I was really pleased finding this last year, but I guess it's time to look for an alternative. I don't get why everything has to have AI shoved into it
Its also the Repo. There's a lot of AI-guided commits. I'm all for using AI in a reliable and safe environment, but letting Claude steer just leads to garbage
I took a look at the repo, but i didn’t see any garbage commits / evidence of sloppy vibe coding.
Care to elaborate? Also, don’t you trust that an author knows what they’re doing with AI in the same way as trusting them with their regular code writing skills?
I looked at the repo and couldn’t even find an example, so it can’t be that many of their commits. But also: this is ridiculous. Whether the commit appears as done by Claude or not is a setting you can change. If they turned it off, you’d never even notice.
These are great developers and they’ve built an incredible tool. I use it a hundred times a day. It is very odd and dogmatic to think that because you saw a commit authored by Claude, whatever skills and qualities let them build something so good are now being thrown out.
Hmm might be great for some. I’m a Unix philosophy guy, one tool for one job. So far atuin was fine to be a better search history. Now it might be time to look for simpler alternative. Any suggestions? (I’m on zsh)
not zsh .. plugging my bash script (and gnome task bar UI) - to start a gnome terminal with a different named history file. https://github.com/appsmatics/gtsh-hist
I have to ask -- why? Atuin has not gotten any worse at its core history search functionality. All of the new features are entirely opt-in. Why switch?
If all you need is basic splits, sessions, and some simple templates/layouts (and like the convenience of knowing that tmux is widely available, and often installed by default) then you're fine to stay on tmux.
Zellij can do things like floating windows, contextual keybinding guidance (helps learn everything that can be done), and a more complex layout schema. You can disable all the UI eye-candy and switch to tmux-style bindings too.
It's worth trying out. I use both so that I can still function on systems without it.
I'm still looking forward to being able to only remember a command for a specific time. I currently block sensitive commands, which completely destroys the ability to just press the up arrow key to quickly edit the command. If we had like a timeout of 1 minute for sensitive commands, we could edit them and still make sure they are not persistent
Atuin AI sounds like a useful addition. The page suggests they're probably using hosted models:
We use the latest frontier models, which already do a good job of generating commands using well-known binaries and CLIs. On top of that, we integrate a dataset powered by man pages and command outputs to ensure you get the correct command first.
This is great, but does it mean we'll need to log in somehow? It doesn't seem reasonable to expect the project maintainers to pay for the tokens.
EDIT: I was unaware of Atuin's 'hub' which does things like sync your shell history across computers. I think they use the same sign-in as they already use for that: https://hub.atuin.sh/register
> On top of that, we integrate a dataset powered by man pages and command outputs to ensure you get the correct command first.
Also makes it sound like they're "providing that dataset", rather than generating that from the users computer. Wouldn't that mean it's potentially a mismatch between various versions of the software available? Not to mention some OSes will have a different version of some software available compared to others, how does it deal with those situations if they're shipping a dataset?
The Data Privacy paragraph would suggest otherwise.
> By default, Atuin AI knows nothing about your machine, other than the operating system and shell. This is the bare minimum required to generate a decent shell command.
> It will soon be able to ask you for access to more data - such as the current directory path, contents, git status, etc - but you must give permission first. This will happen in a similar way to existing agents, and be configurable to an even finer degree in your config file.
And my take! A fork of fish where any command that starts with > or a capital letter is fed to $fish_llm_command: https://github.com/breuleux/fish-shell. With Claude's help, that took all of 30 minutes to make.
Why does every tool on the face of earth try to add AI features ? Good tools are simple and orthogonal. If you want AI, there's already plenty of other tools doing it probably better.
I'm overall fairly disappointed by this announcement. This IMHO doesn't bode well
About the "ai", the announcement is very vague. Is this incorporating a local model on device, something running on your infrastructure or a third party model like Claude? Because to me nowadays adding AI on anything usually means higher running costs equals sooner or latter enshittification.
