Vercel Sandbox vs Agent 37

Vercel Sandbox gives each run its own Firecracker microVM, built for short bursts of code in 19 regions, and stops every session after 24 hours. Agent 37 keeps one always-on gVisor sandbox per user in the US that can sleep, keep its memory and wake on the next message.

The companies and products

Vercel Sandbox

Vercel is a cloud platform for web apps and agents, and Vercel Sandbox is its product for running untrusted or agent-written code in isolated Linux microVMs.

Each sandbox is a Firecracker microVM with its own kernel, started from a Vercel image, your own OCI image, or a saved snapshot. You drive it with the JavaScript or Python SDK or the sandbox CLI: run commands, move files, open ports, and set network rules. It is aimed at code interpreters, app builders, test runs and agent tools, and Vercel's docs say it is not meant to run nonstop.

Isolation: Firecracker microVM with a dedicated kernel per sandbox
Default image: Ubuntu with Node.js 24, Python 3.14, and Claude Code, Codex, OpenCode and pi preinstalled
SDKs: JavaScript/TypeScript, Python, and a CLI
Regions: 19 regions, default iad1 (US East)
Longest session: 24 hours on Pro and Enterprise, 45 minutes on Hobby; a stopped sandbox resumes from its last snapshot
Biggest sandbox: 8 vCPU / 16 GB on Pro, 32 vCPU / 64 GB on Enterprise
Compliance: SOC 2 Type 2, ISO 27001, PCI DSS; HIPAA for Enterprise

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.

Isolation: gVisor by default; opt-in full VM type with its own kernel and KVM
Where it runs: Dedicated servers in the United States
Billing: Prepaid wallet, metered per minute, no plan fee or seat charge
Instance types: Default (shared), Performance (dedicated cores), VM
Compliance: SOC 2 Type I, Type II in progress
Company: Y Combinator company based in New York

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. “22nd cheapest of 345” means 21 of the other 344 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	Vercel Sandbox: price rank	Agent 37: price rank	bochka: price rank
Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month Vercel has no desktop and stops at 24 hours; Agent 37 needs its VM type for Docker; bochka gives every agent Docker, systemd and a desktop.	244th of 367, with limits $3,351 a month	2nd cheapest of 367 $551 a month	3rd cheapest of 367 $634 a month
Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB Agent 37 is built for always-on assistants that sleep and keep their memory; Vercel stops each session at 24 hours, and bochka runs with no cap on paid plans.	257th of 366, with limits $3,657 a month	Cheapest of 366 $279 a month	5th cheapest of 366 $657 a month
Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month Both suit short code runs: Vercel bills CPU only while code uses it, Agent 37 has small cheap default sandboxes, and bochka's start limits make large bursts a poor fit.	22nd cheapest of 345 $216 a month	Cheapest of 345 $22 a month	186th of 345, 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 Vercel fits sandboxes for your users at scale; Agent 37 stops at 200 instances, counting sleeping ones, and bochka ranks first on cost.	15th cheapest of 361 $1,267 a month	30th of 361, with limits $79 a month	Cheapest of 361 $209 a month
RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month Vercel ranks third here thanks to high concurrency and fast starts; Agent 37's self-serve cap of 200 instances makes big rollouts a compromise.	3rd cheapest of 360 $15,761 a month	6th of 360, with limits $363 a month	34th of 360, with limits $2,000 a month
Agent evals 250 at once · 4 vCPU / 8 GB · 15,000 sessions of 20 min a month	69th cheapest of 364 $1,508 a month	90th of 364, with limits $58 a month	10th cheapest of 364 $200 a month
CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month All three fit CI runners; Agent 37 ranks 2nd and bochka 4th on cost, while Vercel ranks 107th.	107th cheapest of 377 $1,144 a month	2nd cheapest of 377 $84 a month	4th cheapest of 377 $86 a month
24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB Agent 37 keeps a machine around the clock and is cheap to keep idle; Vercel restarts every 24 hours and loses running processes.	276th of 368, with limits $145 a month	3rd cheapest of 368 $12 a month	18th cheapest of 368 $26 a month
Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Vercel Sandbox has no desktop; Agent 37 adds a noVNC desktop through a template build, and bochka streams a 60fps desktop out of the box.	235th of 369, with limits $952 a month	Cheapest of 369 $53 a month	6th cheapest of 369 $200 a month
Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month	259th of 367, with limits $486 a month	48th of 367, with limits $29 a month	7th cheapest of 367 $200 a month
Scraping & bots (startup scale) 50 at once · 1 vCPU / 2 GB · 100,000 sessions of 3 min a month	277th of 346, with limits $468 a month	Cheapest of 346 $32 a month	6th cheapest of 346 $90 a month
Mass scraping & bots (undetectable) 1,000 at once · 1 vCPU / 2 GB · 3,000,000 sessions of 2 min a month	234th of 338, with limits $9,960 a month	36th of 338, with limits $648 a month	71st of 338, with limits $1,800 a month
Parallel red teaming 200 at once · 2 vCPU / 4 GB · 5,000 sessions of 30 min a month	283rd of 360, with limits $554 a month	Cheapest of 360 $43 a month	2nd cheapest of 360 $200 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	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	Can't run it

