Beam vs Agent 37
Beam is a serverless GPU platform with gVisor sandboxes you start from code and a separate line of CPU VMs. Agent 37 hosts one agent per user, in a gVisor instance by default, that can sleep when idle, keep its memory and wake on the next message.
The companies and products
Beam
Beam is a serverless cloud for GPU inference, sandboxes, VMs and background jobs, built on its open-source runtime Beta9.
Beam runs functions, REST endpoints, task queues, pods, sandboxes and VMs on CPUs and GPUs, from a Python or TypeScript SDK. A sandbox is a container isolated with gVisor by default. It lives until you terminate it or its time to live runs out, and you can save it as a filesystem image or a memory snapshot. Beam VMs are a separate CPU product: a microVM with its own kernel, systemd, a root disk that stays when it stops, and a browser desktop. A VM can pause with its memory, fork and take disk snapshots. CPU work runs on Spot, Shared or Performance tiers at different prices.
Agent 37
Agent 37 hosts AI agents, one isolated, always-on sandbox per user, behind one API.
You create an instance from a template (Hermes, OpenClaw, Claude Code, Codex, OpenCode, Pi, Grok or your own Docker image) and talk to it through a chat API, files, exec, SSH and HTTPS URLs. The default instance is a container enforced by gVisor. A VM type gives a full Linux machine with its own kernel for Docker, Kubernetes and Android emulators. It is built for companies that give each of their customers an agent.
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. “7th cheapest of 365” means 6 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 workload | Beam: price rank | Agent 37: price rank | boat: price rank |
|---|---|---|---|
| Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month Beam sandboxes run Docker with docker_enabled but have no systemd, while Beam VMs have both. Agent 37 needs its VM type for Docker, and its entrypoint is PID 1 with no init system. boat gives every sandbox Docker, systemd and /dev/kvm. | 7th cheapest of 365 $1,200 a month | 4th cheapest of 365 $551 a month | 5th cheapest of 365 $634 a month |
| Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB Agent 37 ranks 1st here: one assistant per user that sleeps when idle, keeps its memory and wakes on the next message. boat ranks 5th and Beam 12th; a Beam VM keeps its disk and can pause with its memory. | 12th cheapest of 361 $1,495 a month | Cheapest of 361 $279 a month | 6th cheapest of 361 $657 a month |
| Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month Agent 37 ranks 1st and Beam 17th for short code runs. boat ranks 130th, because its start limits make big bursts a compromise. | 21st cheapest of 336 $161 a month | Cheapest of 336 $22 a month | 189th of 336, with limits $2,000 a month |
| Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month Agent 37 ranks 1st, boat 2nd and Beam 12th. Agent 37 allows up to 1,000 awake instances after $1,000 of top-ups, and sleeping or stopped ones do not count. boat plans go up to 2,000 at once, with starts per minute, hour and day capped by plan. | 14th cheapest of 349 $567 a month | Cheapest of 349 $79 a month | 3rd cheapest of 349 $209 a month |
| RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month | 332nd of 349, with limits $984 a month | 9th of 349, with limits $363 a month | 39th of 349, with limits $2,000 a month |
| Agent evals 250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month | 16th cheapest of 359 $258 a month | Cheapest of 359 $58 a month | 10th cheapest of 359 $200 a month |
| CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month boat ranks 2nd for CI and Beam 23rd. Builds that need /dev/kvm, such as Android emulators, work on Agent 37's VM type and in every boat sandbox; Beam documents no /dev/kvm. | 29th cheapest of 374 $300 a month | 227th of 374, with limits $96 a month | 2nd cheapest of 374 $86 a month |
| 24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB | 44th cheapest of 371 $55 a month | 3rd cheapest of 371 $12 a month | 19th cheapest of 371 $26 a month |
| Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Agent 37 ranks 1st, boat 8th and Beam 14th. Beam VMs have a browser desktop with an automation SDK, Agent 37 adds a noVNC desktop through a template, and boat streams a 60 fps desktop with Chrome in every sandbox. | 12th cheapest of 367 $431 a month | 3rd cheapest of 367 $120 a month | 7th cheapest of 367 $200 a month |
| Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month | 73rd of 356, with limits $272 a month | 47th of 356, with limits $29 a month | 7th cheapest of 356 $200 a month |
| Scraping & bots (startup scale) 50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month | 16th cheapest of 343 $196 a month | Cheapest of 343 $32 a month | 6th cheapest of 343 $90 a month |
| Mass scraping & bots (undetectable) 1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month | 44th of 327, with limits $967 a month | 39th of 327, with limits $648 a month | 73rd of 327, with limits $1,800 a month |
| Parallel red teaming 200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month | 6th cheapest of 351 $219 a month | Cheapest of 351 $43 a month | 3rd cheapest of 351 $200 a month |
| GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month Beam ranks 1st here, with GPUs up to H100, H200 and B200. Agent 37 and boat have no GPUs. | Cheapest of 102 $547 a month | Not ranked: misses a requirement | Not 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 requirement | Not ranked: misses a requirement | Not ranked: misses a requirement |
Key features
| Feature | Beam | Agent 37 | boat |
|---|---|---|---|
| Isolation | gVisor container by default for sandboxes, with an opt-in VM-backed mode; Beam VMs are CPU microVMs with their own kernel | gVisor by default; full VM with its own kernel as an opt-in instance type | Every sandbox is a full VM with its own kernel, on physical hosts dedicated to boat (AMD Ryzen 7 9700X and Ryzen 9 9950X); vCPUs are shared |
| Docker inside | Yes, in sandboxes with docker_enabled, and in Beam VMs | Yes on the VM type, at 2.5x the default compute rate; the default gVisor type cannot run a Docker daemon | Yes, in every sandbox |
| systemd | In Beam VMs; sandboxes run without systemd | Your entrypoint is PID 1 with no init system, also on the VM type; you start services from your entrypoint | Real systemd and root, like a normal Ubuntu server |
| Nested virtualization | Not documented for sandboxes or VMs | /dev/kvm on the VM type, with a ready Android template | /dev/kvm in every sandbox; Android emulators work |
| GPUs | GPU sandboxes, serverless GPUs and GPU machines, from RTX 4090 up to H100, H200 and B200 | No GPUs | No GPU on any size |
| Lifetime and idle | Sandbox default 10 minutes, or -1 to run until you terminate it; VMs run until you stop them, with optional idle sleep | Runs until you stop it; opt-in auto-sleep after a quiet period (default 15 minutes), wakes on the next request | Auto-stop defaults to 1 hour; set up to 30 days, or turn it off on a paid plan for no session cap; the timer does not watch for idle |
| State between runs | A sandbox keeps files through a saved image, a memory snapshot or a volume; a Beam VM keeps its root disk (50 GiB by default, up to 100 GiB) while stopped | Memory checkpointed on auto-sleep; disk kept and billed while stopped or asleep | Disk saved every minute and at each stop; no compute charge while stopped; disk kept at least 14 days; memory is not saved |
| Fork and snapshots | Sandbox memory snapshots with running processes; VMs pause with RAM kept, fork and take disk snapshots | Fork any running, stopped or sleeping instance (disk copy, the fork boots fresh); nightly backups in 7 slots plus one on-demand, restored in place | Fork a running sandbox from its last 1-minute snapshot (the fork boots fresh); saved snapshots that new sandboxes start from |
| Built-in agents | You set it up: install agents in the sandbox or VM and run them through process exec; Beam's MCP server plugs into Claude Code, Codex and Cursor | Hermes, OpenClaw, Claude Code, Codex, Grok, OpenCode and Pi, with a chat API per instance | Seven agents installed, including Claude Code and Codex, with a prompt, steer and events API |
| Desktop and browser | Beam VMs have a browser Linux desktop (Chrome on x86) and an SDK for screenshots, clicks, typing and recording; documented for VMs, not sandboxes | You set it up: noVNC desktop through a template build, or the Ubuntu VM template | 1920x1080 desktop and Chrome at 60 fps in every sandbox, with computer-use tools and an embeddable stream |
