Back at the end of June, the ProjectIgnis fork of windbot – that is, the engine behind AI bots for various Yugioh simulators such as EDOPro – got updated with a bunch of new decks you could let your computer play. Things like Yubel, Ryzeal, and Branded that are a bit more modern than the ones that have been in the game at this point. That doesn’t mean they’re in EDOPro proper yet, not until the next client update I assume, but you can install them by hand if you want – see below for a brief tutorial on that.
Anyway, this release came at just about the right time, since I was actually already kind of looking into hooking up the base version of windbot (that already had these decks) to my local instance so I could test in a more up-to-date environment. Having been spared the effort of doing that, I could instead spend some of the time I had available over summer playing with the new bots to evaluate their prowess. The results were … mixed, as you can see in these two videos compiled from some of the duels.
How to update Windbot in EDOPro
- From the latest GitHub release, grab the file named WindBotIgnite-Release-[…].zip.
- Open your ProjectIgnis game directory and move the WindBot subfolder elswhere (or rename it to something like WindBot_old).
- Take the WindBot folder from the .zip you downloaded and place it in the game directory.
- The next time you open EDOPro and launch an AI duel, you should have the new decks available to select.
And that should be all. The specific release I used was 20260628, but the structure is the same for all of them, so (currently) you can just always pick the latest.
Evaluation of the new decks
Now, to prove I was actually doing SCIENCE here and not just screwing around, I should write things down. What decks are these new AIs bringing to the table, are they actually running something competent, and can they pilot it?
These are my impressions from mixed test runs with the Rahkshi and Avohkii decks, one best-of-3 match against each AI with one or the other.
Apophis
Certainly one of the scarier decks on the roster. If you let it go first, it will set a bunch of Traps to negate everything you do; if it goes second, it’ll try the same with literal handtraps including a hefty package of Dominus cards.
Games against this one can be legitimately unwinnable just from the sheer number of disruptions it can throw at you, but it’s not like it does that in a particularly smart way. For example, it will use Dominus Spark on the first Extra Deck monster you bring out, even if that monster has a way to dodge (granted, with custom cards, how would it know?) – and then it’s stuck under the LIGHT/DARK lock and unable to use its engine. Oops.
Blue-Eyes 2025
A brand-new deck with the latest support should, on paper, be the most valuable testing asset out of all the Blue-Eyes variants the AI has, but the games I played with it left me a bit unsure of that. Yes, it has the good cards, but its actual performance with them wasn’t very impressive when faced with any kind of opposition.
Granted, a sample size of one match means I might very well just have gotten lucky enough to shut it down every time, but as it stands I am still considering keeping the Chaos Max OTK Turbo build in the lineup instead. Actually, looking at the decks, it seems like Chaos Max is in fact in the new one as well, what’s up with that?
Branded
Now this isn’t quite the most recent 2026 version where Albaz is all white and shit, but it’s still a pretty modern deck that can pull of some decent combos. Branded Fusion, Mirrorjade, Bystials make for some powerful things to go up against, and all the Main Phase 3 End Phase shenanigans mean it has lots of followup available to keep gameplay going, which is nice when you’re trying to test something.
Also there was a game where it dropped 3 Ash Blossoms on me, one of them on an ASS that would have drawn 5 otherwise, so I shall never forgive this mechanical menace.
Dogmatika
This is a deck that wasn’t added in this update, but rather just refreshed with some more recent support – if memory serves, the previous version just used the regular monsters to set up things like Punishment and try to win with that. This new one has more of a focus on the Ritual aspect, and particularly features the big ugly Alba Zoa it seems to make quite consistently. That makes it a good test of what your archetype does when half of its Extra Deck is gone.
Exosister
One of the things I look for in a testing environment is coverage of the different types of threats you might encounter from different decks, so it’s great to have one representing the concept of “If you touch your grave you get killed IRL”. It’s easy enough to play around since the intelligence on these bots isn’t all that advanced, but being forced to do so in the first place gives some interesting insight on custom card design and deck building decisions.
Labrynth
I had high hopes for this as a fairly interactive Trap-based deck to add to the mix, and while it does provide the expected play style, a lot of the specific decisions made in the games feel very suspect. Why get Torbie of all things with Welcome? Why randomly summon furniture in attack position and leave it there? Why pay all your LP away with Rollback while failing to address the lethal damage staring you in the face on the field? You get what I mean.
