TokenShooter

A WebGPU-powered FPS where a small language model pilots the robot in your browser.

Description

A Doom-inspired first-person shooter that uses a small language model as a 'pilot' to make decisions in real-time. The system runs entirely client-side via WebGPU, employing a Jev-like approach to read action probabilities based on a text-based tactical description of the robot's surroundings.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: Impressive demonstration of using a small language model as a real-time game agent running entirely in-browser via…
Attributed to: YV17Labs@YVanitou
Built with: Qwen3.5 0.8B Jev Qwen3.5 9B
Links: yv17labs.com
View tweet
View GitHub repo
Tags: #webgpu #fps #llm-pilot #open-source

Discovered: X
First posted: X · Sep 28, 2026
Listed: Added 2h ago
Build: Strong
Models: Strong

Prompt

MONSTERS
Watcher · floating eye. It lines up with the robot, its mouth glows, then it spits a fireball. Takes three hits.
Seeker · burning skull. Fast: it charges in and bites. One hit is enough.

They sleep until they see the robot or hear a gunshot, then hunt it down. Shots and fireballs fly straight along the rows and columns of the maze: to hit, you have to line up.

Source

TokenShooter

Description

How it grew

TokenShooter

Description

How it grew