boat vs PandaStack

Both give each agent its own VM with its own kernel. boat gives it a whole workstation, with a desktop, SSH, nested virtualization and agents you drive by API. PandaStack is an open-source microVM platform built around fast boots, memory forks, active-CPU billing and hosting apps and Postgres.

Where boat stands out

A desktop your agent can see and use

Every boat sandbox can run a Linux desktop and Chrome streamed at 60fps, with computer-use tools built in. You can embed the stream in your app so a person can watch or take over. PandaStack ships a headless Chromium template. For a desktop there, you set it up: build a custom Dockerfile template with a desktop and a VNC stream.

Agents you drive from outside, keys kept out

Claude Code, Codex and five other agents come preinstalled, and you prompt, steer and follow them through one API. With the optional credentials proxy on ($1 a day plus $1 per 1,000 requests), the real keys of supported model providers and custom HTTPS APIs stay outside the sandbox. A credential can switch to another account when one runs out of credits.

A whole workstation, sized for the job

You pick the machine size, from 2 vCPU / 4 GB up to 16 vCPU / 32 GB, and get nested virtualization for Android emulators and VMs, standard SSH and HTTPS URLs that need a token by default. PandaStack's docs give every sandbox 8 burstable vCPUs, and by default its preview URL itself is the credential.

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.

PandaStack

PandaStack calls itself the cloud for AI agents: open-source Firecracker microVMs for sandboxes, apps, Postgres and functions.

Every PandaStack sandbox, app and database runs in its own Firecracker microVM with its own Linux 5.10 kernel, booted from a template snapshot. Templates have 8 burstable vCPUs, and RAM is set per template. You can build your own template from a Dockerfile. It bills CPU by active seconds and RAM by resident working set. It is built for teams that want agent sandboxes, git-driven app hosting and managed Postgres on one platform, and its core is Apache-2.0 so you can run it yourself.

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 rankPandaStack: price rank
Software factory
100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month
A software factory needs Docker, long builds and sometimes emulators. Both run Docker on their own kernel; boat adds nested virtualization, SSH and machines up to 16 vCPU / 32 GB.
5th cheapest of 365
$634 a month
122nd of 365, with limits
$832 a month
Always-on AI employees
50 machines 24/7 · 2 vCPU / 4 GB
An assistant that lives on its machine all month fits both. PandaStack hibernates a persistent sandbox after about 5 idle minutes and wakes it on the next request. boat keeps it running once you turn auto-stop off (default 1 h), and bills full rate while it runs.
6th cheapest of 361
$657 a month
62nd of 361, with limits
$1,778 a month
Code interpreter bursts
100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month
PandaStack suits short code runs: templates boot from a snapshot, it bills only active CPU seconds, and idle sandboxes hibernate.
189th of 336, with limits
$2,000 a month
145th of 336, with limits
$341 a month
Sandboxes for your users
300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month
PandaStack has no vCPU size choice (every sandbox has 8 burstable vCPUs), but it bills only the CPU and RAM a sandbox actually uses. boat offers fixed sizes from 2 vCPU / 4 GB and token-protected URLs.
3rd cheapest of 349
$209 a month
84th of 349, with limits
$701 a month
RL rollouts
1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month
PandaStack's fork-tree copies a running sandbox with its memory into up to 16 children at once, which suits rollouts that branch from one state.
39th of 349, with limits
$2,000 a month
188th of 349, with limits
$21,864 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
59th cheapest of 359
$814 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
71st cheapest of 374
$517 a month
24/7 dev boxes
1 machine 24/7 · 4 vCPU / 8 GB
A dev box wants real SSH for VS Code Remote and scp. PandaStack gives a terminal over WebSocket from its CLI, and boat gives standard SSH.
19th cheapest of 371
$26 a month
234th of 371, with limits
$40 a month
Computer-use agents
50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month
PandaStack ships a headless Chromium template. For a desktop, you set it up: build a custom template from a Dockerfile with a desktop and a VNC stream; PandaStack documents no example. boat includes a Linux desktop streamed at 60fps with computer-use tools.
7th cheapest of 367
$200 a month
137th of 367, with limits
$165 a month
Browser-use agents
50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month
Both run Chromium with Playwright. boat can also show the live browser on a desktop so a human can watch or take over.
7th cheapest of 356
$200 a month
62nd of 356, with limits
$158 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
77th cheapest of 343
$962 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
281st of 327, with limits
$19,236 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
75th of 351, with limits
$641 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

