Before anyone raises an eyebrow: I think there are actually some really interesting ethical uses for AI in games.

One example would be a text adventure where the world, characters, and story are all hand-crafted by the developer — but instead of choosing from a fixed list of commands, the player can just type what they want to do naturally.

That idea got me experimenting, and I ended up building this:

• GitHub: love-deploy-local-llm repo
• Windows download: itch.io build

How it works

The game is built with LÖVE (Lua) and ships with everything needed to run a local LLM.

No internet connection, API key, or Ollama install required.

The zip (~402 MB) includes:

• the game executable
• llama-server.exe from llama.cpp plus its runtime DLLs
• a quantized qwen2.5:0.5b GGUF model

When the game starts, it launches llama-server.exe in the background and waits for the /health endpoint to come online.
All communication goes through llama.cpp’s OpenAI-compatible API on localhost:8080.

To keep the game responsive, the HTTP requests run on a separate Love2D worker thread so the main loop never blocks.

The end result is a fully self-contained AI-powered game that runs entirely on the player’s machine.

This is a simple POC in the form of a chatbot. What you use this for is up to you - for me it was a fun experiment. I am thinking about integrating a chatbot into my ascii roguelike to make interactions with npc's more natural. Not sure if I can manage that :D