bochka vs Buildkite

Buildkite is a CI platform, and its hosted agents are machines that run one pipeline job each and are then destroyed. bochka is built for AI agents that work for hours or days on one full Linux VM, with a disk that persists, a desktop, and coding agents you drive from outside.

Where bochka stands out

One machine that keeps its state

A bochka sandbox runs as long as the work does on paid plans, and its whole disk is saved every minute and on stop, so it starts again with every package and file in place. A Buildkite hosted agent is destroyed after each job, which can last 8 hours on Linux and 4 on macOS.

Coding agents you drive from outside

bochka ships Claude Code, Codex and five other agents in each sandbox, and you drive agents inside from outside with a prompt, steer and events API. The credentials proxy keeps real keys for any HTTPS API out of the machine, and can switch a credential to another account when credits run out.

A full Linux computer, reachable from outside

Every bochka sandbox has root, systemd, Docker and /dev/kvm, a 60fps desktop with Chromium for computer use, and HTTPS URLs on up to 50 ports plus raw TCP. Buildkite hosted agents expose no ports, and their desktop access is for a person debugging a macOS job.

The companies and products

bochka

bochka.xyz 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
Hardware: Dedicated AMD Ryzen 9 9950X (Zen 5) hosts
Regions: EU: Germany, Finland, France
Plans: From $20 a month, all of it usage credit, 7-day trial

Buildkite

Buildkite is a CI and software delivery platform, and hosted agents are the build machines it runs for you.

You write Buildkite pipelines, and each job runs on a fresh virtual machine that Buildkite starts for that job and destroys when it ends; only cache volumes carry files to the next job. Hosted agents come in Linux and macOS on Apple M4, with Windows in private preview, and you can also run Buildkite's agent on your own machines. It is a CI service, not an agent sandbox, so this page compares the part where they overlap: a machine where an agent or a job does its work.

Product: CI and software delivery platform; hosted agents, Mac hosted agents and self-hosted agents run pipeline jobs
Isolation: A virtual machine per job, destroyed after the job with all of its data; macOS uses Apple's Virtualization framework
Job time limit: Up to 8 hours on Linux and 4 hours on macOS; longer on request to support
Where it runs: Hosted agents run only in US East Coast data centers
Customers: Logos include Cursor, Shopify, Uber, Reddit, Figma and Contact
Compliance: Maintains a SOC 2 Type II report

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	bochka: price rank	Buildkite: price rank
Software factory 100 at once · 4 vCPU / 8 GB · 2,200 sessions of 8 h a month A coding agent that builds a whole app needs the same machine for hours with its files intact; a Buildkite job VM is destroyed when the job ends, and only cache volume paths carry over.	3rd cheapest of 367 $634 a month	43rd of 367, with limits $11,648 a month
Always-on AI employees 50 machines 24/7 · 2 vCPU / 4 GB An assistant that lives on its machine all month does not fit a CI job that stops after 8 hours; bochka has no session cap on paid plans.	5th cheapest of 366 $657 a month	351st of 366, with limits $17,564 a month
Code interpreter bursts 100 at once · 1 vCPU / 2 GB · 200,000 sessions of 1 min a month	186th of 345, with limits $2,000 a month	315th of 345, with limits $1,587 a month
Sandboxes for your users 300 at once · 2 vCPU / 4 GB · 30,000 sessions of 23 min a month Your users' sandboxes need a URL they can reach; Buildkite hosted agents expose no ports, while bochka gives each sandbox HTTPS on up to 50 ports and raw TCP.	Cheapest of 361 $209 a month	326th of 361, with limits $4,984 a month
RL rollouts 1,000 at once · 2 vCPU / 4 GB · 1,000,000 sessions of 3 min a month	34th of 360, with limits $2,000 a month	316th of 360, with limits $23,881 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	324th of 364, with limits $3,884 a month
CI runners 40 at once · 8 vCPU / 16 GB · 6,000 sessions of 12 min a month CI is what Buildkite is built for, with pipelines, cache volumes and test tooling; bochka runs CI jobs on full VMs with Docker and /dev/kvm for Android emulators.	4th cheapest of 377 $86 a month	203rd of 377, with limits $2,318 a month
24/7 dev boxes 1 machine 24/7 · 4 vCPU / 8 GB A dev box has to stay up and keep its installed packages; on Buildkite everything outside cache volumes resets with each job.	18th cheapest of 368 $26 a month	354th of 368, with limits $715 a month
Computer-use agents 50 at once · 4 vCPU / 8 GB · 20,000 sessions of 15 min a month Buildkite's macOS desktop access is for a person debugging a job, with no API for an agent to click; bochka streams a 60fps Linux desktop with computer use.	6th cheapest of 369 $200 a month	70th of 369, with limits $3,500 a month
Browser-use agents 50 at once · 2 vCPU / 4 GB · 30,000 sessions of 10 min a month	7th cheapest of 367 $200 a month	142nd of 367, with limits $1,772 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	142nd of 346, with limits $2,354 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	317th of 338, with limits $46,814 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	85th of 360, with limits $1,184 a month
GPU eval jobs 10 at once · 8 vCPU / 32 GB · 500 sessions of 30 min a month Neither fits: all of Buildkite's hosted shapes and all of bochka's are CPU only.	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 Only Buildkite fits here: it runs macOS on Apple M4 with Xcode, and bochka runs Linux only.	Can't run it	35th of 36, with limits $2,310 a month

