datalabrotterdam-nova-sdk 1.5.2

Python client for the Nova AI API

Python SDK

Python client for the Nova AI API.

The package is intentionally small:

• sync methods
• no external dependencies
• typed request and response structures
• simple streaming support for chat completions with a normal for loop
• audio helpers for transcription, translation, and speech generation

Install

pip install datalabrotterdam-nova-sdk

Create A Client

from nova_ai import NovaAIClient

client = NovaAIClient(
    api_key="sk_live_..."
)

If you do not pass base_url, the client uses https://api.nova.datalabrotterdam.nl/v1.

Example: Normal Chat

from nova_ai import NovaAIClient

client = NovaAIClient(
    api_key="sk_live_..."
)

completion = client.chat.completions.create({
    "model": "llama3.2",
    "messages": [
        {"role": "system", "content": "You are helpful."},
        {"role": "user", "content": "Explain embeddings in one sentence."}
    ],
    "max_tokens": 256,
    "temperature": 0.7,
})

print(completion["choices"][0]["message"]["content"])
print(completion.get("usage_source"))
print(completion.get("metrics", {}).get("tokens_per_second"))

completion["usage_source"] is:

• "provider" when the upstream provider reported usage
• "gateway_estimated" when Nova AI estimated usage to keep the contract consistent

Example: Streaming Chat

from nova_ai import NovaAIClient, collect_chat_text

client = NovaAIClient(
    api_key="sk_live_..."
)

stream = client.chat.completions.stream({
    "model": "llama3.2",
    "messages": [{"role": "user", "content": "Write three short tips."}]
})

for event in stream:
    if event["type"] == "chunk":
        choices = event["data"].get("choices", [])
        if choices:
            print(choices[0].get("delta", {}).get("content", ""), end="")

print()

print(collect_chat_text(client.chat.completions.stream({
    "model": "llama3.2",
    "messages": [{"role": "user", "content": "Summarize vectors in one line."}]
})))

Example: Embeddings

from nova_ai import NovaAIClient

client = NovaAIClient(
    api_key="sk_live_..."
)

embedding = client.embeddings.create({
    "model": "nomic-embed-text",
    "input": "Nova AI"
})

print(len(embedding["data"][0]["embedding"]))

Example: Audio Transcription

from nova_ai import NovaAIClient

client = NovaAIClient(
    api_key="sk_live_..."
)

transcript = client.audio.transcriptions.create(
    file_path="./sample.wav",
    model="whisper-1",
    language="nl",
)

print(transcript["text"])

Example: Speech Synthesis

from nova_ai import NovaAIClient

client = NovaAIClient(
    api_key="sk_live_..."
)

audio = client.audio.speech.create({
    "model": "Qwen/Qwen3-TTS-12Hz-1.7B-Base",
    "input": "Hello from Nova AI",
    "task_type": "Base",
    "ref_audio": "https://example.com/reference.wav",
    "ref_text": "Exact transcript of the reference audio.",
    "response_format": "wav",
})

with open("speech.wav", "wb") as target:
    target.write(audio)

Included Methods

• client.providers.list()
• client.models.list(limit=None, after=None)
• client.chat.completions.create(payload, config_id=None, request_id=None)
• client.chat.completions.stream(payload, config_id=None, request_id=None)
• client.embeddings.create(payload, config_id=None, request_id=None)
• client.audio.transcriptions.create(file_path=..., model=..., ...)
• client.audio.translations.create(file_path=..., model=..., ...)
• client.audio.speech.create(payload, config_id=None, request_id=None)
• client.audio.voices.list(config_id=None, request_id=None)

Audio Notes

• client.audio.transcriptions.create() and client.audio.translations.create() upload audio as multipart/form-data.
• client.audio.speech.create() returns raw bytes.
• client.audio.voices.list() only exposes non-uploaded voices returned by the gateway.
• The gateway enforces runtime scopes:
  • audio:transcribe
  • audio:translate
  • audio:synthesis

Why streaming uses for, not async for

For this Python SDK, a normal iterator is the simplest way to support streaming without extra dependencies. That means you can stream like this:

for event in client.chat.completions.stream({...}):
    ...

Using async for is also possible, but it usually means introducing an async HTTP client such as httpx or aiohttp, which makes the first version of the SDK heavier.