boat vs Collimate
Both put each job in its own small virtual machine. boat gives an agent a whole Linux computer to live on for as long as it needs, while Collimate is built to clone a running machine, memory and all, in milliseconds for RL and tree search.
Where boat stands out
A whole Linux computer, not a runtime
Every boat sandbox gives the agent root, Docker, systemd, SSH and /dev/kvm for emulators and VMs, and you pick a size up to 16 vCPU / 32 GB. Collimate's templates are language runtimes such as Python and Node, and its docs do not document Docker, root or nested virtualization.
Runs as long as the work takes
A boat sandbox has no session cap on paid plans, and its disk is saved every minute and on every stop. On Collimate Pro a sandbox that passes 24 hours is deleted, and the Demo tier stops at 1 hour.
Hosting and agents built in
boat serves HTTPS on up to 50 ports, raw TCP and a dedicated IP, so users can open what the agent builds. Claude Code, Codex and five other agents come preinstalled with a prompt, steer and events API, and real API keys stay outside in a credentials proxy.
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.
- Isolation
- Full VM per sandbox, own kernel
- Regions
- EU: Germany, Finland, France
Collimate
Collimate is a managed sandbox cloud of Firecracker microVMs whose headline feature is live fork: clone a running sandbox into many copies that keep its memory.
You create a sandbox from a template, run commands in it through an exec API, and fork it into copy-on-write children in a few milliseconds, each with the parent's running processes. You drive it over REST, a Python SDK, a hosted MCP server for agents like Claude Code, or the collimate-rl SDK for training rollouts. It is built for RL post-training labs and for agents that try many actions from one prepared state.
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 workload | boat: price rank | Collimate: price rank |
|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month A software agent needs Docker, root and long sessions; Collimate's docs do not document Docker or root, and Pro sessions end at 24 hours. | 3rd cheapest of 367 $634 a month | 271st of 367, with limits $4,531 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB On Collimate Pro a sandbox that passes 24 hours is deleted, so an always-on assistant has to suspend and resume around the cap; on boat it just keeps running. | 5th cheapest of 366 $657 a month | 47th of 366, with limits $4,672 a month |
| Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month Collimate fits short code runs well: new sandboxes fork from a warm template in milliseconds, and code-only templates have no network at all. | 186th of 345, with limits $2,000 a month | 21st cheapest of 345 $213 a month |
| Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month Your users' apps need public URLs; Collimate reaches a port only through a signed, short-lived tunnel, while boat gives HTTPS URLs and a dedicated IP. | Cheapest of 361 $209 a month | 265th of 361, with limits $1,498 a month |
| RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month Collimate is built for rollouts: it forks a mid-episode sandbox with its memory and can pin a seed so a graded run replays the same way. | 34th of 360, with limits $2,000 a month | 102nd of 360, with limits $6,400 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 | 290th of 364, with limits $1,280 a month |
| CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month | 4th cheapest of 377 $86 a month | 73rd cheapest of 377 $614 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB A dev box needs SSH and a machine that lives all month; Collimate has no SSH and reaps sessions at 24 hours on Pro. | 18th cheapest of 368 $26 a month | 287th of 368, with limits $187 a month |
| Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Computer use needs a desktop to see and click; Collimate offers a browser driven over CDP, while boat streams a full desktop at 60fps. | 6th cheapest of 369 $200 a month | 261st of 369, with limits $1,280 a month |
| Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month Both run a browser an agent can drive; Collimate's browser template is reached over CDP from Playwright on your machine. | 7th cheapest of 367 $200 a month | 96th of 367, with limits $640 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 | 30th cheapest of 346 $320 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 | 178th of 338, with limits $6,400 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 | 63rd of 360, with limits $322 a month |
| GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month | Can't run it | Can't run it |
| macOS agents 5 at once · 4 vCPU / 8 GB · 1,000 sessions of 20 min a month | Can't run it | Can't run it |
Key features
| Feature | boat | Collimate |
|---|---|---|
| Isolation | Full VM with its own kernel on dedicated AMD Ryzen 9 9950X hosts | Firecracker microVM with its own kernel |
| Machine size | You pick the size, up to 16 vCPU / 32 GB on the $100 plan and above | Set by the template; sizes are not published and the create call takes no CPU or RAM setting |
| Session length | No cap on paid plans; you set the time to live | 1 hour on Demo, 24 hours on Pro, then the sandbox is deleted; no cap on Enterprise |
| Pause and resume | Disk is saved every minute and on stop; resume boots the machine again | Suspend saves memory and files; resume puts processes back where they were |
| Fork | Fork a running sandbox with its disk | Live fork into many copy-on-write children with the parent's memory, about 2.4 ms at the median |
| Docker, root and systemd | Yes, root with Docker and systemd in every sandbox | Not documented; templates are language runtimes like Python, Node and PyTorch |
| Nested virtualization | Yes, /dev/kvm for Android emulators and VMs | Not documented |
| Shell access | SSH, plus an exec API | Exec API and MCP tools; no SSH |
| Inbound access | HTTPS URLs for up to 50 ports, token-protected by default; raw TCP; dedicated IPv4 or IPv6 | Signed, time-limited tunnel URL to a port the template exposes, Pro only; no public URLs or IP |
| Outbound rules | Open internet, no provider-side allowlist | Host allowlist per sandbox, per command or per fork branch; cloud metadata always blocked |
| Desktop and browser | Linux desktop streamed at 60fps, Chromium and computer use | Browser template driven over CDP; no desktop |
| Coding agents | Seven agents preinstalled, with a prompt, steer and events API to drive agents inside from outside | Hosted MCP server so an outside agent can use sandboxes as tools; no agents inside |
| Keeping API keys out of the sandbox | Credentials proxy for model providers and any HTTPS API; switch a credential to another account when credits run out | Secret store that injects the secret into the sandbox at create time |
| Deterministic replay | No | Pin the guest's random seed so a run can be replayed the same way |
| Compliance and transparency | SOC 2 is in progress. We're open about where we stand: live status and data retention policy. | Not published |
Every feature Battleships tracks (43)
| Feature | boat | Collimate |
|---|---|---|
| Full VM (own kernel) | Yes | Yes |
| Docker inside | Yes | Not published |
| Nested virtualization | Yes | No |
| Root | Yes | Not published |
| Systemd | Yes | Not published |
| Desktop GUI | Yes | No |
| Browser | Yes | Yes |
| Computer-use API | Yes | No |
| SSH | Yes | No |
| Persistent disk | Yes | Yes |
| Pause/resume | Yes | Yes |
| Automatic snapshots | Yes | No |
| Fork/clone | Yes | Yes |
| Live fork (no pause) | Yes | Yes |
| Memory fork | No | Yes |
| Custom image (Docker or snapshot) | Yes | Yes |
| ≥24 h sessions | Yes | Yes |
| Idle auto-stop | No | No |
| Wake on request | No | No |
| HTTPS preview URLs | Yes | No |
| Raw TCP inbound | Yes | Not published |
| Public IPv4 | No | No |
| Egress allowlist | No | Yes |
| Secret proxy | Yes | No |
| Secret proxy for any API | Yes | No |
| Preinstalled agents | Yes | No |
| Agent harness API | Yes | No |
| MCP server | No | Yes |
| GPU | No | No |
| Windows | No | No |
| Arm64 | No | Not published |
| Self-hosting | No | Yes |
| BYOC | No | Yes |
| Open source | No | No |
| Scoped API keys | Yes | Not published |
| Spend limits | Yes | Not published |
| Usage per sandbox (API) | Yes | No |
| Audit logs | No | Not published |
| SSO | No | Yes |
| SOC 2 | No | Not published |
| HIPAA | No | Not published |
| EU data residency | Yes | Not published |
| Zero data retention | Yes | Not published |
Performance
There is no comparable measurement for Collimate yet: it has not been run on the hpc-sandbox-benchmarks CPU test. boat did 26.69 runs per second on that test at 4 vCPU / 8 GB. Collimate publishes a fork time of about 2.4 ms at the median, which measures how fast it clones, not CPU speed.
Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Collimate 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.
| Scenario | boat, a month | Collimate, 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 | $45 · Pro |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $26 · $20/mo | $187 · Pro · 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 | $2,253 · Pro · 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 | $640 · Pro · 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 | $1,280 · Pro · with limits |
boat: Plans start at $20 a month, and the whole fee is usage credit. You pay per second while a sandbox runs, and a stopped sandbox costs nothing and keeps its disk.
Collimate: Collimate Pro bills per second for the vCPU and RAM a sandbox holds while it is live, plus a running-disk charge whose rate is not published, and charges per GB-month for suspended snapshots. Spawning and forking are free; the pricing page says no platform fee, while the docs mention a monthly base on Pro.
Limits in those scenarios
Collimate Pro deletes a sandbox after 24 hours, so the always-on scenario does not fit without suspending before the cap. Docker inside is not documented, which is why the 50-agent fleet is marked unconfirmed, and its pricing page lists up to 100 sandboxes at once on Pro, which the user-sandbox and burst scenarios go past. boat has no session cap, but at 1,000 short jobs at once its $2,000 plan queues starts.
| Scenario | boat | Collimate |
|---|---|---|
| One coding agent, workdays | None that bite |
|
| One agent that never stops | None that bite |
|
| 50 agents in parallel | None that bite |
|
| Sandboxes for your users | None that bite |
|
| 1,000 short jobs at once |
|
|
Which one should you use?
Use Collimate when
you run RL rollouts or tree search that branch many times from one running state and need each branch to keep its memory.
- Live fork with memory in milliseconds. For tree search and best-of-N sampling, it clones a running sandbox with its live memory in a few milliseconds.
- Built for RL rollouts. RL teams get per-branch egress rules, seeded replay of graded runs, and an SDK made for rollout groups.
- Dedicated nodes and your own cloud. Enterprise customers can get dedicated nodes or run it inside their own cloud account.
Use boat when
your agents need a real computer to work on for hours or days: Docker, root, SSH, a browser desktop, public URLs and coding agents built in.
- A whole Linux computer, not a runtime. A coding agent that has to start a database in Docker, run a full test stack or boot an Android emulator needs those pieces; on Collimate you would have to bake them into your own image and hope they work.
- Runs as long as the work takes. An agent that works for days on Collimate Pro must be suspended and resumed by your code before each 24-hour cap, or it loses its machine and everything not copied out.
- Hosting and agents built in. Collimate reaches a port only through a signed tunnel from your own client and injects secrets into the sandbox, so you build the preview hosting and the agent harness yourself, and the keys sit where the agent can read them.
Questions people ask
Is boat cheaper than Collimate?
In four of the five scenarios on this page, yes: $20 against $45 for one agent on workdays, $26 against $187 for one that never stops, and $200 against $640 for sandboxes for your users. For 1,000 short jobs at once neither fits as listed, and Collimate's estimate is lower at $1,280 against boat's $2,000.
Can boat fork a sandbox with its memory like Collimate?
No. boat forks a running sandbox with its disk, so files carry over but processes start fresh. Collimate's live fork copies the running memory too, in a few milliseconds.
How long can a Collimate sandbox run?
Up to 1 hour on Demo and 24 hours on Pro, after which the sandbox is deleted; Enterprise has no cap. You can suspend before the cap and resume later. boat has no session cap on paid plans.
Can I run Docker in a Collimate sandbox?
Collimate's docs do not say. Its templates are language runtimes, and on Pro you can bring your own container image. boat gives root, Docker and systemd in every sandbox.
Can Claude Code use either one?
Both work with it in different ways. Collimate's hosted MCP server lets Claude Code on your machine create and fork sandboxes as tools, while boat runs Claude Code and six other agents inside the sandbox and lets you drive them from outside with an API.
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.
Start a sandbox
Install the CLI, then run boat new.
7-day free trial