| Inbound access | Sandboxes: a public HTTPS URL per port, no built-in auth. VMs: token-protected URLs and TCP ports on a persistent public IPv6 address | HTTPS URL per port, signed URLs, public ports, custom domains; no public IP, SSH through an authenticated WebSocket tunnel | Up to 50 HTTPS URLs per sandbox, token-gated by default; public IPv6; SSH; the IP changes at each resume |
| Keeping API keys out of the sandbox | Brokered secrets in sandboxes and VMs: the guest sees a placeholder, and Beam puts the real key into HTTPS headers to approved hosts | Managed LLM, search and Composio through a per-instance token, metered at cost; your own keys go in as env vars | Optional AI credentials proxy ($1 a day plus $1 per 1,000 requests) for model providers and HTTPS APIs you register; other secrets still enter |
| Machines at once | 30 vCPUs at once on Developer, 1,000 on Team, across sandboxes, VMs and apps | Awake instances: 3 before your first top-up, then 10, 50, 200 and 1,000 by lifetime top-ups; sleeping and stopped ones do not count | 100 to 2,000 sandboxes at once by plan; starts per minute, hour and day capped by plan |
| Regions | US by default; another region if you choose one (no list published); any region of your own cloud with BYOC | Dedicated servers in the United States | Germany, Finland and France; you cannot pick a region; no US or Asia region |
Every feature Battleships tracks (43)
| Feature | Beam | Agent 37 | boat |
|---|---|---|---|
| Full VM (own kernel) | No documented way | No documented way | Yes |
| Docker inside | Yes | No documented way | Yes |
| Nested virtualization | No documented way | No documented way | Yes |
| Root | Yes | Yes | Yes |
| Systemd | No documented way | Not published | Yes |
| Desktop GUI | No documented way | With your own setup | Yes |
| Browser | No documented way | Yes | Yes |
| Computer-use API | No documented way | No documented way | Yes |
| SSH | Yes | Yes | Yes |
| Persistent disk | With your own setup | Yes | Yes |
| Pause/resume | No documented way | Yes | Yes |
| Automatic snapshots | No documented way | Yes | Yes |
| Fork/clone | Yes | Yes | Yes |
| Live fork (no pause) | Yes | Yes | Yes |
| Memory fork | Yes | No documented way | No documented way |
| Custom image (Docker or snapshot) | Yes | Yes | Yes |
| ≥24 h sessions | Yes | Yes | Yes |
| Idle auto-stop | No documented way | Yes | No documented way |
| Wake on request | No documented way | Yes | No documented way |
| HTTPS preview URLs | Yes | Yes | Yes |
| Raw TCP inbound | No documented way | Not published | Yes |
| Public IPv4 | No documented way | No documented way | No documented way |
| Egress allowlist | Yes | No documented way | No documented way |
| Secret proxy | Yes | Yes | Yes |
| Secret proxy for any API | No documented way | No documented way | Yes |
| Preinstalled agents | No documented way | Yes | Yes |
| Agent harness API | No documented way | Yes | Yes |
| MCP server | Yes | No documented way | No documented way |
| GPU | Yes | No documented way | No documented way |
| Windows | No documented way | No documented way | No documented way |
| Arm64 | No documented way | No documented way | No documented way |
| Self-hosting | Yes | No documented way | No documented way |
| BYOC | Yes | No documented way | No documented way |
| Open source | Yes | No documented way | No documented way |
| Scoped API keys | Not published | Not published | Yes |
| Spend limits | Not published | Yes | Yes |
| Usage per sandbox (API) | No documented way | Yes | Yes |
| Audit logs | Not published | Not published | No documented way |
| SSO | No documented way | No documented way | No documented way |
| SOC 2 | Yes | Yes | No documented way |
| HIPAA | Yes | No documented way | No documented way |
| EU data residency | Not published | No documented way | No documented way |
| Zero data retention | Not published | Not published | Yes |
Performance
Beam and Agent 37 are not in StarSling's run of 3 October 2026 (run 37073680030), so we cannot say which is faster at CPU work. In that run, boat did 28.06 V8 Web Tooling Benchmark runs per second at 4 vCPU / 8 GB, second of 15. Every boat sandbox in that run ran on an AMD Ryzen 9 9950X host. boat also places sandboxes on Ryzen 7 9700X hosts, which that run did not test, and does not promise a CPU model. For start time, ComputeSDK measured a Beam sandbox at 534 ms median with 100 at once (2026-11-06), at Beam's default size. Agent 37 and boat are not in ComputeSDK. Agent 37's default instances share host resources, and its Performance type runs on dedicated cores.