Key features

Feature	bochka	Buildkite
Kind of product	Agent sandboxes you create and drive by API, CLI or SDK	CI build machines that run jobs from Buildkite pipelines
Isolation	Full VM with its own kernel on dedicated AMD Ryzen 9 9950X (Zen 5) hosts	A virtual machine per job; Apple Virtualization framework on macOS
How long a machine runs	No session cap on paid plans; you set the time to live	Up to 8 hours per job on Linux and 4 on macOS, longer on request
Files after a run	The whole disk is saved every minute and on stop; a stopped sandbox costs nothing	The VM and its data are destroyed; cache volumes keep chosen paths after successful jobs, best effort, up to 14 days
Fork	Fork a running sandbox with its disk	Each job gets its own copy of the last committed cache volume
Docker, root and nested virtualization	Docker, root, systemd and /dev/kvm on every machine	Docker on Linux; macOS has the Docker CLI but no daemon; /dev/kvm not documented
Desktop and computer use	60fps Linux desktop with Chromium, computer use and browser control, embeddable stream	Browser-based desktop access to debug a running macOS job; no computer-use API
Reaching the machine	HTTPS URLs for up to 50 ports, token-protected; raw TCP; dedicated IPv4 or IPv6	No exposed ports; a terminal into a running job through Buildkite's UI or bk job ssh
Coding agents	Claude Code, Codex and five more preinstalled, with a prompt, steer and events API	You install Claude Code in your image or step and run it as a pipeline job
Keeping API keys out	Credentials proxy for model providers and any HTTPS API you register; switch a credential to another account when credits run out	Anthropic proxy: jobs call it with the job token and Buildkite sets the API key header
Operating systems	Linux only	Linux (AMD64, ARM64 on Enterprise), macOS on Apple M4 with Xcode, Windows Server 2022 in private preview
Biggest machine	Up to 16 vCPU / 32 GB on the $100 plan and above	Linux up to 8 vCPU / 32 GB on Pro and 64 vCPU / 256 GB on Enterprise; macOS up to 12 vCPU / 56 GB
CI pipelines	No pipeline service; run CI jobs on a sandbox yourself	Full CI platform with pipelines, test tooling and your own self-hosted agents
Regions	EU only (Germany, Finland, France), with EU data residency and zero data retention	US East Coast only
Compliance and transparency	SOC 2 is in progress. We're open about where we stand: live status and data retention policy.	SOC 2

Every feature Battleships tracks (43)

Feature	bochka	Buildkite
Full VM (own kernel)	Yes	Yes
Docker inside	Yes	Yes
Nested virtualization	Yes	No
Root	Yes	Not published
Systemd	Yes	No
Desktop GUI	Yes	Yes
Browser	Yes	Yes
Computer-use API	Yes	No
SSH	Yes	Yes
Persistent disk	Yes	Yes
Pause/resume	Yes	No
Automatic snapshots	Yes	Yes
Fork/clone	Yes	Yes
Live fork (no pause)	Yes	No
Memory fork	No	No
Custom image (Docker or snapshot)	Yes	Yes
≥24 h sessions	Yes	No
Idle auto-stop	No	Yes
Wake on request	No	No
HTTPS preview URLs	Yes	No
Raw TCP inbound	Yes	No
Public IPv4	No	No
Egress allowlist	No	No
Secret proxy	Yes	Yes
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	Yes
Arm64	No	Yes
Self-hosting	No	Yes
BYOC	No	No
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	Yes
HIPAA	No	No
EU data residency	Yes	No
Zero data retention	Yes	Not published

Performance

bochka26.69 runs/s

BuildkiteNot measured

We have no comparable measurement for Buildkite hosted agents: they were not part of the StarSling hpc-sandbox-benchmarks run. On that Node.js build-and-test benchmark at 4 vCPU / 8 GB, bochka did 26.69 runs per second on dedicated AMD Ryzen 9 9950X (Zen 5) hosts.

Node.js build-and-test loop at 4 vCPU / 8 GB, higher is better. Measured with the open hpc-sandbox-benchmarks suite. Buildkite 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	bochka, a month	Buildkite, 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	$166 · Pro · with limits
One agent that never stops 4 vCPU / 8 GB, 50 GB disk, running 24/7 all month	$26 · $20/mo	$715 · 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	$7,677 · 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	$2,204 · 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	$4,787 · Pro · with limits

bochka: You pick a monthly plan from $20, and all of it becomes usage credit spent per second on running sandboxes. A stopped sandbox costs nothing and keeps its disk.

