I'm a big tmux fan, but admittedly I've found myself switching between a lot of sessions recently and wishing it was a little bit more mouse friendly.
If you're a Mac user, and can forgive the shameless plug, you might find belfry.robgough.net useful. Connects to local and remote machines entirely through tmux, and libghostty - populating the sidebar from tmux session info, and optionally adds a little Claude visibility in there for good measure.
There's even an iOS/iPadOS version – though for the moment you'll need to build that yourself. Source is all on Github.
I wanted to mention that `zmx` is a beautiful piece of minimal software that basically has gotten rid of my tmux and zellij use. Uses a ghostty-based terminal emulator for each session, supports pretty much all of these features by leveraging ghostty. I truly don’t see the benefit of something like herdr over this minimal, robust setup. If I really wanted a lot of customization for AI assistants, I’d consider using Orca, but my current zmx approach works OOB and it’s solid. Great software!
From what I can figure it is actively calling itself a multiplexer, so pretty closely linked to something like tmux (one of the review videos on the page calls it a "tmux rewrite").
But they seem to be targeting very different scopes use wise.
I think if you use this to do stuff like, edit files, run htop, drop into an interactive shell, compile some code, my guess is you're gonna be fighting uphill and have a bad time.
On the other hand, if you're only using tmux so that you can have a bunch of terminal windows for agents. Then this is probably going to make your workflow a fair bit smoother.
What? Do y'all even try it before saying things like this?
I use it primarily as a replacement for tmux to do terminal stuff, not use agents. It's awesome. It worked out of the box, has actually good, working mouse support, quickly ships fixes, and the defaults are sane with obvious behaviors represented in an intuitive TUI.
I think that just enables xterm mouse support, and allows for various mouse key bindings in tmux. I think the deal is that it's turned on by default, and lots of the UI elements by default are configured with mouse bindings. Not just selecting and resizing panes, you can right click on a pane or a workspace and one of those display-popups pops up with a "rename" dialog box.
I think this is why it's described as "mouse first" terminal.
The example I saw of "mobile support" was connecting via a mobile terminal emulator. I don't see how that is any different than using tmux with the same mobile terminal emulator.
I used to work at BigTech in a role where I spent a lot of time reviewing new open source projects. It was always a fun game to guess which single person in which department tried out a piece of software that caused our logo to end up on the "as used by..." description on the tiniest of projects.
Fair haha. I'm no trademark lawyer, but I have to assume there's some wiggle room when you're not actively stealing their trademark and just referencing it... but still, the combined value of endorsements from Google/Nvidia/Amazon/ByteDance/Tencent/Alibaba/SalesForce/IBM etc has to be worth... a lot. That "not company endorsements" is rather load-bearing. I don't think I've seen anyone just say that, so maybe they're being a bit less bold haha.
You might as well say Herdr runs on 3 billion devices, go all in! 16 billion devices! Every human on earth installed Herdr twice! (source: study amongst users of herdr, n=5)
I get your point completely. I was (naively, I suppose) pretty surprised when I realized how low most people's threshold is to add a logo and claim a client/brand.
I've seen VCs very much encourage this (and shadier) behavior. There's a line between hustling / being an entrepreneur and actively trying to mislead people... clearly we don't all see that line being in the same place (:
Everyone does that. Some random startup "We're used by Google!". More often than not some random engineer signed up for a trial or some tiny department somewhere is using it for some internal thing
I was under the impression that "individual engineers ..." is basically implied always, if not just outright lies, I don't trust these listings very highly. Except when its clear that the companies are actually customers. Like; how do I even add my company to this list? Email them?
I've been using Herdr for running my AI agents and it has been really nice. I like that I can reconnect to it from multiple sessions (I have one up in a window on my desktop, and when I ssh in from my "after hours" laptop I can also attach to it there and continue one if I'm at my son's Dr appt or the like. I can also attach to it natively from my Mac over SSH transport as well as resuming a remote wezterm connection that is running herdr.
A few small downsides: I can't copy/paste in wezterm using the keyboard/vim keys because it is constantly drawing the screen and unselects my selection. The mouse drag in herdr works very well though. It'd also be nice if you could rebind key mappings in the UI, because I still haven't rebound the keys and am using the mouse.
The primary benefit I've gotten over just a straight tmux session is that there is a collapsable left "tab" bar that shows you your different workspaces, which you can relabel, and below that is a list of the agents you are running (claude code, codex, etc) along with their status (idle, blocked, working).
