Server systems
run by your
coding agent
From side project to billion-dollar company.















Bank-vault secure
- No open web ports: your servers connect out to Cloudflare, which filters out attack traffic before it reaches them
- A private network between your servers
- No production keys on laptops: a stolen laptop can't reach your servers
- Backups locked at a separate storage provider for 14 days by default, so an attacker can't delete them
- Zero open web ports
- Private network
- No keys on laptops
- Locked backups
Massively scalable
- Simple server fleets like these have run billion-dollar companies
- Grow when real load asks for it, not before, so you never pay for servers you don't need yet
- Automatic database failover when the main database server fails
- Need more? Just add servers. Up to thousands of them.














































































A fraction of the cost
- Plain Linux servers from any provider, or your own hardware
- No platform fees, and fully open source
- No lock-in: switch server providers anytime
- Built to need less than an hour of your time in a normal week
Cost of hosting the same app
Shorter is cheaper. A 2 vCPU, 4 GB server costs about $6/month at OVHcloud (US) or Hetzner (Europe), vs. $85/month on Render or $150/month on Heroku — 10 to 30 times more for the same hardware.
Built in, so there's no extra bill
- On-call alerting instead of PagerDuty, incident.io
- Server and website monitoring instead of Pingdom, UptimeRobot
- Status page instead of Statuspage, Instatus
- Outside alarm instead of Healthchecks.io, Dead Man's Snitch
- Database failover and backups instead of Amazon RDS, Heroku Postgres
What you still pay for: your servers, backup storage, phone alerts, and your coding agent.
Nearly runs itself
- Your agent answers alerts first and wakes you only if it can't settle the problem in a few minutes
- Restore drill every week: your backups are restored onto a test server and checked
- An outside alarm that alerts you even if every server is down
An example night
- MMWatchdog sees the app stop answering
- MMHotline hands the alert to the triage agent
- The triage agent finds the cause and settles it
- Nobody gets woken up
If the agent can't settle it in a few minutes, MMHotline wakes the person on call.
What about Kubernetes?
A lot of people think Kubernetes is a whole production system. It isn't, and it doesn't claim to be. Kubernetes runs and scales containers. It's the right tool when you have thousands of services and dozens of teams.
What a production system needs
| The job | The Stack | Kubernetes |
|---|---|---|
| Set up and change your servers | MMX | Kubernetes: Not its job |
| Release your app without downtime | Health-checked deploys | Kubernetes: Yes |
| Grow when the load grows | Tested procedures for adding and replacing servers | Kubernetes: Yes |
| Run a database that survives a server failure | Postgres with a standby and automatic failover | Kubernetes: Not its job |
| Keep backups, and prove they restore | Locked offsite backups, restored in a drill every week | Kubernetes: Not its job |
| Watch every server and service | MMWatchdog | Kubernetes: Not its job |
| Reach a person when something breaks | MMHotline, with on-call schedules | Kubernetes: Not its job |
| Triage alerts before waking anyone | The triage agent | Kubernetes: Not its job |
| Keep attackers out | Cloudflare in front, a private network, no keys on laptops, and more | Kubernetes: Not its job |
| Know what changed, and rebuild from scratch | Every change is a commit | Kubernetes: Not its job |
If all you need is zero-downtime deploys and scaling up and down, the Stack does both. Your coding agent handles them with tested procedures.
“So I think Kubernetes is overused, because people just don’t quite understand what it does.”
Everything you get
Built to be understood
- Maxxed for monkey/human understanding
- Built and run by your coding agent, Claude or Codex
- Ruthlessly simple: plain Linux, WireGuard, SSH, etc.
- Convention over configuration, so no decision fatigue
- Your whole system is plain files in your own Git repository
Cost and getting started
- Hosting 10-30x cheaper than Heroku, Railway, Replit, and similar hosts
- Fully open source (Apache 2.0)
- Set up in a few hours
- Migration guides for 19 platforms (Heroku, Railway, Replit, Vercel, etc.)
- Guides for 11 app frameworks (Rails, Django, Laravel, Next.js, etc.)
