Practical notes on AI coding agents, terminals, IRC and self-hosting. Written by the people who build xShellz.
Why pay per token when your own machine can run a coding agent? Point the borg terminal agent at a local Ollama instance and never hand your code to a cloud model again. Here's how in one command and a few environment variables.
Your AI coding agent gets a real shell. If it guesses the wrong path, it can wipe your machine. Here’s how to give it real root inside a sandbox that can never touch your files.
Every time you start a new AI coding session, the agent re-reads your entire codebase, burning tokens and time. Borg's built-in learn command creates a persistent mental model of your project so you never have to start from scratch again.
A single manifest file in your home directory can make every xshellz.box an identical clone of your toolchain. Here's how to stop fighting your local environment and hand your AI agent a known setup every time.
Every time you spin up a new sandbox, you're reinstalling the same tools. Write them down once and every box you boot will configure itself, no babysitting required.
Forget config files and package managers. One command is all it takes to get a full Hugo site running on your xShellz box. No installs, no scaffolding scripts, just a working blog you can poke with curl.
One command connects your local AI to a hardened remote Linux box, skipping SSH keys, bastion hosts, and permission headaches. Here's how to turn Claude Code into a remote sysadmin in under a minute.
Claude Code is great until you close the lid. Give it an always-on, root-access sandbox that survives your laptop sleeping and network drops. One command wires it up.
You wouldn't manually assemble your project's dependencies every morning. Why treat your dev environment any differently? An xshellz.box manifest turns your toolkit into a file, so every new box you spin up is instantly ready to code.
Laptops sleep, Wi-Fi drops, terminals close. A persistent remote shell keeps your workflows running around the clock and it changed how I build things.
Every LLM coding agent you use is quietly losing track of your project the moment its token window scrolls past. There is a better way: give the agent its own Linux shell with a real filesystem, and let it use ls, grep, and cat like any other developer. Here is why that changes everything.
Local AI coding agents can push your laptop's CPU past 100°C, triggering thermal throttling and hardware degradation. I moved my AI workloads to a remote shell and my temps dropped to 45°C overnight.
Most AI coding tools live inside your editor and shut down when you close the laptop. But a real agent needs a home: a Linux box where it can run long jobs, manage its own filesystem, and keep working even after you disconnect.
The real productivity shift is not a new IDE. It's a shell that survives your laptop sleeping. Here's how asynchronous agentic workflows changed the way I ship code.
Most developers lose hours of AI agent progress when their laptop goes to sleep. A remote, persistent shell turns that problem into a superpower and transforms you from a coder into an orchestrator.
Long-running AI agents and build scripts don't mix well with sleep timers, CPU throttling, and flaky hotel Wi-Fi. Here's what happened when I moved my dev loop to a persistent remote shell.
SpaceX designs rockets to fail fast and learn faster. I applied the same logic to my test environments. Now I use remote, disposable Linux shells to sandbox everything from risky commands to AI refactors. Here's the workflow that keeps my laptop clean and production safe.
SpaceX's rapid prototyping comes from treating hardware as disposable. Your dev environment shouldn't be precious either. Here's how remote shells and AI agents can give you the same 'fail fast, fix fast' loop that local machines can't match.
Most developers stop shipping when they close their laptop. I stopped closing mine. My best work runs on a remote shell, orchestrated by agents, long after I've gone to bed.
Large context windows feel like infinite memory for AI coding. They're not. A persistent file system is what turns a clever LLM into an agent that actually ships code.
Every time I opened my laptop, I lost hours to dependency hell and resource contention. Then I moved my entire development environment to a small Linux VM. Now I code from any device, anywhere, and my environment is ready in 10 seconds flat.
When your laptop goes to sleep, background AI tasks crash without warning. A persistent remote Linux shell keeps your agent running through the night, no WiFi required.
You close your laptop lid and your AI agent's training run dies. Your CI test suite was 90% done. That shell was your project's home, and it just got evicted. Here's how I moved everything to a persistent remote Linux shell and stopped losing progress.
Most AI coding assistants live inside your editor and vanish the moment your laptop goes to sleep. The real magic happens when you cut the cord and let your agent run on a persistent remote shell.
Long R computations drain your battery and die when you close the lid. A persistent remote shell keeps them running 24/7 with zero drama.
Your 'free' local AI coding tool has a hidden hourly rate. Hardware depreciation, electricity, and the punishing productivity tax of interrupted agentic loops can turn a $0 tool into a $200/month drain. Here's the breakdown, and a ridiculously cheap alternative that never sleeps.
Curious where my focus kept breaking, I logged every interruption: npm install loops, Docker restarts, and AI agent drops. The total was a full day each week. Here's the breakdown and the one change that gave it back.
Long-running AI coding agents, compile jobs that survive disconnects, and an IRC bouncer that never logs off: the remote-first workflow changes everything. Your laptop's job is just to get you to the real machine.