So I will start a workspace for each different thing I'm working on, label it "Studio Shed Packet", "Teapot game", "Mux experiment". Then in each one I run a Claude Code. Then I can see the status of my Claude Codes just by glancing at the sidebar, rather than having to switch between screens to see what is waiting next.
You can rename the tabs in tmux but not sure if there's a way to do dynamic status (I suspect there is but it's probably non trivial to get it to play nice with the harness)
Yes but someone else has put in the effort so I don't have to. I could go to the supermarket and buy a bunch of groceries and make food, or I could buy food from a restaurant.
I saw this trending on github yesterday and tried it and liking it so far.
What I like:
- familiar tmux like key binding (configurable) interface
- comes with all the tmux advantages like detach, ssh etc
- mouse
- nice sidebar
I prefer to manage my worktrees manually with a super simple script.
With the long waits for agents to do stuff I really don't see how one can get anything done without multitasking with multiple worktrees in parallel. So I'd want support for listing the worktrees and then have a list of agents within each worktree.
Emdash and Nimbalyst have this kind of UI. Unfortunately both of them want to manage the state of each worktree group themselves; I'm looking for something that just would just call git worktree directly so that I can switch more seamlessly between CLI and IDE/TUI..
Claude can easily do the merges I need (or myself manually for that matter). I guess all codebases and usecases are very different here and hard to give general advice.
But I do have many years of experience working in a larger team and it's the same problem there (just that people want to merge after some days of working). I'm not sure if AI changes the picture much vs working in a team.
Either way one has to plan ahead a bit and select tasks that are not going to trample on each other. I can typically imagine roughly what the code generated is going to be (at least what files will likely be involved in what way) and when selecting tasks to work on I take into account if it's going to likely cause conflicts.
In my experience if different branches work on different things, Claude have no issue doing "trivial" merges where you just ended up changing different aspects of the same lines. Of course, if two branches rewrite the same pieces of code there's a problem -- so don't do that..
I always rebase the worktree back to the source branch before merging, and resolve conflicts on the branch. I have a resolve conflicts skill and just say:
echo “resolve conflicts” | runpi
Where runpi is my pi -p wrapper. I’ve never had a regression from it, but it gives me a report at the end so I can double check the decisions if I need to.
The skill is basically don’t use automatic resolvers, err on the side of including both sides, refer to recent commits, missions and runfiles for context and in your report to me use real branch names not HEAD and incoming because I can never remember what those refer to.
git worktrees are managed by git, not the IDEs.
e.g. at agentastic.dev, we use git's worktree command to create them, and they should be portable to any other IDE or app.
I'm sure they're roughly equivalent. Parent is probably actually asking: is there native support for managing multiple checkouts/branches for parallel work (and I would add: with lifecycle hooks for create/teardown so I can have dedicated test databases etc).
The problem I have with this is if I need to switch branches to merge something. If a work tree for any branch involved exists, it prevents me. So, I have to go clone somewhere else, do whatever, then update everything. I really wish git didn't care about the other worktree folders, since it clearly doesn't anyways, since you can switch a worktree to and branch you want.
I thought for a minute that they had built agent support for all terminals, which would render obsolete https://terminai.app which I just released.
Luckily for me, this is on the other end of the spectrum. Maybe there would even be a use case for running herdr as the AI CLI inside of terminai.
> Each runs in its own real terminal, on a server that keeps it alive when you close the laptop.
How does one describe what's happening with stuff like this? Where a business tries to intercept people who are still learning the lay of some land, to get them to pay for something that they just haven't learned yet can be essentially free to them? Is there a word for it?
It sets up a bunch of panes for you and tracks your agents' status and everything is clickable. You could probably do much of the same by setting up tmux appropriately. Or you could use this, which is already set up.
I tried herdr today, and it's not bad. Better than my previous many terminal tabs. Copy-and-paste is wonky (and it's ridiculous that copying is documented in multiple places in the docs, but pasting is never mentioned). And when an agent is waiting for a shell command, it shows as "idle", which IMO is wrong. Still, seems OK so far, I'll stick with it for a while.
I've been working on Hyperia, which is very similar: https://github.com/deepbluedynamics/hyperia. Hyperia is a fork of Hyper Terminal. Both are open source. Competition is good for this space, I think, especially for the user. There's also cmux and Intelligent Terminal (by Microsoft).
I do separation of concerns with the agent orchestrator (Nemesis8): https://github.com/deepbluedynamics/nemesis8. That can be run with or without Hyperia. I do not suggest anyone run agents on their bare metal. Putting them in a container gets a lot of wins, especially around log aggregation. Working now on a Splunk/Loggly-like interface for searching logs, tool runs (useful in tuning a custom local MoE drafter) and full session suspend, stop, detach, and search. It also does single MCP tool installs for all agents. Nemesis also supports dynamic port exposure to the host metal, for testing agent builds inside their containers.
