Playground

This is the playground. Every workflow here executes for real: it creates live OpenSandbox containers, writes real files into them, streams real command output back, and tears everything down when it finishes. There is no replay and no mock data — what you see is what the SDK does.

It talks to the same alineo sandbox API that powers the standalone dashboard at sandbox.alineo.tech. By default it points at the hosted endpoint; if you'd rather run against your own machine, start a server locally and switch the endpoint below.

Sandbox serveridle

Every workflow in this section runs against a real OpenSandbox instance through the alineo sandbox API. Point it at the hosted endpoint, or at your own local server from bunx alineo-cli init.

—

Endpoint https://sandbox-api.alineo.tech · hard-capped at 3 sandboxes, auto-expiring. Runs are cleaned up when they finish.

Run your own server

The hosted endpoint is hard-capped and shared. For unlimited runs, or to poke at the internals, run the stack locally:

Start OpenSandbox in Docker:

bunx alineo-cli init

From a clone of the alineo repo, start the sandbox API:

cd apps/sandbox && bun run start

It listens on http://localhost:3000. For the agent workflow, also set GEMINI_API_KEY in its environment.

Back here, open Change on the connection panel and pick Local.

Workflows

Each one composes several SDK primitives into a single real task. Start with Run untrusted code — it's the shortest.

~30s

Run untrusted codeExecute an arbitrary Python snippet in a disposable, resource-capped container and prove it is isolated from everything else.

createSandboxwriteFileexecdeleteSandbox

~40s

CI test runnerRun a project's test suite in a disposable sandbox, watch it fail, apply a fix, watch it pass — and turn the raw output into a structured report.

createSandboxwriteFileexecdeleteSandbox

~50s

Checkpoint a multi-stage pipelineRun an extract → transform → load pipeline that snapshots the container after every stage, so a crash resumes from the last good state instead of the top.

createSandboxexeccheckpointlistCheckpoints

~1 min

Fork a sandbox for parallel workDo the expensive setup once, then fork the live container into independent children that share that state but can't see each other's changes.

createSandboxexecforkdeleteSandbox

~2 min

Let an agent fix a failing testLoad a Pi coding agent into a sandbox, hand it a repo with a failing test, and stream every tool call as it debugs, patches, and re-runs the suite until it's green.

createAgentprompt (stream)deleteAgent

How it maps to code

The playground drives the HTTP API, but every step corresponds directly to a call you'd make from the Core SDK or Agent SDK:

Playground step	SDK call
Create sandbox	`client.sandbox({ image, resources })`
Write / read a file	`sb.writeFile(path, content)` / `sb.readFile(path)`
Run a command	`await sb.exec(cmd)` — streamed
Checkpoint	`await sb.checkpoint()`
Fork	`await sb.fork(tag)`
Load an agent	`Alineo.load(spec, { adapter })`
Prompt an agent	`for await (const ev of agent.prompt(text))`