Hey, thanks for responding. I guess I used the prototype then. Definitely don't remember anyone saying "this is a prototype" at the time, so I took the product at face value, and part of the reason I chose it was the fzf support.
I'm sure I recall some unhappy GitHub issues about the shift away...
And the algorithm isn't the value prop for me, not by a long shot. fzf's customizability takes the cake. And now the overall product is way too big and feature-ful for me. I want simple, unix-y software that clicks together like Lego.
You should be proud of the project's success for sure, it's just not for me!
Yep, out of precaution i've never used their sync infrastructure, which I guess was reasonably cheap to run, but the moment you add LLMs to the mix it is obvious that they are in for the free VC money and are soon going to need a lot of investment to keep the lights on.
The thing I found frustrating was I wanted to merge changes with just files without their sync server (i.e. just import this other atuin sqlite dB) so I raised a PR to support that.
They closed it (which is fine) but there is no offline migrate alternative.
It's a shame, and fair enough, their project, but I don't think my wishes and the projects are very aligned.
I keep half meaning to move back to zsh-histdb (I think that's what it was called) but haven't found an impetus to.
I'll probably check if there's a file based sync option next time I switch machines and decide then.
Care to elaborate? Also, don’t you trust that an author knows what they’re doing with AI in the same way as trusting them with their regular code writing skills?
These are great developers and they’ve built an incredible tool. I use it a hundred times a day. It is very odd and dogmatic to think that because you saw a commit authored by Claude, whatever skills and qualities let them build something so good are now being thrown out.
Losing some of the scrollback was a minor nuisance that I kind of lived with until now.
[0]https://github.com/junegunn/fzf
I love tmux and haven't had a reason to switch for a while, but have heard these new Rust based terminal tooling get really popular.
Zellij can do things like floating windows, contextual keybinding guidance (helps learn everything that can be done), and a more complex layout schema. You can disable all the UI eye-candy and switch to tmux-style bindings too.
It's worth trying out. I use both so that I can still function on systems without it.
I switched away from tmux a year or so ago due to one crash I kept getting, but thinking of going back. Really miss the simplicity.
EDIT: I was unaware of Atuin's 'hub' which does things like sync your shell history across computers. I think they use the same sign-in as they already use for that: https://hub.atuin.sh/register
> On top of that, we integrate a dataset powered by man pages and command outputs to ensure you get the correct command first.
Also makes it sound like they're "providing that dataset", rather than generating that from the users computer. Wouldn't that mean it's potentially a mismatch between various versions of the software available? Not to mention some OSes will have a different version of some software available compared to others, how does it deal with those situations if they're shipping a dataset?
"get the correct command first" and "shipping a [external] dataset" are incompatible.
> By default, Atuin AI knows nothing about your machine, other than the operating system and shell. This is the bare minimum required to generate a decent shell command.
> It will soon be able to ask you for access to more data - such as the current directory path, contents, git status, etc - but you must give permission first. This will happen in a similar way to existing agents, and be configurable to an even finer degree in your config file.
Excited to use atuin ai though, especially as it gets more features.
I'm overall fairly disappointed by this announcement. This IMHO doesn't bode well
This release actually adds support for nucleo, which matches with the same algorithm as fzf and was a common request
I'm sure I recall some unhappy GitHub issues about the shift away...
And the algorithm isn't the value prop for me, not by a long shot. fzf's customizability takes the cake. And now the overall product is way too big and feature-ful for me. I want simple, unix-y software that clicks together like Lego.
You should be proud of the project's success for sure, it's just not for me!
Why does this happen mostly?
They closed it (which is fine) but there is no offline migrate alternative.
It's a shame, and fair enough, their project, but I don't think my wishes and the projects are very aligned.
I keep half meaning to move back to zsh-histdb (I think that's what it was called) but haven't found an impetus to.
I'll probably check if there's a file based sync option next time I switch machines and decide then.