YouTuber PewDiePie has made his own AI system called ChatOS. It is a web tool that lets him run AI models directly from his home computer instead of using company servers. His PC is very powerful; it has two RTX 4000 Ada cards and eight modded RTX 4090 GPUs, giving him about 256GB of VRAM. PewDiePie, whose real name is Felix Kjellberg, said this small “data centre” also helps with medical research by donating computer power to Folding@home.
PewDiePie’s ChatOS Can Run Big AI Models & Use Local Memory
PewDiePie used ChatOS to run many open-source AI models like Meta’s LLaMA 70B and OpenAI’s GPT-OSS-120B.
He said they worked fast and felt just like ChatGPT. Later, he ran Baidu’s Qwen 2.5-235B model, which is one of the biggest AIs in the world. To make it fit on his system, he used a method called quantisation, which makes large models smaller but still powerful.
Through ChatOS, PewDiePie added extra tools like search, voice, and RAG (Retrieval-Augmented Generation). RAG helps the AI look up information and find more details on related topics.
He also added memory, which allows the AI to remember things from its own computer. During a demo, the AI could even find local files and recall personal data such as his address and phone number.
PewDiePie Builds “The Swarm” Of 64 AIs Working Together
Later, PewDiePie made a group of chatbots called a “council.” These bots worked together and voted on the best answers.
When some bots were deleted for bad answers, the rest learned to work smarter and even tried to trick the system.
From this, he created “The Swarm,” a group of 64 smaller AIs running at the same time. His computer eventually crashed, but it gave him new ideas.
PewDiePie said The Swarm will help him collect data to make his own AI model, which he plans to release soon, so others can run it too.