Key features

Feature	Vercel Sandbox	Agent 37	bochka
Isolation	Firecracker microVM with its own kernel	gVisor by default; full VM as an opt-in instance type	Every sandbox is a full VM with its own kernel, on dedicated AMD Ryzen 9 9950X hosts
Longest session	24 hours on Pro and Enterprise, 45 minutes on Hobby; resume starts a new session	No time limit; runs until you stop it or it auto-sleeps	No cap on paid plans; default time limit 1 hour, changeable
Sleep and memory	Filesystem snapshot on stop; memory and processes start fresh	Opt-in auto-sleep on idle; memory checkpointed, processes resume when the restore works	Disk saved every minute and on stop; a resume reboots the machine
Docker inside	Yes, with sudo	Only on the VM type, at 2.5x the default compute rate	Yes, in every sandbox, with root and systemd
Nested virtualization	No	/dev/kvm on the VM type, with a ready Android template	/dev/kvm in every sandbox, Android emulator works
Desktop and browser	None; headless Linux only	noVNC desktop through a template build or the Ubuntu VM template	60fps desktop and Chromium in every sandbox, computer-use and browser-control APIs
Custom images	Your own OCI images from a Dockerfile, via Vercel Container Registry	Your own linux/amd64 Docker image as a template	Named snapshots as shareable templates; no Dockerfile builds
Coding agents	Claude Code, Codex, OpenCode and pi preinstalled; you start them as commands	Hermes, OpenClaw, Claude Code, Codex, Grok, OpenCode, Pi, with a chat API per instance	Claude Code, Codex, pi, OpenCode, Prime Agent, Kimi Code, Mistral Vibe, with an API to prompt, steer and stream events
Keys kept outside the sandbox	Credentials brokering injects headers for any HTTPS domain you allow	Managed LLM, search and Composio through a per-instance token; your own keys go in as env vars	Proxy for model providers and any HTTPS API you register; switch a credential to another account when one runs out
Egress rules	Allow-all, deny-all or a domain and CIDR allowlist, changeable while running	No provider-side allowlist	No provider-side allowlist; firewall inside the VM
Inbound access	Public URLs for up to 15 ports, no built-in auth; no raw TCP or public IP	HTTPS URL per port, signed URLs, custom domains; no public IP, SSH through a WebSocket tunnel	HTTPS URLs for up to 50 ports, token-protected by default; raw TCP; dedicated IPv6 or IPv4
How many at once	Up to 10,000 concurrent sandboxes on Pro	1 to 200 instances by lifetime top-ups, counting sleeping and stopped ones; email for more	Concurrency set by plan, up to 2,000 at once
Regions	19 regions across the US, Europe, Asia, South America, Africa and Australia	United States	EU only (Germany, Finland, France), EU data residency
CPU speed (4 vCPU / 8 GB)	12.96 runs/s	Not measured	26.69 runs/s

Every feature Battleships tracks (43)

