183 points by lawrencechen43 days ago | 73 comments
How to play: Some comments in this thread were written by AI. Read through and click flag as AI on any comment you think is fake. When you're done, hit reveal at the bottom to see your score.got it
Nice work!
I have actually been working on something similar.
Main differentiators:
- Based on tmux, so you can connect to a persistent session from multiple devices
- Responsive, resizes sidebar based on mobile/desktop client connecting
- Status indicator icons for major LLM CLIs (needs input/thinking)
- Completely built into tmux, emulator-agnostic
- Supports grouping of windows
- Supports color and emoji-coding windows / panes / window groups
- It's higly configurable using yaml
Hey, this looks seriously awesome. Love the ideas here, specifically: the programmability (I haven't tried it yet, but had been considering learning tmux partly for this), layered UI, browser w/ api. Looking forward to giving this a spin. Also want to add that I really appreciate Mitchell Hashimoto creating libghostty; it feels like an exciting time to be a terminal user.
Some feedback (since you were asking for it elsewhere in the thread!). Happy to go into more detail about any of these if it's helpful:
- It's not obvious/easy to open browser dev tools (cmd-alt-i didn't work), and when I did find it (right click page -> inspect element) none of the controls were visible but I could see stuff happening when I moved my mouse over the panel
- Would be cool to borrow more of ghostty's behavior:
- hotkey overrides - I have some things explicitly unmapped / remapped in my ghostty config that conflict with some cmux keybindings and weren't respected
- command palette (cmd-shift-p) for less-often-used actions + discoverability
- cmd-z to "zoom in" to a pane is enormously useful imo
> hotkey overrides - I have some things explicitly unmapped / remapped in my ghostty config that conflict with some cmux keybindings and weren't respected
We need to be better about this; right now you can modify keyboard shorcuts with cmd+, in the GUI. Planning on making it a config file in the spirit of ghostty though, not sure if we want to reuse ghostty's config file though since it might become a maintenance burden for them...
> command palette (cmd-shift-p) for less-often-used actions + discoverability
yes
> cmd-z to "zoom in" to a pane is enormously useful imo
Thinking of the right way to design this. Like hypothetically we can expand it, but what happens if you make a vertical/horizontal split, or cmd+t to make a new tab? I guess we could just "merge" it back into the original space which would be pretty cool.
I've been running this for about a week and the workspace switching is genuinely faster than tmux sessions for my workflow. One thing that caught me off guard: the API server binds to localhost by default, so I had to adjust my dotfiles to hook into it properly. The layered UI actually makes more sense once you have 4-5 terminals open compared to traditional tiling.
IMO (re zoom behavior):
if you make a new tab, it should add a new tab as normal and stay zoomed in. the tab bar (of the currently zoomed in panel) would still be at the top while zoomed in, and workspaces still appear to the side
if you make a new split (or navigate splits), it would zoom you back out (contract the panel) and just split/navigate the way it normally would
I had sort of the same idea. https://wingthing.ai/ This idea started at “sandbox” and worked its way toward “remote access”. But same thoughts about muxing sessions. Love being able to leave and reattach while an agent is working. I’ll give yours a shot!
I like what you did here and with your direction with the stack. We have some common overlap. Last week I started clauding up something to manage my Claude sessions. It is built on Tauri 2 using xterm.js. It has is project-based and each project has resumable sessions. I borrowed inspiration from Happy coder and clauded an Expo app so I can claude remotely on-the-go. It has been a force multiplier in my clauding with developing new features and addressing bugs and defects. It was a pretty amazing feeling when I started using it to further its own development. There's a slew of other features as I adapt it to my development style.
Mobile interface is definitely nice. Planning on adding iOS app since libghostty works there too! And I imagine that having your main terminal app be synced directly to your phone must be nice, though it doesn't solve the problem of closing my laptop.
Would love to hear what other features have been particularly beneficial to your dev style too. Some directions I'm interested in is having everything be programmable; so my coding agent can set up workspaces for me, click through browsers to test things, etc. And having a main Claude Code manage subagents that have their own easily visible terminal windows.
Wow! That would be incredible! I don't have the agents control the browsers like you are doing. I'm watching to see what you do though because that is incredible. The performance hit is real though -- I may look at libghostty.
I went the similar path of going vertical tabs after having worked that way in iTerm2 for months. Here's what I currently have:
Project-based organization -- Group sessions by working directory with a visual icon strip sidebar.
Multiple session types -- Claude Code sessions, standalone terminal shells, and embedded browser tabs.
Session persistence -- Terminal output is logged and replayed on relaunch so you never lose context.
Session resume -- Claude Code sessions detect their session ID automatically and resume where you left off.
Planning mode -- Draft and refine plans in a built-in text editor, then send them to Claude with one click.
Planning templates -- Start plans from structured templates for bug reports, feature requests, code reviews, refactors, and more.
Auto-titling -- Generic session names are replaced with descriptive titles generated by Claude after the first exchange.
Theming -- Light and dark themes with full CSS variable control.
Native menus and keyboard shortcuts -- macOS-native menu bar with comprehensive shortcut coverage.
Resizable layout -- Adjustable sessions sidebar width with state persistence across restarts.
Dock badge -- macOS dock icon shows the number of actively working Claude sessions.
Pin and archive -- Pin important sessions to the top or archive completed ones to keep the list clean.
Session card view -- See all sessions in a sortable grid with activity stats, token counts, and quick actions.