Buildkite: Pro costs $30 per active user a month and includes 4,000 Linux vCPU minutes; after that Linux is $0.004 per vCPU minute and macOS $0.02, metered to the second with no minimums. A 4 vCPU / 16 GB Linux agent comes to $0.96 an hour, and Enterprise pricing is custom.

Limits in those scenarios

The limit that bites most on Buildkite is the job itself: a hosted agent stops after 8 hours on Linux, and everything outside cache volume paths is gone, so the agent that never stops and the coding agent that installs packages have to rebuild each run. Pro shares one pool of 256 Linux vCPU across all running agents, so 1,000 short jobs at once mostly wait in the queue. On bochka the limits are a sandbox count per plan and starts per minute, so the same burst is queued and needs more starts per day than the $2,000 plan allows.

Scenario	bochka	Buildkite
One coding agent, workdays	None that bite	different kind of product: CI runner files survive a stop only in cache volume paths: packages installed elsewhere reset egress price not published
One agent that never stops	None that bite	different kind of product: CI runner sessions end after 8 h: needs restarts to run 24/7 files survive a stop only in cache volume paths: packages installed elsewhere reset egress price not published
50 agents in parallel	None that bite	different kind of product: CI runner files survive a stop only in cache volume paths: packages installed elsewhere reset egress price not published
Sandboxes for your users	None that bite	Pro: max 128 concurrent Pro: max 256 vCPU running at once (needs 600) different kind of product: CI runner egress price not published
1,000 short jobs at once	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)	Pro: max 128 concurrent Pro: max 256 vCPU running at once (needs 2,000) different kind of product: CI runner egress price not published

Which one should you use?

Use Buildkite when

you need a CI platform with pipelines, macOS on Apple M4 for iOS builds, Windows, or Linux build machines larger than 16 vCPU.

macOS on Apple M4, and Windows. Useful if you build iOS or Mac apps: Buildkite runs macOS on Apple M4 with current Xcode, and Windows in preview.
A whole CI platform. Useful if you want managed CI: pipelines, test tooling and cache volumes, mixing hosted and your own self-hosted agents.
Bigger build machines. Useful for heavy builds: Enterprise Linux shapes go up to 64 vCPU and 256 GB, with ARM64 shapes too.

Use bochka when

your agents work on Linux for hours or days: one full VM that keeps its disk, Docker and /dev/kvm, a 60fps desktop, URLs you can reach, and coding agents you drive from outside.

One machine that keeps its state. An agent on Buildkite loses everything outside its cache volumes when the job ends, and those volumes are best effort and only saved after a successful exit. Your code has to rebuild the machine and its context every run.
Coding agents you drive from outside. On Buildkite an agent is a pipeline step you start and read through logs, with no way to send it a new instruction mid-run. Its key proxy covers Anthropic only, so other keys sit inside the job.
A full Linux computer, reachable from outside. An agent that runs a preview server, an Android emulator or a browser it clicks through has nowhere to show it on a Buildkite agent, and no display API to drive.

Questions people ask

Is bochka cheaper than Buildkite hosted agents?

In the scenarios we priced, yes: one coding agent on workdays comes to $20 a month on bochka and $166 on Buildkite, and one agent running all month to $26 versus $715. 50 agents in parallel come to $317 versus $7,677. For 1,000 short jobs at once neither fits as priced, at $2,000 on bochka and $4,787 on Buildkite.

Can I run Claude Code on Buildkite hosted agents?

Yes, as a pipeline step: you install it in your image or step, and Buildkite's Anthropic proxy lets it call the API with the job token. The run ends when the job ends. On bochka Claude Code is preinstalled, and you prompt and steer it through an API while it works.

How long can a Buildkite hosted agent run?

Up to 8 hours per job on Linux and 4 hours on macOS, and support can allow longer runs. After the job, the machine is destroyed. bochka has no session cap on paid plans.

Do files survive between Buildkite jobs?

Only in cache volumes: paths you declare are saved when a job exits successfully and handed to the next job, on a best-effort basis for up to 14 days. Everything else is deleted with the VM. bochka saves the whole disk every minute and on stop.

Does bochka run macOS like Buildkite?

No. bochka runs Linux VMs on dedicated AMD Ryzen 9 9950X (Zen 5) hosts. If you need macOS with Xcode, Buildkite's Mac hosted agents run it on Apple M4.

Is bochka SOC 2 compliant?

SOC 2 is in progress. We're open about where we stand: live status at status.bochka.xyz and our data retention policy at docs.bochka.xyz/data-retention. Buildkite maintains a SOC 2 Type II report.

Start a sandbox

Install the CLI, then run bochka new.

Product preview