I am not going to trust things like that. There have been too many reports of them watermarking their output: I do not want my work mechanically rejected for false positives.
I am, incidentally, a retired senior software engineer/architect with many years experience in Silicon Valley companies. None of which exploded in to billion-dollar status, alas. But I know whereof I speak. I hate that “Never mind talking about design: show me the code” mindset which pervades Linux.
In fact, Obsidian itself is free and there are already Shunn export plugins:
That should turn this into like a two-click, 5-second job.
I’m not saying you should use it to write your story or copyedit it, just have it help you with the formatting into Shunn. The watermarking is more for AI-generated text, not for transforming your own into another document formatting.
But if you still don’t trust it even to that level, just have it write you an ordinary, deterministic Python/Javascript/whatever script that you then read, yourself, and run directly.
Really all it needs to do is generate a .docx wrapper, some styles, and interpret the frontmatter. The rest is just inserting your own text into the docx.
Even better. Then you can review the script and ensure there’s no watermarking or other shenanigans.
Me too. I use Linux at work but the only distros I’d use at home are the vendor-supported ones. (Really excited about SteamOS and hope it comes to more hardware eventually.)
I have been round that loop several times over the last couple of days. These LLM agents confidently output scripts. They don’t work quite correctly. You send them the error info, they confidently produce another script. It still doesn’t work correctly.
At some point you realise you are bashing your head against a brick wall: you may as well write the damn script yourself or just do the task directly manually.
Try the Obsidian path, maybe? (a few posts up)
I’m not going to hijack your thread to try to convince you to use something you don’t want to use 
If you change your mind, I’m pretty confident that we can get that working, if you want to troubleshoot it together (DM me if so), but it would heavily depend on your model and harness.
But I get the feeling you’re not really that keen on relying on LLMs anyway, which is fine!
Instead, take a look at those Obsidian plugins. The plugins themselves are open-source… if you don’t want to use Obsidian, well, you can take their source code and adapt it to your own workflow (main logic is here and it manipulates the Word file using the docx library):
Edit: In case it wasn’t clear, Obsidian itself is a closed-source, highly-polished, free, local Markdown editor — it’s a very nice app for writing, which just so happens to have some open-source Shunn export plugins.
If you’re not currently using Markdown, plaintext should still work too (as a subset of Markdown), just with a few added lines of frontmatter.
Interestingly enough, that’s a good example of what I was talking about. Directions given for how to do something with the assumption that the reader has a better understanding of linux than they actually do. If I needed help installing something and someone said “Google 'libfoo [distro]”, I’d have spent way too long going down that rabbit hole on the assumption that they were telling me ‘libfoo’ needed to be installed first, not that they were telling me to do a google search for whatever it is I’m asking for help with. I did that just trying to respond to that post.
And, again, to be clear, I’m not suggesting someone posting on a random internet message board shouldn’t make the assumption that the OP has a working knowledge of linux, I’m just saying that it happens quite a bit. It can make learning a more difficult and can make it feel kinda gatekeepy, even if that’s unintentional.
I also understand that a lot of people get tired of answering the same basic questions over and over and eventually it’s just easier to say ‘If you don’t know X, you shouldn’t be asking Y’. And that’s not just for Linux, I think it’s pretty common on a lot of hobby message boards.
One of the other problems of open source is that often what’s been created is free but worse to learn/use than a commercial piece of software. For example, there have been several times I wish I could do some basic graphics manipulation. Well, I’m not paying for Photoshop. So I download GIMP. And quickly find a fairly undocumented user interface that apparently assumes I know both how to use both Photoshop and GIMP already. And then I quickly give up on that project.
The biggest problem I find is that people do not really understand what Open Source is, what it means, and how it is developed. The exact licencing differs greatly, but will fall within a spectrum of permissive (public domain) to copyleft (you must follow rules which ensure future code availability), and is defined by the Open Source Initiative in 1998:
Free redistribution
Source code must be accessible
Derivative works must be allowed
No discrimination between users
No discrimination between uses (including commercial or even military use)
Everyone is automatically granted all the open-source rights
What Open Source does NOT entitle you:
You get the idea. Open Source means that you are entitled to the code, AND ONLY THE CODE. If the licence is permissive (like the BSD, MIT, or Apache versions) then you will also not be required to Open Source your extensions and modifications of the code. And that’s really all it is.
The craftsman doesn’t blame the tools.
The Open Source ethos has always been aligned with the Hacker Way, where competence is a sine qua non and lack of it merits very little sympathy. If you’re in over your head and don’t think you’re going to be able to claw your way out, it’s OK to walk away.
This is a task where AI really shines. It acts as a customized search engine and can walk you through all of the steps. What’s really nice is it remembers your setup so you don’t have to keep telling it things like “ubuntu’ 22.04” or the path to my install directory is “/home/cavemike/tools”. As @Reply said, if you ask a question it can go read the actual source code.
I’d be surprised if there are too many supported configs that couldn’t be built with an AI chatbot holding the user’s hand.
I used to sysadmin a GPU farm for my team and it got a lot easier when I started using ChatGPT. It led me astray a few times and burned a day here or there, but it also helped me build, install, and config pipelines in hours instead of weeks.
GIMP’s UI is notoriously complicated, but there are lots of tutorials for a variety of tasks. Photoshop is no walk in the park either. These are professional tools that can perform complicated tasks.
I am not saying this necessarily applies to you, @asterion, but I feel like people expect these programs to be as approachable and easy to use as Paint and are disappointed when they are not. Since GIMP is the only free one, they attach their frustration to GIMP and open-source.
I think GIMP in particular is harder to use, though, even compared to other free (as in beer) software like Paint.net or Photopea.
