Google. OpenAI. Anthropic. PewDiePie. As we all know, these are the names of the biggest players in the artificial intelligence space.
Yes, the once-most-subscribed person on YouTube-turned-edgelord-turned-semi-retired dad vlogger has a new AI workspace and agent to offer, and it’s basically a Frankenstein’s monster of a bunch of other open-source AI projects. Dubbed Odysseus, the primary selling point (though it’s free) is privacy. PewDiePie promised there is “no tracking, no subscriptions, no funny business. It’s yours and yours forever.”
PewDiePie described the project as “Basically, it’s Claude and ChatGPT’s web UI but self-hosted.” But the closer comparison is probably OpenClaw, with some of the pain points of the self-hosted AI tool (supposedly) ironed out. The web UI part is a key note, though—this thing is accessible via desktop or mobile browser, but there is no macOS or Windows port available—PewDiePie said that was someone else’s problem to deal with.
By running AI models locally instead of pinging prompts and responses back and forth between some data center, there is less of a concern of your behaviors being tracked, shared, and monetized (though it does apparently support using models like Claude, ChatGPT, and others via API if you want to go that route).
“The more you share about yourself with AI, the better it becomes. The more it understands you, your preferences, your past experiences, your workflow, your work, your documents, your computer. The more you give it access, the better it works,” PewDiePie said in a video announcing the project. “The more you do that, the more you’re handing over a huge piece of yourself to these giant tech companies.”
As for what Odysseus offers other than privacy, it has an agent, powered by the open-source OpenCode, that can handle tasks autonomously based on prompts. That includes basic stuff like responding to emails. PewDiePie also gave an example of asking the agent to transcribe an audio file for him, and it was able to find the file and run it through WhisperTranscribe.
PewDiePie also included several other “stolen” features (that’s the way he describes the hodge-podge of open-source projects that he’s equipped Odysseus with), like a Deep Research feature that can collect data across multiple sources and create a visual output to compare and contrast. That feature is a modified version of code made available by the China-based Tongyi Lab, which is owned by Alibaba. There’s also a built-in search feature that taps into SearXNG, a metasearch engine that aggregates results from hundreds of search services and databases. PewDiePie also said there is a document editor that cribs from Claude and a calendar and notes app that he described as a “copy-pasted Google Keep.” There’s also a knock-off of Photoshop for image editing.
Odysseus supports a few chatbot personalities and characters that users can communicate with one-on-one or in a group chat, if that’s a thing you want to do. That seems to be a sort of modified version of a past PewDiePie AI project, his “AI council,” in which he created a collective of AI personalities to provide input on his decision-making.
So is this project actually anything? Maybe! PewDiePie is extremely rich and has a $20,000 rig with eight GPUs, so his local model is considerably more powerful than the average person’s Mac Mini setup. It’s also pretty obviously vibecoded, just like OpenClaw, and that led to a whole lot of security issues.
Reviews from people who have tried it out seem pretty mixed. One Redditor said it is “okay-ish” when using a commercial model like Claude via API, but the experience with a local model is very hardware-specific to your setup. Another said their initial impression of the platform is “pretty good,” but when they tried to install a DeepSeek model, “it crashed my PC.”
If there’s a selling point to Odysseus, it’s that it does seem like it’s been cobbled together with the intention of actually addressing a real person’s workflow needs, so it likely feels pretty intuitive. But in the grand scheme of things, this is just another mostly slopped-up interface made by a guy with a big platform. It’ll get attention and probably lots of forks for that reason, but there’s no reason to think it is any more secure or free of the same problems that have plagued other open-source AI workspaces and agents that came before it.
