frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Built a 1.3M-line agent-native OS in Rust while homeless. What now?

10•jamieoglindsey•14h ago
I’m going to be straight about my situation because I don’t know where else to turn.

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

Comments

Dumblydorr•10h ago
That’s a tragic story, sorry to hear it.

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.

jamieoglindsey•8h ago
Thanks for the kind words.

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)

kingkongjaffa•9h ago
Sorry to hear about your situation.

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!

what•8h ago
It’s just ~8 lines per minute if you work 24/7 for 4 months. Maybe they don’t sleep.
jamieoglindsey•8h ago
10-12 parallel AI coding sessions running simultaneously.

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.

jamieoglindsey•8h ago
I appreciate the thought, genuinely.

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.

kingkongjaffa•8h ago
> 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.

matternous•7h ago
Don't post generated comments or AI-edited comments. HN is for conversation between humans.

https://news.ycombinator.com/item?id=47340079

rl3•58m ago
What makes you think anything this person wrote contains generated or AI-edited comments?

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.

-1•7h ago
I'm going to try to provide what I think is the most practical advice I could give, because I don't think you're being realistic. You are at the bottom of Maslow's hierarchy of needs right now. You need proper shelter, food etc. You need to get yourself above the second level ("safety needs") before spending time on highfalutin efforts like monetizing an OS or trying to get angel investors to swoop in and save you. I think you should try to get a job unrelated to IT (since the job market sucks), perhaps as a tattoo artist since you have experience in that? Get yourself and your dog out of the tent and safe first. Good luck to you.
brettgriffin•4h ago
> That’s what dies if I can’t keep going.

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.

Ask HN: How is AI-assisted coding going for you professionally?

288•svara•14h ago•476 comments

AI tools are making me lose interest in CS fundamentals

55•Tim25659•4h ago•45 comments

Ask HN: How to Learn C++ in 2026?

8•creatorcoder•8h ago•8 comments

Ask HN: How is your team collaborating while working with coding agents?

2•wek•4h ago•1 comments

I'm 60 years old. Claude Code killed a passion

197•fred1268•17h ago•158 comments

Tons of new LLM bot accounts here

12•koolala•11h ago•13 comments

Ask HN: Have you successfully treated forward head posture ("nerd neck")?

48•trashymctrash•1d ago•39 comments

Ask HN: What breaks first when your team grows from 10 to 50 people?

8•hariprasadr•18h ago•6 comments

Built a 1.3M-line agent-native OS in Rust while homeless. What now?

10•jamieoglindsey•14h ago•11 comments

I traced $2B in nonprofit grants for Meta and Age Verification lobbying

91•theseusares•2d ago•20 comments

Ask HN: What was it like for programmers when spreadsheets became ubiquitous?

11•yodaiken•1d ago•11 comments

Tell HN: iPhone 6s still getting security updates

6•uticus•1d ago•2 comments

Ask HN: Med student interested in bci startups..where do I start?

2•Sxouterred•23h ago•3 comments

Ask HN: Got cancer, a new job,new boss in less than a year What do I do now?

19•Goleniewski•2d ago•18 comments

Ask HN: Why is there a lack of useful use cases for OpenClaw?

3•nazbasho•10h ago•5 comments

Ask HN: Why can't we just make more RAM?

25•chatmasta•2d ago•21 comments

How not to fork an open source project

5•jsattler•1d ago•1 comments

MiniMax M2.5 is trained by Claude Opus 4.6?

10•Orellius•2d ago•10 comments

Ask HN: What will you be doing for the next 10 years?

8•smarri•11h ago•9 comments

Ask HN: How do you use Coding Agents/CLIs out of coding?

4•arbayi•1d ago•6 comments

X is selling existing users' handles

197•hac•4d ago•91 comments

Ask HN: What's your biggest pain point when joining a new developer team?

8•KevStatic•2d ago•15 comments

Ask HN: Is Claude down again?

86•coderbants•4d ago•73 comments

Ask HN: Why have co-ops never played a major role in tech?

14•AbstractH24•2d ago•8 comments

Toolpack SDK, an Open Source TypeScript SDK for Building AI-Powered Applications

2•sajeerzeji•1d ago•1 comments

Ask HN: 100k/year individual token usage?

7•alecsmart1•2d ago•4 comments

Prompt to make Claude more autonomous in web dev

5•louison11•1d ago•1 comments

Claude broke a ZIP password in a smart way

7•jgrahamc•1d ago•2 comments

Ask HN: Would this eliminate bots for good?

3•piratesAndSons•1d ago•12 comments

Generate tests from GitHub pull requests

7•Aamir21•2d ago•7 comments