Felix Kjellberg, the YouTuber known as PewDiePie, released a free self-hosted AI workspace called Odysseus on May 31, announcing it in a video bluntly titled "MY trillion $Dollar Project is finally OUT!" The GitHub repo shipped at version 1.0 under an MIT license, and it runs on your own hardware rather than someone's cloud.
The pitch, lifted almost word for word from the project's landing page, is that this is the ChatGPT-and-Claude experience minus the subscription and minus the data handover. The README puts it more honestly than most: a self-hosted workspace "with more jank and fun."
What you actually get
A lot, on paper. Chat against local models or external APIs (vLLM, llama.cpp, Ollama, OpenRouter, OpenAI all listed). An agent that can run shell commands, touch files, and hit the web. A research mode. An email client with IMAP and SMTP. A calendar with CalDAV sync. Notes, tasks, a document editor, persistent memory, model comparison. It even installs as a phone app.
That breadth is the interesting part, and also the part worth squinting at. Email clients, calendars, agent frameworks, and vector memory are each a serious engineering project on their own. Whether one developer's 1.0 holds all of that together past launch week is a fair question, and it's the one a couple of writeups have already raised.
Borrowed parts, credited
Odysseus doesn't pretend to be built from scratch. The agent runs on opencode, the model-recommendation feature called Cookbook is built on llmfit, and the research mode is adapted from Alibaba's Tongyi DeepResearch. The acknowledgments file spells out the licenses, which is more diligence than a lot of AI side projects bother with.
The privacy framing runs through everything. No telemetry, the repo insists, with the recurring joke that there's "no Trojan horse" buried in it. The honesty extends to the security section, which reads less like marketing and more like a warning: this thing has shell access, file uploads, and API tokens, so treat it like an admin console and do not point it at the open internet without HTTPS and a reverse proxy.
Should you run it
If you have the hardware and the patience, the barrier is low. Clone the repo, run Docker Compose, open localhost:7000. There's a one-command Windows launcher and a native macOS path for Apple Silicon, since Docker can't reach the Metal GPU.
The harder question is whether a privacy-first workspace from a YouTuber with 100-plus million subscribers nudges the bigger players. Probably not on its own. But it lands at a moment when the conversation around local AI has shifted from chatbots to full workspaces that remember and act, and Odysseus is a working, deployable version of that idea rather than a manifesto.
The code is live now on GitHub. The roadmap file lists where the project wants help, with fresh-install testing and provider-setup bugs at the top.
