PewDiePie builds Ajax AI to run privately at home

PewDiePie has developed his own artificial intelligence model designed to run on a home computer, pushing his experiments with locally hosted AI into a much more ambitious stage. Felix Kjellberg has named the model Ajax and built it around Qwen3.5-9B, an existing open model that he has fine-tuned for use inside his self-hosted Odysseus AI workspace.

The project is particularly interesting because it takes tasks increasingly associated with huge cloud AI services and attempts to move them back onto hardware controlled directly by the user. That gives the project a very different focus from simply launching another chatbot designed to compete with ChatGPT or Gemini.

What does PewDiePie's Ajax actually do?

Ajax is a nine-billion-parameter model, meaning it contains roughly nine billion adjustable values learned during training. That is relatively small compared with the enormous frontier AI systems operated by the biggest technology companies, but its smaller scale is part of the point.

Kjellberg wants the model to be capable of running locally rather than requiring an enormous remote data centre every time somebody uses it. A smaller model can potentially run on powerful consumer hardware while still being capable enough to handle everyday assistant tasks.

Ajax powers Odysseus, his self-hosted AI workspace for activities including searching, browsing the web, working with email, managing calendar information and organising personal material. Instead of opening several different applications and manually moving information between them, the aim is for the assistant to work across those tasks from one environment.

That makes the wider project closer to an AI agent than a simple chatbot. An agent can use tools and carry out several connected actions rather than only responding with text to an isolated question.

Why does running AI locally matter?

Most popular consumer AI systems operate primarily in the cloud, which means a user's request is sent to powerful remote computers for processing. That approach gives people access to models vastly larger than most laptops could ever run, but it also means relying on another company whenever the assistant is used.

Local AI takes a different approach by keeping more of the processing on hardware controlled by the user. That can give people greater control over their information, how the model behaves and whether their assistant remains available if a provider changes prices, limits or product rules.

Privacy is one of the biggest attractions because sensitive information does not necessarily need to leave the user's computer for every task. Someone using an assistant to organise private files, emails or personal notes may prefer a system that can process more of that information locally.

The trade-off is that local AI requires capable hardware and considerably more technical knowledge than simply opening a website. Users also become responsible for more of the setup, maintenance and security that cloud providers normally handle.

Why did OpenAI suspend PewDiePie's account?

Kjellberg says his OpenAI account was suspended twice while he was developing Ajax. One email he displayed referred to distillation, bringing an increasingly important AI industry dispute into a project being followed by a huge mainstream audience.

In AI development, distillation involves using information from a larger model to help train a smaller one so that some of the larger system's capabilities can be reproduced more efficiently. The technique itself is widely used across the industry, but companies can restrict how outputs from their own models are reused.

The issue becomes particularly sensitive if those outputs are being used to help develop another model that could eventually compete with the original service. That tension is becoming more important as smaller developers try to build capable models without the enormous training budgets available to the biggest AI companies.

The detailed account of the two suspensions comes from Kjellberg, and OpenAI has not publicly provided a separate detailed explanation of his individual case. That distinction matters because his version should not be presented as an independently established finding about why every action against the account occurred.

Is Ajax available for everyone to use?

Ajax remains a developing creator project rather than a mainstream consumer service comparable with ChatGPT, Gemini or Claude. That means readers should not assume there is a polished app ready for anybody to install and use without technical knowledge.

Kjellberg has also experimented with reducing some of the refusal behaviour found in mainstream AI assistants. He argues that locally controlled systems should give users greater authority over how the model behaves, while retaining restrictions around requests involving real-world harm.

Those safety claims have not been independently tested at anything approaching the scale applied to major commercial models. Smaller local systems can give users greater freedom, but they also transfer more responsibility for how that technology is configured and used.

The wider significance may actually be PewDiePie's enormous audience rather than Ajax itself. Local models, fine-tuning, self-hosted assistants and AI agents were specialist developer subjects not long ago, but one of YouTube's best-known creators is now putting those ideas in front of mainstream viewers.

Ajax may therefore matter less because everybody is about to install it and more because it shows where consumer AI could be heading next. Instead of every useful assistant living on somebody else's server, some of the next generation may live directly on the computers people already own.