October 3, 2026, (Inside AI) — Felix Kjellberg, the Swedish YouTuber known as PewDiePie, released Ajax on October 2, a 9-billion-parameter AI model stripped of safety refusals and designed to run on home computers. The launch follows two separate bans from OpenAI, which flagged his account for distillation, the practice of using one model's outputs to train another.
Ajax is fine-tuned from Alibaba's Qwen3.5-9B and operates as an autonomous agent inside Odysseus, a free, open-source workspace Kjellberg launched in May 2026. The model handles web search, email, calendar scheduling, and to-do management. All conversations stay on local hardware rather than cloud servers. With over 110 million YouTube subscribers, Kjellberg brings self-hosted AI to an audience that enterprise-focused open-source projects rarely reach.
OpenAI Bans Expose Distillation Fault Lines
OpenAI's first ban cited distillation explicitly. Kjellberg eventually regained access, but a second ban followed when he ran the model again to generate seed data. Distillation violates OpenAI's terms of service because it turns proprietary outputs into training material for competing products. The bans underscore a widening conflict between large AI companies protecting their models and independent developers building alternatives.
To remove safety refusals, Kjellberg used an open-source tool called Heretic that identifies and strips refused prompts from the model's behavior. He acknowledged the trade-off directly: "Fewer guardrails also mean more room for misuse." (Felix Kjellberg, creator of Ajax) However, he maintained that Ajax retains restrictions against harming others or oneself and is not designed to provide dangerous actionable instructions.
Read: Abu Dhabi AI Institute Releases Six Fully Open-Source Models with Training Data and Code
Early testing showed roughly 90% task completion rates, though Kjellberg described Ajax as a work in progress. His development setup includes 10 GPUs with eight modified RTX 4090 cards carrying 48 GB memory each. Despite this high-end infrastructure, Ajax targets standard home PCs for end-user deployment.
The model weights remain private as of launch. The download link currently directs users to a data contribution page rather than the model itself. The Odysseus workspace operates under an AGPL-3.0-or-later open-source license and remains free to use.
Local AI Gains A Mainstream Face
Ajax joins a growing category of locally deployable AI models that prioritize user privacy and reduced censorship over cloud-based convenience. Meta's Llama, Mistral, and Alibaba's Qwen family already serve this market. Kjellberg's entry differs primarily through his creator's massive audience reach. His promotion of local AI deployment introduces self-hosted models to a mainstream consumer audience that enterprise-focused open-source projects typically miss.
Read: Gnani Unveils Sovereign AI Stack Artha With Open-Weight Model and Enterprise Agents
The timing matters. As AI regulation tightens and cloud providers face scrutiny over data handling, tools like Ajax offer an alternative path. But the model's private weights and unclear download process raise questions about reproducibility and long-term viability. For now, Ajax stands as a test case: can a celebrity-backed, uncensored model thrive outside the walled gardens of Big AI?