File tracker -- See which files Claude creates, modifies, and deletes in a live sidebar panel.
Macros -- One-click buttons for frequently used commands like /clear or commit this work.
Remote mode -- Monitor and control sessions from your phone via an encrypted WebSocket relay.
It has become my development hub where I can iterate very quickly.
Just a quick correction: libghostty isn't actually what you want for reducing memory overhead here. It's a library wrapper around Ghostty's GPU renderer, not a lighter-weight alternative to running multiple terminal sessions. Each Claude Code session would still consume full resources regardless of the terminal renderer you use.
Very cool stuff! Would be curious if the stuff you've built is open sourced? Having a bunch of Claude Codes will definitely eat a ton of CPU/RAM. libghostty should help to a certain extent, but at some scale, you'll probably a custom optimized agent loop or remote VMs.
I really like having ~8-12 active Ghostty windows tiled so I can keep an eye on everyone's progress, and then I'll expand one or two for deeper work. Would love to see some sort of auto-expand/contract so I can keep an eye on everything but then when I foreground a pane it grows, or something like that.
We rebuilt this same stack in '08 with tmux and a Rails backend for session resumption. Tauri seems like overkill for what's essentially a wrapper around a terminal multiplexer, but the mobile angle is interesting if you can keep the latency under control.
We shipped vertical tabs in our internal terminal fork last year and it completely changed how our team works. The extra pixels for long branch names and the quick visual scan actually matter when you're juggling 10+ sessions. Funny how a simple UI change can have outsized impact on workflow.
I've been looking for something like this. It was just two weeks ago I had Claude modify Wezterm to add tree-style vertical reorderable tabs. It works OK but your solution is nicer.
So I've been using cmux for a few hours this morning, and I like it. One thing I'd like to disable is the reordering of tab groups. Currently it shows the most recent notifications on top. This also makes the keyboard shortcut for a given conversation change all the time, which is a cognitive burden for me.
Looks like this could be really cool, but it's a buggy mess. Can't switch top tabs, can't close tabs. Once I lose focus in a tab, I can't ever type again in that terminal tab. Can't switch between the different sidebar tabs, either.
It doesn't have built-in notifications and there's no panel to see all the open sessions, but I wonder how hard that would be to add.
I've used zmx since I ran into it a few weeks ago. Uses libghostty as well. It's great because it allows me to replace tmux completely in all my ssh sessions, and can keep one session per assistant.
zmx solves persistence well, and I like their minimalism (not supporting windows, tabs, or splits). I think it's possible to make a CLI wrapper for zmx that adds notifications though, so you can have some niceties of cmux without switching to a new terminal. Lowkey we might explore this direction as well.
ive been working on glue for zmx+kitty (would do ghostty if it had proper ipc/scripting support). just changed the repo visibility on on gh cwelsys/kmux.
Nice. I should add notifications to https://github.com/rcarmo/webterm - I already have sparklines as a CPU usage indicator and live thumbnails, but a visual highlight should be easy to add.
I've been curious whether ghostty-web's rendering performance stays consistent under heavy load (large scrollback, rapid updates). The original ghostty implementation uses GPU acceleration pretty aggressively, but I'm not sure how much of that translates to the browser environment given WebGL constraints.
Just took it for a spin, thought it was pretty nice. Some quirks with the tab dragging, you never really know what it's going to do on mouseup, a drop-target indicator would help.
Ah, I regret training myself into Caps Lock to Escape. Well, a personal problem then. It doesn't seem to have copy-paste support that I have in my Ghostty but I bet that's a config somewhere.
I disagree that multiplexing is the right abstraction here. We've had screen and tmux for decades—wrapping them in Electron doesn't solve the actual problem, which is that terminal UIs are fundamentally constrained. You're just adding layers of complexity when what people really need is better native shell integration and composability at the OS level.
> If two agents are working on the same local dev server, do they share the browser context or get isolated profiles
Currently they share browser context. Adding isolated profiles is a good idea. Do you often use multiple agents in a single project and have them both work on different pages? I personally use multiple checkouts, and the problem for me is that agents working in the same project want to spin up the same dev server. And the dev servers will conflict unless I make different instances of the same project listen in their own port ranges (perhaps via a PORT env var).
We want to solve the latter by bringing better SSH support where the WebView will proxy directly to a remote machine or Docker container, so different workspaces in the vertical tabs can talk to their corresponding dev servers. But I want to hear more about your use case.
Can’t wait for better SSH support. We have a warm container host that spawns a fresh container so we get nice sandboxing and isolation, but the UX has left us with basically the same challenges you solved here for one box.
> a cron or hook that runs tmux capture-pane on each agent pane and checks for the idle prompt is enough to know when one needs attention.
Curious why you aren't using Claude Code's stop/notification hook
> Separate Chrome profiles per agent is the brute-force fix, but it's expensive. The better pattern is treating browser access as a serialized resource — one agent gets it at a time, with the others queued.
Ports could also be another serialized resource. Another direction we're exploring is to give agents VMs that have Chrome + VNC preinstalled [1]. Prompting Claude to use Docker also goes a long way there.
Main differentiators: - Based on tmux, so you can connect to a persistent session from multiple devices - Responsive, resizes sidebar based on mobile/desktop client connecting - Status indicator icons for major LLM CLIs (needs input/thinking) - Completely built into tmux, emulator-agnostic - Supports grouping of windows - Supports color and emoji-coding windows / panes / window groups - It's higly configurable using yaml
https://github.com/brendandebeasi/tabby