Play On Your Own Machine
Run the model on your own computer. No meter, no provider, no limits.
Part of the Own Key plan, $4.99 a month
Your Rome, Your Hardware
The Own Key plan also plays against a model server running on your computer.
Tools like Ollama and LM Studio run language models on your own computer. Point Legio Aeterna at one and the game talks to it straight from your browser: every scene, every fight round, every letter is written on your own hardware, and there is no provider bill because there is no provider.
No meter at all
No tokens billed, no credit to top up. Your hardware does the writing.
Written at your desk
The game talks to your server straight from your browser. Scenes are written on your computer, not in someone else's building.
Any compatible server
Ollama, LM Studio, llama.cpp, vLLM, Jan, KoboldCpp. If it speaks the standard chat format, it works.
One plan, both ways
The Own Key plan covers hosted keys and your own machine alike. Switch between them in Settings whenever you like.
Set Up In Three Steps
Pick Your Server
They all speak the same protocol. One setting each, and the browser may call them.
Ollama
The quickest start. Install it, then pull a model such as an instruct model of 12B or larger. Browsers must be allowed in: start it with OLLAMA_ORIGINS=https://legioaeterna.com (on Windows, set that under environment variables and restart Ollama). The address is http://localhost:11434/v1.
LM Studio
The friendliest screens. Load a model, start the local server from the Developer tab, and switch on CORS in the server settings so the browser may call it. The address is http://localhost:1234/v1.
Anything compatible
llama.cpp's llama-server, vLLM, Jan, KoboldCpp, and any other server that speaks the standard chat format. Allow this site's origin in its CORS settings and use its /v1 address.
Worth Knowing
Your browser will ask once
The first time the game reaches for your server, Chrome may ask permission to reach devices on your local network. Choose Allow.
Same computer only
The server must run on the machine you play on. Browsers block a plain http address on another machine; localhost is the one exception to that rule.
Size matters
The Game Master leans on structured tags to run the world, and very small models lose track of them. An instruct model of roughly 12B or larger keeps the story on rails.
No key needed
Most local servers take no key at all. If yours demands a token, there is a field for it in Settings.
Try It Now
If your server is running, this panel reaches it right here, no sign-in needed.
How far it strays from the likeliest next word.
How much of its vocabulary it weighs at each word.
This panel goes from your browser to your server and nowhere else. If it answers here, it will answer in the game.
If Something Goes Wrong
Nothing answered at that address
Check the server is running, that its CORS settings allow this site, and that your browser did not quietly deny the local network permission.
The model list stays empty
The address should end in /v1. Ollama answers at http://localhost:11434/v1, LM Studio at http://localhost:1234/v1.
The story falls apart
The model is likely too small for the Game Master's rules. Try an instruct model of 12B or larger.
Slow scenes
Local speed is your hardware's speed. A smaller model or a quantized build of the same model answers faster.
Rather bring a key from a provider instead? That is the other half of the same plan. Any questions, write to [email protected] and we will get you playing.
Every turn on your own hardware, $4.99 a month.