Simon Høiberg
@SimonHoiberg • 165,366 subscribers
Building a portfolio of bootstrapped SaaS products. - https://t.co/XguzG2JTiQ - https://t.co/I9Idzbxkj9 - https://t.co/liUkf3Z2eB - https://t.co/oJEhvc0ZrW
Shorts
Videos
You can now get Opus 5.5 quality out of Qwen 3.8 27B. You just have to teach it the job first. We let a frontier model work through thousands of realistic support cases, keep only the ones that meet our standard, and train a small LoRA on Qwen with them. Now it runs on our own hardware. No API bill, no rate limits, and nobody can change the deal on us 👇
Simon Høiberg97,468 次观看 • 3 天前
We used to pay AWS close to $8,000 a month. Today, our entire company runs on 3 rented bare metal servers at Hetzner. That's 4 SaaS products with more than 50,000 users (apps, databases, background jobs and monitoring included). The bill is around $300 a month. Everyone told me it would be unstable. It's been the opposite. Our uptime is close to 100%. The other day, we had about 15 seconds of downtime. An AI agent caught the alarm, fixed it and sent me the report on Telegram. I read it when I woke up, and not a single user noticed 👇
Simon Høiberg80,057 次观看 • 17 天前
I honestly didn't think anyone would catch up with ElevenLabs this fast. Natural voice cloning took big models, tons of training data and a lot of money. That's why everyone ended up paying for it. Now there's VoiceStudio. Open source, and it runs on your own machine. Clone a voice from a short recording, or design a brand new one just by describing it, in 646 languages. And your agents can use it too. OpenClaw or Claude Code can produce a voiceover or a dub on their own, without anyone ever opening the app. Have a listen and judge for yourself 👇
Simon Høiberg15,642 次观看 • 6 天前
We have a problem! There's a huge token rug pull coming. Right now, Claude and Codex max subscriptions are SUPER generous, and it won't last. We've seen it before: - First, a huge AI credit giveaway. - Then the models get dumber. - Credits start running out faster. - Prices jump / the offer ends. If you rebuilt your company to be entirely dependent on OpenAI's / Anthropic's generosity, you're in for a bad surprise soon 👇
Simon Høiberg58,983 次观看 • 2 个月前
I'm never using stock video again 💩 AI is now perfectly capable of creating custom, high-quality b-roll (with me in it!) Here's the workflow I built: 1️⃣ Custom ChatGPT I describe what I want in my video. It forwards the request to n8n. 2️⃣ n8n workflow n8n takes over and starts working. It uses Replicate API to start the AI tasks. - It creates a starting frame. - It turns it into a video. - It upscales it to 4K. 3️⃣ I get the final result in NocoDB Here I can download a high-quality 4K b-roll shot for my videos. Or I can re-run, tweak and refine if needed. ✨ Models used: - Flux Dev (LoRA) - Kling Pro 2.1 - Video Upscaling (Topaz Labs) All available through Replicate.
Simon Høiberg116,634 次观看 • 1 年前
I'm using Kubernetes to run my products on Hetzner. But not because it needs to scale - but because it's super practical. Whenever I hear people say "Kubernetes is overkill" I think they're mostly missing the point. I want a "PaaS"-like experience, but self-hosted. And my setup does this super well: - Forgejo (self-hosted git) - Docker images - Kubernetes (k3s) And this doesn't have to be insanely complex. It can be very simple and very practical when self-hosting SaaS 👇
Simon Høiberg14,783 次观看 • 1 个月前
For years, I thought DynamoDB was both brilliant and convenient. - Schemaless - Serverless - Cheap What's not to like? Well, turns out, it's only cheap for a while. And the schemaless nature of DynamoDB quickly starts working against you... One of the stupidest decisions I made. Instead, I ditched DynamoDB and went with some boring old tech. Let me explain 👇
Simon Høiberg38,594 次观看 • 4 个月前
Since we moved from AWS → self-hosting on Hetzner, our products have been much more stable/reliable. Our uptime is near-100% and we're never blocked from shipping. So the whole "self-hosting is fun until servers break at 2 AM" is a meme. You know which services go down all the time? - Supabase - AWS - GitHub - Cloudflare Ask my competitors 😬 They would know!
Simon Høiberg25,491 次观看 • 2 个月前
I'm using Kubernetes to run my products on Hetzner. But not because it needs to scale - but because it's super practical. Whenever I hear people say "Kubernetes is overkill" I think they're mostly missing the point. I want a "PaaS"-like experience, but self-hosted. And my setup does this super well: - Forgejo (self-hosted git) - Docker images - Kubernetes (k3s) And this doesn't have to be insanely complex. It can be very simple and very practical when self-hosting SaaS 👇
Simon Høiberg35,569 次观看 • 4 个月前
I quietly launched something 👀 And it changes EVERYTHING. The best part? ...SLOP! I can't fking take this AI slop writing anymore. So we just open-sourced our internal list of AI slop writing patterns that we use in FeedHive, for YouTube script writing, and much more. +120 patterns (and the list is growing). Most comprehensive list on the internet. Available on GitHub (including Claude skill). And a tool you can use for free. See links below 👇 ...let's face it, we're all using AI to write now. At least, let's teach it how to not write total slop.
Simon Høiberg11,250 次观看 • 1 个月前
Grafana is so underrated! We self-host it - not just for system monitoring, but analytics in general. - We track and store raw event data. - We visualize it with Grafana. - We use AI to build the dashboards. Beats Amplitude, PostHog, etc, any day of the week. I've made MUCH better data-driven decisions by hosting my own product analytics.
Simon Høiberg28,155 次观看 • 3 个月前
The single best use of AI right now is building your own internal tooling and custom harnesses. What you need: - Omarchy Linux - OpenClaw/Hermes What you don't need: - Claude Code - Grok Bot - macOS/Windows - Cloud for everything - 19 generic SaaS tools 2025 was a big revenue-growth year for my company. This year has been all about building sovereign company infrastructure and preparing for the next few years of building with AI. Let me show you 👇
Simon Høiberg17,433 次观看 • 2 个月前
This year, I've gradually left AWS. Service by service. Slowly, but surely. If I had to start over today, there is *one* service I would prioritize the highest: DynamoDB. For years, I thought DynamoDB was both brilliant and convenient. - Schemaless - Serverless - Cheap What's not to like? Well, turns out, it's only cheap for a while. And the schemaless nature of DynamoDB quickly starts working against you... Instead, I ditched DynamoDB and went with some boring old tech. Let me explain 👇
Simon Høiberg55,975 次观看 • 11 个月前
I'm blown away by Qwen 3.8 27B. A few months ago when I was testing open-weight models, it seemed clear that only GLM and Kimi class models (+1T parameters) could get close to GPT-5.6 / Opus / Fable levels. And now this tiny 27B model gets almost as close. I gotta say, China got this down to an art. Not just benchmarks. I actually tested this in my companies 👇
Simon Høiberg16,325 次观看 • 2 个月前
I'm leaving the cloud 👋 +80% of my SaaS portfolio now runs on bare metal. Why? I have at least 3 major reasons. One of them is, obviously, cost. This year, I've been able to reduce my monthly cloud bill on AWS from $7800 on average to below $1000 (last month). It's a very noticeable cost reduction. One question I get often: Is it really worth it? Considering the time I now need to spend managing my own dedicated servers. Short answer: Yes. In fact, running your products using OS software is surprisingly "easy" - although it comes with a few gotchas. Let me explain 👇
Simon Høiberg35,928 次观看 • 1 年前