OpenAI SDK

TOKEN is OpenAI-compatible. Most clients work after changing two settings:

Setting	Value
Base URL	`https://api.thetoken.dev/v1`
API key	your `tkn_...` key from the console

Chat Completions

Python
TypeScript

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.thetoken.dev/v1", api_key=os.environ["TOKEN_API_KEY"])

response = client.chat.completions.create(
    model="deepseek/deepseek-v4.1-flash",
    messages=[
        {"role": "system", "content": "Answer in one sentence."},
        {"role": "user", "content": "What is an API gateway?"},
    ],
    max_tokens=2048,
)
print(response.choices[0].message.content)

import OpenAI from 'openai'

const client = new OpenAI({ baseURL: 'https://api.thetoken.dev/v1', apiKey: process.env.TOKEN_API_KEY })

const response = await client.chat.completions.create({
  model: 'deepseek/deepseek-v4.1-flash',
  messages: [
    { role: 'system', content: 'Answer in one sentence.' },
    { role: 'user', content: 'What is an API gateway?' },
  ],
  max_tokens: 2048,
})
console.log(response.choices[0].message.content)

Responses API

/v1/responses works with the same client:

Python
TypeScript

response = client.responses.create(
    model="openai/gpt-oss-120b",
    input="Name three prime numbers.",
)
print(response.output_text)

const response = await client.responses.create({
  model: 'openai/gpt-oss-120b',
  input: 'Name three prime numbers.',
})
console.log(response.output_text)

Environment variables instead of code

The SDKs also read their settings from the environment, which is handy for tools built on them:

export OPENAI_BASE_URL="https://api.thetoken.dev/v1"
export OPENAI_API_KEY="$TOKEN_API_KEY"

Endpoints

/v1/chat/completions: streaming, tool calling, structured output and images on models that support them
/v1/responses
/v1/embeddings (embeddings)
/v1/models (the models your key can use)
the Anthropic-format /v1/messages

Fields and models of each: API reference.

Use TOKEN model IDs such as moonshotai/kimi-k3 or zai-org/glm-5.3. Short aliases such as kimi-k3 work too. OpenAI’s hosted models (for example GPT-5) are not served.

Per-key controls

In the console, under API Keys, each key can have:

a spending limit
an expiry date
a model allow-list
a group (for example the us-eu-hosted privacy group)

Retries

The SDKs retry 429 and 5xx responses twice by default, with backoff. That fits TOKEN’s limits; see Errors and rate limits.