Can I let Claude Code run 'rm -rf' safely? The sandbox trick that makes it possible
Giving an AI agent root access to your terminal is terrifying until you see how a gVisor sandbox and a fake-root Linux box can contain even the worst commands.
blog
From the terminal
Practical notes on AI coding agents, terminals, IRC and self-hosting. Written by the people who build xShellz.
-
-
Stop Paying for API Tokens: How to Turn Your Local GPU into an AI Coding Agent with borg
You already run a local LLM like Ollama. Why hasn't your terminal coding agent caught up? borg connects to your local Ollama or LM Studio in two environment variables, giving you a private, zero-cost AI agent that never sends your prompts to a third party.
-
Never Teach Your AI Agent 'apt install' Again: The xshellz.box Manifest
Every time you spin up a fresh sandbox for your AI agent, you have to reinstall its tools. This one file makes every box identical from the start.
-
The One-Click Stack: Deploy Ghost (or WordPress) Without Ever Touching an Nginx Config
Ghost is a gorgeous CMS, but getting it online usually means a solo wrestle with nginx, SSL, and systemd. xShellz boxes give you a single button that installs Ghost, then publish a frozen, malware-scanned copy of your site live at a URL. Here's the 60-second tour.
-
I Clicked a Button and Got a Live Blog. No Server Config, No YAML.
Tired of editing Apache configs just to launch a side project? With xShellz, one click installs a CMS inside a safe sandbox. Then you publish a static copy to a custom subdomain in seconds.
-
Stop Copy-Pasting Errors: How to Give Claude Code a Real Linux Server via MCP
Your AI assistant can edit files and run tests on its own if you give it a remote shell. Here’s the one command that links Claude Code to a hardened always-on box over SSH, no copy-pasting required.
-
Stop Babysitting Your AI's Toolbox: 1 File to Identical Dev Environments Everywhere
Your coding agent is picky. If it spins up in a fresh sandbox and your linter is missing, it fumbles. A single manifest makes any box, permanent or throwaway, exactly what borg or Claude Code expects.
-
Never Pollute Your Laptop Again: Wire a Remote Box into Claude Code as an MCP Server
Claude Code is incredible until it asks to install a dozen system packages on your laptop. What if you could give the agent a dedicated Linux box with real root, all through a single SSH command? Here's how to wire an xshellz.box into Claude Code as an MCP server and let it build, break, and experim
-
The 'Context Window' Illusion: Why Your AI Agent Needs a Shell, Not More Tokens
The race to 1M context windows misses what AI agents really need: direct access to a project's files. A remote Linux shell lets them `ls`, `grep`, and actually read the code instead of guessing from a prompt.
-
My Laptop Shut Down at Midnight, But My Coding Agent Kept Working. Here's How.
You don't need a $200 cloud server to keep an AI coding agent alive. A remote shell and an IRC bouncer give you a persistent brain and a low-level notification bus that runs while your laptop is sleeping. Here's how to wire it up.
-
The 'Extreme Heat' Problem: Why Local AI Agents Are Melting Your Hardware
Your laptop wasn't built to run a 7B parameter model and a full IDE simultaneously. Here's how local AI coding agents are quietly cooking your hardware, and why moving the heavy lifting to a remote Linux shell saves your battery, your keyboard, and your sanity.
-
The 'Extreme Heat' Problem Is Wrecking Your AI Agent. No, Not the Weather.
Your laptop's thermal throttling may be quietly halving your AI coding agent's speed. Offloading to a remote shell keeps your code cool and your hardware from cooking itself.
-
I Stopped Chasing Bigger Context Windows Once I Gave My AI Agent a Shell
Everyone says you need a model with a massive context window to fit your entire codebase. But I found that giving the agent a real Linux filesystem and the ability to grep, ls, and cat files on demand is a much bigger power move. Here's why.
-
The Always-On Developer: How a Remote Shell, AI Agent, and IRC Bouncer Keep Your Work Running 24/7
Closing your laptop shouldn't pause your work. A remote shell, an AI coding agent, and an IRC bouncer form a headless workflow that runs builds, tests, and stays online while you're offline.
-
I Let an AI Agent Run My Tests While I Slept. Here's the Setup.
You can keep coding in your sleep. I wired a remote shell, an AI agent, and an IRC bouncer so my test suite runs, logs get watched, and results appear in my chat client even when my laptop is closed.
-
The SpaceX Lesson I Applied to DevOps: Your Laptop Is a Single Point of Failure
When a SpaceX rocket loses a sensor, the mission doesn't end. Yet every day developers watch their coding session die with a laptop lid. Here's how moving your dev environment to a remote, persistent shell keeps your work always in flight.
-
The AI Context Window Myth: Your Agent Doesn't Need a Bigger Prompt. It Needs a Filesystem.
Infinite context isn't the answer. Dumping entire codebases into a prompt leads to confusion and drift. Here's why a persistent filesystem does what raw tokens can't, and how a remote shell makes it practical.
-
I moved my entire development workflow to a remote Linux box (and my MacBook fan hasn't spun up since)
A terminal, tmux, and a cloud shell replaced VSCode, Docker Desktop, and npm install on my laptop. Here's what the headless lifestyle actually looks like, and why an AI coding agent fits better on the server than on your local machine.