- Runs on any Linux VPS or bare metal
- No lock-in: switch server providers anytime
Deploys and scaling
- Zero-downtime, health-checked deploys
- Every change is a commit, so everything is auditable
- Runbooks for the common events: add a server, lost laptop, database restore, etc.
- Autoscaling app servers, within limits you set
- A planned scaling path, each step triggered by real load
- Scales to thousands of servers
- Automatic database failover
Security
- No open web ports: everything goes through Cloudflare
- Encrypted private network between servers (WireGuard)
- Compromised-laptop protection: no production keys on laptops, 2FA login
- Break-in response guide, with an emergency lock-out button
Alerting
- Monitoring and alerting built in
- Your agent triages alerts first and wakes you only if needed
- On-call scheduling
- Offsite dead man's switch: alerts you even if every server is down
- Offsite status page that stays up even if every server is down
Backups
- Offsite database backups, restore-tested every week
- Deletion-proof backups: even you can't delete them for 14 days
- Restore the database to any minute
- Offsite backups of user uploads, also deletion-proof
How it works
- 1Your coding agent builds it It sets up the whole Stack in a few hours and explains any part you ask about.
- 2Your coding agent changes it Every change is made with the free tools and saved as a commit you can review.
- 3The triage agent watches it It answers alerts first and wakes you only when it needs you.
Who created the Stack
Matt Jaynes created the Stack. His past customers include:
Over 25 years of production workAuthor of Taste Test: Puppet, Chef, SaltStack, AnsibleFormer #1 karma leader of Hacker News
The Stack is built on what he learned from hundreds of systems and dozens of system rescue projects.
What you need
- A coding agent you trust Claude or Codex. It does the setup and the changes, so you don't need to be a systems engineer.
- An app A web app, a website, or an API. The Stack supports Postgres: your app uses it already, or your agent can move it to Postgres.
- A domain name and a Git host Your own domain, and a place for your code, such as GitHub or GitLab.
Not on Postgres, and want to stay that way? That isn't directly supported yet, but it's perfectly doable. Just instruct your agent to replace the database portions of the Stack to reflect what you already do with your own database.
Set up in a few hours
As fast as 2 hours until your site is live. Building this by hand used to take weeks or months.
The setup goes at your pace
- Fast by default: during the setup, your agent stops only when it needs you, about 20 minutes of your time
- Say "guide me" and it slows down and explains each part before building it
- Say "go fast" to speed up again
- Ask "why?" at any step: every rule in the Stack has its reason written next to it, and your agent reads you that reason
An example question
- You: Why do my backups go to a different storage provider?
- Your agent: So that someone who breaks into your servers can't delete them. They are also locked against deletion for 14 days.
Have your coding agent map out your move
Your agent reads your app and the Stack, then shows you what would move where, what would change in your app, and what it would cost. Nothing is installed or changed.
Paste this into Claude or Codex, in your app's directory
Read the Monkeymaxxing Stack at https://github.com/monkeymaxxing/monkeymaxxing-stack and read this app. Don't install or change anything. Show me how this app would run on the Stack: a before-and-after picture of my setup, what would change in the app, which of the Stack's guides apply, and what the hosting would cost. Ask me what I pay for hosting now, then compare the two.
Starting a new project? Ask it to map out a new setup.
Read the full guide on GitHub Ready? Start the setup
This is an alpha version. The system’s design is solid. The setup may still have a few rough spots, but your coding agent can get you past them.
Free and open source
The Stack and its tools
The Stack
Everything your coding agent needs to build and run a complete production system on your own servers.
The Stack →MMX
Sets up and manages all your Linux servers: software, settings, users, services, and more, all the same way every time.
MMX →MMWatchdog
Monitors your servers, services, and websites, and alerts you when there's a problem.
MMWatchdog →MMHotline
Wakes the person on call for urgent alerts, but first gives a triage agent a few minutes to find the cause and fix it.
MMHotline →doubleshell
Sends SSH through WireGuard instead of an open SSH port. WireGuard ignores anyone without a key, so bots find nothing to attack.
doubleshell →Why the monkey
Reject the galaxy brain. Return to monkey.
- The galaxy brain builds complicated systems: many services, many tools, and parts nobody on the team fully understands.
- The monkey brain builds simple systems the team understands and can fix fast.
Anyone with a coding agent can now build a giant galaxy-brain mess in an afternoon. Complexity doesn't prove engineering skill anymore — keeping a system simple, fast, and easy to understand does.