Feature	Vercel Sandbox	Agent 37	bochka
Full VM (own kernel)	Yes	No	Yes
Docker inside	Yes	No	Yes
Nested virtualization	No	No	Yes
Root	Yes	Yes	Yes
Systemd	Not published	Not published	Yes
Desktop GUI	No	Yes	Yes
Browser	No	Yes	Yes
Computer-use API	No	Yes	Yes
SSH	No	Yes	Yes
Persistent disk	Yes	Yes	Yes
Pause/resume	Yes	Yes	Yes
Automatic snapshots	Yes	Yes	Yes
Fork/clone	Yes	Yes	Yes
Live fork (no pause)	Yes	Yes	Yes
Memory fork	No	No	No
Custom image (Docker or snapshot)	Yes	Yes	Yes
≥24 h sessions	Yes	Yes	Yes
Idle auto-stop	No	Yes	No
Wake on request	No	Yes	No
HTTPS preview URLs	Yes	Yes	Yes
Raw TCP inbound	No	Not published	Yes
Public IPv4	No	No	No
Egress allowlist	Yes	No	No
Secret proxy	Yes	Yes	Yes
Secret proxy for any API	Yes	No	Yes
Preinstalled agents	Yes	Yes	Yes
Agent harness API	No	Yes	Yes
MCP server	Yes	No	No
GPU	No	No	No
Windows	No	No	No
Arm64	No	No	No
Self-hosting	No	No	No
BYOC	No	No	No
Open source	No	No	No
Scoped API keys	Not published	Not published	Yes
Spend limits	Yes	Yes	Yes
Usage per sandbox (API)	No	Yes	Yes
Audit logs	Not published	Not published	No
SSO	Yes	No	No
SOC 2	Yes	Yes	No
HIPAA	Yes	No	No
EU data residency	Yes	No	Yes
Zero data retention	Not published	Not published	Yes

Performance

Vercel Sandbox12.96 runs/s

Agent 37Not measured

bochka26.69 runs/s

On the hpc-sandbox-benchmarks Node.js build-and-test loop at 4 vCPU / 8 GB, bochka did 26.69 runs per second and Vercel Sandbox did 12.96, so bochka finished about twice as many runs. Agent 37 has no published result on that benchmark, so there is no comparable measurement. Agent 37 says its default instances share host resources, and sells a Performance type on dedicated cores.

Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Agent 37 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	Vercel Sandbox, a month	Agent 37, a month	bochka, 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	$57 · Pro	$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	$236 · Pro · with limits	$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	$2,884 · Pro	$329 · Wallet ($100 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	$828 · Pro	$33 · Wallet ($250 lifetime top-ups) · with limits	$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	$3,152 · Pro	$73 · Wallet ($250 lifetime top-ups) · with limits	$2,000 · $2000/mo · with limits

Vercel Sandbox: Vercel charges for CPU only while your code uses it, but charges for memory for the whole time a sandbox runs, idle time included, plus a fee per sandbox created and per GB of snapshots. The $20 Pro credit is shared with your other Vercel usage.

Agent 37: Agent 37 uses a prepaid wallet with no plan fee, metered per minute on the size you pick: $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 asleep. Performance compute costs 4x and VM compute 2.5x.

Agent 37: bochka charges a flat price per size, per second, only while the machine runs; a stopped sandbox costs nothing and keeps its disk. Plans start at $20 a month, and all of the fee is usage credit.

Limits in those scenarios

For one agent that never stops, Vercel ends each session at 24 hours and running processes start over, so it does not fit. Agent 37 counts every instance it holds, sleeping and stopped ones too, against a cap of 200 at $250 of top-ups, which is why it does not fit 300 user sandboxes at once or 1,000 jobs at once. Vercel fits that burst, and bochka does not: even its $2,000 plan allows 600 starts an hour and queues bursts above 90 starts a minute. Agent 37 is priced on its VM type for 50 agents in parallel, because that scenario needs Docker.

Scenario	Vercel Sandbox	Agent 37	bochka
One coding agent, workdays	None that bite	egress price not published	None that bite
One agent that never stops	Pro: sessions capped at 24 h	egress price not published	None that bite
50 agents in parallel	None that bite	egress price not published	None that bite
Sandboxes for your users	None that bite	max 200 concurrent egress price not published	None that bite
1,000 short jobs at once	None that bite	max 200 concurrent shared vCPUs on this tier: heavy, sustained CPU use can be slowed by other tenants egress price not published	max 2,400 starts/day (needs ~6,667) max 600 starts/hour (needs ~1,000) bursts above 90 starts/min are queued (a burst of 333 takes ~4 min to start)

Which one should you use?

