Every computer
your agent needs.

Sandboxes, swarms, desktops and phones for your agents.

Book a call

Developers building on Maritime work at these companies

Govern d'Andorra

Sandboxes, desktops and phones.

Persistent Sandboxes
A workspace that keeps its files, sleeps when idle and wakes in about a second.
From $0.05 / hour awakeRead the docs
Ephemeral Sandboxes
Run untrusted or generated code once. Deleted with its disk when the task ends.
From $0.018 / hour awakeRead the docs
Swarms
Thousands of sandboxes on one command, for evals, RL rollouts and batch jobs.
From $0.018 / hour per sandboxRead the docs
Linux Desktops
A Linux desktop with a screen your agent sees, clicks and types on. A person can take over.
$0.12 / hour awakeRead the docs
Windows Desktops
For the desktop apps that only run on Windows. Windows Server 2022, licence included.
$0.18 / hour awakeRead the docs
Android Phones
For the apps that only exist on a phone. Install any app as an APK.
$0.15 / hour awakeRead the docs

Built for every kind of agent.

train.py

model.py

data.py

import torch

from tqdm import tqdm

from model import Net

from data import loader

net = Net().to("cuda")

opt = torch.optim.Adam(net.parameters(), lr=3e-4)

for epoch in range(5):

bar = tqdm(loader, desc=f"Epoch {epoch + 1}/5")

for x, y in bar:

loss = net.step(x, y)

opt.step()

bar.set_postfix(loss=f"{loss:.2f}")

torch.save(net.state_dict(), "workspace/model.pt")

TERMINALPROBLEMSOUTPUT

$ python train.py

Coding agents. Run generated code in a throwaway sandbox, and keep the repo in a workspace that is still there next week.

Ephemeral Sandboxes Persistent Sandboxes

notes.md

report.pdf

root@atlas: ~+

$ node server.js

listening on :8080

notes.md

report.pdf

root@atlas: ~+

$ node server.js

listening on :8080

notes.md

report.pdf

root@atlas: ~+

$ ./backup.sh

synced 1.2 GB to s3

notes.md

report.pdf

root@atlas: ~+

$ pdflatex paper.tex

Output written on paper.pdf.

notes.md

report.pdf

root@atlas: ~+

$ python ingest.py

48,000 rows written

notes.md

report.pdf

root@atlas: ~+

$ node server.js

listening on :8080

notes.md

report.pdf

root@atlas: ~+

$ ./backup.sh

synced 1.2 GB to s3

notes.md

report.pdf

root@atlas: ~+

$ pdflatex paper.tex

Output written on paper.pdf.

notes.md

report.pdf

root@atlas: ~+

$ python ingest.py

48,000 rows written

notes.md

report.pdf

root@atlas: ~+

$ node server.js

listening on :8080

Evals and RL. Fan one command out to hundreds of sandboxes, let some fail safely, and collect every exit code and output.

Swarms

My Computer

Books 2012

Recycle Bin

Books Desktop 2012 - Enter Bills_□×

FileEditVendorsBankingReportsHelp

VendorAcme Medical Supply

Bill date10/09/2026

Amount due1,284.00

MemoInvoice 4471, Oct

ClearSave & Close

Books Desktop 2012_□×

iBill saved. 12 of 12 entered.

StartBooks Desktop 20129:41 AM

Computer-use agents. A desktop with a screen for apps with no API: a browser behind a login, or the Windows app your customer runs.

Linux Desktops Windows Desktops

9:41

Loadboard

3 loads near Dallas

#8812$1,450

Dallas, TX → Tulsa, OK

257 mi · dry van

#8807$2,180

Austin, TX → Memphis, TN

650 mi · dry van

#8794$1,320

Houston, TX → Mobile, AL

472 mi · dry van

‹ Loads

Load #8812

Accepted · pickup 2:00 PM

PickupDallas, TX

DropTulsa, OK

Rate$1,450

Equipment53' dry van

Weight38,000 lb

Accept load

›open Loadboard app

›tap load 8812, Dallas to Tulsa

›accept at $1,450

✓booked, dispatcher notified

Mobile agents. A phone in the cloud for apps that only exist on mobile, with a person able to take over by link.

Android Phones

atlas · /data

atlas

reports

september.pdftoday

workspace

clients.xlsx08:03

journal.md412 entries

memory

2026-09.md

Ary

online

can you move our 9am?18:03

ok, moving to 10am18:04

done18:05

Type a messageok, moving to 10amdone

mail.google.comcalendar.google.com

signed in as atlas

9:00

10:00

11:00

Sync with Ary

Assistants. Host the agent itself from a template, one per customer: it sleeps when idle and wakes on the next message.

Persistent Sandboxes

paper.tex

paper.pdf

root@atlas: ~/paper+

$ apt install texlive-latex-base

Get:1 texlive-binaries8.1 MB

Get:2 texlive-base21.4 MB

Get:3 texlive-latex-base1.2 MB

Fetching30.7 MB

Setting up texlive-latex-base ...

$ pdflatex paper.tex

Output written on paper.pdf (4 pages).

$ libreoffice paper.pdf

paper.pdf · LibreOffice Draw

Snapshot-Restore Scheduling for Persistent Agent Micro-VMs

Ary Shrivastava, Maria Gorskikh, and Brandon Li

September 3, 2026

Abstract

Long-lived software agents spend most of their lifetime idle, yet the conventional deployment keeps a full virtual machine resident for each of them. We study a scheduler that snapshots an idle guest to disk and restores it on the next inbound message, so that a parked agent costs only its storage. We characterise the restore path, show that its latency is dominated by page-cache warm-up rather than by the hypervisor, and give a bound on the expected monthly cost of an agent as a function of its wake rate.

1 Introduction

A persistent agent is a process with a disk, a set of credentials, and a schedule of its own. Unlike a request handler, it has state that outlives any single interaction, which rules out the stateless scaling model of serverless platforms. We model an agent as an alternating sequence of awake and asleep intervals and write a(t) ∈ {0, 1} for its state at time t. The cost over a horizon T is then

C(T) = c_sST + c_v ∫₀^T a(t) dt(1)

where S is the size of the snapshot on disk and cₛ, cᵥ are the storage and compute prices. For the agents we run the second term is small: a typical agent is awake for eleven minutes a day, and the bound in Section 4 follows from this observation alone.

2 The restore path

We restore a guest in three steps. The memory snapshot is mapped rather than read, so the first instructions execute before the file is resident; the block device is attached as a thin clone of the parked volume; and the network tap is re-created with the address the guest had when it slept.

Research and back-office agents. Install any tool, work through files and hand back a finished report, then let the machine sleep.

Persistent Sandboxes Linux Desktops