boat vs Tinfoil

Tinfoil runs a Docker image you publish inside a confidential-computing enclave, so users can check that nobody, Tinfoil included, can read their data. boat gives each agent a full VM to work in, with root, Docker, a disk saved every minute, a desktop and coding agents you drive from outside.

Where boat stands out

A machine the agent can change

A boat sandbox is a full VM where the agent has root, systemd, Docker and /dev/kvm, and can install whatever the task needs. A Tinfoil container starts with zero Linux capabilities, rejects privileged mode and devices, and runs the image and resources pinned in a published config.

Work that survives a stop

boat keeps the whole disk, saves it every minute and on stop, and lets you fork a running sandbox to try two paths at once. On Tinfoil the enclave filesystem is wiped when the instance stops or updates, and only an encrypted volume you attach, still in beta, keeps data.

Coding agents you drive from outside

boat ships Claude Code, Codex, OpenCode and other agents in every sandbox, with a prompt, steer and events API and a 60fps desktop to watch. Its credentials proxy keeps real AI keys outside the sandbox and can switch to another account when one runs out of credits.

The companies and products

boat

boat.dev runs sandboxes for AI agents: each one is a full Linux VM with its own kernel.

You start a machine with one command or one API call, then SSH in, run Docker, open a desktop, or let a built-in agent (Claude Code, Codex, OpenCode and others) work in it. The disk stays between sessions, and a stopped sandbox costs nothing.

Tinfoil

Tinfoil is a private AI company: it sells an encrypted chat app, a private inference API and Tinfoil Containers, a host for your own apps inside secure enclaves.

Tinfoil Containers is a hosting platform, not an agent sandbox. You publish a tinfoil-config.yml in a public GitHub repo that pins a Docker image, its CPU, memory and network rules, and Tinfoil boots it inside an AMD SEV-SNP or Intel TDX enclave whose memory is encrypted in hardware. Clients use remote attestation to check the exact code running, which suits teams that must prove they cannot see their users' data.

Which use cases each one fits

For each use case, Battleships prices every sandbox provider on the same workload and sorts them from the cheapest monthly bill up. “3rd cheapest of 367” means 2 of the other 366 providers run that workload, with every feature it needs, for less. “With limits” means the provider runs it but misses something, so it is listed after every full match. Open a use case for its whole ranking and what each provider misses.

Use case and its workloadboat: price rankTinfoil: price rank
Software factory
100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month
A software factory needs Docker, root and a disk that keeps installs; Tinfoil containers drop all Linux capabilities and lose local files when they stop.
3rd cheapest of 367
$634 a month
309th of 367, with limits
$6,252 a month
Always-on AI employees
50 machines 24/7 · 2 vCPU / 4 GB
An always-on assistant fits Tinfoil only as a fixed app you redeploy to change, while on boat it is a VM the agent can reshape itself.
5th cheapest of 366
$657 a month
296th of 366, with limits
$6,160 a month
Code interpreter bursts
100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month
186th of 345, with limits
$2,000 a month
339th of 345, with limits
$10,651 a month
Sandboxes for your users
300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month
Tinfoil allows 10 instances per organization, so 300 sandboxes for your users need a raised quota through sales.
Cheapest of 361
$209 a month
325th of 361, with limits
$4,943 a month
RL rollouts
1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month
34th of 360, with limits
$2,000 a month
335th of 360, with limits
$121 a month
Agent evals
250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month
10th cheapest of 364
$200 a month
332nd of 364, with limits
$4,943 a month
CI runners
40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month
CI jobs that build images need Docker inside the machine, which boat has on every VM and Tinfoil does not document.
4th cheapest of 377
$86 a month
362nd of 377, with limits
$4,057 a month
24/7 dev boxes
1 machine 24/7 · 4 vCPU / 8 GB
A dev box on Tinfoil means an admin container with SSH and an encrypted volume for anything that must survive a stop.
18th cheapest of 368
$26 a month
187th of 368, with limits
$269 a month
Computer-use agents
50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month
Tinfoil has no desktop or browser to drive, so computer-use agents fit only as a compromise there.
6th cheapest of 369
$200 a month
358th of 369, with limits
$6,584 a month
Browser-use agents
50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month
7th cheapest of 367
$200 a month
355th of 367, with limits
$4,943 a month
Scraping & bots (startup scale)
50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month
6th cheapest of 346
$90 a month
305th of 346, with limits
$872 a month
Mass scraping & bots (undetectable)
1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month
71st of 338, with limits
$1,800 a month
271st of 338, with limits
$17,067 a month
Parallel red teaming
200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month
2nd cheapest of 360
$200 a month
307th of 360, with limits
$861 a month
GPU eval jobs
10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month
Only Tinfoil fits: it rents H200, B200 and B300 GPUs in confidential mode, and boat has no GPUs.
Can't run it98th of 98, with limits
$51,201 a month
macOS agents
5 at once · 4 vCPU / 8 GB · 1,000 sessions of 20 min a month
Can't run itCan't run it