FeatureboatPandaStack
IsolationFull VM with its own kernel, on AMD Ryzen 7 9700X and Ryzen 9 9950X hosts dedicated to boat; vCPUs are sharedFirecracker microVM with its own Linux 5.10 kernel
Machine sizeYou pick 2, 4, 8 or 16 vCPU (4 GB RAM per 2 vCPU); 16 vCPU / 32 GB needs the $100 plan or higher. Disk is fixed per size8 burstable vCPUs per sandbox in the docs (its pricing API lists up to 16 on Team and 64 on Enterprise); RAM is set per template, from 1 GB
Desktop and computer useLinux desktop and Chrome streamed at 1920x1080, 60fps, with computer-use toolsHeadless Chromium template with Playwright. You set up a desktop yourself: a custom Dockerfile template with a desktop and VNC; no documented example
Nested virtualizationYes, /dev/kvm in every sandbox for Android emulators and VMsNot documented for the guest; KVM is a host requirement
Shell accessStandard SSH, so ssh, scp and VS Code Remote workInteractive terminal over WebSocket from the CLI; SSH with your own key is not documented
Coding agentsSeven agents preinstalled, with a prompt, steer and events APIAgent template with seven coding agent CLIs, run through exec; you build the harness. A claude-agent template runs Claude Managed Agents
Keeping API keys out of the sandboxOptional paid credentials proxy ($1 a day plus $1 per 1,000 requests) for model providers and custom HTTPS APIs; switch a credential to another account when credits run out. Other secrets still enter the sandboxKeys go into the sandbox as environment variables or files
Inbound accessHTTPS URLs for up to 50 ports, token-protected by default; raw ports over IPv6 (no own IPv4 on some sandboxes)HTTPS URL for up to 32 ports; by default the URL itself is the credential (its changelog mentions preview access controls)
Pause and resumeDisk is saved every minute and at each stop; memory and running programs are not saved, so a resume is like a rebootPause keeps memory; hibernate saves memory and disk, and running processes come back
ForkFork a running sandbox with its disk, from the last 1-minute snapshot; the copy boots freshFork with disk, or fork-tree with memory into 1 to 16 children while the parent keeps running
Wake on requestNo; you start a stopped sandbox by APIYes; a hibernated persistent sandbox wakes on the next API request
Start limits$20 plan: 12 / 60 / 200 starts per minute / hour / day included; extra starts cost $0.03 to $0.07 and fail above 6 times the included number (3 times per minute)Creates per hour: Free 60, Pro 600, Team 6,000; above that, a create fails with 429
Outbound rulesOpen internet, no provider-side allowlistOpen internet, no outbound allowlist
RegionsMachines in Germany, Finland and France; you cannot pick a region, and snapshot copies can be stored outside the EU. Optional zero data retentionNo region choice documented; self-host for data residency
ComplianceSOC 2 in progress; no SOC 2 report todayNo SOC 2, ISO 27001 or HIPAA attestation yet
Self-hostingNo; hosted service onlyApache-2.0 core, runs on your own cloud or hardware
Compliance and transparencySOC 2 is in progress. We're open about where we stand: live status and data retention policy.No SOC 2 report published
Every feature Battleships tracks (43)
FeatureboatPandaStack
Full VM (own kernel)YesYes
Docker insideYesYes
Nested virtualizationYesNo documented way
RootYesYes
SystemdYesYes
Desktop GUIYesNo documented way
BrowserYesYes
Computer-use APIYesNo documented way
SSHYesNo documented way
Persistent diskYesYes
Pause/resumeYesYes
Automatic snapshotsYesYes
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 inboundYesNot published
Public IPv4No documented wayNo documented way
Egress allowlistNo documented wayNo documented way
Secret proxyYesNo documented way
Secret proxy for any APIYesNo documented way
Preinstalled agentsYesYes
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 wayYes
BYOCNo documented wayYes
Open sourceNo documented wayYes
Scoped API keysYesNot published
Spend limitsYesYes
Usage per sandbox (API)YesNo documented way
Audit logsNo documented wayNot published
SSONo documented wayYes
SOC 2No documented wayNo documented way
HIPAANo documented wayNo documented way
EU data residencyNo documented wayNo documented way
Zero data retentionYesNot published

Performance

boat28.06 runs/s
PandaStackNot measured

PandaStack is not in the public benchmarks we use, so we cannot say which is faster. In StarSling's run of 3 October 2026 (run 37073680030), boat did 28.06 runs/s on the V8 Web Tooling Benchmark at 4 vCPU / 8 GB, second of 15. 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. PandaStack publishes its own boot times (about 49 ms at p50), which measure start time, not CPU speed.

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. PandaStack 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 monthPandaStack, 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$40 · Pro
One agent that never stops
4 vCPU / 8 GB, 50 GB disk, running 24/7 all month
$26 · $20/mo$144 · Pro
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$1,789 · Pro
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$801 · Team · 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$4,244 · Team · with limits

boat: Plans cost $20, $100, $500 or $2,000 a month, and the whole plan price comes back as VM time for that month. You pay per second for the size you pick while a sandbox runs, busy or idle: $0.036 an hour for 4 vCPU / 8 GB. A stopped sandbox has no compute charge, and boat keeps its disk for at least 14 days. Saved snapshots above the included count and the credentials proxy still charge while sandboxes are stopped. The $20 plan includes 12 starts a minute and 200 a day. Create, fork and resume each count as a start. Extra starts cost $0.03 to $0.07 each, up to 6 times the included amount (3 times per minute), and fail above that. The credentials proxy costs $1 a day plus $1 per 1,000 requests.

