boat vs Firecrawl

Firecrawl is a web data API: search, scrape and crawl, plus Interact, a cloud browser session of up to one hour. You drive it with Playwright code or a plain-language prompt. boat gives each agent a full Linux VM with root, Docker and a disk that boat saves every minute. Each VM has a 60 fps desktop, Chrome and coding agents that you drive from outside.

Where boat stands out

A whole machine, not only a browser

Every boat sandbox is a full VM with root, SSH, systemd, Docker and nested virtualization, on dedicated AMD Ryzen hosts (Ryzen 7 9700X and Ryzen 9 9950X). Chrome runs on a desktop inside it. A Firecrawl session is a container built around one Chromium, where you run Playwright, Python, Node or bash against that browser.

Runs as long as the task, keeps its disk

On a paid plan you can turn boat's auto-stop off (default 1 h), and then there is no session cap. boat saves the disk every minute and at each stop, and keeps a stopped sandbox's disk for at least 14 days. A resume is like a reboot: memory and running programs are not saved, and a machine failure loses the writes after the last snapshot. A Firecrawl session ends after one hour at most and closes after 5 idle minutes by default. Its files end with the session; a profile keeps cookies and localStorage.

Coding agents that you drive from outside

boat installs seven coding-agent harnesses in each sandbox, among them Claude Code, Codex and OpenCode. You prompt, steer and follow them through one API with an event stream. Firecrawl's prompt mode runs its own browser agent and returns one answer per call. boat's optional AI credentials proxy keeps the real keys of supported providers outside the sandbox. It costs $1 for each day it is on, plus $1 per 1,000 requests.

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.

Firecrawl

Firecrawl calls itself the web data API for AI agents: one API to search, scrape and interact with the web and turn pages into clean Markdown or structured data.

Its main products turn web pages into Markdown or JSON per call. Interact and the Browser Sandbox add a cloud Chromium session with Playwright and agent-browser preinstalled. You drive it with code, bash commands or a prompt to Firecrawl's own browser agent, and you watch it through a live view. It is a browser and web data product, not a general sandbox. Files on the session disk end with the session, unless your code copies them to your own storage first. This page compares where the two products overlap: agents that browse.

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 rankFirecrawl: price rank
Software factory
100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month
A coding agent needs a shell, Docker and a disk that stays. A Firecrawl session is a browser container that ends within an hour, and its docs do not mention Docker. Builds and tests fit boat.
5th cheapest of 365
$634 a month
126th of 365, with limits
$846 a month
Always-on AI employees
50 machines 24/7 · 2 vCPU / 4 GB
An always-on assistant hits Firecrawl's one-hour session cap. Between sessions it keeps a browser profile, and your code copies files to your own storage. On a boat paid plan you can turn auto-stop off (default 1 h), and then there is no session cap.
6th cheapest of 361
$657 a month
287th of 361, with limits
$3,779 a month
Code interpreter bursts
100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month
189th of 336, with limits
$2,000 a month
290th of 336, with limits
$749 a month
Sandboxes for your users
300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month
Firecrawl's interactive live view lets your users click inside a browser. boat gives each user a whole machine, with up to 50 token-gated HTTPS URLs for hosting.
3rd cheapest of 349
$209 a month
239th of 349, with limits
$984 a month
RL rollouts
1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month
39th of 349, with limits
$2,000 a month
262nd of 349, with limits
$5,716 a month
Agent evals
250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month
10th cheapest of 359
$200 a month
269th of 359, with limits
$749 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
273rd of 374, with limits
$399 a month
24/7 dev boxes
1 machine 24/7 · 4 vCPU / 8 GB
A dev box has to keep its files. Firecrawl sessions end within an hour and keep only a browser profile, so this fits boat. On boat, programs and memory do not survive a stop: a resume is like a reboot.
19th cheapest of 371
$26 a month
260th of 371, with limits
$82 a month
Computer-use agents
50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month
Firecrawl's agent works on web pages through the page structure. boat streams a 60 fps Linux desktop, where an agent can use any desktop app, not only a browser.
7th cheapest of 367
$200 a month
196th of 367, with limits
$399 a month
Browser-use agents
50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month
Both fit. Firecrawl gives a managed Chromium that you drive with a prompt or Playwright. boat gives Chrome on a full VM, where the same agent can also run code and keep files.
7th cheapest of 356
$200 a month
21st cheapest of 356
$399 a month
Scraping & bots (startup scale)
50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month
Firecrawl's scrape API returns clean Markdown for about a credit a page, so many pages need no browser session. On boat, you run Chrome or your own scraper on the VM.
6th cheapest of 343
$90 a month
64th cheapest of 343
$569 a month
Mass scraping & bots (undetectable)
1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month
Firecrawl's scrape API retries blocked requests through enhanced proxies at no extra credit. boat has no proxy service. You set it up: route Chrome on the VM through a proxy provider that you pay for.
73rd of 327, with limits
$1,800 a month
242nd of 327, with limits
$11,449 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
306th of 351, with limits
$749 a month
GPU eval jobs
10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month
Not ranked: misses a requirementNot ranked: misses a requirement
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

