OpenAI-Compatible API Quickstart: cURL, Python, and Node.js

Most API quickstarts stop after printing a model response. A safer first-call workflow also limits the test key, discovers the exact model ID, verifies the request in usage logs, and handles common errors deliberately.

This guide shows that workflow with HangToken. I’m part of the HangToken team, and candid technical feedback is welcome.

Before production use: verify the endpoint, model availability, feature support, and pricing for your own account. Never expose a real API key in screenshots, repositories, browser code, or support messages.

1. Create a scoped API key

1. Sign in to the HangToken dashboard.
2. Open API key management and create a new key.
3. Give it an application-specific name, such as laptop-quickstart-dev.
4. Set a small initial quota and short expiry while testing.
5. Store the key in an environment variable or secret manager.

2. List the models available to your account

export HANGTOKEN_API_KEY="sk-your-key"

curl https://global.hangtoken.com/v1/models \
  -H "Authorization: Bearer $HANGTOKEN_API_KEY"

Copy one exact data[].id value from the response. Do not assume a display name from a pricing page is the API model ID. Availability may differ by account, group, or region.

3. Send one non-streaming request with cURL

export HANGTOKEN_MODEL="model-id-from-v1-models"

curl https://global.hangtoken.com/v1/chat/completions \
  -H "Authorization: Bearer $HANGTOKEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"'"$HANGTOKEN_MODEL"'",
    "messages":[{"role":"user","content":"Reply with: connection successful"}],
    "stream":false
  }'

A successful basic test should return HTTP 200 and a response body containing model output. Then confirm the request appears in your dashboard usage log.

Python example

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["HANGTOKEN_API_KEY"],
    base_url="https://global.hangtoken.com/v1",
)

response = client.chat.completions.create(
    model=os.environ["HANGTOKEN_MODEL"],
    messages=[{"role": "user", "content": "Reply with: connection successful"}],
)

print(response.choices[0].message.content)

Node.js example

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.HANGTOKEN_API_KEY,
  baseURL: "https://global.hangtoken.com/v1",
});

const response = await client.chat.completions.create({
  model: process.env.HANGTOKEN_MODEL,
  messages: [{ role: "user", content: "Reply with: connection successful" }],
});

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

Troubleshooting checklist

Status	Likely cause	What to check
401	Incorrect or disabled key	Recopy the key and verify the Bearer header
402	Insufficient balance or key quota	Check the account balance and key limits
403	Model or group is not allowed	Use an ID returned by /v1/models
404	Incorrect final URL	Avoid a duplicated /v1 path
429	Rate or concurrency limit	Reduce concurrency and use bounded backoff
5xx	Temporary upstream or service error	Record the time/request ID and retry cautiously

A practical first-call checklist

• Use a dedicated, low-limit test key.
• Discover the exact model ID with the same key.
• Start with a non-streaming request.
• Confirm HTTP status and response structure.
• Confirm the call in usage logs.
• Test tool calling, structured output, image input, context limits, and streaming separately before production use.

Ready to test the flow? Start at HangToken Global and tell us where the first-call path breaks.

Comments

[Alex Shev]

The first-call checklist is usefully cautious, especially the distinction between a model display name and the ID returned to the same key. For the troubleshooting section, a concrete addition could be to log the HTTP status, request ID, and a redacted endpoint/model pair together—enough to correlate support incidents without ever capturing Authorization headers or prompt payloads.