On the Loose with Odysseus
Its likely I wasnt the first to predict the AI future being personal, small highly efficient models are evolving daily with gushing TikToks to match.
I'm also sure it's not just me creeped out by the idea of my AI life floating around on some corporate American data center, I have a golden rule with Claude, we never ever paste or display API keys or other sensitive strings in the context window. Claude instead has one key to an encrypted vault and has to check creds in and out like a library, I pass in new values by a shared text file that gets wiped after being processed. Am I paranoid? you tell me but I'm a little more relaxed around key rotation row. Which brings us at last, (no snoring in the back Akshay), to Odysseus.
We are definitely going off-piste with the author of this creation, legendary gaming Youtuber Pewdiepie, has released Odysseus, a personal AI workspace that runs 100% locally with a big focus on privacy and security. There's something in the water in Sweden; like an AI Avicii, Pewdiepie aims to singlehandedly batter the evil forces of recurring subscriptions, freeing us from the tyranny of the AI giants who are making out like ketamine powered robbers.
So this sounds like my bag, 100% local, secure, personal. I'd feel comforatble sharing things knowing it would never leave the building, and I can chat with it on my phone. The bells and whistles continue, a calendar, email mangement, tasks, Deep Research, Gallery Library, Notes and more. Its like ChatGTP, Claude, Claude Code, Claude Cowork and Copilot in one interface with all the tools you need for changing models, inference engines and all the tech stuff.
As an example, the Cookbook scans your system, for CPU, GPU and other factors and recommends the best models with a download and install all in one UI.
The setup is pretty straightforward, I went with the reccommended approach of running Odysseus in Docker though you can also not. I next chose the Qwen2.5-Coder-7B-Instruct-AWQ model as the best fit for my machine, I have a RTX5050 8GB GPU but Odysseus witl struggling bravely along on RAM and disk alone. And I think this may be the point, I dont need a machine with a brain the size of a planet to summarise my emails and suggest a response.
So it's brand new, bound to have a few bugs, but is quite exciting, the neat thing is it's easy to build on, After installing Qwen, I'll think I'll plug-in Context7, its an MCP documentation server that will contain trhe entire current React Development documentation, this then injects itself into context so any coding decision effectively gets run through the learning manual first.
This should give you a good idea about Odsseus, it's not a magic-bullet, life-changing OOTB deal, but it's a well laid out and solid framework which will allow you to organise your life and AI life together. You'll need to take the time to build it out to your own specs. Need image generation?,transcription?, coding - there's a model for that.
Ultimately, truly day-to-day, personal AI is still the playground of the data centres and tech-savvy enthusiasts with bucks, that's changing and Odysseus is a good start.
Comments
Post a Comment