Using the OpenAI SDK | Tutujin API Docs

The official OpenAI SDKs need two changes: set base_url to https://open.tutujin.com/v1 and use your Tutujin key as api_key. One client covers chat, Responses, embeddings, images and audio. To switch models you only change model, and that includes non-GPT models such as Claude, Gemini and DeepSeek wherever the catalog lists Chat Completions support.

Install

pip install openai

npm install openai

Create a client

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://open.tutujin.com/v1",
    api_key=os.environ["TUTUJIN_API_KEY"],
)

import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'https://open.tutujin.com/v1',
  apiKey: process.env.TUTUJIN_API_KEY,
})

Or leave the code alone and set environment variables — the SDKs read OPENAI_BASE_URL and OPENAI_API_KEY by default:

cURL

export OPENAI_BASE_URL="https://open.tutujin.com/v1"
export OPENAI_API_KEY="sk-your-key"

Chat

resp = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[
        {"role": "system", "content": "You are a concise assistant."},
        {"role": "user", "content": "Explain RAG in three sentences."},
    ],
)
print(resp.choices[0].message.content)
print(resp.usage)  # input and output tokens for this call

const resp = await client.chat.completions.create({
  model: 'gpt-5.4-mini',
  messages: [
    { role: 'system', content: 'You are a concise assistant.' },
    { role: 'user', content: 'Explain RAG in three sentences.' },
  ],
})
console.log(resp.choices[0].message.content)
console.log(resp.usage) // input and output tokens for this call

Parameters are covered in Chat Completions.

Streaming

stream = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[{"role": "user", "content": "Write a 60-word product description"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

const stream = await client.chat.completions.create({
  model: 'gpt-5.4-mini',
  messages: [{ role: 'user', content: 'Write a 60-word product description' }],
  stream: true,
})
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '')
}

Embeddings

emb = client.embeddings.create(
    model="text-embedding-3-small",
    input=["What is the return policy?", "How soon do orders ship?"],
)
print(len(emb.data), len(emb.data[0].embedding))

const emb = await client.embeddings.create({
  model: 'text-embedding-3-small',
  input: ['What is the return policy?', 'How soon do orders ship?'],
})
console.log(emb.data.length, emb.data[0].embedding.length)

Passing an array as input handles many texts in one request — faster than one request per text and less likely to hit rate limits.

Calling other vendors' models

Keep the client, change model. Any model the catalog lists for Chat Completions works this way:

Python

for model in ["claude-sonnet-5-5", "deepseek-v4.1-flash"]:
    resp = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "Hello"}],
    )
    print(model, resp.choices[0].message.content)

Vendors support optional parameters to different degrees. For native Claude features, use the Anthropic SDK with the native Messages endpoint.

Timeouts and retries

The SDKs retry 429s, 5xx responses and network errors automatically, twice by default. In production, tune retries and timeouts for your workload:

client = OpenAI(
    base_url="https://open.tutujin.com/v1",
    api_key=os.environ["TUTUJIN_API_KEY"],
    max_retries=4,
    timeout=60.0,  # seconds
)

const client = new OpenAI({
  baseURL: 'https://open.tutujin.com/v1',
  apiKey: process.env.TUTUJIN_API_KEY,
  maxRetries: 4,
  timeout: 60_000, // ms
})

An empty balance returns 403, which the SDK does not retry; alert on it in your own code. Details in Errors and retries.

Async calls (Python)

For lots of concurrent requests use AsyncOpenAI; it takes the same options:

Python

import asyncio
import os

from openai import AsyncOpenAI

client = AsyncOpenAI(base_url="https://open.tutujin.com/v1", api_key=os.environ["TUTUJIN_API_KEY"])


async def main():
    resp = await client.chat.completions.create(
        model="gpt-5.4-mini",
        messages=[{"role": "user", "content": "Hello"}],
    )
    print(resp.choices[0].message.content)


asyncio.run(main())

For choosing a concurrency cap, see Rate limits.

Troubleshooting

Symptom	Cause and fix
404	`base_url` is missing `/v1`; it should be `https://open.tutujin.com/v1`
401	Wrong or invalid key; check the environment variable holds the full key
503 with `model_not_found`	The key's group lacks this model; switch groups or check the name — see Choosing a group
Slow responses or timeouts	Stream long outputs and raise `timeout`

Still stuck? Add our WeChat support JsnonoChat and send the request id from the error message, and we will look into it.

