DocumentationAPI v1Preview

Build on Nexpod.

From an API key to a running server: nexctl, the REST API at api.nexpod.ai/v1 and the SDKs.

⌘ K

nexctl — first server

$ nexctl login

✓ logged in · project ml-research

$ nexctl servers create --family h100 --region eu-ams-1

→ reserving 8× H100 in eu-ams-1

→ health checks 8/8 GPUs

✓ ready in 44s · billing per second

$ nexctl servers ssh h100-ams-01

1 · log in2 · create3 · connect

Quickstart

Four commands from nothing to GPUs. Each step bills by the second and stops when you stop it.

01~1 min
Install and log in
One binary for macOS, Linux and Windows.
shell
```
brew install nexpod/tap/nexctl
nexctl login
```
02~1 min
Start a server
A whole 8-GPU machine, ready over SSH.
shell
```
nexctl servers create \
  --family h100 --region eu-ams-1
```
03~1 min
Deploy an endpoint
Serverless workers that scale to zero.
shell
```
nexctl endpoints create \
  --image chat-13b:1.4 --family h100
```
04~1 min
Launch a cluster
Several nodes on one fabric for training.
shell
```
nexctl clusters create \
  --nodes 4 --family h200
```

Install nexctl

nexctl wraps the whole API. It reads your key from NEXPOD_API_KEY or from the profile that nexctl login writes.

1brew install nexpod/tap/nexctl
2nexctl version   # nexctl 2.0.0

Authenticate

Create a key in the console under Settings, then export it. Keys are scoped to one project and can be limited to read-only.

.env

1export NEXPOD_API_KEY="npk_live_…"
2nexctl whoami   # ml-research · eu-ams-1

Products

Overview pages for each product, with the concepts the API uses.

gpus: 8× B300
status: ready
rate: €48.00/hr

servers

Cloud Servers

A whole 8-GPU machine over SSH, billed from ready to stop.

workers · 24 hmin 0 · max 24

endpoints

Serverless

Workers that follow the queue and scale to zero.

clusters

Instant Clusters

Several 8-GPU nodes on one fabric for training.

VLvllm-openaillama 70B · FP8deploy
COcomfyuiimage pipelinesdeploy
PYpytorch2.9 · cuda 13deploy

templates

Nexpod Hub

Ready-to-deploy templates and open models.

API reference

REST over HTTPS, JSON in and out. Every request needs a bearer token; every response carries a request id.

Base URL https://api.nexpod.ai/v1JSON

GET/v1/serversList servers in the project
POST/v1/serversCreate a server from a family and region
DELETE/v1/servers/{id}Stop billing and hand the node back
POST/v1/endpoints/{id}/runSend a job to a Serverless endpoint
POST/v1/clustersCreate a multi-node cluster
GET/v1/billing/usageSeconds and euros per resource

1curl https://api.nexpod.ai/v1/servers \
2  -H "Authorization: Bearer $NEXPOD_API_KEY" \
3  -H "Content-Type: application/json" \
4  -d '{"family": "B300", "region": "eu-ams-1"}'

SDKs and tools

brew install nexpod/tap/nexctl
nexctl servers list -o json | jq '.[].status'

CLI · v2.0.0

nexctl

Servers, endpoints and clusters from one binary. JSON output on every command with -o json.

import nexpod
nexpod.servers.create(family="h100")

Python

nexpod

Typed client with waiters, retries and a Serverless handler decorator.

import { Nexpod } from "@nexpod/sdk";
const np = new Nexpod();
await np.servers.list();

TypeScript

@nexpod/sdk

Promise-based client for Node and edge runtimes, with full types.

Guides

Longer walkthroughs from the blog.

Help

Changeloglatest 3

# add it to your ticket
x-request-id: req_01JB7Q4M2X

Talk to support

Questions about a server or a bill go to the team that runs the hardware.

Building something new?Startup programCredits and a direct line to the team while you build.

Docs preview

This landing shows the structure of the docs. Full reference pages follow.

Ready

Key in hand? Start your first server.

Pick a server, pick a region, and connect over SSH. Stop it whenever you are done; the meter stops with it.

Get started Talk to sales

On-demand ratesEUR · excl. VAT

B300· 288 GB HBM3e€6.00/GPU·hr
B200· 180 GB HBM3e€5.00/GPU·hr
H200· 141 GB HBM3e€4.00/GPU·hr
H100· 80 GB HBM3€3.00/GPU·hr

Build on Nexpod.

Install and log in

Start a server

Deploy an endpoint

Launch a cluster

Key in hand? Start your first server.