Key features

FeatureboatTinfoil
What it isA sandbox VM per agent, created and driven by APIA host for one Docker app per enclave, deployed from a config in a public repo
IsolationEvery sandbox is a full VM with its own kernelEach instance is a confidential VM (AMD SEV or Intel TDX) with memory encrypted in hardware
Proving what runsNo attestation of the running codeRemote attestation: clients check the enclave runs the exact published config and image digest
Root, Docker and nested virtualizationRoot, systemd, Docker and /dev/kvm on every machineContainers start with zero capabilities; privileged and devices are rejected; only an admin container runs as root; no documented Docker inside
Files after a stopThe whole disk is kept, with snapshots every minute and on stop; a stopped sandbox costs nothingThe enclave filesystem is wiped on stop; only attached encrypted volumes (beta) keep data
Fork and snapshotsFork a running sandbox; named snapshots to start new onesNo fork or snapshot; new instances boot from the saved config
Changing the machineThe agent installs and changes what it needs while it runsResources and image are fixed in the measured config; changes ship as a new release
Shell accessSSH and exec into any sandboxNo SSH by default; a debug instance (not attested) or an admin container with port 22
Reaching the machineHTTPS URLs for up to 50 ports, token-protected; raw TCP; dedicated IPv6 or IPv4One HTTPS upstream port with listed paths, on a tinfoil.dev subdomain or your own domain; you build the auth
Outbound internetOpen internet on every plan; no provider-side allowlistClosed by default; per-network hostname allowlist or open, set in the attested config
Keeping keys out of the machineAI credentials proxy for model providers and any HTTPS API you register; switch a credential to another account when credits run outSecrets are injected into the container as environment variables, released after attestation
Coding agents and desktopClaude Code, Codex, OpenCode and more preinstalled with a prompt, steer and events API; 60fps desktop with ChromiumNone in the containers; its coding-agents guide points your own agents at its private inference API
GPUsNoneH200, B200 and B300 in confidential mode, 1 to 8 per instance, on request
Machines at once100 to 2,000 by plan10 instances per organization; more through contact
Compliance and transparencySOC 2 is in progress. We're open about where we stand: live status and data retention policy.SOC 2
Every feature Battleships tracks (43)
FeatureboatTinfoil
Full VM (own kernel)YesYes
Docker insideYesNo
Nested virtualizationYesNo
RootYesYes
SystemdYesNot published
Desktop GUIYesNo
BrowserYesNo
Computer-use APIYesNo
SSHYesYes
Persistent diskYesYes
Pause/resumeYesNo
Automatic snapshotsYesNo
Fork/cloneYesNo
Live fork (no pause)YesNo
Memory forkNoNo
Custom image (Docker or snapshot)YesYes
≥24 h sessionsYesYes
Idle auto-stopNoNo
Wake on requestNoNo
HTTPS preview URLsYesYes
Raw TCP inboundYesNot published
Public IPv4NoNo
Egress allowlistNoYes
Secret proxyYesNo
Secret proxy for any APIYesNo
Preinstalled agentsYesNo
Agent harness APIYesNo
MCP serverNoNo
GPUNoYes
WindowsNoNo
Arm64NoNo
Self-hostingNoNo
BYOCNoNo
Open sourceNoYes
Scoped API keysYesNot published
Spend limitsYesNot published
Usage per sandbox (API)YesYes
Audit logsNoNot published
SSONoYes
SOC 2NoYes
HIPAANoYes
EU data residencyYesNo
Zero data retentionYesNot published

Performance

boat26.69 runs/s
TinfoilNot measured

There is no comparable measurement. boat did 26.69 Node.js build-and-test runs per second at 4 vCPU / 8 GB in StarSling's public benchmark, but Tinfoil has no result in that benchmark, so we cannot say which is faster.

Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Tinfoil has no published run in this suite yet, so there is no like-for-like number to compare.

Pricing in five scenarios

Each bill is priced from the providers' public price lists by the open Battleships engine, with the cheapest plan that runs the scenario.

Scenarioboat, a monthTinfoil, a month
One coding agent, workdays
4 vCPU / 8 GB, 50 GB disk, 8 hours a day, 22 days a month, files kept between days
$20 · $20/mo$91 · Tinfoil Containers · with limits
One agent that never stops
4 vCPU / 8 GB, 50 GB disk, running 24/7 all month
$26 · $20/mo$269 · Tinfoil Containers · with limits
50 agents in parallel
50 machines of 4 vCPU / 8 GB at once, 8-hour sessions, 1,100 sessions a month, Docker inside
$317 · $20/mo × 3 seats$3,136 · Tinfoil Containers · with limits
Sandboxes for your users
2 vCPU / 4 GB, 30,000 sessions of 10 minutes a month, up to 300 at once
$200 · $100/mo × 2 seats$4,943 · Tinfoil Containers · with limits
1,000 short jobs at once
2 vCPU / 4 GB, 1,000 at once, 3 minutes each, 200,000 jobs a month
$2,000 · $2000/mo · with limits$32,842 · Tinfoil Containers · with limits

boat: You pick a monthly plan from $20, and all of it becomes usage credit spent per second on running sandboxes. A stopped sandbox costs nothing and keeps its disk.