Install

pip install openai

npm install openai

Create a client

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://open.tutujin.com/v1",
    api_key=os.environ["TUTUJIN_API_KEY"],
)

import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'https://open.tutujin.com/v1',
  apiKey: process.env.TUTUJIN_API_KEY,
})

Or leave the code alone and set environment variables — the SDKs read OPENAI_BASE_URL and OPENAI_API_KEY by default:

cURL

export OPENAI_BASE_URL="https://open.tutujin.com/v1"
export OPENAI_API_KEY="sk-your-key"

Chat

resp = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[
        {"role": "system", "content": "You are a concise assistant."},
        {"role": "user", "content": "Explain RAG in three sentences."},
    ],
)
print(resp.choices[0].message.content)
print(resp.usage)  # input and output tokens for this call

const resp = await client.chat.completions.create({
  model: 'gpt-5.4-mini',
  messages: [
    { role: 'system', content: 'You are a concise assistant.' },
    { role: 'user', content: 'Explain RAG in three sentences.' },
  ],
})
console.log(resp.choices[0].message.content)
console.log(resp.usage) // input and output tokens for this call

Parameters are covered in Chat Completions.

Streaming

stream = client.chat.completions.create(
    model="gpt-5.4-mini",
    messages=[{"role": "user", "content": "Write a 60-word product description"}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="", flush=True)

const stream = await client.chat.completions.create({
  model: 'gpt-5.4-mini',
  messages: [{ role: 'user', content: 'Write a 60-word product description' }],
  stream: true,
})
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '')
}

Embeddings

emb = client.embeddings.create(
    model="text-embedding-3-small",
    input=["What is the return policy?", "How soon do orders ship?"],
)
print(len(emb.data), len(emb.data[0].embedding))

const emb = await client.embeddings.create({
  model: 'text-embedding-3-small',
  input: ['What is the return policy?', 'How soon do orders ship?'],
})
console.log(emb.data.length, emb.data[0].embedding.length)

Passing an array as input handles many texts in one request — faster than one request per text and less likely to hit rate limits.

Calling other vendors' models

Keep the client, change model. Any model the catalog lists for Chat Completions works this way:

Python

for model in ["claude-sonnet-5-5", "deepseek-v4.1-flash"]:
    resp = client.chat.completions.create(
        model=model,
        messages=[{"role": "user", "content": "Hello"}],
    )
    print(model, resp.choices[0].message.content)

Vendors support optional parameters to different degrees. For native Claude features, use the Anthropic SDK with the native Messages endpoint.

Timeouts and retries

The SDKs retry 429s, 5xx responses and network errors automatically, twice by default. In production, tune retries and timeouts for your workload:

client = OpenAI(
    base_url="https://open.tutujin.com/v1",
    api_key=os.environ["TUTUJIN_API_KEY"],
    max_retries=4,
    timeout=60.0,  # seconds
)

const client = new OpenAI({
  baseURL: 'https://open.tutujin.com/v1',
  apiKey: process.env.TUTUJIN_API_KEY,
  maxRetries: 4,
  timeout: 60_000, // ms
})

An empty balance returns 403, which the SDK does not retry; alert on it in your own code. Details in Errors and retries.

Async calls (Python)

For lots of concurrent requests use AsyncOpenAI; it takes the same options:

Python

import asyncio
import os

from openai import AsyncOpenAI

client = AsyncOpenAI(base_url="https://open.tutujin.com/v1", api_key=os.environ["TUTUJIN_API_KEY"])


async def main():
    resp = await client.chat.completions.create(
        model="gpt-5.4-mini",
        messages=[{"role": "user", "content": "Hello"}],
    )
    print(resp.choices[0].message.content)


asyncio.run(main())

For choosing a concurrency cap, see Rate limits.

Troubleshooting

Symptom	Cause and fix
404	`base_url` is missing `/v1`; it should be `https://open.tutujin.com/v1`
401	Wrong or invalid key; check the environment variable holds the full key
503 with `model_not_found`	The key's group lacks this model; switch groups or check the name — see Choosing a group
Slow responses or timeouts	Stream long outputs and raise `timeout`