Steven Gonsalvez

Software Engineer

← Back to Tools & Tips
Free|

Remote Coding: Running AI Agents From Anywhere (The Full Stack)

From DIY tmux+Tailscale to dedicated platforms like conductor.build and vibetunnel. Every approach to running coding agents remotely, rated.

Visit tool →

The itch

You're on the train. Or on the sofa. Or in a meeting that could have been an email. And you know your agent is grinding away on your dev machine at home but you can't see what it's doing, can't approve anything, can't nudge it. Flying blind until you're back at your desk.

Everyone hit this wall around the same time in early 2026, and a load of tools showed up to fix it. Here's what's out there, from DIY bodges to proper platforms.

DIY: tmux + Tailscale + Mosh

Still the most reliable setup, honestly. SSH into your machine over Tailscale (or Mosh if your connection is rubbish), reattach your tmux session, carry on. Free, works everywhere, been around forever. The downside is it's all terminal and you need to know your way around. Not exactly mobile-friendly either. Typing SSH commands on a phone keyboard is proper painful.

Add Termius on your phone and it gets a bit more bearable. Decent SSH client with snippets, port forwarding, and a keyboard that doesn't make you want to throw the phone. Switch to Mosh instead of SSH if you're travelling. Handles dodgy WiFi and mobile network handoffs way better.

Gotchas: scrolling back is not supported in most mobile terminal clients, which makes reviewing agent output a pain. Voice input doesn't work in terminal clients either, so you're thumb-typing everything. vibetunnel solves both of these since it's browser-based, which is a proper win.

vibetunnel

vibetunnel is the one that clicked for me. Proxies your Mac terminal into any browser. No SSH setup, no port forwarding, just a URL. Built specifically for watching Claude Code and other agents from anywhere. Has a Git Follow mode that tracks your branch switching across worktrees, which is mint if you're running multiple agents. Records sessions in asciinema format too. 4,400 stars and growing.

The mobile lot

Omnara came out of YC S25 with a voice-first angle. Your agent runs on your machine, Omnara gives you an iOS app to talk to it. Speak your coding instructions while walking the dog, basically. They pivoted from a plain CLI dashboard to this voice interface, which I reckon was the right call.

CodeAgentsMobile is a SwiftUI iOS client that connects to Claude Code over SSH. Can even provision a DigitalOcean or Hetzner server straight from the app. More hands-on than Omnara but gives you full control.

The funded platforms

conductor.build raised $22M and built a macOS app for running parallel agents in isolated worktrees. Linear, Vercel, Stripe, and Notion are on it. Closed source, but the UX is polished and the merge tooling is proper good.

Terragon Labs was free in beta and I used it loads. Absolutely brilliant. This was before Claude and Codex had their own web interfaces, so a cloud-hosted sandbox you could access from anywhere was properly useful. The writing's on the wall for standalone remote tools though. Now that the platforms ship their own web access, the dedicated remote wrappers are becoming infrastructure that gets absorbed. I reckon remote access gets baked into every coding agent within the year.

Where I'm at (April 2026)

The Discord-based agent management approach is getting interesting. Run your agents on a server, manage them through Discord on your phone. No custom app, just the Discord client you've already got.

But honestly? I reckon remote access gets baked into the agent tools themselves within six months. Anthropic, OpenAI, and Google all know this is the gap. The standalone tools are building bridges that the platform companies will build natively. Use them now, enjoy them, don't get too attached.

What I'm actually watching is the shift toward talking to your agents like colleagues rather than typing commands. Conversational agentic engineering, if you want the wanky phrase for it. That's where this lot is heading. Going to write a proper long-form piece on this because it deserves more than a paragraph.

One last thing: beware the anxiety. Running agents from your phone is way more addictive and way more annoying than you'd expect. You check it on the bus. You check it in the queue at Tesco. You check it at 2am because you can't sleep and you're wondering if that migration finished. It's a proper dopamine trap. Set up notifications (see ntfy + PingMe) and then put the phone down. Trust the process.

Share𝕏in

Comments & Reactions