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. Beam and Agent 37 are 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.
| Scenario | Beam, a month | Agent 37, a month | boat, 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 · Developer | $6 · Wallet (any top-up) | $20 · $20/mo |
| One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month | $58 · Developer | $12 · Wallet (any top-up) | $26 · $20/mo |
| 50 agents in parallel 50 machines of 4 vCPU / 8 GB at once, 8-hour sessions, 1,100 sessions a month, Docker inside | $836 · Team | $329 · Wallet ($50 lifetime top-ups) | $317 · $20/mo × 3 seats |
| Sandboxes for your users 2 vCPU / 4 GB, 30,000 sessions of 10 minutes a month, up to 300 at once | $312 · Team | $33 · Wallet ($1,000 lifetime top-ups) | $200 · $100/mo × 2 seats |
| 1,000 short jobs at once 2 vCPU / 4 GB, 1,000 at once, 3 minutes each, 200,000 jobs a month | $527 · Team · with limits | $73 · Wallet ($1,000 lifetime top-ups) · with limits | $8,800 · $2000/mo · with limits |
Beam: Beam bills sandboxes per millisecond while they run, for each vCPU and GiB, on Spot, Shared or Performance tiers; new accounts default to Performance. For 4 vCPU / 8 GiB that is $0.28 an hour on Performance, $0.088 on Shared and $0.036 on Spot. VMs bill per second while running, and nothing for compute while stopped or paused. Developer is $20 a month with $20 of credit; Team is $89 a month plus usage. There are no egress fees, and storage is not billed yet.
Agent 37: A prepaid wallet with no plan fee, metered per minute: $0.80 per vCPU and $0.70 per GB of RAM per month while running, and $0.09 per GB of disk per month even while stopped or asleep. VM compute costs 2.5x and Performance compute 4x the default rate. New accounts get a one-time $5 credit after adding a card.
Agent 37: Plans cost $20, $100, $500 or $2,000 a month, and the whole plan price comes back as VM time for that month. A default 4 vCPU / 8 GB sandbox costs $0.036 an hour, billed per second while it runs. The $20 plan includes 12 starts a minute, 60 an hour and 200 a day; extra starts cost $0.03 to $0.07 each. A stopped sandbox has no compute charge. Saved snapshots above the plan's count and the AI credentials proxy still bill while stopped.
Limits in those scenarios
Agent 37 caps awake instances by lifetime top-ups: 3 before the first, then 10, 50 at $50, 200 at $250 and 1,000 at $1,000; sleeping and stopped instances do not count. Every Agent 37 type stops at 4 vCPU. Its 50-agent scenario needs Docker, so it is priced on the VM type at 2.5x. Beam's Developer plan allows 30 vCPUs at once, so 50 agents of 4 vCPU need Team, and Beam does not bill storage yet, so the scenarios count no disk cost for it. For 1,000 short jobs at once none of the three fits as listed: Beam Team allows 1,000 vCPUs where the burst needs 2,000, Agent 37's default type shares vCPUs, and boat's $2,000 plan charges for starts above 90 a minute and refuses them above 270.
| Scenario | Beam | Agent 37 | boat |
|---|---|---|---|
| One coding agent, workdays |
|
| None for this scenario |
| One agent that never stops |
|
| None for this scenario |
| 50 agents in parallel |
|
| None for this scenario |
| Sandboxes for your users |
|
| None for this scenario |
| 1,000 short jobs at once |
|
|
|
Which one should you use?
Use Beam when
your agents need GPUs, when you want sandboxes and VMs that keep memory, strict outbound rules or a SOC 2 report, or when you want to self-host.
- GPUs in the same platform. For work that needs a GPU, Beam offers GPU sandboxes, serverless GPUs and GPU machines up to H100, H200 and B200.
- VMs with memory, outbound rules and compliance. Beam VMs pause with their memory, a sandbox or VM can block outbound traffic or allow up to 64 hosts, and Beam has a compliance report.