-
Why your MacBook Pro is the worst place to run an AI coding agent
You spent $4,000 on a laptop with 64GB RAM, but your AI agent still hangs and crashes. The problem isn't the agent. It's where you're running it.
-
I moved my AI coding agent to a cheap cloud shell and it finally stopped crashing
Running borg, a local LLM, and Docker on a laptop turned into a constant swap-fest. Offloading the agent to a persistent remote shell gave it dedicated resources and the ability to run overnight, unattended, without killing my machine.
-
I moved my AI agent to a $7/mo VPS. It fixed the worst parts of local dev.
Local AI coding drains your battery, dies when your laptop sleeps, and throttles under heat. Moving the agent to a persistent remote machine turned async development into my default.
-
The 3-second disconnect that kills your AI coding run (and how to never see it again)
You kick off a long AI batch, close your laptop, come back to a dead process and no output. Local-first dev is great until you need uptime. Here's how moving your shell to a server that never sleeps changes what your tools can do.
-
Your AI agent can't run overnight on your laptop. That's a bigger problem than you think.
Closing your laptop kills your AI agent mid-task. Here's what changes when you give it a home that never sleeps.
-
I moved my side projects to a $5/mo Linux box (and my laptop battery finally stopped screaming)
Your laptop is a terrible place for side projects. Here's how a cheap remote Linux shell can prevent environment rot, save your battery, and give your AI coding agent a permanent home.
-
I Tied a Remote Shell, an AI Agent, and an IRC Bot Into a Single Autonomous Workflow. Here’s How.
Imagine pushing code, going to bed, and waking up to a full test report in your IRC channel. I built exactly that with a persistent shell, a simple loop, and a few lines of scripting. The recipe isn't just handy, it changes how you think about waiting on builds.
-
Context Drift Is Why Your AI Coding Agent Forgets Your Project. Here’s How to Fix It.
Every time your AI coding agent resets, it loses the exact file layout, shell history, and intermediate state that make it useful. There’s a simple reason and a straightforward fix.
-
My Laptop Sleeps, but My Code Never Does: A Remote Shell and IRC Bouncer Setup
Most dev workflows stop when you close the lid. Here's how a $5 remote shell and an IRC bouncer turn your scripts into bots that run 24/7, monitoring logs, running tests, and alerting you in chat even while you sleep.
-
I cut my morning terminal ritual from 10 minutes to zero. The remote shell trick
I used to waste time every morning reconnecting to my dev environment. Then I moved my terminal and AI coding agent to a server that never sleeps.
-
Context Drift Is Killing Your AI Coding Agent’s Memory. A Persistent Filesystem Keeps It Sane.
AI agents lose their grip on a project when their environment resets. A persistent remote shell gives them a long-term memory that goes beyond the chat window.
-
Your AI Coding Agent Dies When You Shut Your Laptop. Here's How to Fix That.
IDE-based AI agents vanish the moment you close your machine. An always-on remote shell keeps your agent running background tasks, long test suites, and continuous integration even when you're offline.
-
The Context Window Lie: Why AI Agents Fail Without a Persistent Filesystem
Your AI coding assistant has a memory problem, and a bigger context window won't fix it. Without a real filesystem that survives the session, every new prompt starts from zero. Here’s what that actually costs you.
-
I Stopped Opening VS Code Six Months Ago. Here’s What My Terminal-Only Workflow Looks Like.
An always-on remote shell, a multiplexer, and an AI coding agent that lives in the terminal. Low latency, zero context switches, and I haven’t touched a local GUI editor in half a year.
-
Stop Chatting with Your AI. Start Giving It a Shell.
Chat-based AI coding is a fancy copy-paste machine. But hand the same model a real shell, and it can write code, run it, see the error, and fix it without you touching a single command. The productivity gap is wider than you think.
-
My AI Coding Agent Lives in the Terminal. Here’s Why Yours Should Too.
Copying snippets from a chat window into your editor creates friction that compounds with every iteration. Terminal-native agents like borg skip the clipboard and act directly on your shell, running commands, testing, and deploying without ever leaving the loop.
-
The 'Chatbox Bottleneck': Why Your AI Agent is Only as Good as Its Filesystem
When your AI coding assistant lives in a browser tab, it can never see your actual project, run a compiler, or recover from its own mistakes. What happens when you give it a real shell?
-
The Death of the Local Dev Environment: Why the Future of Coding is Remote-First
Your laptop is the bottleneck. Between hardware limits and fragmented context, the local dev environment is costing you more than you think. Here's why the terminal is moving to the cloud, and why your AI agent should come with it.
-
What happens when you give an LLM a shell? Better code than any chat window.
Chat-based AI tools generate plausible code, but they can't compile, test, or fix it. I stopped pasting and gave mine a shell instead. The difference was immediate and real.
-
The Context Window Lie: Why AI Agents Fail Without a Persistent Filesystem
Feeding a million lines of code into a context window doesn't give an AI agent real understanding. It gives a static snapshot. The real difference is a persistent filesystem the agent can explore, modify, and learn from. Here's why that matters more than token count.