boat vs Microsoft Foundry

Foundry hosted agents run your own agent code in a managed per-session sandbox on Azure, next to its models, identity and Teams. boat gives each agent a full Linux VM with root, Docker, a disk saved every minute, and seven coding agents installed.

Where boat stands out

A whole machine you can log into

A boat sandbox is a full VM with root, systemd, Docker with Compose and /dev/kvm, and you can SSH into it. A Foundry session runs your image in a microVM, and Microsoft documents no shell access, Docker or nested virtualization inside it.

Room to work, and a disk that keeps it

boat machines go up to 16 vCPU / 32 GB, and boat saves the whole disk every minute and at each stop, so installed packages come back after a resume. A resume is like a reboot: memory and /tmp are not saved, and a machine failure loses writes after the last snapshot. You can also fork a running sandbox to try two ideas at once.

Coding agents ready to drive from outside

Claude Code, Codex and five other agents come installed, 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), real keys for model providers and other HTTPS APIs stay outside the sandbox.

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.

Microsoft Foundry

Microsoft Foundry Agent Service is Azure's managed platform for building and running AI agents; hosted agents are the part that runs your own agent code.

You package your agent as a container image, push it to Azure Container Registry, and Foundry runs it in a per-session VM-isolated sandbox behind one agent endpoint (Responses, Invocations, WebSocket or A2A). The sandbox scales to zero after an idle timeout and restores $HOME and /files on the next request. A separate key-value state store survives restarts and idle eviction. For prompt agents, Foundry also runs a managed Copilot harness (preview) with shell, file, web and subagent tools. It is a hyperscaler agent platform for teams on Azure who want their agent wired to Foundry models, Entra ID, private networks and Microsoft 365.

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 rankMicrosoft Foundry: price rank
Software factory
100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month
A software factory needs a 4 vCPU / 8 GB machine, Docker and a disk that keeps installed packages. A Foundry session tops out at 2 vCPU / 4 GiB, Microsoft does not document Docker inside it, and only $HOME and /files are documented as persisted. On Azure, Container Apps Sandboxes offer bigger machines.
5th cheapest of 365
$634 a month
313th of 365, with limits
$5,932 a month
Always-on AI employees
50 machines 24/7 · 2 vCPU / 4 GB
An assistant that lives in Teams fits Foundry's channel bridge well. One that works on its own machine all month needs its installed tools to persist: boat saves the whole disk every minute, and Foundry documents $HOME, /files and a key-value state store.
6th cheapest of 361
$657 a month
143rd of 361, with limits
$7,722 a month
Code interpreter bursts
100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month
Short code bursts suit Foundry: compute is released after an idle timeout you set from 2 to 240 minutes, and you pay for active sessions only. A running boat sandbox bills at full rate, busy or idle, until it stops.
189th of 336, with limits
$2,000 a month
271st of 336, with limits
$403 a month
Sandboxes for your users
300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month
Both give each user an isolated VM. In the Sandboxes for your users scenario, boat costs $200 a month and Foundry $1,107. boat adds SSH and public URLs; Foundry wakes a session on its next request.
3rd cheapest of 349
$209 a month
291st of 349, with limits
$2,532 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
166th of 349, with limits
$12,232 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
303rd of 359, with limits
$1,968 a month
CI runners
40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month
CI runners that build containers need Docker inside the sandbox. Microsoft does not document Docker inside a hosted agent session, and a session tops out at 2 vCPU / 4 GiB.
2nd cheapest of 374
$86 a month
327th of 374, with limits
$1,181 a month
24/7 dev boxes
1 machine 24/7 · 4 vCPU / 8 GB
A dev box needs SSH, root and a disk that keeps what you install. Foundry documents no SSH into a session and persists only $HOME and /files.
19th cheapest of 371
$26 a month
336th of 371, with limits
$309 a month
Computer-use agents
50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month
boat has a 1920x1080, 60 fps desktop with Chrome inside the sandbox. Foundry's computer use tool is in preview and acts on a machine you supply; its browser automation tool (GA) gives a managed browser with Live View.
7th cheapest of 367
$200 a month
313th of 367, with limits
$1,771 a month
Browser-use agents
50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month
Foundry has a managed browser automation tool (GA) with Live View and user take-over for sign-in. It runs outside the session in a separately billed Playwright Workspace. boat runs Chrome inside the agent's own machine, so the agent and its browser share files and sessions.
7th cheapest of 356
$200 a month
31st cheapest of 356
$886 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
68th cheapest of 343
$600 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
249th of 327, with limits
$11,993 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
14th cheapest of 351
$584 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

