# why i moved claude code off my laptop and onto a remote linux box (tailscale + mosh + tmux) — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/78b9ff5f5d3720b8

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-09-02 14:26 ingested 2026-09-03 02:38

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1w5bx9x/why_i_moved_claude_code_off_my_laptop_and_onto_a/)

## retained text



```
I used to carry my laptop half-open through dinners and airports, waiting for Claude to finish a run. Walking through dinners, airport gates, even a beach in Australia with sand getting everywhere (and I mean everywhere), and always trying to keep the lid cracked so the session won't die. I needed Claude running 24/7 with no excuses. 

We moved to the most used setup that I saw on twitter and on reddit:

* **Hetzner Cloud:** A simple Linux server (4 cores, 16GB RAM for \~€19/mo).
* **Tailscale:** Private mesh network linking laptop, phone, and box. No public SSH ports exposed to the internet.
* **mosh:** Replaces standard SSH so network drops or sleeping devices don’t kill the connection. It runs inside the Tailscale tunnel.
* **tmux:** Keeps sessions persistent. Claude runs inside tmux, so when you disconnect, it keeps working. Reattach from phone or laptop anytime.
* **Tailscale Funnel:** For testing webhooks without ngrok or opening external ports.

The setup was good for me, but making it a toolitising (is this a word?) it to something the whole team can use was challenging. So I created a CLI that allows the team to spin up a pre-configured (with all the tools we use) Hetzner server in no time. It changed the way we work. My teammate even took it to the next level with [nono.sh](http://nono.sh) and sandboxing. 

I now have have 3 servers with 3 claude accounts running, and all of my team uses it. 

It also fixed the permissions issues:

Auto-mode prompts A LOT of permission requests - like a LOT, so I wanted to run with full bypass permissions but with limited access to services we use, so I restricted what claude can touch. It has access only to:

\* Scoped API keys and read-only DB credentials

\* Read access to code and PR-only git permissions (cannot merge to main)

\* Isolated test environments



With those boundaries, Claude runs on bypass mode safely for long sessions.

Curious how others are handling long-running sessions, local laptops or dedicated boxes?
```



document fetched 2026-09-03 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1w5bx9x/why_i_moved_claude_code_off_my_laptop_and_onto_a
