seems like solving a problem that doesn’t need to be solved.
-
Sunday 2nd August 2026 04:38 GMT Jou (Mxyzptlk)
Next ist Ghospix
Ghospix:So you can run multiple pixel sessions!
Ghoswin: So you can run multiple graphical application sessions!
Ghosaud: So you can have multiple audio stream sessions!
Ghoswin²: So you can have multiple graphical-audio application session in parallel on one screen, and have multiple screens!
So within the next 20 years it will be somewhere between X-Windows, NT 4.0 Terminal Server Edition (and its 'improved' follow-ups), and Citrix/Parallels. Of course with Agentic AI!
Then:
Ghosframe²: New name 'cause Microsoft sued and won the lawsuit over "win".
Ghosbox²: New name 'cause Shittrix still has a © of "Metaframe", and too similar.
Ghosx²: New name 'cause X is cooler!
Ghosy³: Spoken "GhosWhyCube", 'cause why?
-
Sunday 2nd August 2026 16:42 GMT Crypto Monad
Sounds rather like mosh
Mosh lets the application write into a remote text screen buffer, and then syncs it to the client. The client can disconnect and reconnect, even on a different IP address, and the screen refreshes and continues where it was.
If the mosh client notices keystrokes being echoed then it starts doing speculative local echo - which can be undone automatically with the next screen refresh. This means that it feels very fast for interactive use on long-latency connections. Also, if the server splurges large amounts of data to the console, you can hit ctrl-C and it finishes immediately, because you're not having to wait for the whole stream to catch up.
The main downside of mosh is that the remote screen buffer doesn't support scrollback, so you're basically forced to use something like screen or tmux as well - in which case you might as well just use that instead. The feature request has been open for 14 years.
As far as I know, mosh does not support GPUs or AI. Thankfully.
-
Sunday 2nd August 2026 17:53 GMT MrReynolds2U
The new new new
It feels like this is something with VC ambitions, rather than genuinely commercial.
I also feel like many companies continually have to buy or invest in the new tech, or else someone will stop and ask whether the current tech was a good investment. So there's plenty of companies in the industry that rebadge or create the new shiny to enable this behaviour. Meanwhile, laughing all the way to the bank.
-
Sunday 2nd August 2026 18:58 GMT MultipleMonomials
Not that new of an idea
The idea of running a "terminal emulator" server-side is not a new one -- mosh (Mobile Shell) has been doing it for over a decade already. I imagine that using mosh + screen/tmux would get one pretty close to what they are trying to make here, but it's free and already available.