---
title: "S&Box LLM Proof of Concept - s&box game"
description: "Proof of concept for running a real LLM entirely inside s&box using a custom C# GPT-Neo inference engine. No APIS, fully local"
canonical: "https://sbox.watch/games/jeffskitchen/llm_poc/"
generated_at: "2026-08-20T13:51:15.800Z"
---

# S&Box LLM Proof of Concept

Proof of concept for running a real LLM entirely inside s&box using a custom C\# GPT-Neo inference engine. No APIS, fully local

- Type: game
- Ident: jeffskitchen.llm_poc
- Creator: jeffskitchen
- Canonical: https://sbox.watch/games/jeffskitchen/llm_poc/
- Markdown: https://sbox.watch/games/jeffskitchen/llm_poc.md
- sbox.game: https://sbox.game/jeffskitchen/llm_poc/
- Generated: 2026-08-20T13:51:15.800Z

## Stats

- Playing now: 0
- 24h player peak: 0
- 24h momentum: +1
- Terry score: 19%
- Favorites: 2
- Upvotes: 3
- Downvotes: 4
- Total players: 15
- Total sessions: 19
- Cumulative playtime: 1.4 hours
- Updated: Aug 17, 2026
- Created: Aug 17, 2026
- Size: 32.1 MB
- Files: 41
- Tags: llm

## Description

!!!!NOTE!!!!!!:
This is not meant to be even a tiny bit close to an actual useable ChatGPT or Claude clone, the model (TinyStories-Instruct-1M) is quite literally 1/2000 the size of any flagship model. The findings in this project can be used to build larger language model projects. Next step is implementing Qwen3-0.6B which could actually have some uses.

S&Box LLM POC is a proof of concept demonstrating that a transformer language model can run entirely inside the s&box game runtime without external AI software, servers, APIs, Python, Ollama, or separate executables.

The project includes a custom C\# inference engine capable of loading packaged model weights, GPT-2 tokenization, GPT-Neo transformer inference, global and local attention, and autoregressive text generation directly within s&box.

The current implementation runs TinyStories-Instruct-1M locally on the CPU and has been numerically validated against the original PyTorch/Hugging Face model. Generated token sequences match the reference implementation while inference runs on an s&box worker thread to keep the game responsive.

The end goal of the POC is a simple in-game chat interface powered entirely by an LLM running inside s&box itself.

From this, I hope other developers can see what is possible and potentially use small LLM to power NPCs, game functions or even whole games.