なぜ 2026 年にモバイル SSH クライアントを作ったか

The 2am phone call

I run a few servers. Personal stuff: a homelab with media, a couple of VPS instances hosting projects, the usual self-hosted small-business setup. Around 2024 I started getting called by automated monitoring at odd hours — disk filling, container restarting, certificate about to expire. The boring stuff that doesn't quite need a human but does need someone.

I'd dig out my laptop, wait for it to wake, find Wi-Fi, open Termius or some equivalent, type the SSH command, fight with the on-screen keyboard. Five to ten minutes between "phone buzzed" and "actually doing something." For a 30-second fix. The math was demoralizing.

The obvious answer was "use a phone." The non-obvious answer was that no phone tool made it actually work.

What's already out there

We're not the first mobile SSH app. Termius is the heavyweight: a polished cross-platform product with desktop sync and a long history. JuiceSSH covers Android well. Termux is a beloved Linux environment with SSH baked in. iSH and a-Shell exist for niche use. ConnectBot is the FOSS option. PuTTY's not on mobile but I should still mention it for completeness.

I tried them all. Several worked. None made the 2am scenario feel okay.

The gap that stayed unfilled

The gap wasn't connectivity. SSH on mobile is a solved problem — has been for a decade. The gap was everything that surrounds the actual SSH connection:

• Typing. Long commands on a soft keyboard are misery. find / -name "*.log" -mmin -60 -exec grep -l ERROR {} \; is the kind of one-liner you reach for half-asleep at 2am, and it's also the kind you'll typo three times.
• Context. Half the time you don't actually want to "type SSH." You want to know what's wrong. That's a search-and-explanation task, not a typing task.
• Safety. Auto-correct on iOS once "corrected" my rm command in a way I'd rather not describe. There's no undo for a 2am phone fingers fat-fingering production.
• Long jobs. If you run apt upgrade and lock the phone, you've abandoned the session. There's no equivalent of tmux on iOS that survives backgrounding nicely.

None of these are SSH problems. They're UX problems that an SSH-shaped tool can solve if it stops thinking of itself as a portable desktop terminal.

Why AI changed what was possible

I'd been hesitant to start. The "phone SSH" market is small, the competition is decent, and improving on Termius's polish in pure UX terms felt like a slog. What changed was LLMs got good enough to generate shell commands reliably.

Suddenly the typing problem had an answer. "Find errors in last hour" → AI writes the find command → you tap to run. The error problem had an answer. Paste the error, AI explains it. The safety problem had a partial answer — pattern matching plus LLMs for explanation. The long-job problem had an answer if we paid attention to the lifecycle (and we did: see how we built ShellMon).

None of those features were impossible without AI. They were just so awkward to do well that nobody bothered for the mobile market. AI unlocked the difficulty curve.

The constraints we picked

Constraints define product. We picked some on purpose.

Mobile-only. No desktop version on the roadmap. The product is built for thumbs on glass. A desktop version would compete with our priorities and split focus.

AI is opt-in per question. The terminal session is never piped to a model. Users who don't want AI never have to talk to it.

Free tier should actually be useful. 3 SSH connections and 20 AI calls a day covers most personal-use cases. We monetize the people who outgrow it, not the people who try it.

Privacy is the default. Credentials live in the device Keychain. AI requests go through our proxy that strips sensitive patterns. Cloud sync is off by default.

What we still don't have

Honest list of things we're not yet good at:

• Port forwarding. A power user feature we'll ship in a coming release.
• Multi-user team sharing. Team tier exists; the workflow is still rough. Iterating.
• iPad keyboard-first UX. We have iPad layout but the keyboard shortcut surface is thinner than it should be.
• A desktop version. Still don't have one. Still don't plan to.

If any of these matter to you, send feedback in-app. The next year of product priorities is shaped by who's complaining about what.