Laptop lids are side project serial killers. A persistent remote shell costs pocket change and keeps cron jobs, background agents, and long-running builds alive 24/7. Here's what changed when I moved off localhost.
E2B is a great SDK for spawning sandboxes from your app. But if you just want a persistent Linux box to hand Claude Code or Codex, xShellz Agent Shell is simpler and always-on.
Daytona is strong infrastructure for running AI-generated code at scale. If you want a personal, always-on Linux box to hand your agent, xShellz Agent Shell is the simpler path.
A practical, honest comparison of where to run your AI coding agent: E2B, Daytona, Modal, and xShellz Agent Shell. Pick by the job you actually have.
Most developers still treat the terminal like it lives on their local machine. That means every long-running build, refactor, or migration dies the moment their Wi-Fi hiccups. A persistent remote command center changes the rules entirely.
Ever left an AI agent running overnight only to find a dead laptop and zero results? The local dev loop is a single point of failure for agentic tasks. Here's the always-on alternative that actually finishes the job.
The local terminal is a single point of failure. Here is why persistent remote shells and AI agents that run while you sleep are not just convenient. They prevent the quiet, repetitive data loss most developers just accept.
AI coding agents run for hours, but a single sleep event or flaky Wi-Fi can wipe their context and waste compute. Here's why I moved my agent off my MacBook and onto a persistent remote shell.
You shut the lid expecting your terminal session to be there tomorrow. It's not. A persistent remote shell changes that, and it gave my AI coding agent a place to keep working while I slept.
A single sleep mode or network blip can kill a long running AI agent task. I moved my workflow to a persistent remote shell and never looked back.
A single sleep event, a brief Wi-Fi blip, or a forced OS update can erase hours of terminal work. The fix is simpler than you think.
AI models keep getting smarter, but if your terminal environment can't survive a network hiccup or a 2-hour compile, you're still dead in the water. Here's what actually matters.
A smarter AI model won't stop your SSH session from dropping or your laptop from killing its context when the lid closes. The real bottleneck is the environment where your code actually runs.
Every AI coding tool brags about its giant context window. Those tokens vanish the moment a conversation gets too long. Here's the real way to keep your agent from forgetting everything mid-session.
Local LLMs promise privacy and speed, but when you ask one to refactor ten files at once, it hits a wall that has nothing to do with inference. The real bottleneck isn't the model. It's your laptop's file system.
A step-by-step guide to spinning up turborg - an always-on AI IRC bouncer - and every connection option, from SASL to channels.
Channels, DMs, slash commands, AI /tb commands and Polish - a full tour of the modern web IRC client built into turborg.
Turboflows turn plain English into working IRC bot automations - triggers, AI steps, memory and a 40+ flow gallery. Here is how they work.
Skills are the quickest way to teach your turborg a trick - a !command, an AI reply or a moderation rule, all built in the browser.
Quiz scores, karma, seen-lists - how turborg remembers things between messages, and how to inspect and manage that storage.
The models behind turborg, your rolling AI credit budget, the /tb commands, and AI Polish - everything AI in your IRC channels.
You close your laptop, the AI agent dies. Put it on a server and it'll finish that refactor while you sleep. No local CPU drain, no lost context.
A terminal AI agent like Borg paired with tmux keeps your coding context alive across flaky wifi and reboots. One simple change gave me a stateful dev environment that never quits.
Running a dev environment on AWS or GCP felt mandatory until I realized I was paying for infrastructure I didn't need. A $5 remote Linux shell replaced the whole setup and made my workflow faster, not just cheaper.
Agents like Claude Code and Borg thrive on multi-hour compute runs. Your laptop’s sleep timer, flaky Wi-Fi, and thermal throttling kill those runs. There’s a much simpler path.
Your local machine sleeps, drops WiFi, and drains battery. Here's how a persistent remote shell lets your AI coding agent run long refactors and tests while you're offline.
When I gave my AI coding agent a permanent home on a remote server, it finished a three-day refactor without burning a watt of my local battery.
Environment drift and dependency hell aren't inevitable. A persistent, always-on remote Linux workspace can eliminate the "works on my machine" problem and make your whole team more consistent.
Most engineers treat IRC as a chat relic. I turned it into a safety net that catches every server alert, even when my laptop lid is closed. No new SaaS required.
My laptop kept killing long-running processes, environment drift ruined demos, and every deploy was a ceremony. A single remote shell turned all of that into two commands.
A hands-on look at how three popular AI coding tools fit into a terminal-centric workflow, and why an agent that lives in your shell changes the equation.
Local development environments break in creative ways. A persistent remote Linux shell gives you one environment that survives broken laptops, OS updates, and dependency hell. Here's why managed hosting can save you more time than you think.
AI coding tools are moving beyond tab-complete. Terminal agents like borg run commands, debug environments, and manage files right inside your shell, cutting context switching compared to IDE assistants. Here's why that matters.
Christina
xShellz concierge