I Tried PewDiePie’s Odysseus AI So You Don’t Have To (Its FREE)

Last person on earth I’d expect to release open software? Kim Jong Un. 😅

Those who watch pewdiepie videos already knew he was making local AI vrs.

self-hosted and on-premise models and solutions are the only future that makes sense if companies want to save the horrific overblown costs. there is no need for state of the art technologies when there are free and opensource alternatives that are just "good enough"

Basically, LM Studio + Opencode....

I’ve played with it a little, it’s cool and will probably get better 😊

The "Together AI" Fallacy: At [09:01], Matt asks the local model Gemma 3 12B who made Odysseus. The model hallucinations say it was developed by Together AI. Matt notes that the app "doesn't know thyself" [09:14], but he attributes this to a failure of Odysseus. In reality, this is purely a limitation of the local model's pre-training cutoff and its complete lack of internal self-context. Local weights do not inherently know what codebase interface they are running inside without explicit system-prompt injection or RAG-based context parsing. BUT the video is a realistic, practical critique of the current state of advanced, local-first AI tooling. It rightly concludes that Odysseus is a fantastic playground blueprint for developers, power users, and data-sovereignty hobbyists who want an all-in-one ecosystem [28:25], but it still possesses the natural dependency friction and edge-case errors common to bleeding-edge, open-source codebases [01:40]. Thanks fr this and all your takes and info, Mr. W :)

The inpainting is about as complicated as Leonard's canvas feature

8:30 needs a bit more scepticism when using an LLM, especially a low tier one. Just because it sounds like it knows doesn't mean it actually knows, it was just highly probable to output an answer that sounds like it knows.

Funny timing, Matt. I've been stuck on exactly this problem for months. Too many AI tools, too many tabs, and too much context switching trying to remember which model does what best. That rabbit hole turned into a design project I call "Ariadne", an AI librarian that helps navigate the growing AI ecosystem. Then I wake up this morning and Odysius lands in my feed. 😂 Not quite the same vision, but definitely scratching the same itch. Now I just need to see if my poor GTX 1050 Ti can survive the experience without bursting into flames.

It's a good start and more technically minded people will help the project, so it will surely get better.