Safeera Cloud V2 SDK — GPU Buyer Guide

Programmatic SDK and REST API for provisioning live 30-second heartbeat-verified NVIDIA Blackwell RTX PRO 6000 GPUs (0.5x 48 GB CUDA MPS, 1x 96 GB, and 2x 192 GB) with Safeera's 6-Stage Ephemeral-OS + Persistent /data architecture.

Live 30s-Verified GPU Inventory

listAvailableConfigs() returns strictly configurations backed by physical hosts verified online within the last 30 seconds. During Beta, all compute configurations default to $0.00 / GPU-hr.

1-Click or Custom SSH Keys

Pass your own ssh-ed25519 public key, or set generateKeypair: true to have Safeera generate a fresh keypair and return generatedPrivateKey once for immediate SSH login.

Ephemeral OS + Persistent /data

Every start and reboot boots a pristine OS root from the pre-built GPU template and re-runs Stage-1 & Stage-2 init while keeping your persistent /data volume intact.

1. Quickstart (Node.js ESM SDK)

import { SafeeraClient } from '@safeera/sdk';

const client = new SafeeraClient({
  baseUrl: 'https://master.safeera.ai',
  origin: 'https://master.safeera.ai'
});

// Authenticate session
await client.authenticate({ username: 'owner', password: process.env.SAFEERA_PASSWORD });

// 1. List strictly live, 30s-verified GPU configurations ($0.00/hr Beta default)
const configs = await client.listAvailableConfigs();
console.table(configs);

// 2. Option A: Generate a fresh Ed25519 SSH keypair on the platform (private key returned once)
const { profile, generatedPrivateKey } = await client.createSshProfile({
  name: 'my-auto-key',
  username: 'safeera',
  generateKeypair: true
});
console.log('Save your private key (.pem):\n', generatedPrivateKey);

// 3. Launch 1x2 GPU (192 GB VRAM), 2x1 GPU (96 GB each), or 4x0.5 GPU (48 GB CUDA MPS each)
const launch = await client.launchInstance({
  name: 'trainer-blackwell',
  family: 'rtxpro',
  region: 'eu-north',
  count: 2,          // 0.5 (48 GB) | 1 (96 GB) | 2 (192 GB)
  quantity: 1,
  profileId: profile.id,
  diskGib: 64
});

const [vm] = await client.waitForReady(launch.createdIds);
console.log('SSH ready:', vm.sshCommand);

2. SSH Key Options (Paste Public Key vs. Auto-Generate Keypair)

Mode	Request Payload	Behavior
User-Supplied Public Key	`{ username: "safeera", publicKey: "ssh-ed25519 AAAA..." }`	Validates public key format (rejects private keys) and injects into `~/.ssh/authorized_keys` during Stage-1 boot.
Platform-Generated Keypair	`{ username: "safeera", generateKeypair: true }`	Generates a fresh `ed25519` keypair, injects the public key into the VM, and returns `generatedPrivateKey` once in the response for `.pem` download.

3. 6-Stage Ephemeral-OS VM Lifecycle Reference

Action	SDK Method / REST Endpoint	OS Root Disk Behavior	Persistent `/data` Volume Behavior
Launch	`client.launchInstance(opts)` `POST /api/compute/launch`	Deploys fresh OS root from pre-built GPU template + runs Stage-1 init (`/var/lib/safeera/stage1.done`).	Creates/formats ext4 volume, mounts at `/data`, runs `/data/.safeera/on-restore.sh` (`stage2.done`).
Stop	`client.machineAction(id, 'stop')` `POST /api/compute/machines/:id`	Gracefully syncs & unmounts `/data`, then destroys/expunges the OS root VM (compute billing stops).	Detached and preserved intact in storage pool.
Start	`client.machineAction(id, 'start')` `POST /api/compute/machines/:id`	Deploys a brand-new OS root from the pre-built template and re-runs Stage-1 init.	Re-attaches existing `/data` volume and re-runs Stage-2 `on-restore.sh`.
Reboot	`client.machineAction(id, 'reboot')` `POST /api/compute/machines/:id`	Replaces OS root from pre-built template (new `boot_id`) and re-runs Stage-1 init.	Preserves and re-mounts `/data`, re-running Stage-2 init.
Snapshot	`client.machineAction(id, 'snapshot', { snapshotName })`	No impact on running OS.	Records point-in-time snapshot of persistent `/data` volume.
Delete	`client.machineAction(id, 'delete', { deleteDisk: true })`	Expunges OS root VM and releases NAT/elastic IP rules.	Deletes volume if `deleteDisk: true`, or keeps detached volume if `false`.

4. Verified Hardware Profiles (Helsinki `eu-north`)

Profile	GPU Count	VRAM per VM	Isolation Mode	Beta Default Price
`1x2 GPU`	`2`	192 GiB	Full dual-GPU allocation	$0.00 / hr
`2x1 GPU`	`1`	96 GiB	Dedicated single-GPU allocation	$0.00 / hr
`4x0.5 GPU`	`0.5`	48 GiB (49,152 MiB)	CUDA MPS 50% active thread & 48 GiB memory cap	$0.00 / hr