I’m building BoxLite: computers for AI agents. Run code in isolated microVMs. Keep files between sessions. Put what your agent builds online. Open-source runtime + managed cloud. Follow for the engineering behind it. Try: https://boxlite.ai Demo: https://blog.boxlite.ai/from-one-prompt-to-a-playable-game-on-boxlite-cloud…

BoxLite — The managed cloud for AI agents

Source: https://boxlite.ai/ Build and ship with your coding agent. BoxLite runs your app in the cloud and gives you a public URL to share.

what you get

A runtime that hosts your app. Compute, network, filesystem, volumes.

cpus · memory_mib

Compute in a real microVM

Every box is a virtual machine with its own Linux kernel, sized when you create it — up to 4 vCPU and 32 GiB of memory. Hardware isolation, not a shared kernel.

preview-url

A public URL for your app

Publish a box and every service port gets a URL you can open and share. Outbound traffic can be pinned to an allowlist, so the box reaches only what you named.

/home/boxlite

A disk that survives a restart

Around 10 GiB of real disk per box. Packages, files and environment are still there after a stop and a start, so you are not rebuilding the world on every boot.

managed volumes

Storage that outlives the box

Mount a managed volume by a name you choose and the data survives the box that wrote it. Volumes are not metered — you pay for the box, not the bytes it keeps.

quickstart

A URL, a key, and a thousand boxes.

Create an API key in the console, export two environment variables, and the SDK you already know points at the managed runtime. Handing/agent.mdto a coding agent works too — every recipe in it is executed against a live deployment before it ships, and says which one and when.

01from boxlite import Boxlite, BoxliteRestOptions02from boxlite import ApiKeyCredential, BoxOptions03import os0405rt = Boxlite.rest(BoxliteRestOptions(06    url=os.environ["BOXLITE_REST_URL"],07    credential=ApiKeyCredential(os.environ["BOXLITE_API_KEY"]),08))0910box = await rt.create(BoxOptions(image="python"))11await box.start()1213result = await box.exec("python", args=["-c", "print(2 + 2)"])14print(result.stdout)

104 endpointsthe SDK wraps the common ones; REST reaches all of them

POST/api/boxCreate a microVM.

GET/api/box/\{id\}State, image, network, and whether it is public.

GET/api/box/\{id\}/ports/\{port\}/preview\-urlA public address for a port inside it.

POST/api/box/\{id\}/public/\{bool\}Publish it, or take it back.

POST/api/box/\{id\}/autostop/\{seconds\}Idle lifecycle. Zero disables it.

POST/api/volumesStorage with a lifetime of its own.

GET/api/box/\{id\}/telemetry/logsWhat the box has been doing.

use cases

Built for the agentspeople actually ship.

RUN CLAUDE CODE IN A BOX

An autonomous coding agent confined to a microVM.

The agent gets root inside its own kernel and nothing at all outside the box.

BUILD A CODE INTERPRETER FOR YOUR LLM

A model writes Python; it executes in a disposable microVM.

Generated code never touches your host, and the box is thrown away when it is done.

REVIEW UNTRUSTED PULL REQUESTS

Tests and an automated review of code you have not read.

Clone the PR into a box, run the suite, then read the diff — with no access to your machine.

BUILD A DATA ANALYSIS AGENT

CSV in, model-written pandas analysis and a chart out.

The model installs what it needs and writes files; you get the artifact back.

SCRAPE AND ANALYZE THE WEB SAFELY

A browser box plus a parsing box, feeding clean data to a model.

Playwright and CDP are ready in the browser box; parsing runs in a separate one.

GIVE EACH USER A PERSISTENT SHELL

A live, stateful shell per user, branch, or agent.

State survives stop and restart, so every user or branch keeps its own environment.

local or cloud

Same runtime. Local or Cloud.

Cloud is the same SDK and the same core API as the local runtime — reached over a URL instead of loaded in-process.

LocalCloudWhere boxes runYour laptop, your servers, air-gappedBoxLite’s managed fleetHow you reach itIn-process library, no accountREST URL + SDKYou operateHosts, image cache, capacityNothingIsolationHardware microVM, own kernelHardware microVM, own kernelStorageYour own diskManaged volumes and cloud disks that outlive the boxBox lifecycleYou manage itIdle-stop and wake-on-accessConcurrencyAs far as your own hardware goesThousands of concurrent boxesSupportDiscord and GitHub issuesDedicated support for Cloud customersBest forLocal dev, embedding in your app, on-premProduction agents, bursty scale, teamsExplore the open source runtime→

pricing

Pay as you go, or take a plan.

Plans come with a usage allowance that resets each cycle; anything beyond it draws from a prepaid wallet. You can also run wallet-only with no subscription at all.

Starter

$19per month

Get an agent into production.

• $30 included usage each cycle
• 20 concurrent boxes

Start building $149per month

For agents carrying real traffic.

• $250 included usage each cycle
• 100 concurrent boxes

Start building

Max

$499per month

Fleet-scale parallel agents.

• $900 included usage each cycle
• 1,000 concurrent boxes

Start building

Per-box ceilings apply on every plan: 4 vCPU, 32 GiB memory, 120 GiB storage. Included quota covers usage first; a prepaid wallet funds the rest. Paid top-ups are refundable within 14 days of purchase.Refund terms·Full billing docs

faq

The questions people actually ask.

Yes. The docs describe Cloud as a hosted agent runtime built on BoxLite — the same SDK and the same core API, reached with a URL and an API key, with managed storage and a managed box lifecycle.