MabiPro: The Authentic Old-School Mabinogi Experience

Hi everyone, I want to talk about some of MabiPro's long term goals, finances, and AI.

MabiPro's AI usage policy

First, I want to talk about our use of AI in the project. To preface this conversation, I want to iterate what MabiPro's internal AI use policy is:
AI is allowed, but has these restrictions:

• It may not be used for art, at all.
• It may not be used for story, quests, puzzle/dungeon/shadowmission creation.
• All AI written code must be reviewed by a human.

This really is a massive majority of what developers on MabiPro actually work on.

But, here are some examples of things it CAN be used on, and real world examples you have seen lately:

• It may be used for core engine development. For example, the creation of "Me" and "Decorate" tab.
• It may be used for bug fixes in existing content, especially for complex bugs that have plagued MabiPro for almost 9 years.
• It may be used for backporting newer content to MabiPro, such as Cursed Labyrinth (which is G16 content) where the content, story, quests, and puzzles are all pre-existing, made by humans, and simply mechanically backported. All changes and updates to this content were done by human hands.

The reason that I bring this up and clarify how we use AI is because, understandably, people may be concerned about code quality, content quality, and many other things concerning the ethical use of AI ^{(and whether AI is even ethical to use at all, given it is trained on the corpus of every piece of text and code in humanity's written history).}
Most of MabiPro's developers do not use AI at all, in fact, I am basically the only one myself.
I also want to clarify that NONE of MabiPro's donation money has ever been used to fund AI usage, at all.

MabiPro's long term goals

That being said, AI has limits. Right now, core engine changes built by AI are created and added to the game through complex reverse engineering, function hooking, and manipulation of existing compiled code.
It isn't possible to do much more than what we have already done in a way that is maintainable and keeps the code quality standard we strive for internally.

I have two long term goals for MabiPro

• I want to convert MabiPro to 64 bit, and replace DirectX9 with OpenGL. This should have incredible performance improvements.
• For fun, I want to copy what some World of Warcraft private servers have done, and make a web-based version of MabiPro.

Both of these goals sound well, impossible right? I mean, you cannot just magically convert a 32 bit program to 64 bit, even with the existing frameworks we have for modifying the game.

Now, how can I actually achieve my goal here? Let's take a look at some other projects:

• Human-made GTA3 Re3 matched decompile
• AI-assisted Super Mario 64 DS reconstruction reports 99.9% of its functions matched
• Keldurn (World of Warcraft 1.21.1 matched decompile, ported to WebAssembly and playable in your browser)
• Rewriting Bun in Rust "Claude, rewrite Bun in Rust."
  _{We dont endorse any of the above projects, they are just inspirational to us.}
  

The first project, GTA3 decomp, took humans 5 years to do by hand. The other two - done within this year alone....if you are even a little bit familiar with software engineering and reverse engineering, that timeline should make you go WTF...?

For those who are unfamiliar with the current state of AI, I want you to watch this video:
https://fixvx.com/jadenjtw/status/2106068013283971483

If you can't beat em, might as well join em.

With that context, let's revisit my long term goals: Both of these goals are proven to be possible using the techniques already established above. So, why not actually do it?

MabiPro matched decompile project

I recently began the project of decompiling MabiPro's binaries to original, compileable matched C++ source code, using an AI swarm of over 1000+ agents. It is pinned to the exact MSVC2008 compiler used to compile the original client. Right now, as of writing this message, the game is 76% decompiled. That is massive progress, and just having 76% of the source code is a game changer. Two of our binaries reached 100% and are fully compiled from source at this point.
As I approach 100%, the progress slows down as the AIs have to unravel every single line of code in the game, with functions getting more and more complex.
I won't go into detail about exactly how this works, if you are curious you can send me a DM.

But the bottom line is: This is a massive deal for MabiPro's long term maintainability and feature development.
Every ASM patch, every hooked mod/feature patch can be converted to REAL, native Mabinogi features. It unlocks a world of capability that we have only dreamed about over our 9 years running this server.

MabiPro's finances

MabiPro is privately financed by me directly, and has always been. Historically, donations do not reach me, they are distributed to our content developers as tokens of appreciation for the great work they do for the server.
This allows MabiPro players and developers to never worry about whether we will shutdown due to cost of hosting and infrastructure. There are only a handful of times that donation money has ever been used by me - a couple of times 8 years ago in our early days (About $100 USD), and another time recently to finance our EU server (about $300).

Over our 9 year history, MabiPro has expensed over $30,000 USD ranging from dedicated servers, VPSes, DDoS protection, bandwidth, domain costs, hard drives for physical backups, and whatnot.
This is how MabiPro has always been, and how it always will be. I only give you these numbers to give you a perspective, but at the end of the day, they dont matter. We are privately financed and it does not really matter how much we expense. It's a small price to pay for something that means as much to me as MabiPro.

And, like I mentioned before, we have never used any donation money for AI use, including the decompilation project. However....

Financing the MabiPro matched decompile project

We are burning through billions of tokens, right now that is about 140B cached input tokens and 50M output tokens.
And, we will need double, or even triple the amount just to hit 100% as complexity grows.

Tokens are expensive. So far, I have spent $1200 of my own money on this project, and I estimate it will cost me another $2000 to complete.

To help fund this, I am opening donations towards this process. Just like any other donation drive, donators will be given the cosmetic Donator title.
All funds from this round will be used for the decompilation project.
The money will be used to buy Anthropic, Deepseek, and OpenAI tokens. If you do not want your money going to an AI lab, do not donate during this round.

For those who are itching to get a Donator title but do not want to contribute funds towards this project, we will have another donation drive in the near future that will be dispersed to content developers just like previous rounds.

If you would like to contribute to the cause, contact Darecrow on Discord for details on how to donate.

can we like donate usage of our gpus or something ?~

I always feel gratitude to Drahan, who runs the whole server
thank you for your work
Erinn is the only place where my soul can rest
its a very meaningful way to spend some savings to un- binarizing massive server files
i would have done that too and thank you agin for everything you've done
(I will contact Mr crow and please Make Erinn Great Again!)

Based position.
MabiPro server is unreachable under my ISP anymore unfortunately, but I wish you tons of luck and fortune.

im not really one to argue for or against LLMs because I do think that stuff like "telling the funny robot to parse this engine made by people that arent natively english around the 2000-2004 era" is very understandable. but I am a bit concerned about the idea of making a web-based client of this server. i've seen other ~funny servers~ that didn't otherwise get noticed by legal departments suddenly come under fire. something about that seems to just catch their attention. the 64 bit conversion would be fantastic theoretically, though.

and im sure you're already aware of this but keep in mind that for a long term project like this, the AI industry is comically unstable and is propped up by the corporations hoping the other corporations they're scamming don't notice they're being scammed. the instant any large companies decide they wanna stop being scammed, prices of tokens for the users might go up significantly. or the entire provider might just shut down.