Hyperia has a lot of extra features as well that I have found personally useful:
- Sticky notes (search too)
- addressable panes in addressable tabs, tabs in windows, multiple windows
- full ACLs across panes, notes, tabs, windows
- Poke-a-pane to keep an agent going (any agent, not just CC which has a timer function)
- webpanes with markdown extraction, JavaScript injection
- directory pickers for people who find cd'ing to things confusing or those weary of typing nearly the same directory path over and over again in new terminals (not perfect, but I'm iterating on it)
- a built in agent loop (in the Rust sidecar) that allows using local models for tool calls (needs a trained drafter to make it viable) or using a local model for token maxxing (compresses reads of panes by frontier models)
- pane splits down/up/left/right and quick layouts.
As for whether it was "vibe coded" or not, or Herdr for that matter, I don't think that term is useful, other than for quick judgment. No, this is not a one-prompt project. I've spent 100s of hours on it, started out with Hyper, and did a crazy amount of planning on how to architect it. I have done systems architecture for a living before, and have a strong search background. People who hate on AI, and therfore projects done with AI, are threatened. Nothing more. That's why they shortcut with "AI slop" or "Vibecoded. Nope". That's just ignorance speaking from a standpoint of fear.
Slop, whether AI or human, is an effort problem: https://deepbluedynamics.com/blog/ai-slop-effort-problem. Looking at Herdr, it looks solid. Judge the product by it's outcomes, it's use, not whether or not AI wrote it or not. That's the moment we're in though, for now, so downvote or not. I don't care.
Cool that you're using Hyper Terminal as a starting point! What was the reason why you've added MCP to this? How is the intended use? I launch Claude Code on Hyperia and then Claude Code can open as many tabs as it wants inside of Hyperia? Or m I missing something
If you're a Mac user, and can forgive the shameless plug, you might find belfry.robgough.net useful. Connects to local and remote machines entirely through tmux, and libghostty - populating the sidebar from tmux session info, and optionally adds a little Claude visibility in there for good measure.
There's even an iOS/iPadOS version – though for the moment you'll need to build that yourself. Source is all on Github.
- everything is mouse clickable
- tmux style display-popups are used for friendly UI interactions everywhere
- it has a UI for agents running in panes, with a cool status (idle/working) display
- has opinionated defaults like automatic clipboard copy on mouse text select
- makes nested sessions easier & has default affordances for remote SSH attach
- is generally prettier
- uses display-popups for notifications
Otherwise it seems exactly like tmux.
So for a new worktree the agent can create a tab in zellij with multiple panes, properly named etc.
It works really nicely.
Although since I got the 52 inch Dell monitor I tend to just put everything visible at all once most of the time.
But they seem to be targeting very different scopes use wise.
I think if you use this to do stuff like, edit files, run htop, drop into an interactive shell, compile some code, my guess is you're gonna be fighting uphill and have a bad time.
On the other hand, if you're only using tmux so that you can have a bunch of terminal windows for agents. Then this is probably going to make your workflow a fair bit smoother.
I use it primarily as a replacement for tmux to do terminal stuff, not use agents. It's awesome. It worked out of the box, has actually good, working mouse support, quickly ships fixes, and the defaults are sane with obvious behaviors represented in an intuitive TUI.
I think this is why it's described as "mouse first" terminal.
You might as well say Herdr runs on 3 billion devices, go all in! 16 billion devices! Every human on earth installed Herdr twice! (source: study amongst users of herdr, n=5)
I've seen VCs very much encourage this (and shadier) behavior. There's a line between hustling / being an entrepreneur and actively trying to mislead people... clearly we don't all see that line being in the same place (:
A few small downsides: I can't copy/paste in wezterm using the keyboard/vim keys because it is constantly drawing the screen and unselects my selection. The mouse drag in herdr works very well though. It'd also be nice if you could rebind key mappings in the UI, because I still haven't rebound the keys and am using the mouse.
So I will start a workspace for each different thing I'm working on, label it "Studio Shed Packet", "Teapot game", "Mux experiment". Then in each one I run a Claude Code. Then I can see the status of my Claude Codes just by glancing at the sidebar, rather than having to switch between screens to see what is waiting next.
I've been using it around a week so far.
I prefer to manage my worktrees manually with a super simple script.
