Documentation

Your deployment speaks the OpenAI Chat Completions API on its own address. Change the base URL; change nothing else.

Authentication

Inference requests require a Bearer token in the Authorization header. That token belongs to the deployment you are calling — it authenticates you to your own machine, and it is not a Ratel API key.

Your access token: The portal shows it once, when you deploy. Copy it then: we do not keep it, so we cannot show it again.

API endpoint

The base URL is the deployment's own address — your GPU, serving vLLM directly. Ratel operates no inference endpoint, so requests go from your client to your machine and nothing of ours sits between them. The chat completions path follows OpenAI's standard:

POST /v1/chat/completions

It accepts the same request format as OpenAI's Chat Completions API, making it drop-in compatible with existing code. You may also point your own domain at the machine and terminate TLS with your own certificate.

Quickstart

Choose your language below to see how to call your deployment. Replace YOUR_API_KEY with the access token you copied when you deployed, and YOUR_BASE_URL with the deployment's address, which is on its page in the portal.

curl -X POST YOUR_BASE_URL/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "your-model-key",
    "messages": [
      {"role": "system", "content": "You are a helpful assistant."},
      {"role": "user", "content": "Hello, how are you?"}
    ],
    "temperature": 0.7,
    "max_tokens": 500
  }'
from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="YOUR_BASE_URL/v1"
)

response = client.chat.completions.create(
    model="your-model-key",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello, how are you?"}
    ],
    temperature=0.7,
    max_tokens=500
)

print(response.choices[0].message.content)
import OpenAI from 'openai';

const client = new OpenAI({
  apiKey: 'YOUR_API_KEY',
  baseURL: 'YOUR_BASE_URL/v1',
});

const response = await client.chat.completions.create({
  model: 'your-model-key',
  messages: [
    { role: 'system', content: 'You are a helpful assistant.' },
    { role: 'user', content: 'Hello, how are you?' }
  ],
  temperature: 0.7,
  max_tokens: 500,
});

console.log(response.choices[0].message.content);

Rate limits

Ratel applies no rate limit to inference, because Ratel is not in the path to apply one. The machine is dedicated to you and serves until its own capacity is reached; throughput and concurrency are properties of the GPU and the model you chose, not of a plan tier. Rate limits do apply to the management API in the portal.

Supported models

Each deployment serves the one model you deployed to it. Query it with GET /v1/models against that deployment's own address — vLLM answers it directly. To see everything running across your workspace, use the portal or the management API.

GET /v1/models

Need help?

There is no ticket system yet. Write to us and a person answers — usually the one who wrote the thing you are asking about.