Game Engine
Unity 6000.0.46

3rd Party Assets Used
Assets used:

2D Pixel Quest Vol.3 - The UI/GUI
Text Animator
DoTween Pro

NobodyWho (Local LLM for Unity and Godot)
https://github.com/nobodywho-ooo/nobodywho

UIOutliner
https://github.com/PandaArcade/UIOutline

Editor tools:

Odin Inspector
QuickNav
PlayModeSave

AI Used
NPC Portraits and background images were AI generated using Stable Diffusion (using the pixel-art-xl model and lora:pixel_art_style_1_1 lora)

Story generation was done by generating multiple json files from a template I gave the LLM. I'd hoped to generate these on the fly and make the game entirely procedural but the speed and inconsistency of the results meant that I had resort to a set of 50 pre-crafted stories, each with a set of characters and initial interactions. If you choose to continue speaking with any of the character, the local NobodyWho LLM (using a qwen2.5-1.5b-instruct-q8_0.gguf model) then handles these responses on the fly, enabling endless exploration of each character's story arch. A system memory is enabled so that each NPC remembers what you have told it, to give context to its responses.

ChatGPT and Github CoPilot were used with code generation and editing, json file validation, and python scripts for automating art asset creation.

** NobodyWho, Rust and SIMD instruction‐sets **

Every project's a learning opportunity, and never more so than a game jam. I've been using Undream's LLMUnity (https://github.com/undreamai/LLMUnity​) for my local LLM needs in Unity but have had a lot of problems with it crashing and hanging the Unity Editor, although builds have been working fine... enter NobodyWho. Thought I'd give it a go as I heard about it on reddit just as the MessyCoder's jam was announced. It's been mostly great in the Editor, some teething problems but stable. So I ran with it. Sadly it turns out that it will only work on a CPU that supports the same SIMD instruction-set the Rust binary it was compiled with.

I managed to stop the build crashing on unsupported CPUs but it still won't load the actual LLM model. So if your game constantly tells you that "You bore me too much to speak." and "I really should be going now.", that's just the fallback method I coded to handled the LLM failing to return a valid response. Once or twice is just the model falling over, but every time means your PC isn't compatible. Sorry, but it is what it is. I may get around to porting the game to LLMUnity. Shouldn't be too much of a headache, I just ran out of time (or didn't leave enough time) on this game jam.​