FeatureboatFirecrawl
Kind of productA full Linux VM per sandbox, with a desktop and Chrome insideWeb data API (search, scrape, crawl) plus managed browser sessions
IsolationFull VM with its own kernel on dedicated AMD Ryzen hosts (Ryzen 7 9700X and Ryzen 9 9950X); vCPUs are sharedEphemeral cloud container per browser session
What the agent can touchThe whole machine: root via sudo, SSH, any package, any processPython, Node and bash run in the session against its browser; the docs show no SSH
Docker and nested virtualizationDocker with BuildKit and Compose, systemd and /dev/kvm on every sandboxNot documented; sessions run a preset Chromium image with Playwright and agent-browser, and the docs do not mention Docker
Session lengthAuto-stop defaults to 1 hour; set up to 30 days, or turn it off on a paid plan for no cap10 minutes by default, 1 hour at most, 5 minutes idle by default
Files after a sessionboat saves the disk every minute and at each stop. A stopped sandbox keeps its disk for at least 14 days; a saved snapshot keeps it longer. Memory and /tmp are not saved.Files end with the session. You set it up: read files in the session and save them to your own storage. A named profile keeps cookies and localStorage.
Fork and pauseStop and resume with the disk kept (a resume is like a reboot); fork a running sandbox from its last 1-minute snapshotNot in the docs: sessions are create, execute, list and delete. A new session can load a saved profile.
Browser and computer use1920x1080 60 fps desktop with Chrome, computer use on any desktop app, embeddable streamChromium with CDP, a live view, and an interactive live view that users can click in
Web data without a browserYou run your own scraper, or call a web data API from the VMScrape, crawl and search to Markdown or JSON, with ads blocked by default
Blocked sitesNo proxy service; you set it up: route Chrome through a proxy providerScrape retries through enhanced proxies on 401, 403 and 429, at no extra credit
AgentsSeven coding-agent harnesses preinstalled (Claude Code, Codex, OpenCode and others), with a prompt, steer and events APIIts own browser agent from a prompt, one answer per call; an MCP server and skills let outside coding agents call Firecrawl
Reaching the machineSSH and up to 50 HTTPS URLs per sandbox, token-gated by default; public IPv6 addressCDP WebSocket and live view links; the docs show no way to host a port
Keeping keys out of the sandboxOptional AI credentials proxy ($1 a day while on, plus $1 per 1,000 requests); with it on, real keys of supported providers stay outside the sandboxNot in the docs; code in a session uses the keys that you pass to it
Where it browses fromMachines in Germany, Finland and France; no region choice, no US or Asia regionUS by default, or any country that you name per session
Outbound rulesOpen internet; no provider-side allowlistThreat Protection allow and block lists by domain, on Enterprise
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)
FeatureboatFirecrawl
Full VM (own kernel)YesNo documented way
Docker insideYesNo documented way
Nested virtualizationYesNo documented way
RootYesNot published
SystemdYesNot published
Desktop GUIYesNo documented way
BrowserYesYes
Computer-use APIYesNo documented way
SSHYesNo documented way
Persistent diskYesNo documented way
Pause/resumeYesNo documented way
Automatic snapshotsYesNo documented way
Fork/cloneYesNo documented way
Live fork (no pause)YesNo documented way
Memory forkNo documented wayNo documented way
Custom image (Docker or snapshot)YesNot published
≥24 h sessionsYesNo documented way
Idle auto-stopNo documented wayYes
Wake on requestNo documented wayNo documented way
HTTPS preview URLsYesNo documented way
Raw TCP inboundYesNot published
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 wayNo documented way
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 keysYesNot published
Spend limitsYesNot published
Usage per sandbox (API)YesNo documented way
Audit logsNo documented wayNot published
SSONo documented wayYes
SOC 2No documented wayYes
HIPAANo documented wayNo documented way
EU data residencyNo documented wayNo documented way
Zero data retentionYesYes

