Persistent agent workspaces

Run AI Agents
in the Cloud

Isolated compute, durable storage, and extensible tunnels.

Bring your own agents

Run Claude Code, Codex, OpenCode, Cursor, or your own, on the credentials you already pay for. No token resale.

  • Antigravity
  • Kiro
  • Goose
  • Cline
  • GitHub Copilot
  • Claude Code
  • Codex
  • Cursor
  • opencode
  • pi

Repo to running agent in three steps

No servers to babysit, no dotfiles to re-create. Paperboat turns a git URL into a persistent agent workspace.

  1. 01 · Create

    Start from any repo

    Point Paperboat at a git repository. It becomes a project: cloned, configured, and ready for your agents.

  2. 02 · Provision

    Get a dedicated machine

    Every project runs on its own isolated VM with its own persistent volume. Pick the machine size and disk that fit the job.

  3. 03 · Connect

    Reach it from anywhere

    Attach from the desktop app, your phone, or the terminal. Agents keep working in the cloud while your laptop sleeps.

One session, on every screen

Every client attaches to the same live session through the same tunnel. Start at your desk, steer from your phone, finish in the terminal.

paperboat
my-applive

Idle machines resume when a client connects. In the terminal, paste a screenshot and the agent gets a VM path.

Built for how agents get used

Long runs, odd hours, many projects at once. Work that doesn't fit a laptop with the lid closed.

Let it run overnight

Hand an agent a long task and close the laptop. It keeps working in the cloud; the machine stops itself when the work is done.

  1. 18:40
    You sign off

    the agent keeps working in the cloud

  2. 02:10
    Task done

    nothing left to run, the machine falls idle

  3. + idle
    Machine sleeps itselfidle timeout5 min

    no activity from you or the agent, so it stops and metering ends

  4. 07:15
    You're back

    reconnect and it resumes where it left off

Don't want to wait for the timeout? End the session yourself.End session
Steer from your phone

Check on a run from the couch, the train, another country. Review, redirect, and approve from the full session on your phone.

acme-store live
Message acme-store…
Run a fleet in parallel

Every project gets its own machine, so agents never trip over each other. Run three refactors at once; sleeping projects cost nothing.

New session
Sessions · 9 live
api-server1h 32m
my-app47m
ml-experiments2h 08m
web-dashboard18m
data-pipeline3h 04m
auth-service52m
mobile-app1h 11m
infra-scripts26m
analytics4h 20m
docs-siteasleep
Share work in one link

When an agent stands up a dev server, it hands back a preview URL you can open on any device or send to anyone.

https://my-app.preview.paperboat.dev
1

Every preview has a review layer. Anyone with the link can comment on the running app, and the agent gets the notes back.

One tunnel between you and your agents

Every client reaches your projects through one agentunnel tunnel. Nothing on your VM is exposed as a public port.

your devicesagentunnelyour projectsacme-storebilling-apidocs-site

Pay for runtime, not uptime

Machines stop themselves when you and the agent go quiet, and resume when you connect. Credits meter only while a machine is running.

  • Auto-stop on idle, with a timeout you set per project
  • Instant resume on connect, from any client
  • Run several projects in parallel; sleeping ones cost nothing
my-app · one day00:00·zero spend · machine stopped
working · meters idle · meters until auto-stop asleep · free

Machines are disposable. Your setup isn't.

Before a machine is torn down, Paperboat diffs your tracked config and pushes it to a private GitHub repo it manages for you. Every new machine boots with your dotfiles, agent settings, and tools already in place.

  • Fully automated, no agent involvement, no manual step
  • Your config, in your GitHub account, as a normal repo
  • Project code lives on its own persistent volume, untouched
  1. machine stopping

    config diffed automatically

  2. pushed to your config repo

    dotfiles · agent settings · private

  3. new machine starting

    latest config cloned on boot

  4. everything where you left it

    no setup, no agent involved

3af912c sync: dotfiles, agent settingsyou/config · private

Everything a workspace needs, per project

Machine shape, disk, idle behavior, tooling. Each project is tuned independently from one pool of plan resources.

Machines in the right size

Pick a shape per project: scale up for heavy builds, down for docs. Bigger shapes meter more, and only while running.

One storage pool, your split

Your plan gives you a pool of disk; you decide how much each project gets. Delete a project and its share returns instantly.

Idle on your terms

Set how long a machine waits before stopping itself. Per project, not one global setting.

Tooling preinstalled

Pick agents and tools from the preset catalog, or drop in your own setup script. New machines come up ready to work.

A volume that remembers

Project code lives on its own persistent volume. Stop for a week, resume, and the working tree is exactly as you left it.

Every client, same session

Every client attaches to the same workspace through the same tunnel. Start on one, finish on another.

no public ports·your agent credentials, never resold·config repo in your GitHub account·one isolated VM per project

Frequently asked questions

Your agents keep working while yousleep

A git URL is all it takes. Isolated machine, persistent volume, one tunnel to everywhere. Your agents, your credentials, no lock-in.