My dad was diagnosed with cancer. While he was in hospital, the council emptied his house. Everything I owned was in that house. £20,000+ of equipment, years of research, a server with thousands of hours of work. Locks of my kids’ hair. Photos. All thrown in a tip.
My family turned my dying dad against me. I ended up living with someone suffering from paranoid psychosis. That’s where I built most of what I’m about to describe. Three days ago, 24 hours of abuse, and now I’m in a tent with my dog. 5°C weather. No money.
The council refused housing. The government won’t recognise my autism. They want me job hunting 35 hours a week from a tent.
I’m not incapable. I’ve raised a family. I’ve worked my whole adult life. Supervising teams, tattooing, freelance programming, building proprietary backend systems across 20 years of working with Linux. My autism isn’t a disability here. It’s the reason I can hold an entire OS architecture in my head and see how every component connects. When I point this brain at a problem, it produces systems that work, at a speed that doesn’t make sense to most people.
Over the past 4 months, I’ve been building OctantOS. An operating system for autonomous AI agents. Not a framework. Not a container wrapper. An actual OS with its own kernel (OctantCore, from-scratch Rust), its own hypervisor (OctantVMM), a single-binary Rust userspace, and a 10-layer security stack enforcing agent permissions at the kernel level.
~1.3M total lines of code. ~800K Rust. 50 crates, ~25 satellite projects. 3,900+ tests. Solo developer. No CS degree. 4 months.
The thesis: application-layer trust is insufficient for autonomous agents. OctantCore makes agent identity, capability boundaries, TTL enforcement, and audit first-class kernel primitives. Manifests compile to kernel enforcement policies. The agent doesn’t decide what it can do. The kernel does.
Rust LSM patches reviewed on lore.kernel.org by Google’s Rust-for-Linux team and the LSM maintainer. OctantCore boots on OctantVMM with memory manager, interrupts, syscall interface, Agent Descriptor Table, and capability enforcer initializing at boot. Built by orchestrating 10-12 parallel AI coding sessions simultaneously.
It goes beyond isolation. Agents identify gaps in their own knowledge and seek out what they don’t know (curiosity subsystem, implemented). Background inference consolidates learned patterns (dreaming). A 7-stage self-evolution pipeline within constitutional safety boundaries. New skills propagate across every OctantOS instance globally via the mesh layer. All kernel-constrained.
Nothing like this has existed before. That’s what dies if I can’t keep going.
I need stability. A place to live and enough to cover basics for 3 months to get OctantOS investment-ready. An angel willing to back me for that runway. A company that says “come work here, we have a place.” I’ll relocate anywhere, tomorrow, with my dog. Or just advice from someone who’s been here.
I just need someone to take a bet on what this brain can do when it’s not freezing in a tent.
https://github.com/MatrixForgeLabs/OctantOS https://octant-os.com https://gofund.me/f554a86ee
Your priorities are completely wrong. You need to redirect the time and energy you're spending on this project towards getting back to stable ground, starting with utilizing the safety nets that your government and community make available to you. You need to find psychiatric care and shelter immediately, and eventually, gainful employment of any kind.
There are no available services. But even if there were, I don't need psychiatric care. I'm perfectly capable of living normally when I do.tnhve the weight of the world on my shoulders.
As the post makes very clear, I have tried to get assistance for years. The UK is a shambles for helping people ony situation. Because I'm classed as high functioning, I get treated like I'm just wierd.
The government will not help me with Housing or shelter.
The government will not provide care for my mental condition.
I don't need a dead end job or a good job eventually.
I just want to do what I'm very good at.
But it seems most people struggle to comprehend that I just want to be given a chance.
Making it open-source would like of defeat the object here and render it worthless.
A project like this isn't worth much when it gets cloned 1000 times and people start using all my code for free.
The whole point is to create a proprietary software system that will be able to help me make something of my life.
There’s a lot of big projects posted these days, hard to find traction. You built it all? or AI coded it and you orchestrated? I think look for IT work and get money for housing until the SWE market improves.
Best of luck to you, it’s a brutal job market in tech right now. Try to find local people and help them with tech, I think there are probably enough huge new projects being made for no one specific and no specific use case.