Performance

boat28.06 runs/s
FirecrawlNot measured

Firecrawl is not in the public benchmarks we use, so we cannot say which is faster. It does not publish the vCPU or RAM behind a browser session. For reference, a 4 vCPU / 8 GB boat sandbox did 28.06 runs/s on the V8 Web Tooling Benchmark in StarSling's run of 3 October 2026 (run 37073680030), second of 15. That sandbox ran on an AMD Ryzen 9 9950X host. boat also places some sandboxes on other hosts.

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. Firecrawl 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 monthFirecrawl, 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$19 · Hobby · with limits
One agent that never stops
4 vCPU / 8 GB, 50 GB disk, running 24/7 all month
$26 · $20/mo$82 · Hobby · 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$399 · Growth · 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$749 · Scale · with limits
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$942 · Scale · with limits

boat: Plans cost $20, $100, $500 or $2,000 a month. The whole plan price comes back as VM time for that month, billed per second while a sandbox runs; unused time does not carry over. A card is required, also for the 7-day trial. The $20 plan includes 100 sandboxes at once and 12 starts a minute, 60 an hour and 200 a day. An extra start costs $0.03 to $0.07, up to 1,200 starts a day. A stopped sandbox has no compute charge; saved snapshots above the included count and the AI credentials proxy still charge.

Firecrawl: Plans sell a monthly pool of credits shared with scraping and search: 5,000 on Hobby ($19), 100,000 on Standard ($99), 500,000 on Growth ($399) and 1,000,000 on Scale ($749). Paid yearly, they cost $16, $83, $333 and $599 a month. A Free plan gives 1,000 credits a month. A browser session costs 2 credits a minute with code only and 7 with a prompt, billed per second with a one-minute minimum. Extra credits come in $5 top-ups. Zero Data Retention is Enterprise only and adds 2 credits a minute to a session.

Limits in those scenarios

The Firecrawl prices in the table are for its browser session. Firecrawl does not publish its size, and the session keeps no files, so none of these prices is for a 4 vCPU / 8 GB machine. The one-hour session cap is why no scenario in the table fits on Firecrawl: a workday or always-on agent must start a new browser every hour, and the fleet scenario also needs Docker. Concurrency is another limit. The plan table lists 50 sessions at once on Growth and 100 on Scale, but Firecrawl's browser docs cap every plan at 20 for the launch period, so the real cap on those plans may be 20. On boat, the 1,000-job burst does not fit. The top plan includes 90 starts a minute, 600 an hour and 2,400 a day. Starts above that cost $0.03 to $0.07 each, and starts above 270 a minute (3 times the included number) fail with 429. boat never queues a start. A boat sandbox keeps its disk for at least 14 days after a stop; keep it longer with a saved snapshot.

