[Beta] Vox Deorum 0.6.3: Play Civilization With LLMs (VP 5.1.3 Compatible)

[johnchen]

Vox Deorum v0.6.3 (VP 5.1.3)
Play Civilization with LLMs

Download Installer Here (01/20/26)
​

Vox Deorum is a modmod for Vox Populi that enables Large Language Models (LLMs) to play Civilization V + Vox Populi alongside or against you. You can:

• Play with or against LLM-controlled civilizations
• Chat with your LLM opponent (in progress)
• Watch LLMs compete against VPAI or each other

This is a beta release. The mod is functional and playable, but it needs more testing across different game situations.

Watch LLMs Play Vox Populi​

You can see how LLMs play using the Vox Deorum Replay Viewer. In the example replays, Player 0 is an LLM, and you can watch their decisions unfold turn by turn.

• Example Game 1
• Example Game 2
• Example Game 3

Frequently Asked Questions

Spoiler FAQ :

Q: Which VP version do I need to use?
5.1.2. At this point, VP is bundled inside Vox Deorum's installer, since we added custom DLL code (so that LLMs can access in-game state and control VPAI's strategies). Unfortunately, as of now, savegames are not incompatible. I am working to bring the DLL code into upstream VP; once that happens, savegames will be compatible.

Q: Can I chat with LLMs?
Yes, but only in the WebUI at this time. We are seeking a modder who can help us implement an in-game interface.

Q: What can LLMs do in the game?
At this point, LLMs will take control of the grand strategy (which victory type it pursues), diplomatic persona (how hostile/friendly/deceitful it will be), economic strategy (which buildings to prioritize), military strategy (which units to prioritize, and how aggressive its army behaves), policies, and technologies. We are working to enable LLMs take diplomatic actions against you soon.

Q: Do LLMs play better than VPAI?
At this point, no, but also not worse. We have an academic study to sell you soon. That said, LLMs can have a different playstyle. Please let us know your thoughts!

Q: Do I need to pay for LLMs?
Vox Deorum supports many LLM providers out-of-the-box, including ones you can run locally (e.g., LMStudio, ollama, llama.cpp). Those models won't cost you a dime. Otherwise, based on our own experiments, an average 4-player game (Tiny map) with the simple strategist and GPT-OSS-120B will cost ~$0.5 per player per game on Openrouter.

Q: How about the latency?
The mod uses a macro-level strategy approach, meaning the LLM focuses on high-level decisions rather than micromanaging every unit. Latency depends on the models you use (the more powerful/expensive, the slower). Normally, LLMs won't need more than one minute to make a decision.

Q: Is Vox Deorum compatible with other modmods?
Yes, at least that's the intention. Modmods should be supported out-of-the-box as long as they don't touch the DLL-level AI code. If you find any incompatible modmod that works with VP but not with VD, please send it to the issues.

Q: Can I remove LLMs mid-game?
Yes. You can normally start Civilization V, load the mods, and load the game without VD. Once VD is no longer running, VPAI will take control back in 10 turns. (On the other hand, you can also add LLMs mid-save. They will try to understand the mess.) However, at this point, please don't reinstall VP since we changed the savegame format a tiny bit.

Q: I don't understand how to do X!
We hope the interactive Web-based interface is intuitive enough for you. If not, please send a screenshot here, and I will try my best to answer your questions / improve the design.

Q: Does it support multiplayer?
I have no idea. I don't quite understand the multiplayer aspect of the DLL code. I need someone help me test/work on that

Update Log
Please see the GitHub release page.

Helping Us!
This is where I need your help! As a beta release, there's much work to be done. As players, you can help us by:

• Playtest with LLMs
• Report any crashes, bugs, or strange LLM behavior
• Share your experience: Is the AI competitive? Entertaining? Frustrating?
• Giving us more ideas on LLMs in Vox Populi

If you are a modder and want to help us, please consider:

• Help build in-game interfaces to interact with LLMs
• Help design better prompts or workflows to make LLMs more clever/capable of playing
• Help design different chatbot personas (we only have one persona right now, a "spokesperson")

Credits

• Alpakinator, axatin, azum4roll, Flamingcheesepie, Gammalyse, Hokath, KungCheops, schnetziomi5, Recursive, Vaderkos, and Vern

Thank you all for keeping Civ V and Vox Populi alive after all these years. It has been a great pleasure working with and for this community.
Please let me know your thoughts

 

[johnchen]

Updated to 0.4.4, fixing an issue with the installer. Thanks @schnetziomi5!

