qvac-sdk 0.1.0

Python SDK for Tether QVAC — local-first AI inference on any device

qvac-sdk

Python SDK for Tether QVAC — local-first AI inference on any device.

QVAC runs AI models entirely on-device with no cloud dependency. This SDK provides a Pythonic interface to QVAC's inference capabilities: LLM completions, image generation, speech-to-text, translation, TTS, OCR, and embeddings.

Installation

pip install qvac-sdk

With OpenAI-compatible client support:

pip install qvac-sdk[openai]

Quick start

Direct API

from qvac_sdk import QVACClient

client = QVACClient()  # connects to local QVAC instance

# LLM completion
response = client.complete("Explain quantum computing in one paragraph")
print(response.text)

# Streaming
for chunk in client.complete_stream("Write a haiku about edge AI"):
    print(chunk.text, end="")

Async

from qvac_sdk import AsyncQVACClient

async def main():
    client = AsyncQVACClient()
    response = await client.complete("Hello from async Python")
    print(response.text)

OpenAI-compatible mode

QVAC exposes an OpenAI-compatible API via qvac serve openai. This SDK wraps it natively:

from qvac_sdk.openai_compat import QVACOpenAI

client = QVACOpenAI()  # points to local QVAC OpenAI endpoint

response = client.chat.completions.create(
    model="local",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Speech-to-text

transcript = client.transcribe("meeting.wav")
print(transcript.text)

Image generation

image = client.generate_image("A cat riding a bicycle", width=512, height=512)
image.save("cat.png")

Embeddings

vectors = client.embed(["sentence one", "sentence two"])
print(vectors.shape)  # (2, 384)

Configuration

client = QVACClient(
    base_url="http://localhost:8080",  # QVAC instance URL
    timeout=30.0,                       # request timeout in seconds
    model="default",                    # model name override
)

Environment variables:

Variable	Default	Description
QVAC_BASE_URL	http://localhost:8080	QVAC API endpoint
QVAC_TIMEOUT	30.0	Request timeout (seconds)
QVAC_MODEL	default	Default model name

Supported capabilities

Capability	Method	Status
LLM completion	complete() / complete_stream()	Planned
Chat (multi-turn)	chat() / chat_stream()	Planned
Speech-to-text	transcribe()	Planned
Translation	translate()	Planned
Text-to-speech	speak()	Planned
Image generation	generate_image()	Planned
OCR	ocr()	Planned
Embeddings	embed()	Planned
LoRA fine-tuning	finetune()	Planned
OpenAI compat	qvac_sdk.openai_compat	Planned

Development

git clone https://github.com/kinance/qvac-sdk
cd qvac-sdk
pip install -e ".[dev]"
ruff check .
mypy src/
pytest

License

Apache 2.0 — same as QVAC.