11:47pm. Disk at 98%. App down. 14 users still online.
ChatGPT gave me the fix in 8 seconds.
rm -rf /var/log/*
Looked fine. I was exhausted. My finger was on Enter.
Then I read it again.
Wrong path. Would've taken out more than logs.
I didn't run it.
Not because I'm a better sysadmin. Because I had to see the command, approve it, or kill it — no auto-run, no "trust me bro" from an AI.
That's the whole reason I built Ohuriya.
If you run your own VPS and you're not a full-time ops person, these are the 3 tasks where I never hit Enter blind:
1. Anything with rm
Logs, caches, old deploys — one typo and you're writing a postmortem.
2. Anything that restarts a servicesystemctl restart nginx on the wrong box = your SaaS is offline while you sleep.
3. Anything that touches SSL / nginx config
"Quick cert renew" turns into 45 minutes of 502s because one line was wrong.
Ohuriya workflow:
Connect your VPS in ~30 seconds (one curl line)
Say what you need in plain English
See every command before it runs
Approve · edit · reject
No OpenAI keys. No monthly plan. Prepaid credits — you pay when work actually runs.
I'm looking for 5 founding users — indie hackers / solo founders who manage their own Hetzner, DO, or Linode box.
Free 15-min onboarding on your real server. We'll run a harmless check together, then one fix you actually need.
If you've ever hesitated before Enter on prod — you're exactly who this is for.
Hello. I read your post.
I strongly resonated with both the story and the product's direction.
In particular, the concept of "preventing the danger of directly executing commands generated by AI" feels very grounded in real-world situations and quite essential.
I develop AI, API integration, and infrastructure as a full-stack developer, and I'm also interested in VPS environments and DevOps automation tools.
I believe that a design like Ohuriya, which "inserts a safety layer into AI execution," can be implemented quite quickly at the MVP stage.
If you're interested, I'd love to be involved not just as a user, but also in building the product together, including the technical aspects.
First, If you are interested in this, feel free to contact me.
Thank you.
Hi Yuki... sounds good. let's connected.
One thing I'd be careful with:
The interesting question may not be whether founders want to approve commands before they run.
It may be what specific risk they're actually trying to avoid when they do.
Those sound similar, but they can lead to very different decisions about positioning, onboarding, and who the first customers really are.
I wouldn't make that call too quickly from early user conversations.