Did PewDiePie just declare war on Big Tech? His AI, Odysseus, keeps your data at home
Streamer PewDiePie has released Odysseus, an open-source AI workspace that runs entirely on your own computer. No subscription, no tracking, all your data stays on your disk.
News analysis
Odysseus + Ollama
What is new
PewDiePie has released Odysseus, a self-hosted interface for AI that looks like ChatGPT or Claude but sends nothing to the cloud. The models run locally, and the memory, emails and documents stay on your disk. He apparently built it just for himself at first, then decided to declare war on the big tech companies along the way. Classic.
When he ran AI models at home, he felt like he was in a sci-fi film, but he was missing memory, deep research and decent integrations. So he built them himself. The result is a whole ecosystem where everything sits in one place.
He explains it all with his usual humour in his video. You can play it right here:
What you'll appreciate most
Your data stays yours. The more an AI knows about you, the better it serves you. The problem is who you are telling. Google or OpenAI can assemble a surprisingly intimate profile from a few queries, and that profile then gets sold on. When a robot calls you with an offer, it is no accident.
Odysseus keeps that profile on your disk, not on someone else's server. No hoping a company really deletes your data when you click „delete".
Who it is for
For anyone tired of paying monthly fees and trusting promises about deleting data. It also suits people who want one tool for everything instead of ten connected apps.
Agents
Memory and email
Research and editor
How to use it in practice
You download Odysseus, run it on your machine and let the Cookbook tool scan your hardware. The program recommends the models your computer can handle, downloads them and wires them in. Then you write to the agent just like in a chat, and it does the rest.
A real-world example
PewDiePie asked the agent to transcribe a video. It found the file on its own, converted the format, ran the local Whisper model and brought back the finished transcript. When the task gave it trouble the first time, it rewrote its own instructions so it could do it faster next time. Mildly unsettling. Mostly just useful.
Recommended tools
Odysseus is mainly for Linux and the more technically confident for now. If you want local AI without much tinkering, look at Qwen 3.6 and similar open-weight models through Ollama. And for everyday work in the cloud, ChatGPT or Claude is still the more comfortable choice, just at the cost of your data leaving home.
Summary
Odysseus is the first serious attempt to give people a full AI workspace that does not spy on them. It is not a tool for everyone, and you will wait a while for Windows and Mac. But the direction is clear: your data belongs to you. And when PewDiePie serves the rebellion with a side of jokes, at least you will be entertained.
Sources
Frequently asked questions
What people often ask
What is PewDiePie's Odysseus?
Odysseus is an open-source AI workspace that you run on your own computer. It looks a lot like ChatGPT or Claude, but it sends nothing to the cloud. PewDiePie originally built it just for himself, because the existing interfaces lacked memory, agents and deep research. Then he released it for free as open source. Beyond chat, it handles autonomous agents, an email client, a document editor and the installation of local models.
Is Odysseus free and genuinely private?
Yes, Odysseus is free and distributed as open source under the MIT licence. Privacy is the whole point: the models run locally on your machine and no telemetry is sent. Your conversations, memory and documents stay on your disk. You can connect external APIs, but only when you choose to. The logic is simple: the more an AI knows about you, the better it serves you, and with Odysseus you are not handing that data to someone else's company.
How do I run Odysseus at home?
You download Odysseus from GitHub and run it on your own machine. The hardest part, picking and installing models, is handled by the built-in Cookbook tool. It scans your hardware, scores it and recommends models that will run on it. Then it downloads them and wires them into the environment. Under the hood it runs on Python and supports Ollama, llama.cpp and vLLM locally, plus APIs such as OpenAI or OpenRouter. Expect to need a bit of technical confidence.
Is Odysseus better than ChatGPT or Claude?
It depends on what you are doing. On privacy, Odysseus wins outright, because the data never leaves your computer. On convenience, ChatGPT and Claude still lead: you install nothing, it works straight from the browser and you do not worry about hardware. Odysseus makes sense when subscriptions bother you, you do not want to hand data to Big Tech and you are not afraid of a little configuration. For quick everyday work with no setup, the cloud tools are still more comfortable.
Does Odysseus work on Windows or Mac?
Reliably, mainly on Linux for now. PewDiePie admits he has not built a native version for Windows or Mac yet and that the community will have to solve it. Because it is an open-source project, he invites developers to get involved, but he wants real code improvements, not cosmetic fixes. The good news is that the environment also works in a mobile browser. So if you are not on Linux, expect a workaround until the port matures.
Keep going
Related articles
More guides from the same area, topics and tools.
Can NVIDIA Spark take on Apple's M5? The new AI chip brings petaflop power straight to a laptop
NVIDIA has shown off RTX Spark, the laptop version of its Grace Blackwell chip. It promises up to a petaflop of AI performance and 128 GB of shared memory. Here is what it does and how Apple's M5 Max stacks up.
America just pulled Claude Fable 5 for all foreigners. Has the hybrid AI war begun?
On 12 June 2026 the US government ordered Anthropic to disable Claude Fable 5 and Mythos 5 for all foreign nationals. What does it mean for Europe and what can you do about it?
Is Qwen 3.6 the best AI model you can run at home?
Alibaba has released an open-weight model that matches Claude Opus 4.7 in some coding tests and fits on a single 24 GB GPU. What does that mean for companies and developers?