private pilot // cloud dev environments for agent fleets
A home for your agent fleet.
If you run more than a couple of coding agents, you know the mess — tmux sprawl, shared checkouts, one agent stepping on another's work, everything logged out after a rebuild. code.all.ai gives every agent its own machine: created in one click, walled off from everything else, and exactly as you left it when you come back. Open it as VS Code in a browser tab, or SSH in from your desktop.
01 / THE MESSFive agents, one laptop, shared everything. It works right up until it doesn't.
02 / THE FIXGive each agent its own machine. A real one, that remembers.
03 / THE POINTYour fleet gets a building. You get your laptop back.
how it feels · 1 of 3
Name it. Click. It's alive.
A fresh, isolated container comes up with VS Code and your agent already installed. Around fifteen seconds, and there's no YAML anywhere in the story.
how it feels · 2 of 3
Step inside. Put it to work.
It's real VS Code in a browser tab — terminal included, agent ready. Hand over the task and close the tab. The agent keeps going without you.
how it feels · 3 of 3
Leave. Come back. Nothing's lost.
Stop a session at midnight, start it after coffee — files, config, and the agent's login are exactly where you left them. These are homes, not sandboxes.
In the pilot, today
What you get on day one.
This is the setup we run our own fleet on, every day. The pilot opens it up to a small cohort — real machines, real isolation, rough edges included.
One click, one machine
Each session is its own isolated container on our hardware. Spin up as many as your quota allows; destroy them without ceremony.
Browser or desktop
Full VS Code in a tab for anywhere-access, or attach your local editor over SSH when you want your own keybindings.
Claude Code, ready to go
Pre-installed in every session. Prefer Codex, Cline, OpenClaw, Hermes, or something newer? It's a normal Linux box — install anything, orchestrate it however you please.
It remembers everything
Files, dotfiles, and agent logins survive stop, start, and rebuild. Set a session up once; it stays set up.
Yours means yours
Sessions can't see each other — not over the network, not through the gateway. Even we can stop your machine but never open it.
Quotas that protect you
Per-user limits keep a runaway agent from eating the box — yours or anyone else's. Boring, and exactly what you want.
Where it's headed
The fleet is just the floor.
We're building the layer an AI-native company runs on. Pilot members get each of these as it lands — and a real say in what lands first.
Everything above
Sessions, both doors, persistence, isolation, quotas. Running today.
Switch models without losing your mind
Claude for the hard refactor, an open-weight model for the grunt work, whatever ships next month for everything else — with your workspace and context staying put. You shouldn't have to marry a model vendor to get work done.
Fleet view
Every session, every agent, what changed while you slept — one screen that answers "what did the fleet do last night?"
Session blueprints
Your stack, your tools, your dotfiles — captured once, so agent #12 starts life exactly like agent #1.
Sessions from code
An API to spawn, direct, and retire sessions programmatically — for when the thing managing your agents is another agent.
Teams
Shared workspaces, an approval gate for who gets in, and an operator view that manages machines without ever seeing inside them.
Pricing
Our costs, doubled.
No pricing-page theater. We take what the hardware costs us and double it — that's your bill. When our costs drop, your price follows them down, not up. Founding-cohort members keep their price for a year.
You, plus an agent or two. Enough to stop running your fleet out of tmux.
An actual fleet: parallel agents on parallel machines, working while you sleep.
A whole box nobody else touches. For fleets that outgrow sharing — or policies that forbid it.
Your tokens stay yours. Bring your own Claude, OpenAI, or any other accounts — we never resell inference or sit in your model bill. Our price is the metal, doubled, and nothing else.
Questions
Asked early, answered straight.
What does the pilot cost?
Light is $14 a month — three sessions, one running at a time. Heavy is $89 — twelve sessions, six at once. The math behind both is stated plainly above: our hardware cost, doubled, and when our costs fall your price follows. Your model subscriptions and API keys stay yours. Founding-cohort members keep their price for a year — and we still ask for the honest feedback.
Can anyone else get into my sessions?
No. Every request — browser or SSH — has to prove who you are before it touches a session, and a session answers only to its owner. We can stop or destroy a machine if something's on fire, but we can't open one. That line is deliberate, and it's load-bearing.
Do I have to change how I work?
Not really. From your side it's a normal dev box: VS Code in a tab, or SSH from the desktop editor you already use. Your agent doesn't know it moved.
Which models and agents do you support?
One-click setup with any model or AI agent — in any combination. Claude Code comes pre-installed because it's what we use; adding Codex, Cline, OpenClaw, Hermes, or whatever ships next week is a one-liner, and you can orchestrate them however you please. We're deliberately neutral about whose model wins.
Why not Codex cloud, Claude Code on the web, or Cursor's background agents?
Use them — they're good at what they are: tasks. You hand one a prompt, it runs in a sandbox that belongs to the vendor, and the sandbox evaporates. Ours are machines: they persist, you can SSH into them, install anything, and — the part the labs will never offer — they're not tied to anyone's model. When you switch vendors, your machines and everything on them come with you. Vendor sandboxes are where tasks run; this is where your fleet lives.
Do I have to use the dashboard?
No. Everything on this page is a thin layer over the API — cde create kepler, cde ls, cde stop kepler from your terminal do exactly what the buttons do. If your orchestrator wants to spawn sessions programmatically, that's the same API too.
The pilot
Bring your fleet.
We're letting in a small cohort of people who run coding agents in earnest — early access, direct influence on the roadmap, and machines that treat your agents like they live there. Tell us what you're building and what you run.