I’ve been messing around with something for the last several months called SeanOS.
The easiest way to describe it is as a Knowledge Operating System. It’s not another AI. It’s the information layer that sits between me and whatever AI I’m using.
The idea isn’t to make the AI smarter. The idea is to make any AI start from the same place.
Right now every model is its own little sandbox. Claude thinks one way, ChatGPT another, Gemini another, and local models are different again. None of them remember things the same way, prioritize the same way, or even access the same information. Every time I switch models I feel like I’m starting over.
SeanOS is my attempt to fix that.
The analogy I keep coming back to is a medical record.
The AI isn’t the doctor, it’s the specialist.
Before the specialist walks into the room, the nurse has already taken your vitals, the nutritionist has updated your diet, the counselor has written their notes, and your medication history is sitting there waiting. The specialist doesn’t have to figure all of that out. They just use it.
That’s what SeanOS is supposed to be.
It’s more than a PKM. It stores hard facts, but it also tracks decisions, reasoning, project history, priorities, relationships between information, and even why I made certain decisions. The whole thing is designed to constantly audit and improve itself so it drifts less over time.
The goal isn’t to improve the AI.
The goal is to reduce the gap between:
-
What Sean knows.
-
What Sean would remember.
-
What Sean would prioritize.
-
What Sean would decide.
-
What the AI actually produces.
Kind of like a universal system prompt backed by an always-growing knowledge base instead of trying to cram everything into context every time.
The dream is that I can hand the same SeanOS repository to Claude, ChatGPT, Gemini, Grok, Gemma, Qwen, or whatever model shows up five years from now, and they all start with the same “medical file.” Then I can bounce between models depending on what each one is good at instead of being locked into one ecosystem.
Getting there has been… interesting.
I originally thought NotebookLM was going to be the librarian. That didn’t last long. Then I tried Google Drive. Claude worked pretty well with it. Gemini wasn’t crazy about writing to it. ChatGPT and Grok could sort of read it, but not consistently enough to trust.
Then I moved everything into GitHub.
That didn’t magically solve the access problem either, but Git itself has been a massive improvement. The structure, version history, append-only workflow, and ability to audit changes are light years ahead of what I was doing in Drive.
Long term, the whole thing will probably live behind my firewall. At some point I want local models to have access to everything without sensitive information ever leaving my network.
I’ve also had Claude and Grok do architecture audits and pre-mortems on the project. Interestingly, neither really attacked the architecture. Their biggest concern was me… being human. Skipping maintenance, getting lazy, or letting the data drift over time. The other big issue they both pointed out is one I already knew: every AI company keeps building bigger sandboxes, so getting them all to see the same source of truth is still the hardest part.
I’m curious if anyone else here has gone down this rabbit hole. I know there are quite a few AI projects floating around this forum, and I’d love to hear how other people are handling shared context between models.
GitHub repo: GitHub - Sully2582/SeanOS: Personal knowledge system and retrieval source for SeanOS v3.0 · GitHub
and yes @Tokoloshe I had ChatGPT help me write this. LOL
)