- Open source and your own cloud. Its Beta9 runtime is AGPL-3.0 and free to self-host, and Beam can also run in your own cloud for a fee.
Use Agent 37 when
you host one mostly idle assistant per customer in the US that should sleep with its memory kept and wake on the next message, at a low price.
- Sleeps when idle and keeps its memory. For assistants that sit idle most of the day, an instance sleeps, keeps its memory and wakes on the next message.
- One ready agent per customer. Templates for Hermes, OpenClaw, Claude Code and others come with a chat API, so each customer gets an agent quickly.
- Lowest bill here in four of five scenarios. Agent 37 meters per minute from a prepaid wallet with no plan fee. It costs $12 a month for one agent that never stops, against $26 on boat and $58 on Beam.
Where boat fits
Beam and Agent 37 both start you in a gVisor container in the US by default, and both offer a VM option. If your agents do real work for hours, with installs, Docker, services and emulators, boat gives each one a full VM with systemd, /dev/kvm, a desktop and agents on every sandbox. boat runs in Germany, Finland and France, with no GPUs.
A full VM by default
A Beam sandbox has no systemd and documents no /dev/kvm; Beam VMs add systemd and Docker. On Agent 37, Docker needs the VM type at 2.5x the compute rate, and even that has no init system, so services do not start on their own.
No compute charge while stopped
A Beam sandbox ends with its time to live, 10 minutes by default, unless you save an image or attach a volume; a Beam VM keeps its disk. Agent 37 keeps the disk and bills $0.09 per GB a month while stopped or asleep. boat saves the disk every minute and keeps it at least 14 days with no compute charge. Saved snapshots above the plan's count and the AI credentials proxy still bill.
Coding agents with a steer and events API
Beam leaves agent setup to you. Agent 37 ships agents with a chat API. boat ships seven harnesses with a prompt, steer and events API, and an optional paid proxy that keeps the real keys of supported AI providers out of the sandbox.
Use boat when your agents do real work for hours: installs, builds, Docker, emulators and a browser on a full VM with systemd, whose disk is saved every minute and kept with no compute charge while stopped. Compare it one to one: boat vs Beam and boat vs Agent 37.
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
Questions people ask
Is Beam cheaper than Agent 37?
No. Agent 37's bill is lower in every scenario we priced. One agent that never stops costs $12 a month on Agent 37 against $58 on Beam, and 50 agents in parallel cost $329 against $836. Sandboxes for your users cost $33 against $312. boat comes to $26, $317 and $200 for the same three. Beam's figures use its default Performance tier; its Spot tier costs $0.036 an hour for 4 vCPU / 8 GiB but can be interrupted with 30 seconds of notice.
Which is faster, Beam or Agent 37?
Neither is in the public CPU benchmarks we use, so we cannot say. ComputeSDK measured a Beam sandbox start at 534 ms median with 100 at once (2026-11-06). boat did 28.06 V8 Web Tooling Benchmark runs per second at 4 vCPU / 8 GB in StarSling's run of 3 October 2026, on an AMD Ryzen 9 9950X host.
Can Beam and Agent 37 run Docker?
Beam runs Docker in a sandbox with docker_enabled, and in its VMs. Agent 37 runs Docker only on its VM type, at 2.5x the default compute rate; the default gVisor type cannot run a Docker daemon. boat runs Docker in every sandbox, with systemd and /dev/kvm.
Do Beam or Agent 37 have GPUs?
Beam does, up to H100, H200 and B200, in sandboxes, serverless endpoints and GPU machines. Agent 37 has no GPUs, and boat has no GPU on any size.
What happens when the agent is idle?
Agent 37 can put an instance to sleep after a quiet period, save its memory and wake it on the next request. A Beam sandbox runs until you terminate it or its time to live ends, and a Beam VM can sleep when idle and pause with its memory. boat stops a sandbox at its auto-stop time (default 1 hour), does not watch for idle, and keeps its disk with no compute charge.
Where do Beam and Agent 37 run?
Agent 37 runs on dedicated servers in the United States. Beam runs in the US by default, or in another region if you choose one, or in your own cloud. boat sandboxes run on machines in Germany, Finland and France, and you cannot pick a region.
Try boat
Install the CLI, then run boat new.
7-day free trial