Compute infrastructure to run coding agents beyond the laptop.

Give every developer a persistent, isolated computer, provision ephemeral sandboxes on demand, and control every environment through one API.

4 vCPU 8 GB hardware-isolated from $0.033/hr

or install the CLI
1 / 6Create a sandbox
Search sandboxes…
M

Provisioning a machine

Its own microVM, booting. Usually a few seconds.

remote · checkout-api
console.remotehost.ai

Sandboxes

SandboxAgentAge
running1
codex-migrate
codex42m
stopped1
docs-preview
codex260m

Path A

For you or your team.

As many sandboxes as each engineer has work in flight, on one org and one bill.

Search sandboxes…
M
claude
checkout.ts
cart.ts
schema.sql
Matthew@remotehost:/code$ claude
Claude Codev2.1.233

Welcome back Matthew!

Sonnet 5 with high effort · Claude API ·Matthew's Org/code
Tips for getting startedRun /init to create a CLAUDE.md file with instructions for Claude
What's newCost estimates (`/cost`, status line, `--max-budget-usd`) now include the 1.1x US-only-inference premium for data-residency orgsAdded the one-time fullscreen renderer offer on Bedrock, Vertex, Foundry and other previously excluded setupsAdded `/claude-api upgrade` to migrate Python projects from `anthropic` 0.x to 1.x/release-notes for more
/model to try Opus 4.5remotehost sandbox claude-pr-481 · /code
>Try "fix typecheck errors"
?for shortcuts
pnpm dev
Matthew@remotehost:/code$ pnpm dev ▲ Next.js 15.5.23 - Local: http://localhost:3000 - Network: http://10.0.0.4:3000 ✓ Ready in 1.4s ○ Compiling /checkout … ✓ Compiled /checkout in 812ms
pnpm test
Matthew@remotehost:/code$ pnpm test --watch ✓ src/checkout/total.test.ts (12) ✓ src/checkout/tax.test.ts (7) Test Files 2 passed (2) Tests 19 passed (19) Waiting for file changes…
remote
$remote new claude --name claude-pr-481
claude-pr-481 running

Path B

For your product.

A hardware-isolated microVM per user, per session, or per task, from your own code.

api.remotehost.ai
POST /v1/orgs/$ORG/sandboxes
{ "endUserId": "u_8321", "mode": "ephemeral" }

201 Created
{
  "sandbox": {
    "id": "sb_9f2k",
    "end_user_id": "u_8321",
    "status": "provisioning"
  }
}

Persistent

A sandbox lasts until you delete it. Sleep one and wake it days later, mid-thought.

Most affordable

4 vCPU and 8 GB from $0.033/hr, below every comparable price sheet we checked. Asleep, it's free.

Secure

Every sandbox is its own hardware-isolated microVM. No shared kernel, no upgrade tier.

Compute infrastructure

Underneath everything is a hardware-isolated microVM: root, Docker, real ports, a persistent disk, and a snapshot when it sleeps. The API is the front door, whether the machines are for your engineers or for the people using your product.

Ready when you attach

A microVM typically boots in under a second. The harness is up, your repo is cloned, and a tmux session is waiting: work in progress, never an empty box.

Root

Full systemd, Docker, background processes, real ports, its own filesystem. Anything that needs a machine rather than a request handler runs here unchanged.

Stocked

Docker, the GitHub and Linear CLIs, language servers, and the runtimes a repo expects ship in the image. The agent's first move is the work, not twenty minutes of apt.

Nothing is thrown away

Ephemeral sandboxes take the work with them when they go. These sleep and wake from their own snapshot: same filesystem, same branch, same half-finished dev server.

Any client, one fleet

The CLI, the TUI, the desktop app, and the web console all attach to the same live sandbox and read the same fleet. The API drives any of it from your own code.

Many people, one session

A teammate attaches to the sandbox you are already in and lands in the same session: your panes, your scrollback, your running processes. No screen share, no second copy.

Remote, without giving up localhost.

Start pnpm dev and open localhost from the sandbox, exactly as you would on your laptop. RemoteHost detects the port, opens the app in your browser, and keeps hot reload intact. No tunnel, port configuration, or new workflow.

claude-pr-481
$ pnpm dev

  ▲ Next.js 15.5.23
  - Local:   http://localhost:3000

 ✓ Ready in 1.4s
$ open localhost:3000
$ 
:3000https://3000--3f2b9c1e-7a4d-4e6b-9c2a-5d8f1b3e7a90.previews.remotehost.ai

Same commands

Start Next, Vite, Storybook, Rails, or another dev server the usual way. RemoteHost detects every bound port automatically.

Same feedback loop

Hot reload and its WebSocket travel through the preview connection. Save a change and the browser updates as it would on your laptop.

Easier to share

Every preview has an authenticated HTTPS URL for your team. Public review links can expire and be revoked without disrupting the sandbox.

Where it runs.

