At some point, it's too much information for the human brain to process, even more than 3 agents is hard to manage and context switch. What is the point of managing herd of swarm of whatever I wonder?
It's something you pick up as a skill. I used to struggle with 3 and would genuinely exhaust me at the end of the day, now I can do 6 to 7 without too many problems.
Herdr has been my daily driver for about a month now. It cant auto-name sessions like every other multiplexer though. That one stupid feature is so valuable I'm considering moving on.
Thanks for mentioning it. Glad to see it's by @mitchellh. Excited! Personally I'm a fan of the inline textbased window(tab)/session style in Tmux and not native tab bars etc. but we'll see.
They have a plugin ecosystem, one of which is auto-named sessions. I imagine implementing your own wouldn't be too difficult either, if that one doesn't meet your standards of trust.
You could even just make a harness hook in Claude or Codex after installing the Herdr skill, telling the LLM itself to update names on start.
I've had no problem just asking an agent to do this. Just run !herdr --skill and ask any agent tab to rename all the sessions to something meaningful (use a cheap model if cost is an issue).
Or vibe code a script for this. I use pi so I just use a random cached session for this and rewind using /tree after (using local models so I don't care much about tokens).
Yeah, the way it's aesthetically framed like it's official immediate skeeves me out. I doubt it's intentional, given they clearly state "community" at the top, and more a case of poor judgement and AI look-aliking.
My setup is ghostty with herdr and this is so much better than using tmux or anything else. The one issue I have with herdr is its default key bindings are not familiar with anything I’ve used before. But it is easy to make Claude configure this - I found you can bind cmd-<key> sequences n addition to ctrl and alt modifiers.
The only thing I’m missing is viewing rendered markdown and mermaid charts as part of the agent outputs.
It looks amazing in the beginning but with work done on 5 remote machines herdr just gets laggy after couple of hours and basically becomes almost unsable.
Seems they should be making it more clear they are not associated with Herdr. Just as I use a minimal harness (Pi) I appreciate the minimal features of Herdr. Much of the features this adds, I prefer to bring my own. In many cases, I let my agents go and don't much look at Herdr.
The minimal decisions is partly what gains my trust. If Herdr were to come up with something like this, then I would give it more attention. Since it is not associated with Herdr, I'm no more likely to give it attention than any other the other overwhelming AI coding things that hit Github in too short of a period.
Thanks for mention they are not associated w/ Herdr.
I've banned myself from installing any more software for half a year. Otherwise, I'll never get rid of this anxiety caused by all the available options :)
19 comments
[ 0.33 ms ] story [ 31.1 ms ] threadYou could even just make a harness hook in Claude or Codex after installing the Herdr skill, telling the LLM itself to update names on start.
Heck, I think this works via the plugin system?
Or vibe code a script for this. I use pi so I just use a random cached session for this and rewind using /tree after (using local models so I don't care much about tokens).
Use at your own risk.
Feels wildly misleading.
The minimal decisions is partly what gains my trust. If Herdr were to come up with something like this, then I would give it more attention. Since it is not associated with Herdr, I'm no more likely to give it attention than any other the other overwhelming AI coding things that hit Github in too short of a period.
I've banned myself from installing any more software for half a year. Otherwise, I'll never get rid of this anxiety caused by all the available options :)