boat vs SF Compute

SF Compute is a GPU cloud. It sells H100 and newer GPU time, plus Autoresearch nodes and RL sandboxes that agents drive over MCP. boat is an agent sandbox: a full Linux VM per agent, with Docker, a desktop, hosting and coding agents already inside. boat has no GPUs.

Where boat stands out

A whole machine the agent can build on

Each boat sandbox is a full VM with root, systemd and Docker, plus /dev/kvm for Android emulators and VMs. You can SSH in. boat saves the whole disk every minute and at each stop; memory and running programs are not saved.

Agents you drive from outside

Claude Code, Codex and five other agents come preinstalled. You prompt, steer and follow them through one API. With the optional paid credentials proxy on, the real keys of supported model providers stay outside the sandbox.

A desktop and hosting in the box

Every sandbox has a built-in desktop and Chrome at 60 fps for computer use, and up to 50 HTTPS URLs. An agent can open a browser, build an app and serve it from the same machine.

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.

SF Compute

SF Compute is a GPU cloud from San Francisco. It builds its own clusters and runs a market where teams buy and resell GPU time.

The main product is GPU capacity: VMs, bare metal and managed Slurm on 8-GPU nodes, bought on an order book and resold when plans change. Autoresearch is its platform for agents. It adds GPU and CPU nodes (containers with a persistent disk), batch GPU jobs, object storage, and small microVM sandboxes that bake once and fork with their memory for RL rollouts. It is made for teams that train, fine-tune and evaluate models. You drive it over MCP, an HTTP API or a CLI, not an interactive shell.

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. “5th cheapest of 365” means 4 of the other 364 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 rankSF Compute: price rank
Software factory
100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month
A coding agent that builds a full app needs Docker, kept packages and ports to host from. boat gives every sandbox all three. On Autoresearch you bake packages into your own container image and expose HTTPS ports; its docs do not cover Docker inside.
5th cheapest of 365
$634 a month
43rd of 365, with limits
$1,762 a month
Always-on AI employees
50 machines 24/7 · 2 vCPU / 4 GB
An assistant that lives on its machine all month needs its files kept, a desktop and its own agents. boat saves the whole disk and ships both. An Autoresearch node keeps its home folder across stops, so you install into that folder or bake packages into your image.
6th cheapest of 361
$657 a month
136th of 361, with limits
$4,520 a month
Code interpreter bursts
100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month
SF Compute sandboxes fork in milliseconds and park after about 1.5 seconds idle, so short code runs that wait between calls pay little. A running boat sandbox bills at the full rate, busy or idle.
189th of 336, with limits
$2,000 a month
141st of 336, with limits
$305 a month
Sandboxes for your users
300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month
3rd cheapest of 349
$209 a month
181st of 349, with limits
$7,691 a month
RL rollouts
1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month
Autoresearch sandboxes are built for RL: bake an environment once, then fork up to 256 copies per call with memory and running processes. boat forks copy the disk only.
39th of 349, with limits
$2,000 a month
3rd cheapest of 349
$7,200 a month
Agent evals
250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month
Both run evals. SF Compute can also run them on H100 GPUs. boat gives each eval a full VM with Docker and coding agents preinstalled.
10th cheapest of 359
$200 a month
39th cheapest of 359
$576 a month
CI runners
40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month
2nd cheapest of 374
$86 a month
82nd cheapest of 374
$691 a month
24/7 dev boxes
1 machine 24/7 · 4 vCPU / 8 GB
A dev box you sit in needs SSH and a disk that persists. boat has both. Autoresearch nodes take commands over the API; SF Compute market VMs have SSH but no persistent storage.
19th cheapest of 371
$26 a month
263rd of 371, with limits
$90 a month
Computer-use agents
50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month
Computer-use agents need a desktop and a browser in the machine. boat ships one, streamed at 60 fps. On SF Compute you set it up: install Xvfb, a browser and a VNC server in your own image.
7th cheapest of 367
$200 a month
48th cheapest of 367
$528 a month
Browser-use agents
50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month
7th cheapest of 356
$200 a month
110th of 356, with limits
$576 a month
Scraping & bots (startup scale)
50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month
6th cheapest of 343
$90 a month
108th cheapest of 343
$446 a month
Mass scraping & bots (undetectable)
1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month
73rd of 327, with limits
$1,800 a month
223rd of 327, with limits
$8,915 a month
Parallel red teaming
200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month
3rd cheapest of 351
$200 a month
26th cheapest of 351
$350 a month
GPU eval jobs
10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month
SF Compute rents H100 and B300 nodes by the minute and larger GPU clusters on its market. boat has no GPU on any size.
Not ranked: misses a requirement40th of 102, with limits
$1,008 a month
macOS agents
5 at once · 4 vCPU / 8 GB · 1,000 sessions of 20 min a month
Not ranked: misses a requirementNot ranked: misses a requirement

