A tiny experience running local LLMS to drive agent behavior via WebGPU.

Controls

  • WASD --- Move camera
  • Mouse --- Interact with UI

What is it

Office Sim is a proof of concept demo that showcases the use of running LLMs locally for high-frequency and low-latency interactions using WebGPU.  This allows developers to explore use cases not possible with traditional server-based inference pipelines due to it either being slow or costing to much to run for thousands of players/users. 

Our goal is to provide a drop-in inference engine that works across all edge devices and platforms (Unity, Godot, three.js, JavaScript, etc) to enable developers to create experiences and tools not possible today. Like dynamic games or web apps where you need a LLM "brain" to process data in real-time. 

Troubleshooting

This is fairly experimental and utilizes browser features that are not widely supported yet across all browsers. If things aren't loading, try updating to the latest version of Chrome and ensure you are running on a device that has a modern integrated/dedicated GPU. 

If still encountering issues, feel free to use the link below to report the bug/issue that you are experiencing. 




---

If you find this interesting/looking for access/have suggestions, we plan to post some updates on our progress eventual:

https://forms.gle/kJGQHeEwykebZEbC6


Published 5 days ago
StatusPrototype
PlatformsHTML5
Rating
Rated 5.0 out of 5 stars
(1 total ratings)
AuthorNoumena Labs
GenreSimulation
Made withUnity
Tagsdemo, llm