boat

Persistent Ubuntu VMs for AI agents, billed per second, running in EU data centres.

Paid#sandboxes#vms#ai-agents#cli

What is boat?

boat rents persistent Linux machines it calls sandboxes. Each one is a full Ubuntu KVM virtual machine with root, SSH and SCP, Docker, Chrome and a desktop you can watch in the browser. Billing runs per second while it runs. Stopping one snapshots the disk, and you can resume it or fork copies from that snapshot. You drive it with the boat CLI, a REST API or the Python and TypeScript SDKs, and boat prompt runs Claude Code, Codex, OpenCode and other agent harnesses inside a sandbox.

boat is made by Dedale AI, Corp., a Delaware company in Y Combinator's Fall 2026 batch. The service is proprietary; only the API-generated SDKs are published, under MIT. Plans start at $20 a month, and every plan dollar comes back as VM time. Overage is paid from a prepaid wallet.

The product was called Box until September 15, 2026. As of November 2026 it runs only in Germany, Finland and France, has no GPU sizes, and lists SOC 2 as in progress. Its list price per hour is well below Fly.io Sprites, E2B and Daytona, and unlike Sprites it bills the size you reserve rather than the CPU you use.

Our review

3.5

A good, cheap place to give coding agents a real computer, if you work in Europe and can live with a very young vendor. $0.036 an hour for 4 vCPU and 8 GB, billed per second with no egress charge, buys a full VM with Docker, systemd and a desktop, and forking a prepared snapshot makes parallel agent runs easy. The trade-offs are just as concrete. There are only EU regions and no GPU, memory isn't kept when a sandbox stops, the IP changes on every resume, and SOC 2 is still in progress. The company is a three-person YC team that renamed the product two months ago. Independent numbers are mixed: StarSling's October run puts boat's CPU near the top, a real-world dependency install mid-pack, and memory bandwidth last of 15. There's almost no independent discussion yet, so who gets paged when it breaks is an open question.

Pros

+ Flat, low hourly rates you know before you run ($0.018 to $0.200 per hour by size), billed per second, with no egress charge
+ Full KVM machines with root, Docker, systemd, Chrome and a streamable desktop; fork a prepared snapshot to fan out many agents
+ Unusually candid docs that state the limits, retention rules, data location and when cheaper rivals win, with a complete llms.txt index

Cons

− Germany, Finland and France only, no GPU, and snapshot copies can be stored outside the EU; SOC 2 is still in progress
− There's no pause: memory and processes are lost on stop after a 10-minute grace window, the IP changes on resume, and stopped disks are only guaranteed for 14 days
− A very young vendor (3-person team, renamed from Box in September 2026) with no public issue tracker and almost no independent user reports yet

Reviewed by SeekDeploy Editorial · updated 2026-11-15. Reviews are independent — we don’t accept payment for coverage or ratings.

Should you use boat?

A good fit if you're…

Developers who want a coding agent to work for hours in a full Linux machine with Docker and systemd, at a fixed hourly price
Teams fanning out many parallel agent runs, CI jobs or eval episodes from one prepared snapshot
Agent workflows that need a browser or desktop you can watch, record or embed
Users in or near Europe: every machine is in Germany, Finland or France

Probably not for you if you're…

Anything that needs a GPU: no size has one
Teams that need US or Asia regions, or strict EU-only data residency: snapshot copies can be stored outside the EU, and SOC 2 is still in progress
Hosting a production web app: the IP changes on resume, public URLs carry a token by default, and the terms disclaim any uptime warranty
Occasional use under about $20 a month: boat's own pricing page says providers with free tiers or credits cost less there, and even the trial needs a card

Pick something else if…

you need agent sandboxes or VMs outside Europe (Fly Machines run in 18 regions), or want Sprites' checkpoint and restore next to a full app platform → Fly.io
you need GPU, Windows or macOS sandboxes, or US regions, and can pay for it: 4 vCPU / 8 GiB is $0.3312 an hour there against $0.036 here → Daytona
your agents run many short-lived sandboxes rather than long VM sessions, or you want an open-source runtime you can self-host; E2B bills only while a sandbox runs, and paused sandboxes keep memory without a compute charge → E2B

Gotchas to know before you commit

There's no pause. Stopping saves the disk (boat also snapshots it every minute while running) but not memory or processes. A resume within 10 minutes usually lands on the same machine with processes alive; after that they're gone and enabled systemd services start again. If a host fails, you lose writes since the last good snapshot.
Auto-stop is on by default, 1 hour after each create or resume (2 hours at most during the trial). A running agent prompt pushes the deadline. Use boat extend or --no-auto-stop for long jobs, and remember a sandbox without auto-stop bills until you stop it.
The IP address changes at every resume, and some sandboxes have no IPv4 address of their own (raw ports are IPv6 only). Refresh SSH config and host keys after each resume.
Stopped disks are kept for at least 14 days. After that, automated cleanup can delete them; as of 2026-11-15 it only reports candidates in production. Keep state with a saved snapshot: 10 come with each $20 of plan, extras cost $0.056 a day, and you can't save any during the trial.
Start limits are hard. Create, fork and resume each count as a start, extra starts cost $0.03 to $0.07, and boat returns 429 above 3x the per-minute or 6x the hourly or daily limit. It never queues a start.

boat pricing

Checked Nov 15, 2026 against the official pricing page

A monthly plan (from $20) that comes back in full as per-second VM time, plus a prepaid wallet for anything past it. It gets expensive when sandboxes are left running around the clock, on the $0.200/h xlarge size, or with lots of saved snapshots. Included time doesn't carry over, so below about $20 a month of use, providers with free tiers cost less.

