@computesdk/arker

Arker provider for ComputeSDK — run code in Arker sandboxed VMs with persistent per-VM filesystems.

Installation

npm install @computesdk/arker

Setup

1. Get your Arker API key (starts with ark_) from arker.ai.
2. Set the environment variable:

export ARKER_API_KEY=ark_live_your_api_key_here

By default the provider targets the aws-us-east-1 region. Select another region with region / ARKER_REGION.

Quick Start

Configure compute with the Arker provider and create a sandbox:

import { compute } from 'computesdk';
import { arker } from '@computesdk/arker';

compute.setConfig({
  provider: arker({ apiKey: process.env.ARKER_API_KEY }),
});

const sandbox = await compute.sandbox.create();

const result = await sandbox.runCommand('node -v');
console.log(result.stdout); // v18.x

await sandbox.destroy();

Or call the provider factory directly when you only need one provider:

import { arker } from '@computesdk/arker';

const sdk = arker({ apiKey: process.env.ARKER_API_KEY });
const sandbox = await sdk.sandbox.create();

How sandboxes are created

Arker disables direct VM creation — every sandbox is forked from a source VM. create() forks the ubuntu-small golden (Node.js + Python preinstalled) by default. The source can be chosen two ways:

• By name — a golden/template, via source in config or templateId in create options.
• By id — a specific existing VM, via snapshotId in create options. The sandboxId returned by create() is itself a valid source, so this branches a sandbox from its current state (Arker's equivalent of a snapshot restore). If both are given, snapshotId wins.

// Default: forks the `ubuntu-small` golden
const sandbox = await sdk.sandbox.create();

// Pick a golden globally for this provider
const heavy = arker({ apiKey, source: 'ubuntu-full' });

// ...or per sandbox, by name
const other = await sdk.sandbox.create({ templateId: 'ubuntu-full' });

// Fork an existing sandbox by id — branches from its current state
const forked = await sdk.sandbox.create({ snapshotId: sandbox.sandboxId });

Configuration

Environment Variables

export ARKER_API_KEY=ark_live_your_api_key_here
export ARKER_REGION=aws-us-east-1     # optional, this is the default
export ARKER_SOURCE=ubuntu-small      # optional default golden

Configuration Options

interface ArkerConfig {
  /** Arker API key (starts with `ark_`). Falls back to ARKER_API_KEY. */
  apiKey?: string;
  /** Region, e.g. `aws-us-east-1`. Falls back to ARKER_REGION, then the us-east-1 default. */
  region?: string;
  /** Golden source VM to fork on create(). Falls back to ARKER_SOURCE, then `ubuntu-small`. */
  source?: string;
}

API Reference

Command Execution

// Shell command
const result = await sandbox.runCommand('echo hello');

// Run a script via heredoc
const result = await sandbox.runCommand(`python3 - <<'PY'
print("Hello from Python")
PY`);

// With environment and working directory
const result = await sandbox.runCommand('npm test', {
  cwd: '/app',
  env: { CI: '1' },
});

runCommand returns { stdout, stderr, exitCode, durationMs }.

Filesystem Operations

Filesystem operations run as shell commands in the VM and work over arbitrary paths.

await sandbox.filesystem.writeFile('/tmp/hello.txt', 'hi');
const content = await sandbox.filesystem.readFile('/tmp/hello.txt');
await sandbox.filesystem.mkdir('/tmp/data');
const entries = await sandbox.filesystem.readdir('/tmp');
const exists = await sandbox.filesystem.exists('/tmp/hello.txt');
await sandbox.filesystem.remove('/tmp/hello.txt');

Sandbox Management

const sandbox = await sdk.sandbox.create();
const info = await sandbox.getInfo();      // { id, provider, status, createdAt, ... }
const all = await sdk.sandbox.list();      // VMs visible to the API key
const existing = await sdk.sandbox.getById(sandbox.sandboxId);
await sandbox.destroy();

Features

• ✅ Command Execution — full shell, Node.js & Python preinstalled (ubuntu-small)
• ✅ Filesystem Operations — persistent per-VM filesystem (read/write/list/mkdir/exists/remove)
• ✅ Fast cold start — forking the default golden boots a sandbox in well under a second
• ✅ Automatic retries — the underlying @arker-ai/sdk retries transient failures (HTTP 429/502/503/504 and transient backend errors)

Limitations

• Creation is fork-only — direct VM creation is disabled; create() forks a golden by name (templateId/source) or an existing VM by id (snapshotId). There's no separate snapshot store, but any sandboxId can be re-forked as a snapshotId.
• getUrl is not supported — Arker does not expose per-port URLs; getUrl throws. VMs forked with network reachability enabled get a stable per-VM hostname — see the Arker SDK fork network options.
• getInfo status — always reports running; idle sandboxes resume automatically on their next command.

Support

• Arker
• Arker SDK
• ComputeSDK Issues

License

MIT