Quickstart

Two paths: launch a GPU deployment in your cloud account, or register an endpoint you already have.

1. Create an account and an API key

Sign up, open API keys in the console and create a key. Copy it now; it is shown once.

2a. Launch a model on an NVIDIA GPU

1. Follow cloud account setup to create a scoped access key and check your GPU quota.
2. For gated models (Llama, Mistral, Gemma) accept the license on Hugging Face and add your token under Settings → Secrets.
3. Open Deployments → New, choose a model, and accept the recommended instance or pick another one that fits.
4. Choose “Launch in my account”. The stack takes 2–4 minutes; the first boot then downloads the model (5–20 minutes).
5. When the status turns to Ready, try it in the Playground.

Prefer to launch it yourself? Choose “Generate template only”, download the CloudFormation template, create the stack in your console, then paste the Endpoint output into the deployment page.

2b. Register an existing endpoint

Any server that speaks the OpenAI API works: your own vLLM, TGI or Triton OpenAI frontend, or a hosted model from the NVIDIA API catalog (https://integrate.api.nvidia.com/v1 with an nvapi- key). Open Deployments → New → External endpoint, enter the base URL, the upstream model name and its key.

3. Call the API

from openai import OpenAI

client = OpenAI(base_url="https://awsbee.com/api/v1", api_key="bee_sk_...")

stream = client.chat.completions.create(
    model="llama-3-1-8b-instruct",   # your deployment name
    messages=[{"role": "user", "content": "Write a haiku about GPUs."}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://awsbee.com/api/v1", apiKey: process.env.BEE_API_KEY });

const res = await client.chat.completions.create({
  model: "llama-3-1-8b-instruct",
  messages: [{ role: "user", content: "Hello!" }],
});
console.log(res.choices[0].message.content);

Next: the API reference.