With the long waits for agents to do stuff I really don't see how one can get anything done without multitasking with multiple worktrees in parallel. So I'd want support for listing the worktrees and then have a list of agents within each worktree.
Emdash and Nimbalyst have this kind of UI. Unfortunately both of them want to manage the state of each worktree group themselves; I'm looking for something that just would just call git worktree directly so that I can switch more seamlessly between CLI and IDE/TUI..
https://gist.github.com/iaindooley/cc8a61a1ff0fe23526c850906...
You include a script .worktree in your repo that does any copying or symlinking to setup the target directory.
It also has a headless mode so that it does the worktree operations without the tmux which I use for executing pi -p prompts in worktrees.
How do you manage that? How do you successfully navigate complex merges using ai?
But I do have many years of experience working in a larger team and it's the same problem there (just that people want to merge after some days of working). I'm not sure if AI changes the picture much vs working in a team.
Either way one has to plan ahead a bit and select tasks that are not going to trample on each other. I can typically imagine roughly what the code generated is going to be (at least what files will likely be involved in what way) and when selecting tasks to work on I take into account if it's going to likely cause conflicts.
In my experience if different branches work on different things, Claude have no issue doing "trivial" merges where you just ended up changing different aspects of the same lines. Of course, if two branches rewrite the same pieces of code there's a problem -- so don't do that..
echo “resolve conflicts” | runpi
Where runpi is my pi -p wrapper. I’ve never had a regression from it, but it gives me a report at the end so I can double check the decisions if I need to.
The skill is basically don’t use automatic resolvers, err on the side of including both sides, refer to recent commits, missions and runfiles for context and in your report to me use real branch names not HEAD and incoming because I can never remember what those refer to.
But I think git worktrees are a bit more ergonomic, I don't have to think about local vs upstream there's just one place to push.
I like to organize my projects like this:
How does one describe what's happening with stuff like this? Where a business tries to intercept people who are still learning the lay of some land, to get them to pay for something that they just haven't learned yet can be essentially free to them? Is there a word for it?
Doesn't tmux and zellij do all of these things that 'herdr' does?
It sets up a bunch of panes for you and tracks your agents' status and everything is clickable. You could probably do much of the same by setting up tmux appropriately. Or you could use this, which is already set up.
I tried herdr today, and it's not bad. Better than my previous many terminal tabs. Copy-and-paste is wonky (and it's ridiculous that copying is documented in multiple places in the docs, but pasting is never mentioned). And when an agent is waiting for a shell command, it shows as "idle", which IMO is wrong. Still, seems OK so far, I'll stick with it for a while.
I do separation of concerns with the agent orchestrator (Nemesis8): https://github.com/deepbluedynamics/nemesis8. That can be run with or without Hyperia. I do not suggest anyone run agents on their bare metal. Putting them in a container gets a lot of wins, especially around log aggregation. Working now on a Splunk/Loggly-like interface for searching logs, tool runs (useful in tuning a custom local MoE drafter) and full session suspend, stop, detach, and search. It also does single MCP tool installs for all agents. Nemesis also supports dynamic port exposure to the host metal, for testing agent builds inside their containers.
Hyperia has a lot of extra features as well that I have found personally useful:
- Sticky notes (search too) - addressable panes in addressable tabs, tabs in windows, multiple windows - full ACLs across panes, notes, tabs, windows - Poke-a-pane to keep an agent going (any agent, not just CC which has a timer function) - webpanes with markdown extraction, JavaScript injection - directory pickers for people who find cd'ing to things confusing or those weary of typing nearly the same directory path over and over again in new terminals (not perfect, but I'm iterating on it) - a built in agent loop (in the Rust sidecar) that allows using local models for tool calls (needs a trained drafter to make it viable) or using a local model for token maxxing (compresses reads of panes by frontier models) - pane splits down/up/left/right and quick layouts.
As for whether it was "vibe coded" or not, or Herdr for that matter, I don't think that term is useful, other than for quick judgment. No, this is not a one-prompt project. I've spent 100s of hours on it, started out with Hyper, and did a crazy amount of planning on how to architect it. I have done systems architecture for a living before, and have a strong search background. People who hate on AI, and therfore projects done with AI, are threatened. Nothing more. That's why they shortcut with "AI slop" or "Vibecoded. Nope". That's just ignorance speaking from a standpoint of fear.
Slop, whether AI or human, is an effort problem: https://deepbluedynamics.com/blog/ai-slop-effort-problem. Looking at Herdr, it looks solid. Judge the product by it's outcomes, it's use, not whether or not AI wrote it or not. That's the moment we're in though, for now, so downvote or not. I don't care.