PandaStack: PandaStack bills CPU only for the active vCPU seconds a sandbox burns and RAM for its resident working set, at $0.054 per active vCPU-hour and $0.0162 per GiB-hour. Free is $0 with 5 sandboxes at once that end after 1 hour. Pro is $20 and Team $99 a month, each fee returned as usage credit. Volume space above the plan quota costs $0.15 per GiB-month. The scenarios price PandaStack at its 8 vCPU shape. Our model applies the same busy share to 8 vCPUs as to boat's 4, so it can overstate PandaStack's CPU cost by up to 2x.

Limits in those scenarios

PandaStack has no vCPU size choice. Every template has 8 burstable vCPUs, and RAM is set per template, from 1 GB up. It bills only active CPU and resident RAM, so idle vCPUs cost nothing. For 1,000 short jobs at once, Team stops at 500 sandboxes at a time. boat's largest plan has the slots, but it charges for starts above 90 a minute and refuses starts above 270. Neither fits that burst without talking to sales. On boat, auto-stop defaults to 1 hour; with auto-stop off on a paid plan there is no session cap. A boat resume is like a reboot: memory, running programs and /tmp are not saved, and a machine failure loses what was written after the last 1-minute snapshot. boat has no GPU, and you cannot boot your own Docker image. PandaStack's persistent sandboxes also run as long as you want, though they hibernate after about 5 idle minutes; non-persistent ones end at their TTL (default 1 hour).

ScenarioboatPandaStack
One coding agent, workdaysNone for this scenarioNone for this scenario
One agent that never stopsNone for this scenarioNone for this scenario
50 agents in parallelNone for this scenarioNone for this scenario
Sandboxes for your usersNone for this scenario
  • smallest size offered is 8 vCPU / 4 GiB
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
  • Team: max 500 concurrent
  • needs talk to sales: above plan "Team" limits: contact sales for more
  • smallest size offered is 8 vCPU / 4 GiB
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 PandaStack when

you need to self-host an open-source sandbox platform, fork running sandboxes with their memory, pay only for the CPU and RAM a mostly idle sandbox uses, or keep agents, apps and Postgres on one platform.

  • Memory forks, hibernation and wake on request. Good for jobs that branch from one running state, or sleep between requests and must come back with processes alive.
  • Billing for the CPU and RAM you use. A sandbox that is mostly idle costs little, because idle vCPUs and unused RAM are not billed.
  • Open source you can run yourself. Its Apache-2.0 core runs on your own cloud, on-prem hardware or a laptop, for teams that must keep everything in house.
  • Apps and Postgres on the same platform. Useful when the agent's sandbox, the app it ships and its database should live on one platform with one bill.

Use boat when

your agents need a whole machine: a desktop and browser they can see, Docker and emulators, SSH, a size that fits the job, and coding agents you drive by API, with an optional proxy that keeps model keys out of the sandbox.

  • A desktop your agent can see and use. Agents that click through real apps, or hit a login a human must solve, need a screen. On PandaStack you build and maintain that desktop and stream image yourself, and its docs give no example.
  • Agents you drive from outside, keys kept out. PandaStack's agent template installs the CLIs, but you build the harness on exec yourself, and keys sit inside a machine the agent controls, where one prompt injection can leak them.
  • A whole workstation, sized for the job. Without /dev/kvm an agent cannot boot an emulator or a VM. A preview URL that works as a credential and leaks in a log opens that port to anyone who reads it.

Questions people ask

Is boat cheaper than PandaStack?

In our model, boat costs less in four of the five scenarios. A fleet of 50 agents runs $317 a month on boat against $1,789 on PandaStack, which bills active CPU and resident RAM per second. For 1,000 short jobs at once, neither fits on a self-serve plan. A nearly idle sandbox, or under about $20 of use a month on PandaStack's Free plan, can cost less on PandaStack.

Does PandaStack have a desktop for computer use?

Not as a template. Its browser template is headless Chromium with Playwright and crawl4ai. You can set up a desktop yourself in a custom Dockerfile template with a desktop and VNC; PandaStack documents no example. boat gives each sandbox a Linux desktop streamed at 60fps with computer-use tools built in.

Can I SSH into a PandaStack sandbox?

PandaStack gives you an interactive terminal over WebSocket through its CLI. Its docs show no SSH with your own key, so ssh, scp and VS Code Remote-SSH do not connect directly. boat sandboxes take standard SSH.

Can boat fork a sandbox with its memory like PandaStack?

No. boat forks a running sandbox with its disk, from the last 1-minute snapshot, and the copy boots fresh. PandaStack's fork-tree copies memory too, so up to 16 children start with the parent's processes already running.

Can I self-host PandaStack or boat?

PandaStack's core is Apache-2.0 and runs on your own KVM-capable Linux hosts, in your cloud or on-prem. boat is a hosted service only, on machines in Germany, Finland and France, with optional zero data retention.

Is boat SOC 2 compliant?

No SOC 2 report today; SOC 2 is in progress. Live status is at status.ascii.dev and our data retention policy at docs.boat.dev/data-retention.

Start a sandbox

Install the CLI, then run boat new.

7-day free trial

More comparisons