PewDiePie Introduces Ajax, a Local AI Model Built for Everyday Tasks

PewDiePie has announced Ajax, a local AI model trained for his Odysseus workspace, with release scheduled for October 3, 2026, at 08:25 JST. The project focuses on everyday tasks such as browsing, email, and calendar management. For creators and affiliate teams, its appeal is practical: using a smaller model on their own hardware to help manage routine research and administration.

Key Takeaways: Ajax Brings a Dedicated Model to Odysseus

• Ajax is a fine-tuned Qwen 3.5 9B model designed to work with Odysseus.
• PewDiePie says he chose a smaller model to make local AI accessible to more users.
• Its intended tasks include web research, inbox management, notes, and calendars.
• The practical question for affiliate teams is how reliably it can handle those workflows on their hardware.

What Ajax Adds to PewDiePie’s AI Workspace

The announcement follows PewDiePie’s launch of Odysseus, the self-hosted workspace that brings AI chat, research, documents, email, and other tools into one interface.

Ajax is the model being trained to operate within that environment. The official Ajax page identifies it as a fine-tuned Qwen 3.5 9B model and sets out its planned release time.

In the announcement video, PewDiePie explains that users were running local models that struggled with Odysseus. Training a dedicated model, he says, would give him a way to address those problems directly.

This also makes the distinction between the two projects useful. Odysseus provides the workspace and tools; Ajax is intended to help users work with them.

Why He Chose a Smaller Model

PewDiePie says he initially planned to train a larger model, then reconsidered because fewer people would be able to run it. His stated aim is to give more users the opportunity to try AI on their own computers.

PewDiePie, creator of Odysseus and Ajax:

I really strongly believe in the small model trained for a specific harness.

Here, the harness means the software and tools around the model. Ajax’s focus is on handling tasks within Odysseus, including using its email, browsing, and calendar features.

He says the first version works around nine times in 10, but that is his informal assessment. The transcript does not provide a benchmark methodology or a defined set of test results.

Hardware requirements and task performance will need to be assessed when the model becomes available.

What This Could Mean for Affiliate Teams

For teams exploring AI tools for affiliate marketing, the potential applications are familiar parts of the working day:

• A publisher could gather product documentation and organize source material before preparing a review brief.
• An affiliate manager could summarize partner emails and draft responses for review.
• A creator could turn research notes into tasks and reminders while keeping the final editorial decisions in their own hands.

These are possible uses based on the tasks described in the announcement. Ajax’s performance on affiliate workflows has not been established.

The appeal of local AI is having more control over where model processing takes place. However, Odysseus also supports external API models and connected services. Teams would still need to understand which services their chosen setup uses and where information goes.

Product claims, offer terms, and commercial details would also need to be checked against current sources before publication.

What the “Uncensored” Description Means

PewDiePie describes Ajax as uncensored, and the official page says its refusal behavior has been altered. In practical terms, that is intended to make the model less likely to reject requests.

In his video, he says he drew a line around causing harm to other people or to oneself and asks users to act responsibly.

That description does not establish better accuracy or more reliable tool use. For an affiliate team, the useful measures remain whether the model reads information correctly, follows instructions, and handles errors.

An assistant connected to email or a calendar also needs appropriate permissions and human review for consequential actions.

Finding Training Data Proved Difficult

Training Ajax required examples of the tasks it was meant to handle. In the video, PewDiePie says Odysseus does not automatically collect user data, so he asked users to contribute examples voluntarily. He built a submission website and tutorials, but says the response was smaller than expected.

He then continued generating task examples and refining Ajax through reinforcement learning. He describes running the same task 16 times and using successful attempts to reinforce the behavior he wanted. When progress stalled, he says he had to generate more tasks to continue training.

PewDiePie says this repeated process delayed the announcement as he kept trying to improve the model. He presents Ajax as a first version, with further updates planned.