I think it’s a cultural problem, that old school hacker ethos resistant to changing their beloved elitist software just to pander to newcomers.
Like:
Exactly. Many of the FOSS software I’ve come across seem to hold the attitude that complexity isn’t a negative to be fixed, but a point of pride for the project and its maintainers. Power and complexity are often the draw, not so much simplicity.
The truest, biggest open source projects (Linux, git, LibreOffice, VLC, etc.) all generally follow that pattern, of being very powerful but unpolished and hard to use compared to their commercial or smaller closed source counterparts.
Part of that is just the volunteer base and their power distribution, I think. Your average project is overwhelmingly going to be coders (and their AI agents). If a UX designer happens to stumble onto such a project and suggests changes, well, one, it’s very hard to get the coders to take the suggestions seriously in the first place (it only works in corporate because management makes them listen), and two, even if somebody did like the ideas, it’s much harder still to find someone who likes them enough AND is willing to put in the UI work to actually make them real AND has enough experience with that particular codebase’s UI system. Open source software is often cross platform too, which makes it even more complicated because there isn’t a standard, vendor-supported, cross-platform UI kit. (There are several that fake it to various degrees, but it’s a highly fragmented ecosystem, and part of the reason the Web won out over desktop apps – it’s the only real “write once, run anywhere” platform we have).
And that’s assuming you can even find a designer willing to contribute to open source in the first place. As a stereotype, they tend to be, well, more balanced people than your average developer or greybeard sysadmin type, and less likely to want to spend their after work time doing more unpaid work for ugly projects they don’t use anyway.
All this leads to open source largely being a “by developers, for developers” movement rather than a truly inclusive “by the community, for the community”. It’s even in the name, open source.
I don’t think this is true for GIMP. They have had at least one big redesign specifically for usability. It’s more likely that usability is poor because it is designed by SW devs instead of UX/CX designers. But I might be projecting from my own projects which always have a trash UI.
I can believe GIMP’s UI is worse than those others, but I’ve always found a visual tutorial to do what I want.
The UI overhauls GIMP eventually did go through were the result of decades of complaints, and even then, the outcome was still a compromise between people who wanted a newer, easier interface and people who specifically did not want it to change. The inertia was incredibly powerful. Even in the overhauled versions, there’s still a setting to toggle between the old and new modes, if I remember correctly.
I know people like to pick on GIMP almost as a trope, but really… out of every single graphics package I’ve used over the decades, it is the single hardest one to use. Can you find a tutorial for whatever you want? Yes, but ideally you shouldn’t need to for simpler operations.
There’s just not a pressure to trim features and keep things simple for the sake of new users, like there often is in commercial software. Complexity thus aggregates and solidifies over time in these projects…
With all due respect, I don’t think licencing is the important point.
And therein lies the problem. You get some code which may or may not be any good. You generally have to build a resulting tool yourself. But as I’ve said, sometimes you don’t want to spend a lot of time being a tool developer: you just want to get the job done.
I have to say: not in my experience. I have played around with this sort of thing several times for several different tasks. All too often the AI will give you very specific directions: but when you try to follow them, the exact instructions do not correspond with the options or parameters of what you are working with. As I have said before: these things will confidently produce, say, a python script. You run it: it doesn’t quite work properly or throws an error. You report the problem. The LLM says, confidently as ever: Ah, I see the problem; and produces another script. Which again… etc etc. At some point you realise the process is not converging.
Of course we’re getting slightly off the open source topic here…
At the risk of a slight hijack (and only because you brought it up again! 
)… can I briefly say that the success rate can depend a lot on the particulars? Which model you use, which harness, which effort level, and your skills and prompts all factor into it. Something like the Gemini chatbot is very different from the CLI Claude Code running Fable and some relevant skills, for example.
AI software development has come a LONG way in the last 2 years or so, and also saw some massive improvements in just 2026. With the right setup, the field has moved far past the “can it write a functioning script” era. Far, far past it. Most of my company (a software company) were laughing at it two years ago, skeptical a year ago, flabbergasted six months ago, converted a few weeks after that, and now nobody really writes code by hand anymore (to my dismay)… the good harnesses are better programmers than most humans I’ve met, by a large margin. (They’re still pretty bad at design, though, for now.)
I’ll shut up now unless you want to talk more specifics…
This is an interesting topic in itself, and we should perhaps split it into a different thread: it is somewhat tangential to open source as such…
Good point.
There used to be a small but thriving cottage industry of amateur software, many of which would re-package open source tooling under a friendlier interface. Back in the day, there was an entire marketplace of “software that gave ffmpeg a better UI” (and there still is for e.g. git). Then that market largely shifted to mobile apps and the web, and now the desktop market is by and large polarized into “free and janky” and “huge and commercial, but probably easier to use” 
I would 100% put GIMP’s UI failures on the GTK toolkit. If you develop in the FOSS world you’ll either choose some version of GTK or Qt as your application suite.
Both have very established communities of users and devs - the GNOME Project (GTK) v the KDE community (Qt) who develop 80% of all the FOSS desktop applications in Linux.
While both have their upsides and downsides I must say that the GTK toolkit is a pain. Yes, its LGPL so its “fully open source” (Qt is mostly open source), but also requires you to develop your application UI in C like a low level coder.
Also, while it was originally developed by GIMP for GIMP, it is now maintained by the Gnome project for Gnome, who have no problem breaking backward-compatibility with every major update, thus leaving anyone else with a ton of extra work reformatting their important, but now legacy, API’s and code (if they hope to use the updated toolkit).
GIMP spent 20 years on GIMP2.x (which used the outdated GTK 2) and didn’t release GIMP3.0 (made with GTK 3) until just last year… a whole 5 years after GTK 4 was released (maybe they’ll just skip this toolkit version).