---
title: "AI Decision Maker (GLiNER 2.5) - s&box game"
description: "Native GPU-accelerated GLiNER 2.5 decision workbench for testing structured AI choices directly inside s&box."
canonical: "https://sbox.watch/games/jeffskitchen/aidecisionmakergliner25/"
generated_at: "2026-09-24T01:27:02.030Z"
---

# AI Decision Maker (GLiNER 2.5)

Native GPU-accelerated GLiNER 2.5 decision workbench for testing structured AI choices directly inside s&box.

- Type: game
- Ident: jeffskitchen.aidecisionmakergliner25
- Creator: jeffskitchen
- Canonical: https://sbox.watch/games/jeffskitchen/aidecisionmakergliner25/
- Markdown: https://sbox.watch/games/jeffskitchen/aidecisionmakergliner25/index.md
- sbox.game: https://sbox.game/jeffskitchen/aidecisionmakergliner25/
- Generated: 2026-09-24T01:27:02.030Z

## Stats

- Playing now: 0
- 24h player peak: 1
- 24h momentum: +0
- Terry score: --
- Favorites: 1
- Upvotes: 0
- Downvotes: 0
- Total players: 1
- Total sessions: 1
- Cumulative playtime: 16 minutes
- Updated: Sep 23, 2026
- Created: Sep 23, 2026
- Size: 284.2 MB
- Files: 43
- Tags: 

## Description

AI Decision Maker (GLiNER 2.5) is a native s&box proof-of-concept that lets you test structured AI decision making directly in-engine.

Enter a context, task/question, and a set of candidates, then let GLiNER 2.5 Small choose the best option and rank the rest. The project runs natively inside s&box, with a GPU-accelerated backend for fast interactive inference and a scalar CPU backend retained as a reference path.

Features
- Native GLiNER 2.5 Small integration in s&box
- GPU backend for fast decision inference
- CPU backend for validation and comparison
- Clean interactive decision workbench UI
- Candidate ranking, scores, and diagnostics
- Designed for testing AI decision-making workflows in games and tools

What it’s for
This project is a technical demo / workbench, not a full game. It is designed for:
- experimenting with AI-driven decision selection
- testing structured prompts and candidate options
- exploring future NPC and gameplay AI workflows in s&box

Example uses
- choosing the next action for an NPC
- ranking tactical options
- testing survival / combat / strategy decisions
- prototyping AI utility-style systems with language models

This build is intended as a showcase of native in-engine AI inference and as a foundation for future gameplay experiments.




## More by jeffskitchen

| Package | Ident | Players | Favorites | Upvotes | Score | Markdown |
| --- | --- | --- | --- | --- | --- | --- |
| S&Box LLM Proof of Concept | jeffskitchen.llm_poc | 0 | 4 | 2 | 9% | https://sbox.watch/games/jeffskitchen/llm_poc/index.md |