Spinning up a cloud virtual machine shouldn't feel like assembling IKEA furniture without instructions. Yet there you are: picking a provider, wrestling with networking configs, fumbling through SSH key rituals, installing dependencies, then—if you remember—tearing the whole thing down before the bill arrives.
For developers running AI agents or tinkering with machine learning workloads, that friction compounds. Barnaby Malet built machine0 to collapse the ceremony into a single line of code.
The YC Summer 2026 company went live on June 15, 2026, with a pitch stripped to its essentials: persistent cloud VMs from the command line, per-minute billing, and a static public IP for every machine. No NAT gymnastics, no tunneling headaches. Two commands—machine0 new my-vm and machine0 ssh my-vm—deliver a running instance with an authenticated HTTPS endpoint at <vm>.mac0.io.
That's the promise, anyway. Whether it's enough to pry developers away from entrenched alternatives is the open question.
What You Actually Get
machine0 is a CLI tool that provisions dedicated KVM/QEMU virtual machines running NixOS or Ubuntu. Every VM arrives preconfigured with Docker, Node, Python, and agent CLIs like Claude Code and Codex—the kind of setup that usually means an afternoon lost to Stack Overflow threads.
Sizes start at 1-vCPU with 1GB of RAM for $0.013 per hour, roughly $9 monthly if you leave it humming. At the other end sits a 60-vCPU, 240GB beast at $3.71 per hour. GPU instances begin at $0.836 per hour and scale up to configurations with eight H200 chips, billing out at $30.27 per hour—serious hardware for serious workloads.
The billing model is per-minute with no long-term commitment, and prices hold steady across all four regions: New York, London, Amsterdam, Singapore. Machines keep the meter running even when stopped, since resources stay reserved. But there's a suspend feature that snapshots your state and halts billing entirely. After that, you're only covering image storage at $0.078 per GB monthly.
The AI Agent Angle
What separates machine0 from the crowded VPS field is its integration with the Model Context Protocol. A remote MCP server lets AI agents provision, manage, and snapshot machines programmatically—no human intervention required.
The documentation walks through examples: coding agents like Claude Code running four to eight hours at a stretch, or autonomous agents like OpenClaw and Hermes that might grind away for days. Short coding sessions with isolated credentials, or long-running agents needing persistent environments with stable IPs. Those are the two use cases Malet zeroed in on.
He sketched out the company's origin story in a June LinkedIn post, mentioning that he initially built an MCP gateway before pivoting to the CLI-first approach. Perhaps the market pulled him there, or perhaps the developer experience just made more sense that way.
A June 15 review from The Agent Post positioned machine0 in an interesting middle ground—somewhere between traditional VPS providers like Hetzner or DigitalOcean, where you bring your own provisioning scripts, and ephemeral sandboxes like Modal or E2B. The review suggested machine0 is betting that managed CLI layers, suspend-and-resume semantics, and a stated 99.99% uptime SLA justify a pricing premium. Time will tell if developers agree.
Open Source Foundations, Rapid Iteration
The base images are open source, living under the fdmtl GitHub organization. NixOS configurations use flakes for declarative, reproducible setups; Ubuntu images lean on Ansible for provisioning. Both repositories showed activity as recently as June 2026, suggesting active development rather than abandonware.
Feature velocity appears brisk. On June 30, version 1.0.93 introduced environment variable sets that attach to VMs at creation. May 12 brought a web-based SSH terminal for managed-key VMs—a small convenience that matters when you're hopping between machines. April 28 saw three new premium Intel sizes roll out: 4xl, 5xl, and 6xl, aimed at heavier workloads.
There's also file sync with --watch mode, native support for versioning snapshots, and the ability to clone images across different instance sizes. All CLI commands output JSON if you need to script around them. These are the kinds of details that matter once you're past the "does it work?" stage and into "can I actually use this in production?"
The Founder's Second Act
Barnaby Malet is no stranger to Y Combinator. He was co-founder, CTO, and Chief Product Officer at Upflow, a YC Winter 2020 company that raised a $15 million Series A in 2021. Upflow's team roster showed around 50 people in its YC directory listing at that time—though those figures likely reflect an earlier snapshot rather than current headcount.
machine0, by contrast, is a solo operation. As of early July 2026, the team size sits at one. That's either admirably lean or a sign that scaling hasn't yet become necessary.
Malet studied computer science at Imperial College London. His personal site positions him as building tools "for machines that work" under the Fundamental Software brand—a suitably abstract tagline for infrastructure aimed at autonomous systems.
Reception and Remaining Questions
The June 15 Show HN thread drew interest around GPU availability and the prospect of sharing preconfigured "golden images" across teams. One commenter asked whether static IPs persist if you delete and recreate a VM. Malet clarified they're retained while the VM exists but released on deletion—a detail that matters if your workflow depends on stable endpoints.
The product is live at machine0.io, with installation scripts for macOS and Linux. Minimum top-up is $5, and unused credits are refundable—a small gesture that signals confidence in retention.
Whether this managed-CLI approach proves compelling enough to justify a premium over bare-bones VPS providers hinges on how much friction it actually removes. And that, in turn, depends on whether AI agents evolve into a genuinely large use case or remain a niche fascination.
For now, machine0 represents a bet that the future involves more machines provisioning machines—and that the humans overseeing them would rather type one command than ten. It's a reasonable wager. Whether the market rewards it is another matter entirely.