Key features

FeatureboatSF Compute
Kind of productAgent sandbox: one full Linux VM per agentGPU cloud and market, plus Autoresearch nodes, jobs and RL sandboxes for agents
Isolation and CPUFull VM with its own kernel on hosts dedicated to boat (Ryzen 7 9700X and Ryzen 9 9950X); vCPUs are shared; root, Docker and systemdSandboxes are microVMs with their own guest kernel and shared cores, commands run as root; cpu-2 and cpu-8 nodes are containers on reserved physical cores
Docker and nested virtualizationDocker with Compose inside, and /dev/kvm for Android emulators and VMsYou can import a Docker image as the sandbox or node filesystem; the docs do not cover running Docker inside, and do not offer /dev/kvm
GPUsNo GPU on any size; graphics run on the CPUH100 and B300 nodes (1 or 8 GPUs) by the minute, batch GPU jobs, and H100 and H200 nodes on the market
What survives a stopThe whole disk, saved every minute and at each stop; memory, running programs and /tmp are not saved. Kept at least 14 days, longer with a saved snapshotNodes keep their home folder on a 250 GiB volume; packages installed elsewhere reset. Sandbox snapshots keep memory, processes and disk until you delete them
ForkFork a running sandbox from its last 1-minute snapshot; the copy boots fresh, without memoryFork 1 to 256 sandboxes per call from a snapshot with memory and running processes, free copy-on-write
Idle billingBilled per second while running, busy or idle; auto-stop defaults to 1 hourA sandbox parks after about 1.5 seconds idle and stops paying the RAM rate on memory it hands back; a node bills through a 15-minute idle window
Shell accessFull SSH and SCPNo SSH on Autoresearch: commands run through the API, MCP or CLI; market VMs have SSH
Inbound accessUp to 50 HTTPS URLs per sandbox, token-gated by default; raw ports through the sandbox firewall (IPv6 only on sandboxes without their own IPv4)HTTPS URL per port with an unguessable hostname and an optional bearer token; raw TCP is not documented
Outbound rulesOpen internet; no API setting for network rules, you add firewall rules inside the sandboxOpen internet, or no network device at all for a hermetic sandbox, set when you bake it
Coding agentsSeven agents preinstalled, with a prompt, steer and events API you drive from outsideYou connect your own agent over MCP from your machine, or install one in your image; a managed researcher agent is in private preview
Keeping API keys out of the sandboxOptional paid credentials proxy ($1 per day on, plus $1 per 1,000 requests): real keys of supported model providers stay outside. Other secrets enter the sandboxSecrets passed as environment variables on each command, redacted in logs; data and registry credentials are used on the host and never enter the node
Desktop and browserBuilt-in desktop and Chrome at 1920x1080, 60 fps, with computer-use toolsNone ships. You set it up: run Xvfb, a browser and a VNC or noVNC server in your own image, reached through an HTTPS port
RegionsGermany, Finland and France; you cannot pick one; no US or Asia regionMarket capacity in the US and EU North (europe-north1)
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)
FeatureboatSF Compute
Full VM (own kernel)YesYes
Docker insideYesNot published
Nested virtualizationYesNo documented way
RootYesYes
SystemdYesNot published
Desktop GUIYesWith your own setup
BrowserYesWith your own setup
Computer-use APIYesNo documented way
SSHYesNo documented way
Persistent diskYesYes
Pause/resumeYesYes
Automatic snapshotsYesNo documented way
Fork/cloneYesYes
Live fork (no pause)YesYes
Memory forkNo documented wayYes
Custom image (Docker or snapshot)YesYes
≥24 h sessionsYesYes
Idle auto-stopNo documented wayYes
Wake on requestNo documented wayYes
HTTPS preview URLsYesYes
Raw TCP inboundYesNo documented way
Public IPv4No documented wayNo documented way
Egress allowlistNo documented wayYes
Secret proxyYesNo documented way
Secret proxy for any APIYesNo documented way
Preinstalled agentsYesNo documented way
Agent harness APIYesNo documented way
MCP serverNo documented wayYes
GPUNo documented wayYes
WindowsNo documented wayNo documented way
Arm64No documented wayNo documented way
Self-hostingNo documented wayNo documented way
BYOCNo documented wayNo documented way
Open sourceNo documented wayNo documented way
Scoped API keysYesYes
Spend limitsYesYes
Usage per sandbox (API)YesYes
Audit logsNo documented wayYes
SSONo documented wayNo documented way
SOC 2No documented wayYes
HIPAANo documented wayNo documented way
EU data residencyNo documented wayYes
Zero data retentionYesNot published

