no, the game does not require A.I., the NPC system does but you don't need to talk to the NPCs to play the game or progress through the game, that's an optional feature.
The NPC system requires ollama (if you want to talk to the NPCs): https://ollama.com/, it always was like that since the A.I. powered NPC system was implemented (nothing has changed in this regard).
The model can be changed (by hardcoding it) but by default it uses "qwen3:b". More info on https://glitchapp.codeberg.page/bookleft/
By default it uses a local ollama server (on the same computer where the game runs), but you can use an external server if you want (optional).
Some information about the licenses used by Ollama and Qwen that may be useful:
Qwen uses the Apache-2.0 license for its local, downloadable models (this does not apply to cloud-hosted versions).
Source: Qwen wikipedia
Ollama is released under the MIT license.
Source: Ollama git source
It is also worth noting that neither Ollama nor Qwen is bundled with the game. You must download them separately if you choose to use them. Their purpose is solely to enhance the gameplay experience—specifically for those interested in AI-powered NPCs. You can completely skip this feature and still enjoy the full game without any limitations.
The reason Qwen is suggested by default is that it is commonly used for AI-driven NPCs across multiple engines, including experimental Unreal Engine projects. However, if you do not like Qwen or feel it does not meet your definition of open source, you are free to replace it with another model such as LLaMA or any alternative you prefer. Instructions for changing the model are provided in the game booklet linked above.
One detail that may be useful for people who aren’t familiar with these models: the “3B” at the end of a model name (for example, qwen3:3b) refers to the number of parameters in the model—about 3 billion. In general, larger models require more memory, compute, and inference time, but they also tend to produce higher-quality and more consistent results.
For an AI-powered NPC system, I chose a 3B model because it offers a good balance between performance and resource usage. It’s large enough to generate believable dialogue and behavior, while still being lightweight enough to run efficiently on most consumer hardware.
That said, there are smaller models (around 1.8B) that can work well on lower-end systems, and larger ones (7B or more) that can deliver better results if your GPU or PC has sufficient memory and processing power. You can adjust the model size based on your hardware constraints and the level of NPC intelligence and realism you’re aiming for.