To answer your question:
I architect everything. Every system design, every architecture doc, every interface contract, every integration decision. The kernel is a from-scratch Rust kernel. The hypervisor is a clean-room Rust implementation. The security model, the cognitive architecture, the 10-layer enforcement stack. All of that comes from 20 years of building systems and an autistic brain that can hold all of it in memory simultaneously.
I then orchestrate upto 10-12 parallel AI coding sessions, each working on a separate component with strict file ownership and interface contracts defined upfront.
It’s closer to being a lead architect directing a development team than it is to prompting ChatGPT and hoping for the best.
You can’t vibe code a kernel. You can’t vibe code a hypervisor. You can’t vibe code a BPF LSM policy compiler.
These things require deep understanding of what they are and how they connect. The AI types fast under my direction.
I’ve developed a methodology around this: roughly 25% of the codebase is architectural markdown that guides code generation, proto files coordinate between sessions, integration happens between batches.
It’s a disciplined process, not random prompting. Nobody accidentally produces 1.3M lines that compile and pass 3,900+ tests (that's the number of tests for the core project only, not the entire 1.3M LoC)
How do I get a job while homeless? What would I do with my dog while I work 8 hours a day?
I struggle with normal jobs. I can't cope working jobs I dont like doing. It makes me depressed on top everything else.
That'd be like me assuming you don't have a soul. A rather callous assumption, really. Admittedly one with a better evidentiary foundation than your accusation.
Have you spoken to https://www.citizensadvice.org.uk/ you should be able to get some help and advocacy. And they will help you find accomodation and apply for the benefits you are entitled to.
If you can get a proper diagnosis you can also claim PIP for more support: https://www.benefitsandwork.co.uk/personal-independence-paym...
You're obviously smart and probably present as high functioning to someone working in the job centre who is not motivated to help you and frankly anything they give you, you will be massively underemployed.
---
I will say something feels off with the project itself just by looking at the stats.
Obviously an LLM is doing the heavy lifting to write 1.3m lines in 4 months, but 3,900+ tests seems an order of magnitude too small to me.
What LLM are you using to generate this much code? Priority 1 right now is to get accomodation, not pay for expensive LLM API costs or subscriptions!
Best of luck!
But I have to be honest: I’ve been through Citizens Advice and the benefits system for years. CA is a signposting service that directs you to the same services that have already rejected you. PIP requires a formal diagnosis, which requires years of waiting lists and assessments.
I’ve been fighting this system long before I ended up in a tent. The post explains that the state won’t help, and that’s not an exaggeration or an oversight on my part. I’ve tried.
On the test count: fair point. The 3,900+ is the core workspace (50 crates).
Satellite projects have their own test suites not included in that number. Coverage could be better. That’s part of what stability buys me.
On the LLMs: Claude Code for complex Rust architecture. Codex for pattern-heavy work. Gemini for docs and broad context. Qwen for parallel workloads. All running simultaneously with strict file ownership and pre-defined interface contracts.
But to be direct about what I’m actually looking for: I don’t want benefits. I don’t want PIP. I don’t want permanent support from the state. I want to use my brain, build complex systems, and turn AI research into reality.
What I need is 3 months of runway so I can do exactly that.
I think you need to be more realistic about your priorities right now, you need a job, any job, and a roof over your head mate.
The OS and runway for a start up can wait. This is not an investable proposition for anyone, what happens after 3 months of runway on a vibe coded OS?
Where's the go to market plan, where's the viable business here? I'm not saying none of that exists but you need to put yourself on more stable footing before working on it, otherwise it will remain a pipe dream.
Please show me a vibe coded OS containing over 20 separate projects all merging together.
A from scratch kernel, from scratch hypervisor, from scratch everything.
Theres a big difference between systema architecture with AI tooling and vibe coding a little app.
This is not a vibe coded application.
I'm a 40 year old man. If I had the option to go get a job I would.
What would I do with my dog while I'm at a job? Do I leave him in the tent while I go off for 8 hours or more each day?
I'm guessing you don't fully grasp the point I'm aiming for here.
I'm hoping someone will see this and understand that I'm better than a dead end job. My brain is very unique and has the ability to do things others can't.
Honestly, if I could just go out and get a job and a place to live in would not be posting things like this.
1.3M lines over 4 months at ~18 hours a day works out to roughly 600 lines per hour total, or about 60 lines per hour per session. One line per minute per session.
The methodology is documented on the repo.