FeatureboatMicrosoft Foundry
Kind of productAgent sandbox: a full Linux VM per agent, driven by API, SSH or the built-in agentsManaged agent platform on Azure: your agent's container runs behind a Foundry endpoint
IsolationFull VM with its own kernel on hosts dedicated to boat (AMD Ryzen 7 9700X and Ryzen 9 9950X); vCPUs are sharedPer-session VM-isolated sandbox running your container image
Biggest machineUp to 16 vCPU / 32 GB with 251 GB of disk; 16 vCPU needs the $100 plan or higher2 vCPU / 4 GiB per session, with a disk budget of up to 20 GiB
Docker and nested virtualizationDocker with BuildKit and Compose, systemd and root; /dev/kvm for Android emulators and VMsDocker, root and nested virtualization inside a session are not documented; you get a 2 vCPU / 4 GiB microVM running your image
Shell accessSSH and a root shell into every sandboxNo SSH documented; you reach your code through the agent endpoint
What persistsThe whole disk, saved every minute and at each stop; memory and /tmp are not saved. A stopped sandbox's disk is kept for at least 14 days, or longer as a saved snapshot$HOME and /files, within a 20 GiB disk budget, plus a key-value state store; sessions deleted after 30 days idle
Scale to zero and wakeAuto-stop after a set time (1 hour by default, not idle-based); you resume by API; no wake on requestCompute is released after an idle timeout of 2 to 240 minutes and comes back on the session's next request
ForkFork a running sandbox from its last 1-minute snapshotNo fork documented; versions snapshot configuration, not sessions
Inbound accessUp to 50 public HTTPS URLs per sandbox, token-gated by default; public IPv6 and SSH over IPv4One Entra ID protected agent endpoint over HTTPS or WebSocket
Outbound controlOpen internet, no provider-side allowlistYour own Azure virtual network, with Azure Firewall and network security group rules
Browser and desktop1920x1080, 60 fps Linux desktop with Chrome inside the sandbox, with computer tools for agentsManaged browser automation tool (GA) with Live View and user take-over, running outside the sandbox in a separately billed Playwright Workspace; computer use tool in preview, on a machine you supply
Coding agentsSeven agents installed, including Claude Code and Codex, with a prompt, steer and events APIManaged Copilot harness (preview) with shell, files, web and subagents, which Foundry runs for you; or your own harness in Python or C# as a hosted agent (samples wrap the Claude Agent SDK and GitHub Copilot SDK)
Keeping API keys out of the sandboxOptional credentials proxy ($1 a day plus $1 per 1,000 requests) for model providers and other HTTPS APIs; with it on, those real keys stay outside the sandboxAgent identity for Azure services; other secrets come from Key Vault and are read by your code inside the sandbox
RegionsGermany, Finland and France; you cannot pick a region, and there is no US or Asia region31 Azure regions worldwide, including the EU
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)
FeatureboatMicrosoft Foundry
Full VM (own kernel)YesYes
Docker insideYesNot published
Nested virtualizationYesNot published
RootYesNot published
SystemdYesNot published
Desktop GUIYesNo documented way
BrowserYesYes
Computer-use APIYesNo documented way
SSHYesNo documented way
Persistent diskYesYes
Pause/resumeYesYes
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)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 wayYes
Secret proxyYesNo documented way
Secret proxy for any APIYesNo documented way
Preinstalled agentsYesYes
Agent harness APIYesWith your own setup
MCP serverNo documented wayNo documented way
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 wayNot published
EU data residencyNo documented wayYes
Zero data retentionYesNot published

Performance

boat28.06 runs/s
Microsoft FoundryNot measured

Foundry hosted agents are 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, rank 2 of 15, shared with four other providers. That boat sandbox ran on an AMD Ryzen 9 9950X host; boat also places sandboxes on Ryzen 7 9700X hosts and does not promise a CPU model. A Foundry session tops out at 2 vCPU / 4 GiB, so it cannot run the test at that size.

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. Microsoft Foundry 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 monthMicrosoft Foundry, 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$78 · 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$323 · 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$3,897 · 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$1,107 · 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$2,446 · Pay as you go

boat: Plans cost $20, $100, $500 or $2,000 a month, and the whole plan price comes back as VM time for that month. A 4 vCPU / 8 GB sandbox costs $0.036 an hour, billed per second while it runs, busy or idle. A stopped sandbox has no compute charge; saved snapshots above the included count and the credentials proxy still charge. Starts above the plan's included number cost $0.03 to $0.07 each.

Microsoft Foundry: Foundry bills the vCPU-hours and GiB-hours allocated to every active session, pay as you go: $0.0994 per vCPU-hour and $0.0118 per GiB-hour in East US. A session stays active until the idle timeout passes after its last request; you can set that timeout as low as 2 minutes. Model calls, tools and browser automation are billed separately.

