The PewDiePie Ajax AI model is a small, locally run assistant being developed for his self-hosted Odysseus workspace—but it is not yet a conventional one-click consumer app. In his October 2, 2026 announcement, Felix “PewDiePie” Kjellberg described Ajax as a fine-tuned Qwen3.5-9B model built for tasks such as web research, browsing, email and calendar work. As of October 7, the project’s public data page was still asking contributors for training examples, and the available evidence did not establish a finished model card, independent benchmarks or a clearly licensed public weights download.
What is the PewDiePie Ajax AI model?
Ajax is intended to be the model layer inside Odysseus, an open-source workspace associated with Kjellberg. Odysseus connects chat, research, documents, notes, email and calendars in one self-hosted interface. Ajax is the smaller model being tuned to understand those tools and decide how to use them.
That distinction matters. Odysseus is the workspace or “harness”; Ajax is the language model Kjellberg says he is training for it. Ajax is based on Alibaba’s Qwen3.5-9B, according to Kjellberg’s video and independent coverage by Tom’s Hardware. The nine-billion-parameter size is designed to make local use more practical than very large frontier models, although actual performance will still depend on hardware, quantization and the software stack.
Can you download PewDiePie’s Ajax AI model?
The safest answer is: not as a clearly documented, finished standalone release yet. The official-looking Ajax path cited by reports led to PewDiePie’s Odysseus training-data page when checked on October 7. That page asked users to contribute example conversations and displayed an in-progress dataset indicator. It was not a normal model download page.
Odysseus itself is available as source code, but installing the workspace is different from downloading finalized Ajax weights. Readers should avoid unofficial “Ajax download” pages until the project publishes a model card, file hashes, an explicit weights licence and an unambiguous release link from Kjellberg or the canonical repository. Those checks protect users from both misleading mirrors and software-supply-chain risks.
What Ajax is supposed to do locally
Kjellberg presents Ajax as an always-on model for everyday tool use inside Odysseus. The stated targets include search, web browsing, email and calendar management. Running a model locally can keep prompts and generated text on the user’s own computer, but “local” does not automatically mean every connected service stays offline. Email providers, search engines, websites and optional hosted APIs may still receive data when a user enables those integrations.
It is also too early to treat the model’s reliability as independently established. Public benchmark results, a detailed model card and reproducible safety evaluations were not available in the sources reviewed for this article. The practical question is therefore not only whether a home PC can run Ajax, but how consistently it can use tools without sending the wrong email, changing the wrong calendar event or exposing sensitive information.
Why PewDiePie calls Ajax “uncensored”
In the video, Kjellberg said he used an open-source technique commonly called refusal ablation to reduce the model’s tendency to reject ordinary requests. He also said he kept boundaries around requests involving harm to other people or the user. “Uncensored” is therefore a project label, not proof that the model has no safeguards—and it should not be read as an independent safety certification.
Kjellberg also said OpenAI suspended his account twice during development and showed an email that cited “distillation” for one deactivation. Tom’s Hardware reported what appeared in the video. OpenAI did not provide an independent account-specific confirmation in the material reviewed here, so the suspension narrative remains Kjellberg’s documented claim rather than a jointly verified account of events.
What the PewDiePie Ajax AI model changes
The noteworthy part of Ajax is not that a creator has suddenly replaced the largest AI labs. It is that one of YouTube’s best-known figures is turning local-model experimentation into mainstream creator content. A small model trained around one workspace could be useful if it performs a narrow set of jobs reliably and gives users more control over where their data goes.
But the release should be judged on evidence: downloadable files from an official source, a clear licence, hardware requirements, benchmarks, a threat model and transparent limitations. Until those arrive, Ajax is best described as an active creator-led local-AI project, not a finished ChatGPT replacement.
The project also marks a sharp change from the personal-video questions surrounding the creator. For that separate story, see our explanation of why PewDiePie ended his family-vlog series. For another technology angle affecting creators, read our guide to the latest YouTube membership update.
Confirmed facts and open questions
- Confirmed by the creator: Ajax is being built for Odysseus and is based on Qwen3.5-9B.
- Independently reported: Tom’s Hardware reviewed the announcement and the account-suspension material shown in the video.
- Publicly visible: the Odysseus data page was still collecting training examples on October 7.
- Not independently established: final benchmark performance, complete safety behavior and the exact circumstances of each account suspension.
- Still needed for a mature release: a clear weights licence, model card, checksums, hardware guidance and a canonical download page.