Performance

boat28.06 runs/s
SF ComputeNot measured

SF Compute is not in StarSling's CPU run of 3 October 2026, so we cannot say which is faster on CPU. In that run (run 37073680030), boat did 28.06 runs/s on the V8 Web Tooling Benchmark at 4 vCPU / 8 GB, second of 15, on an AMD Ryzen 9 9950X host; boat also places some sandboxes on other hosts. ComputeSDK measured Autoresearch sandboxes at a 335 ms median start with 100 at once (2026-11-06, default size). boat is not in ComputeSDK.

V8 Web Tooling Benchmark at a 4 vCPU / 8 GB target, runs per second, higher is better. Source: StarSling's run of 3 October 2026 (run 37073680030), median of 5 sandboxes. Providers ran on different hardware in different regions. StarSling gives a shared rank when the run cannot separate two providers. Every boat sandbox in this run ran on an AMD Ryzen 9 9950X host. boat also places some sandboxes on AMD Ryzen 7 9700X hosts, which this run did not measure. SF Compute is not in the public benchmarks we use, so we cannot say which is faster.

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. Start limits, per-start fees and plan minimums are in the price, boat's included.

Scenarioboat, a monthSF Compute, 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$23 · Prepaid credit (pay as you go) · with limits
One agent that never stops
4 vCPU / 8 GB, 50 GB disk, running 24/7 all month
$26 · $20/mo$95 · Prepaid credit (pay as you go) · 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$1,140 · Prepaid credit (pay as you go) · 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$648 · Prepaid credit (pay as you go)
1,000 short jobs at once
2 vCPU / 4 GB, 1,000 at once, 3 minutes each, 200,000 jobs a month
$8,800 · $2000/mo · with limits$1,440 · Prepaid credit (pay as you go)

boat: Plans cost $20, $100, $500 or $2,000 a month. The whole plan price comes back as VM time for that month, and it does not carry over. A 4 vCPU / 8 GB sandbox costs $0.036 an hour, billed per second while it runs. A stopped sandbox has no compute charge; saved snapshots above the included count and the AI credentials proxy still charge. The $20 plan includes 12 starts a minute and 200 a day; each extra start costs $0.03 to $0.07. A card is required, also for the 7-day free trial.

SF Compute: SF Compute runs on prepaid credit with no plan fee and no free credit. A cpu-2 node (2 reserved physical cores, 8 GiB) costs $0.144 an hour, billed per minute through a 15-minute idle window, with 33% off on Fridays and 20% off on weekends; a stopped node parks its disk at no charge. Sandboxes bill $0.04572 per GiB of RAM per hour while running, with 4 GiB counted per core, and much less for memory handed back while parked. Object storage is free to 100 GB, then $0.10 per GB-month. Market GPU prices are set by the order book.

Limits in those scenarios

The table prices SF Compute on an Autoresearch cpu-2 node: 2 reserved physical cores and 8 GiB, counted as 4 vCPU, at $0.144 an hour. It is the closest match that keeps a disk. Only its home folder survives a stop, so packages installed elsewhere reset; this is why the agent scenarios do not fully fit. By default its docs allow 256 live CPU nodes per member, so the 300 and 1,000 machine scenarios need a raised limit. On boat, the $20 plan runs 100 sandboxes at once and includes 200 starts a day. Even the $2,000 plan includes only 90 starts a minute and charges $0.03 to $0.07 for each start above that; starts above 270 a minute fail with 429, and boat never queues a start.