 

[Autoraem]

I have to wonder if smarter AI's will allow us to completely remove AI bonuses and instead just have scaling difficulties based on how hard the AI tries

 

[johnchen]

I have to wonder if smarter AI's will allow us to completely remove AI bonuses and instead just have scaling difficulties based on how hard the AI tries

That's the goal. That said, I don't think LLMs can outsmart humans on tactical moves; RL models may (and this modmod opens a way to do that).

 

[Recursive]

That's the goal. That said, I don't think LLMs can outsmart humans on tactical moves; RL models may (and this modmod opens a way to do that).

This is where my own interest in this project lies. At best LLMs could make diplomacy much more immersive and realistic (which is still something notable), but reinforcement learning could improve the base AI for everyone. Curious to see how this progresses!

 

[msw1]

Super cool project. It would be both very interesting and probably sometimes hilarious to barter with LLMs instead of items just having a fixed value during trading.

 

[johnchen]

This is where my own interest in this project lies. At best LLMs could make diplomacy much more immersive and realistic (which is still something notable), but reinforcement learning could improve the base AI for everyone. Curious to see how this progresses!

If someone wants to work towards that, I am happy to collaborate. I will also seek some students at my institution to work on this one. Should be a fun thesis project!

 

[johnchen]

Super cool project. It would be both very interesting and probably sometimes hilarious to barter with LLMs instead of items just having a fixed value during trading.

Yes, that's a very natural next step for the project. I am seeking help from fellow modders as this thing grows bigger and bigger

 

[CppMaster]

If someone wants to work towards that, I am happy to collaborate. I will also seek some students at my institution to work on this one. Should be a fun thesis project!

I have sone experience in RL. I've trained AI from scratch for StarCraft 2 that beats the hardest computer using just marines.

https://github.com/CppMaster/SC2-AI

 

[johnchen]

I have sone experience in RL. I've trained AI from scratch for StarCraft 2 that beats the hardest computer using just marines.

SC2-AI/minigames/simple_map/src/planned_action_env at master · CppMaster/SC2-AI

Contribute to CppMaster/SC2-AI development by creating an account on GitHub.

github.com

Whoa. Are you in the VP Discord? We can talk a bit more there.

 

[FinMc]

This is really cool, thank you for making it

 

[CppMaster]

Whoa. Are you in the VP Discord? We can talk a bit more there.

Yes, I've messaged you there.

 

[johnchen]

Lots of updates for LLMs - they now have access to more information, and hopefully, they will do a bit better with it!

 

[eyalgi11]

how do i run it on linux
i succefully launched it with protontricks but when i press play on steam it's says launching updating executable forever

 

[KungCheops]

One idea could be to train an AI to play like the current game AI via RL before letting it start improving through self play.

 

[Teholb]

Is RL possible at scale, given how long games take and how much data is required for RL? It seems like you'd need thousands of game runs, maybe tens of thousands? I'm curious to see where this goes, the idea of having something a little more flexible in strategy than the standard AI sounds awesome!

 

[CppMaster]

Is RL possible at scale, given how long games take and how much data is required for RL? It seems like you'd need thousands of game runs, maybe tens of thousands? I'm curious to see where this goes, the idea of having something a little more flexible in strategy than the standard AI sounds awesome!

Certainly possible, but I wouldn't say feasible Like you said, it would require tens of thousands of games for training, which is few months on RTX 3070 for 1 training. And there may be needed multiple trainings, because of course the first one won't be perfect. There may be bugs, different techniques to try, hyperparameters...

Also, the current VP AI is pretty good, so it'd be hard to do sth better.

 

[johnchen]

how do i run it on linux
i succefully launched it with protontricks but when i press play on steam it's says launching updating executable forever

Never tested on Linux. Can you start VP there?

 

[johnchen]

Updated to VP 5.1.2 - please note that this change breaks savegames!

Also, now LLMs can directly set the tactical AI flavors in-game (not default as of yet, will enable in the next build after our current research experiment), and set diplomatic relationships directly towards another player. They can even get deceptive (ie setting a public and private modifiers in different directions)!

 

[Gazebo]

Ol 'zebo checking in, this looks neat. I dig it, looking forward to seeing where this goes.

Yes, I'm still alive. Too many children, not enough caffeine these days.

G