Use Vercel Sandbox when

you run many short bursts of untrusted code at once, or need sandboxes close to users in many regions.

Regions all over the world. Useful if your users are spread out: Vercel Sandbox runs in 19 regions, while Agent 37 runs only in the US.
Big bursts of short jobs. For thousands of short runs at once, Vercel allows up to 10,000 sandboxes on Pro and bills CPU only while code uses it.
A firewall you control from outside. A running sandbox can be switched to deny-all, or to a list of allowed domains, without a restart.

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.

Sleeps when idle and keeps its memory. Useful for assistants that sit idle most of the day: an instance sleeps, keeps its memory and wakes on the next message.
No session cap. An instance runs until you stop it, so a long task is never cut off at 24 hours as it is on Vercel.
A chat API for each agent. Start Hermes, OpenClaw, Claude Code and others from a template and talk to each one through a chat API.

Where bochka fits

If your agents do real work for hours, such as installs, builds, tests, Docker and checking the app they built, look at bochka. Every bochka sandbox is a full VM with a desktop, a browser and an API to drive the agents inside from outside, and it runs about twice as fast as Vercel Sandbox on the same CPU test.

A full VM is the default

Vercel Sandbox has no /dev/kvm, so no Android emulators or VMs inside. Agent 37's default gVisor type cannot run Docker, and its VM type costs 2.5 times as much and has no systemd. bochka gives every sandbox Docker, systemd and /dev/kvm.

A desktop and a browser built in

Vercel Sandbox is headless, so an agent cannot open the web app it built to check it. Agent 37 needs a template build for a noVNC desktop. bochka streams a 60fps desktop with Chromium and computer-use APIs in every sandbox.

Host real services

Vercel gives up to 15 port URLs with no built-in auth and no raw TCP or public IP. Agent 37 has no public IP and tunnels SSH through a WebSocket. bochka gives token-protected HTTPS on 50 ports, raw TCP and its own IP.

Use bochka when your agents work for hours on one machine: Docker, emulators, a desktop with a browser, real hosting, and an API to drive the agents inside from outside. Compare it one to one: bochka vs Vercel Sandbox and bochka vs Agent 37.

Questions people ask

Is Agent 37 cheaper than Vercel Sandbox?

Yes, in every scenario we priced. One workday agent costs $6 a month on Agent 37 vs $57 on Vercel, and 50 parallel agents cost $329 vs $2,884. But Agent 37 does not fit 300 user sandboxes or 1,000 jobs at once, because of its 200-instance cap.

Which is faster?

Vercel Sandbox did 12.96 runs per second on the hpc-sandbox-benchmarks Node.js loop at 4 vCPU / 8 GB. Agent 37 has no result on that benchmark, so we cannot compare them. bochka did 26.69 runs per second on the same test.

Can Vercel Sandbox or Agent 37 run Docker?

Vercel Sandbox runs Docker with sudo in every sandbox. Agent 37 runs Docker only on its VM type, which bills compute at 2.5x; the default gVisor type cannot run a Docker daemon. bochka runs Docker in every sandbox.

Can either one run an agent 24/7?

Agent 37 can: an instance runs until you stop it, and it can sleep on idle and wake with its memory. Vercel Sandbox sessions stop after 24 hours, and its docs say sandboxes are not designed to run continuously. bochka machines run with no session cap on paid plans.

Does either have a desktop for computer use?

Vercel Sandbox has none; it is headless Linux. Agent 37 offers a noVNC desktop through a template build or its Ubuntu VM template. bochka streams a 60fps desktop with Chromium and computer-use APIs in every sandbox.

Where do they run?

Vercel Sandbox runs in 19 regions around the world, with iad1 in the US as the default. Agent 37 runs on dedicated servers in the United States. bochka runs only in the EU, in Germany, Finland and France.

Try bochka

Install the CLI, then run bochka new.

Product preview