Felix Kjellberg, the YouTuber known as PewDiePie, released Odysseus on May 31: a free, MIT-licensed AI workspace that runs entirely on a user’s own hardware. The GitHub repository crossed 44,000 stars by June 4, according to Mindwired AI. For context, that is more community attention in one week than most venture-funded AI startups accumulate in a year.
Kjellberg announced the launch in a video titled “MY trillion $Dollar Project is finally OUT!” with the tagline: “The war on big tech has just begun,” as The Business Standard reported.
What Odysseus Actually Is
Odysseus is a self-hosted interface for interacting with language models, bundling several capabilities into one local-first workspace. It ships no new AI model. Kjellberg describes it as “the self-hosted version of the UI you get from ChatGPT and Claude, but with more jank and fun,” according to Mindwired AI.
The feature set includes chat (connecting local models via Ollama, llama.cpp, or vLLM, or cloud APIs via OpenAI, Claude, and OpenRouter), autonomous agents that plan and execute multi-step tasks with tool use, deep research flows powered by SearXNG (a self-hosted search engine), AI-assisted email triage and calendar management, and persistent memory using ChromaDB that retains context across conversations. A “Model Cookbook” scans available VRAM and RAM, then recommends and installs models from a catalog of 270+ options.
The platform has no tracking, no subscriptions, and no telemetry, according to both Mindwired AI and The Business Standard. “No sales team, no demo request, no Trojan horse,” the Odysseus website states.
How Kjellberg Got Here
This project did not appear from nowhere. Kjellberg, currently semi-retired in Japan, spent the past year publicly documenting his move into AI and privacy technology: de-Googling his digital life, building his first PC, and learning to code. In February 2026, he fine-tuned a Qwen 32B model that scored 39% on the Aider Polyglot coding benchmark, beating GPT-4o (23.1%) and Gemini 2.0 Pro Exp (35.6%), according to Mindwired AI.
His stated motivation is straightforward: he did not want his data going to large technology companies. The GitHub star count suggests that motivation is widely shared.
The Self-Hosted Agent Market
Odysseus enters a growing field of self-hosted AI tools, but with a distribution advantage that no competitor in the space can match. Kjellberg has 110 million YouTube subscribers. Local AI has been a developer-only conversation. When someone with that audience ships a privacy-first agent workspace as a personal project, it changes who the conversation reaches.
The software is one week old, built primarily by one person, and has rough edges. Local models still require meaningful hardware (RAM and VRAM). It will not beat frontier cloud models on hard reasoning tasks. But for the use case it targets, a private workspace that replaces monthly AI subscriptions with locally run models, it is already more complete than many funded alternatives at first release.
The project’s technical requirements may limit mainstream adoption: Docker or terminal comfort is needed, and GPU resources determine model quality. But the signal from 44,000 stars in a week is not about Odysseus specifically. It is about demand for AI tools that do not require handing data and subscription fees to cloud providers.