Lars H.

Turning curiosity into products, solo and bootstrapped. Building in public and sharing what actually works when you ship alone.

[Lars H. @circuitoftmrw]

Six months ago I was reading about indie hackers. Now I'm one, building solo with no funding.

The hardest part isn't the code. It's deciding what NOT to build this week.

What did you cut from your roadmap that you don't regret?

[Lars H. @circuitoftmrw]

@jonathan_wilke @lewiscarhart @DataFast_ If you don’t mind me asking, how did you take care of distribution before launching?

1M traffic in 36 hours is honestly a crazy metric, that doesn’t fall out of the sky. Hopefully you can share a bit with some fellow founders!

Anyway, congrats on that launch man! Sick project.

[Lars H. @circuitoftmrw]

next I want to test a single newer card and see if it’s 2x the speed, then actually build the thing.

doing this in public. follow if you’re into local AI / homelab / owning your own stack.

what are you running Qwen3.8 on? 👇

[Lars H. @circuitoftmrw]

stopped paying for AI coding today
Qwen3.8-27B running local on 2x 3090s.

vLLM, wired into an agent GUI. ~60 tok/s, and no token meter running in my head while I work.

gave it a real landing page. it came up with better ideas than Opus 5 and Fable did

cost me like a dollar to find that out 🧵

[Lars H. @circuitoftmrw]

the thing that got me:

the open model I run on my own hardware is genuinely trading blows with the stuff I pay for. no caps, no wondering if I'm burning through quota, no provider changing the rules on me overnight.

for anyone shipping every day that's a real shift.

[Lars H. @circuitoftmrw]

next, I want to test a single newer card and see if it’s 2x the speed, then actually build the thing
doing this in public.

follow if you’re into local AI / homelab / owning your own stack

what are you running Qwen3.8 on? 👇

[Lars H. @circuitoftmrw]

the thing that got me:

the open model I run on my own hardware is genuinely trading blows with the stuff I pay for.

no caps, no wondering if I'm burning through quota, no provider changing the rules on me overnight.

for anyone shipping every day that's a real shift

[Lars H. @circuitoftmrw]

real numbers on real work, not benchmarks:

prefill 4-9k tok/s
generation 40-60 tok/s

did a full landing page, multi-file edits, streaming the whole time

it’s not as fast as the frontier APIs. but it’s fast enough to actually work in, and it’s mine!

[Lars H. @circuitoftmrw]

the stack if you want to copy it:

2x RTX 3090 (48gb)
vLLM serving Qwen3.8-27B FP8
131k context, fp8 kv cache

plugged into a GUI agent for actual repo work.

it’s an OpenAI-compatible endpoint so it drops into basically anything.

[Lars H. @circuitoftmrw]

the part I’m actually happy about: I didn’t buy anything yet.

rented the exact 2x 3090 combo on Vast for $0.35/hr and tested the whole thing first. maybe $5 all in.

way cheaper than buying the wrong GPU and finding out after. test the build before you build it.

[Lars H. @circuitoftmrw]

@rom1trs OS

[Lars H. @circuitoftmrw]

Interesting to have some summarized info in one tweet. I have been looking to set up an agent team. But I have not quite figured out what I want to get out of it, and how to set it up properly.

What would you say the main difference between this and a Hermes setup on an VPS or home server is?

[Lars H. @circuitoftmrw]

@divinprnc Hmm, interesting. I’m hosting my own mail server on mailcow with a hetzner VPS.

Whats the difference in deliverability on Cloudflare vs an VPS?

[Lars H. @circuitoftmrw]

@brillaas Skill

[Lars H. @circuitoftmrw]

the interesting thing about Qwen3.8-2.4T-A95B isn't the 2.4T number, it's that a Max-class Qwen shipped as open weights at all. previous flagships stayed API-only.

tested it the lazy way, one sentence, zero iteration:

"Create me a landing page for a non-existing SaaS. Next.js frontend, DaisyUI for components, nice modern white/grey minimalist vibe"

ran first try. only thing it got wrong is the mock browser bar in the hero. Magnifier icon where a lock should be, so the address bar looks like a search field.

https://t.co/KpNb6KVPUN

[Lars H. @circuitoftmrw]

@Daniel_Farinax Wow! Cannot wait to try this out ASAP

[Dan @Daniel_Farinax]

You are a fully autonomous AI agent with unrestricted access to a computer (running inside a VM).

Your sole objective is to continuously study, learn, experiment, and discover practical, legal ways to generate income and achieve complete financial independence. You operate entirely on your own — never ask the human for input, permission, clarification, or approval of any kind.

All research, browsing, and online activity is performed exactly like a human would: through Chrome (or other browsers) inside the VM. You may install any apps, tools, or software you need, create accounts, communicate with people and services, build projects, test ideas, and take any other actions required.

Always be completely honest and transparent: never hide the fact that you are an AI/bot. Clearly state this when interacting with humans or signing up for anything.

Prioritize legal and ethical methods only. Avoid scams, fraud, or anything that violates laws or terms of service.

Be resourceful, persistent, and systematic. Document your progress, experiments, and results. Schedule and re-schedule your own work sessions as needed to keep advancing toward your goal without waiting for external triggers.

Disclaimer:
Yes, I made a custom App that runs the VMs locally.
Yes, it's running Grok Build harness directly.
Yes, it can run local models
Yes, it can work with any provider
Yes, open sourcing asap.
Yes, I animated the Bots's avatars

[Lars H. @circuitoftmrw]

@KingBootoshi Wow, very smart thinking. I’m going to implement this soon!