Hi, the GitHub repository contents have me suspecting Logseq developers have been using LLMs in developing the software.
Since I have major ethical issues with the “AI” industry I would like to know which version of Logseq I could use to not run any LLM-generated code, if any.
Thank you for your work and understanding.
The devs are heavily using AI agents since it speeds up their work tremendously.
I dont know when exactly they started, but in february of this year Tienson said that “for the past 3 months, LLMs have been generating 80 percent of the code”, so at the very least since December of 2025. I started using the DB version around July of 2025, and I dont remember there being any mention of agents in the commits then. The issue is that you wont find a release from that time, since the first beta release of DB came this year, and before that we just used Github artifacts which have expired.
So your options are
I’m also uneasy by the volume of AI generated code in logseq db.
I’ll still give it a go, especially since tienson has been very vocal about using AI, and how he checks everything himself (we’ll have to take his word for it).
On the other hand, it does make me anxious that such a relatively new technique is being used so intensely, and for such profound changes (major refactoring etc). I wouldn’t be surprised if in a few months the project has become a mess that no human would dare (or care) to touch.
Which is why I think we should all keep raising questions like yours, just to avoid the devs getting too comfortable inside their AI bubble.
Sure that’s a possibility, but it’s no different than merging community contributions. AI contributing does not mean that code quality or cohesion suffers especially now where code output can be of very high quality. I’m saying that having been a developer for 13 years while also experimenting with LLMs since their beginnings.
You should try Tine outliner. It works on Logseq OG Markdown data and both applications can work on the same graph. But the code is entirely written by AI.
And the software is a mess. I don’t talk about source code, but the UX itself. A lot of new features are added, the website is quite good. But the usability is disappointing. So many features are unusable when used outside of the basic workflow.
I think many applications will be released following this development pattern. And people will get disappointed. And good well tested software will be in high demand.
I am sure you as a mature software developer can wield this tool to great effect, but people cobbling together their “dream projects” with it are gonna learn the hard way that making good software is extremely difficult and requires incredible expertise on many fronts, not just hot make a nice UI (like @damien.flament seems to suggest).
That however discards ethical and environmental concerns of the technology, which is what primarly motivates me here. Even if LMMs wrote best code ever (which I am sure they would ultimately be capable of, maybe requiring right guidance, maybe not - similarly to how computers are better at chess nowadays than even the best people are) I would not want to run that code because it’s patched together from codebases that had specific licensing, which is disregarded for the greater good of “progress”. Apparently piracy is a crime only when individuals do it.
I strongly disagree.
It’s easy to spot bad code by a human because their personality, style and skills (or lack thereof) will be consistently reflected throughout all their contributions and discussions.
AI has no personality - it’s sampling from the work done by thousands of people. It’s always confident, and usually produces code that looks elegant and professional - regardless if it’s correct or not. Which makes it much harder to spot errors, hallucinations and problematic code - which does happen, a lot.
The above is reason enough for me to be skeptical.
On top of that, there’s the fact that the next natural step for AI will be to completely abandon human language or logic. Certainly code that doesn’t need to be readable (or understandable) by us silly humans will run much faster. Then what?