ScenarioboatFirecrawl
One coding agent, workdaysNone for this scenario
  • Hobby: sessions capped at 1 h
  • Hobby: Scale size not allowed
  • missing persistent disk
  • machine size not published
One agent that never stopsNone for this scenario
  • Hobby: sessions capped at 1 h
  • Hobby: Scale size not allowed
  • missing persistent disk
  • machine size not published
50 agents in parallelNone for this scenario
  • Growth: sessions capped at 1 h
  • Growth: Scale size not allowed
  • missing persistent disk, Docker inside
  • machine size not published
Sandboxes for your usersNone for this scenario
  • Scale: max 100 concurrent
  • machine size not published
  • browser-session product, not a general sandbox
  • different kind of product: browser session
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
  • Scale: max 100 concurrent
  • machine size not published
  • browser-session product, not a general sandbox
  • different kind of product: browser session
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 Firecrawl when

your agent mostly needs web data: pages turned into Markdown or JSON, a short browser session for clicks behind a login, a browser agent that you prompt in plain language, or browsing from a chosen country.

  • Web data without starting a browser. Useful when the job is mostly reading pages: scrape, crawl and search return clean Markdown or JSON per call.
  • A browser agent that you prompt in plain language. Useful for clicks and forms without writing selectors: send a prompt, get the agent's answer back, with a live view to watch.
  • Browse from any country. Useful when target sites show different content by country: each session can browse from a country that you pick.

Use boat when

your agent browses and also works: installs, builds, Docker and dev servers on a full VM. The VM keeps its disk between runs and, with auto-stop off on a paid plan, runs as long as the task takes. Coding agents are preinstalled, and an optional paid proxy keeps model keys outside.

  • A whole machine, not only a browser. An agent that reads a page and then must clone a repo, run docker compose or start a dev server needs a second provider next to Firecrawl. It must also move files between the two.
  • Runs as long as the task, keeps its disk. On Firecrawl, a browsing job longer than an hour runs as several sessions, and your code copies downloads to your own storage before each one closes.
  • Coding agents that you drive from outside. With Firecrawl, you use its browser agent as it is, or you run your own harness somewhere else and call Firecrawl from it.

Questions people ask

Is boat cheaper than Firecrawl?

It depends on the scenario. The table sets a 4 vCPU / 8 GB boat VM against a Firecrawl browser session of unpublished size that keeps no files and stops after an hour. boat costs less in three of the five scenarios: one agent that never stops ($26 against $82), 50 agents in parallel ($317 against $399) and sandboxes for your users ($200 against $749). Firecrawl costs less in the other two: a workday agent ($19 against $20) and 1,000 short jobs at once ($942 against $8,800, and neither fits). For light use, Firecrawl's Free plan (1,000 credits a month) costs less than boat's $20 minimum.

Can Firecrawl run Docker or a full shell?

A Firecrawl session runs Python, Node and bash in a container built for its browser, with Playwright and agent-browser preinstalled. Its docs do not mention Docker, SSH or root access. Files end with the session unless your code saves them to your own storage. Every boat sandbox has root via sudo, SSH, Docker, systemd and /dev/kvm.

How long can a Firecrawl browser session run?

10 minutes by default and 1 hour at most. It closes after 5 minutes without activity unless you change that. A closed session cannot be reopened; a new one can load a saved profile with cookies and localStorage. On boat, auto-stop defaults to 1 hour. You can set up to 30 days, or turn it off on a paid plan. A stop keeps the disk for at least 14 days, but not memory or running programs.

Can I use boat and Firecrawl together?

Yes. An agent on a boat sandbox can call Firecrawl's scrape and search APIs or its MCP server, or open an Interact session over CDP. Its code, files and tools stay on the boat VM. That puts web data on Firecrawl and the rest of the work on a machine that keeps its disk.

Is boat SOC 2 compliant?

No, not today. SOC 2 is in progress, and boat has no SOC 2 report yet. Firecrawl lists SOC 2 Type II on every plan. boat publishes live status at status.ascii.dev and its data retention policy at docs.boat.dev/data-retention. Sandboxes run in Germany, Finland and France; snapshot copies can be stored outside the EU.

Start a sandbox

Install the CLI, then run boat new.

7-day free trial

More comparisons