Free trial

$0 for 7 days (card required)

First plan only; organisations get no trial

25 compute hours in total
small and default sizes

Limit: 2 sandboxes at once; 5 starts per minute, 25 per hour, 75 per day
Limit: Auto-stop required, 2 hours at most
Limit: No saved snapshots; Stripe charges the plan when the 7 days end

$20 plan

$20 / month (per seat in an organisation)

Individuals and small teams

$20 of VM time: 555 h of default (4 vCPU / 8 GB)
100 sandboxes at once; 12 / 60 / 200 starts per minute / hour / day
10 saved snapshots; no egress charge

Limit: Overage: small $0.018/h, default $0.036/h, large $0.072/h
Limit: Included time expires at the end of the billing period
Limit: Extra starts $0.03 to $0.07 each; extra saved snapshots $0.056 per day

$100 plan

$100 / month (per seat in an organisation)

Teams running many agents, or needing xlarge

$100 of VM time: 2,777 h of default
300 sandboxes at once; 30 / 210 / 840 starts per minute / hour / day
xlarge size (16 vCPU / 32 GB) at $0.200/h; 50 saved snapshots

Limit: Same overage rates from the wallet
Limit: $500 and $2,000 plans raise concurrency to 1,000 and 2,000

What it costs in practice

One agent, workdays	$20 / month One default sandbox 8 h a day for 22 days = 176 h × $0.036 = $6.34. It fits inside the $20 plan, and the unused $13.66 of included time expires.
One sandbox always on	~$26 / month One default sandbox 24/7 = 730 h × $0.036 = $26.28: the $20 plan plus $6.28 from the wallet. On large it's 730 × $0.072 = $52.56.
Ten agents, workdays	~$63 / month 10 default sandboxes 8 h a day for 22 days = 1,760 h × $0.036 = $63.36: the $20 plan plus $43.36 from the wallet. 10 starts a day is far under the 200-a-day limit. As a 2-seat organisation the plan is $40, but each seat also adds $20 of included time, so the total is the same.

Estimates in USD from list prices. Your bill depends on usage, region, and discounts.

Health & activity

Updated Nov 15, 2026 · last 90 days

No public repoThe product is closed source, so we can only track public package downloads.

Downloads

npm · @boatdev/sdk355k last 30 days
PyPI · boat-sdk118k last 30 days

Benchmarks

Node.js web tooling (StarSling sandbox benchmarks)

Source: StarSling hpc-sandbox-benchmarks, run 37073680030

V8 Web Tooling Benchmark throughput on a 4 vCPU / 8 GiB sandbox, median across 5 fresh sandboxes (10 trials) · runs/s, higher is better

Brezel30.1 runs/s
boat (default size)28.1 runs/s
Namespace24.9 runs/s
Daytona (VM)21.5 runs/s
Modal (VM)15.5 runs/s
Vercel Sandbox12.7 runs/s
E2B11.1 runs/s

How this was measured

Third-party run by StarSling, a CI-runner company that isn't one of the providers ranked. It uses the versioned Phoronix Test Suite profile, requests 4 vCPU / 8 GiB / 40 GB disk from every provider, and reports medians with cluster-bootstrap intervals. The raw dataset and methodology are public. StarSling marks this run partial (800 of 840 cells). boat is ranked 2nd but statistically tied with Namespace, Daytona (VM), Microsandbox Cloud and Novita. Every measured boat sandbox ran on an AMD Ryzen 9 9950X host; boat says its fleet also has Ryzen 7 9700X hosts and doesn't promise a CPU model. This is the table boat's homepage features. In the same run boat placed 15th of 15 on STREAM Triad memory bandwidth (32,500 MB/s, a table flagged 'too few sandboxes' for 14 of 15 providers), and its network tests failed or were skipped. Seven of 15 providers shown; copied from the leaderboard on 2026-11-15.

Versions: boat default size as of 2026-10-03

Published Oct 3, 2026 · source

Mastra cold install (StarSling sandbox benchmarks)

Source: StarSling hpc-sandbox-benchmarks, run 37073680030

Wall-clock time for a cold dependency install of the Mastra repository on a 4 vCPU / 8 GiB sandbox, median across 12 sandboxes · s, lower is better

Namespace37.2 s
Daytona (VM)39.3 s
Blaxel42 s
boat (default size)44.4 s
Modal (VM)46.4 s
Vercel Sandbox59 s
E2B72.9 s

How this was measured

Same run and methodology as the table above. This is the headline 'real-world' metric, which StarSling puts first because it's what a developer or CI job actually waits on. boat ranks 4th, statistically tied with Novita, Modal (VM), Microsandbox Cloud, Brezel and Vercel Sandbox. Seven of 15 providers shown; copied from the leaderboard on 2026-11-15.

Versions: boat default size as of 2026-10-03

Published Oct 3, 2026 · source