A Firecracker microVM is the same machine wherever it boots, so the region is a placement decision rather than a different product. Tell us where your data has to live and we will put a fleet there. On Enterprise, swap in your own compute and run the same sandbox on Firecracker hosts you already own. Talk to us about either.

Agent sandboxUS East (N. Virginia)us-east4 · Firecracker microVM
vCPU4RAM8 GBDisk80 GB

Command line.

The same lifecycle from a shell. Create, list, attach, and sleep from wherever your agents already run, with nothing in the loop but remote.

$curl -fsSL https://remotehost.ai/install | shcurl -fsSL https://remotehost.ai/install | sh

One command creates the
machine, another gets you in.

remote new is the base primitive. Add an agent name and Claude Code or Codex boots inside the sandbox as it provisions, so it's ready the moment you remote attach.

remote · checkout

Simple enough to memorize.

Verbs, not config. remote ssh is plain OpenSSH and every command exits non-zero on failure, so the lifecycle drops straight into a Makefile or a CI job.

CLI reference
CommandDoes
remote new [agent]Create and enter a sandbox
remote claudeBoot Claude Code inside one
remote lsList every sandbox you can reach
remote attach <host>Attach by name or id
remote ssh <host> -c '…'A shell or command over OpenSSH
remote sleep <host>Snapshot it and stop billing
remote wake <host>Bring it back from that snapshot
remote metrics <host>CPU, memory, and disk pressure

Pricing

Starter, Pro, and Max scale one engineer from a first sandbox to agents that never stop. Business gives the whole team a shared org on dedicated capacity, priced by the seat. Overage is off until you turn it on, and you can cap machine-hours on the org or a single project below that.

Starter
$20/mo

For an engineer proving the workflow before rolling it out to the team.

  • 606 included machine-hours a month
  • 35 concurrent sandboxes included
  • Additional usage billed by machine time
  • Scale to 1,200 concurrent sandboxes
  • Up to 24-hour sandbox sessions
  • Machine-hour spend cap, overage opt-in
  • Community support
Pro
$100/mo

For one engineer keeping several agents busy around the clock.

  • Everything in Starter
  • 3,030 included machine-hours a month (5x Starter)
  • 100 concurrent sandboxes included
  • Scale to 2,000 concurrent sandboxes
  • Repository templates
  • Additional usage billed by machine time
Max
$200/mo

For the engineer whose agents never stop.

  • Everything in Pro
  • 6,061 included machine-hours a month (10x Starter)
  • 250 concurrent sandboxes included
  • Up to 72-hour sandbox sessions
  • Custom base images and repository templates
  • Priority support
  • Additional usage billed by machine time

Every plan runs on the same control plane. When you outgrow ours, Business adds a shared org, and Enterprise can run sandboxes on hardware you own. Your projects, agents, repos, and history stay exactly where they are either way.

What an hour costs.

One sandbox with 4 vCPU and 8 GB, running for an hour, at each provider's published rate. Read from their price sheets on 2026-09-11.

The full comparison: a month at your scale, and what each sandbox can do

ProviderPer hourPublished rate
RemoteHost, Central EU$0.033/hrAgent Standard profile in Germany, France, 80 GB disk included
RemoteHost, East US$0.033/hrAgent Standard profile in North Virginia, 80 GB disk included
RemoteHost, Canada Central$0.033/hrAgent Standard profile in Quebec, 80 GB disk included
Box (EU only)$0.036/hr$20 buys 2,000,000 seconds of 4 vCPU / 8 GB, in Germany, Finland, or France
Northflank$0.133/hr$0.01667 per vCPU-hour + $0.00833 per GB-hour
E2B$0.331/hr$0.000014 per vCPU-second + $0.0000045 per GiB-second
Blaxel$0.331/hr$0.0000115 per GB of RAM per second, CPU included
Daytona$0.331/hr$0.0504 per vCPU-hour + $0.0162 per GiB-hour
Cloudflare Sandboxes$0.360/hr$0.00002 per vCPU-second while busy + $0.0000025 per GiB-second
Modal Sandboxes$0.476/hr$0.00003942 per core-second (a core is 2 vCPU) + $0.00000667 per GiB-second
Fly Sprites$0.630/hr$0.07 per CPU-hour + $0.04375 per GB-hour
Runloop$0.634/hr$0.108 per CPU-hour + $0.0252 per GB-hour
Vercel Sandbox$0.682/hr$0.128 per active-CPU-hour + $0.0212 per GB-hour
  • Compute only: CPU plus memory, by the clock, for one hour. Disk is left out because some providers include it and some meter it; ours is included.
  • Cloudflare and Vercel bill CPU only while it is busy, so their rows are the price at full load. Work that mostly waits on I/O costs less there than shown; memory is billed the whole time either way.
  • A RemoteHost sandbox that is asleep is billed nothing: no runtime, no disk, no snapshot fee.