Still, it at least knows how to use its engine in a vacuum and it has handtraps to back that up, so it’s good enough to be worth playing against I guess.
Maliss
There are two different versions of Maliss among the new bots, and let me just say right away that this is the one that actually knows how to play the deck. It banishes the monsters, it sets the Traps, it uses them to dodge stuff, it makes the Links, it splashes in the Cyberse slop, and so on. This is pretty hard to beat if it manages to go first and pop off, and hard to stop from doing that … unless you happen to side in silver bullets like Lancea or Chaos Hunter (who, as a high-Level DARK Fiend, was always part of the Rahkshi side deck!). Truly the authentic Maliss experience.
Maliss OCG
… and this is the one that, for some reason, is literally incapable of using the deck it’s named after. I’ve seen it use the effects of main deck Maliss monsters, but that’s it. Doesn’t make Maliss Link monsters, doesn’t activate the Maliss Traps under any circumstances, doesn’t even use the effect of March Hare in the hand. I have to assume this one is just literally defective, unless they’re doing things far beyond my understanding over there in the OCG.
Ryzeal
With a notoriously simple and straightforward deck, it’s fair to expect that even the simple AIs in this game could run it well, and that is indeed the case. In fact, it even impressed me by knowing tricks such as using Ghost Ogre against Duo Drive in the mirror match, or passing on Detonator if faced with Fuwalos.
Sacred Beast
This deck technically wasn’t part of the latest update, but just happened to be added in a few months before. It features the new Sacred Beast support from Chaos Origins, which theoretically gives it access to a bunch of big guys with a whole lot of recursion. In practice, it turns out, its body is a machine, by which I mean it tends to just make S:P pass and kind of get stuck on that.
I’m not familiar enough with the new Sacred Beast cards to judge if that’s an issue with the deck or something’s fishy in the code, but whatever, it does still provide enough resistance to be interesting for casual test games at least.
Superheavy Samurai
A modern “Superheavy Samurai” build, which is to say it’s Synchro spam using the engine around Wakaushi. This doesn’t just make a bunch of negates, but also uses Omega to handrip a bit, so it’s a decent challenge if it goes off. Unlike many of the other recursion-heavy decks on this list, however, this one seems to run out of gas at some point, when it just can’t really do anything with the material it puts out anymore.
Yubel
Also one of those decks that presents a unique thing to play around, in this case damage reflection. Where putting up a board of big monsters is usually part of how you win, here it’s a big part of how you can lose … and also kind of a required one, because this AI does not seem to be capable of any lines that do damage against an empty field. And also likes to out its own setup with the End Phase upkeep costs of various Yubel forms.
Yummy
Like Sacred Beasts, this is a bot that was added shortly before the big update, and also like that one, I didn’t see all too much of it fully executing its combo during testing. But in this case it’s kind of my fault, because when I went second against it with the Avohkii deck I just so happened to open the turn 0 play that is Graalok + the mask to reveal. Let me tell you, Yummy combos do not get started when there’s a hungry hungry Ash Bear ready to gobble up the first little snack that dares step on the field every turn.
It also doesn’t have any concept of playing around Charmies, and I’m pretty sure it didn’t use half the cards it could have used in a given situation. Oh well, they can’t all be [AI] Ryzeal.
Zefra
Finally (in alphabetical order), a name that stands out among the various recently released decks here is Zefra, which is an archetype that last got stuff 10 years or so ago if memory serves. What’s hiding behind it is actually more of a Pendulum pile, with Endymion cards and Halqifibrax and whatnot. It puts up a good fight, is what I wrote down in my notes.
Final Verdict
14 bots were tested, and 13 of them worked, in the sense that they made use of the cards in the deck given to them. I have no idea what Maliss OCG is doing.
Among those 13, performance varied, but just having access to modern cards generally makes them good enough at producing interesting games that I’ll definitely integrate them into the regular testing routine instead of some of the ancient AIs I’ve been using for years. Branded, Ryzeal, Superheavy, and Zefra especially felt like there was a lot of meaningful interaction going on.
In terms of recorded footage, that means you can look forward to seeing more of these new decks and less of … Lightsworn? Yosenju? Not sure what’s going to drop out yet, will have to see where I even want the total number to be during the next round of new card testing.