ScenarioboatSF Compute
One coding agent, workdaysNone for this scenario
  • files survive a stop only in the home folder: packages installed elsewhere reset
  • disk beyond 0 GiB: price unknown
One agent that never stopsNone for this scenario
  • files survive a stop only in the home folder: packages installed elsewhere reset
  • disk beyond 0 GiB: price unknown
50 agents in parallelNone for this scenario
  • files survive a stop only in the home folder: packages installed elsewhere reset
  • disk beyond 0 GiB: price unknown
Sandboxes for your usersNone for this scenario
  • disk beyond 0 GiB: price unknown
1,000 short jobs at once
  • 2,400 starts/day included, paid above that (needs ~6,667)
  • 600 starts/hour included, paid above that (needs ~1,000)
  • starts above 90/min are paid, and starts above 270/min fail with 429
  • disk beyond 0 GiB: price unknown
boat's own limits
  • Plans start at $20 a month. The plan price comes back as VM time for that month. Unused time does not carry over. Source
  • Starts (create, fork, resume): the $20 plan includes 12 a minute, 60 an hour and 200 a day. The $2,000 plan includes 90 a minute, 600 an hour and 2,400 a day. Extra starts cost $0.03 to $0.07. A start fails with 429 above 3 times the included starts per minute, or 6 times per hour or per day. Source
  • A stopped sandbox has no compute charge. boat keeps its disk for at least 14 days. A saved snapshot keeps it longer; extra saved snapshots and the AI credentials proxy are paid. Source
  • Auto-stop defaults to 1 hour. You can turn it off on a paid plan. The trial allows 2 hours at most. Source
  • boat saves the disk every minute and at each stop. Memory, running programs and /tmp are not saved: a resume is like a reboot. A machine failure loses writes after the last good snapshot. Source
  • Machines run in Germany, Finland and France only. Snapshot copies can be stored outside the EU. Source
  • Public IPv6 and outbound IPv4. No dedicated IPv4. The IP changes at each resume. Source
  • Hosts are AMD Ryzen 7 9700X and Ryzen 9 9950X. vCPUs are shared. No GPU. You cannot boot your own Docker image. Source

Which one should you use?

Use SF Compute when

the work needs GPUs, from one H100 to a cluster, or when you run RL rollouts that fork many copies of one running environment with its memory.

  • Real GPUs, from one to a cluster. For training, fine-tuning and GPU evals: H100 and B300 nodes by the minute, and larger clusters you can resell.
  • Memory forks for RL rollouts. For RL and evals that branch from one state: fork up to 256 sandboxes per call with memory and processes intact.
  • Hermetic sandboxes, storage and US capacity. A sandbox can have no network device at all, nodes get object storage and private addresses, and GPU capacity runs in the US as well as the EU.

Use boat when

your agents need a whole machine to work in: Docker, emulators, a built-in desktop and browser, apps they host, coding agents you drive from outside, and a disk saved every minute.

  • A whole machine the agent can build on. On an Autoresearch node only the home folder survives a stop, so installed packages reset, there is no SSH, and its docs do not cover running Docker or docker compose inside.
  • Agents you drive from outside. SF Compute lets your local agent call its platform over MCP. To run an agent inside a sandbox, you install it in your image, and secrets reach it as environment variables.
  • A desktop and hosting in the box. SF Compute machines ship headless with no desktop stream, so a computer-use agent needs a display you set up yourself, and only HTTPS ports can be exposed, not raw TCP.

Questions people ask

Is boat cheaper than SF Compute?

For CPU agent work, in four of the five scenarios on this page, yes. One workday coding agent is $20 a month on boat and about $23 on an SF Compute cpu-2 node at $0.144 an hour. The $20 boat plan includes 200 starts a day; more starts cost $0.03 to $0.07 each. For 1,000 short jobs at once, SF Compute is lower: about $1,440 against $8,800 on boat, because boat charges for starts above the included number. Idle-heavy loops can cost less on SF Compute sandboxes, which park after about 1.5 seconds idle. For GPU work there is no comparison, because boat has no GPUs.

Does SF Compute have GPUs?

Yes, that is its main business. Autoresearch rents H100 and B300 nodes with 1 or 8 GPUs by the minute and runs batch GPU jobs. The SF Compute market sells H100 and H200 nodes. boat has no GPU on any size.

Can I SSH into an SF Compute machine?

Not on Autoresearch: its docs say there is no SSH. Commands run through MCP, the HTTP API or the autor CLI. SF Compute market VMs support SSH. Every boat sandbox has full SSH and SCP.

Can I run Docker in each?

boat runs Docker with Compose inside every sandbox and also exposes /dev/kvm for Android emulators and VMs. SF Compute can import a Docker image as a sandbox or node filesystem. Its docs do not cover running Docker inside a sandbox or a node.

Can boat fork a sandbox with its memory like SF Compute?

No. boat forks a running sandbox from its last 1-minute snapshot, with its disk, and the copy boots fresh. SF Compute sandboxes fork from a snapshot with memory and running processes, up to 256 per call.

Is boat SOC 2 compliant?

SOC 2 is in progress; boat has no SOC 2 report today. SF Compute lists a SOC 2 Type 2 report on its trust center. Live boat status is at status.ascii.dev and our data retention policy is at docs.boat.dev/data-retention.

Start a sandbox

Install the CLI, then run boat new.

7-day free trial

More comparisons