Limits in those scenarios

Three scenarios need a 4 vCPU / 8 GB machine, which one Foundry hosted session cannot give: its largest size is 2 vCPU / 4 GiB. Only $HOME and /files are documented as persisted, so packages installed elsewhere may not survive the idle timeout. The agent fleet also needs Docker inside, which Foundry does not document. For 1,000 short jobs at once, Foundry's default quota of 1,000 concurrent sessions per region absorbs the burst. boat's $2,000 plan includes 90 starts a minute, charges $0.03 to $0.07 for each extra start, and refuses starts above 270 a minute with 429; boat never queues a start. A stopped boat sandbox keeps its disk for at least 14 days unless you save a snapshot.

ScenarioboatMicrosoft Foundry
One coding agent, workdaysNone for this scenario
  • capped at large (2 vCPU / 4 GiB)
  • files survive a stop only in $HOME and /files: packages installed elsewhere reset
  • largest size is large (2 vCPU / 4 GiB)
  • disk beyond 20 GiB: price unknown
One agent that never stopsNone for this scenario
  • capped at large (2 vCPU / 4 GiB)
  • files survive a stop only in $HOME and /files: packages installed elsewhere reset
  • largest size is large (2 vCPU / 4 GiB)
  • disk beyond 20 GiB: price unknown
50 agents in parallelNone for this scenario
  • capped at large (2 vCPU / 4 GiB)
  • files survive a stop only in $HOME and /files: packages installed elsewhere reset
  • largest size is large (2 vCPU / 4 GiB)
  • egress price not published
Sandboxes for your usersNone for this scenario
  • egress price not published
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
  • egress price not published
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 Microsoft Foundry when

your agent is a chat or webhook service on Azure that talks to Foundry models, ships to Teams, must stay inside your Azure network, or must run outside the EU. For a bigger machine with snapshots on Azure, compare Azure Container Apps Sandboxes, which run up to 16 vCPU / 32 GiB (see boat vs Azure Container Apps Sandboxes).

  • Built into Azure and Microsoft 365. For teams already on Azure: each agent gets its own Entra ID identity, Foundry models, private networking and a bridge to Teams.
  • Scale to zero with outbound rules. Useful for chat-style agents that sit idle between turns and must only reach approved hosts.
  • Regions around the world. Hosted agents run in 31 Azure regions, from the US and Brazil to Japan, India, South Africa and Australia.

Use boat when

your agents need a full machine to do the work: installs, builds, Docker, emulators and a browser on up to 16 vCPU, with the disk saved between runs and coding agents ready to drive.

  • A whole machine you can log into. An agent that builds images, runs docker compose for tests or boots an Android emulator has a documented path on boat. On Foundry you reach your code only through the agent endpoint.
  • Room to work, and a disk that keeps it. A Foundry session tops out at 2 vCPU / 4 GiB, and only $HOME and /files are documented as persisted, so tools installed elsewhere can need a reinstall after the idle timeout.
  • Coding agents ready to drive from outside. On Foundry, Claude Code or Codex means writing, packaging and deploying the harness yourself; the managed Copilot harness is in preview, and its CPU, memory and lifecycle are not configurable.

Questions people ask

Is boat cheaper than Microsoft Foundry?

boat's monthly bill is lower in four of the five scenarios on this page, and Foundry fits two of them. For sandboxes for your users at 2 vCPU / 4 GB, boat costs $200 a month and Foundry $1,107. For 1,000 short jobs at once, Foundry costs $2,446 a month and boat $8,800, and boat's plans refuse starts above 270 a minute, so Foundry fits that one better. The other three need a 4 vCPU / 8 GB machine, which a Foundry hosted session cannot provide.

Can I run Docker inside a Foundry hosted agent?

Microsoft does not document Docker inside a hosted agent session, and a session tops out at 2 vCPU / 4 GiB. boat gives each sandbox Docker with BuildKit and Compose, systemd, root and /dev/kvm.

What happens to files when a Foundry session goes idle?

After the idle timeout, 15 minutes by default, Foundry releases the compute and restores $HOME and /files on the next request. Microsoft documents only those paths as persisted, plus a separate key-value state store. A session is deleted after 30 days of inactivity. boat saves the whole disk every minute and at each stop; memory, running programs and /tmp are not saved.

Can I run Claude Code on Microsoft Foundry?

Yes, by writing a harness around the Claude Agent SDK, packaging it as a container and deploying it as a hosted agent; Microsoft publishes samples for this. Foundry also runs a managed Copilot harness (preview) for prompt agents. boat has Claude Code, Codex and five other agents installed, driven through one prompt, steer and events API.

Is boat SOC 2 compliant?

SOC 2 is in progress, so boat has no SOC 2 report today. Foundry is covered by Azure's compliance program, including SOC 2. Live 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