Tinfoil: Tinfoil Containers costs a $20 monthly base fee on top of usage, not as credit, plus $0.06 per core-hour and $0.01 per GB-hour of memory, billed in whole hours with no shorter intervals. Encrypted volumes cost $0.30 per GB-month of provisioned size until you delete them, even while the container is stopped, and GPUs bill hourly with a monthly commitment.

Limits in those scenarios

Tinfoil does not fit any scenario cleanly because it is an app host, not a sandbox: there is no Docker inside, and only attached volumes survive a stop. Its quota is 10 instances per organization, so 50 agents, 300 sandboxes for your users or 1,000 jobs at once all need a raised limit through sales. Whole-hour billing also makes short jobs costly, since a five-minute run bills a full hour. On boat, the 1,000-job burst is queued above 90 starts a minute and needs more starts per day than the $2,000 plan allows.

ScenarioboatTinfoil
One coding agent, workdaysNone that bite
  • adjacent product, not a like-for-like sandbox
  • different kind of product: app platform
  • files survive a stop only in attached volumes: packages installed elsewhere reset
  • egress price not published
One agent that never stopsNone that bite
  • adjacent product, not a like-for-like sandbox
  • different kind of product: app platform
  • files survive a stop only in attached volumes: packages installed elsewhere reset
  • egress price not published
50 agents in parallelNone that bite
  • Tinfoil Containers: max 10 concurrent
  • adjacent product, not a like-for-like sandbox
  • missing Docker inside
  • different kind of product: app platform
Sandboxes for your usersNone that bite
  • Tinfoil Containers: max 10 concurrent
  • adjacent product, not a like-for-like sandbox
  • different kind of product: app platform
  • needs talk to sales: fleet of 300 needs a quota increase: new accounts get 10 machines; only through sales
1,000 short jobs at once
  • max 2,400 starts/day (needs ~6,667)
  • max 600 starts/hour (needs ~1,000)
  • bursts above 90 starts/min are queued (a burst of 333 takes ~4 min to start)
  • Tinfoil Containers: max 10 concurrent
  • adjacent product, not a like-for-like sandbox
  • different kind of product: app platform
  • needs talk to sales: fleet of 1,000 needs a quota increase: new accounts get 10 machines; only through sales

Which one should you use?

Use Tinfoil when

you host an app that must prove to its users, by attestation, that no one can read their data, or when you need confidential GPUs for private inference.

  • Encrypted memory you can prove. Useful if users must verify that no operator can read their data: Tinfoil runs code in attested, memory-encrypted enclaves.
  • Confidential GPUs. Useful for private model inference or training: Tinfoil rents H200, B200 and B300 GPUs inside confidential enclaves.
  • Network rules in the attested config. Useful when outbound traffic must be locked down: Tinfoil containers have no internet unless a hostname allowlist opens it.

Use boat when

your agents need a machine to work in: root, Docker and a disk that persists, with forks, a desktop and coding agents you drive from outside while their keys stay out.

  • A machine the agent can change. An agent that runs docker compose, apt installs or an emulator hits a wall in a Tinfoil container. The only way to change it is a new release of the config, which is fine for an app and too slow for an agent.
  • Work that survives a stop. On Tinfoil every stop throws away installed packages and anything written outside the volume, so an agent pays the setup again on each start. There is no fork, so trying two ideas means two cold deploys.
  • Coding agents you drive from outside. Tinfoil has no agent runtime in its containers: you build the image, the control API and the log stream yourself, and secrets arrive inside as environment variables that code the agent runs can read.

Questions people ask

Is boat cheaper than Tinfoil?

In every scenario we priced, yes, though none of them is a clean fit for Tinfoil. One coding agent on workdays comes to $20 a month on boat and $91 on Tinfoil, one agent running all month to $26 versus $269, and 50 agents in parallel to $317 versus $3,136. Tinfoil's numbers include volume storage and assume a quota above its 10 instances.

Is Tinfoil a sandbox for AI agents?

Not really. Tinfoil Containers hosts a Docker app you publish, inside a confidential enclave, with its resources and network rules fixed in a public config. It suits private services and inference servers, while boat gives each agent its own VM to install, build and test in.

Can I run Docker inside a Tinfoil container?

Tinfoil does not document it. Regular containers start with zero Linux capabilities and reject privileged mode and devices; only an admin container runs privileged as root. boat has Docker, root, systemd and /dev/kvm on every machine.

Do files survive a stop on Tinfoil?

Only on an attached encrypted volume, which is in beta. The enclave's own filesystem and memory are lost when it stops or is replaced in an update. A boat sandbox keeps its whole disk, saved every minute and on stop.

Does Tinfoil have GPUs?

Yes. Tinfoil rents NVIDIA H200, B200 and B300 GPUs in confidential-computing mode, hourly with a monthly commitment, once it enables GPU access for your organization. boat has no GPUs.

Is boat SOC 2 compliant?

SOC 2 is in progress. We're open about where we stand: live status at status.ascii.dev and our data retention policy at docs.boat.dev/data-retention. Tinfoil's trust portal lists SOC 2, HIPAA and GDPR.

Start a sandbox

Install the CLI, then run boat new.

7-day free trial

More comparisons