silvol 0.2.1

Python SDK for the Silvol inference API — OpenAI-compatible, decentralised GPU.

silvol-python

Python SDK for Silvol — an OpenAI-compatible inference API running on Nosana's decentralised GPU grid.

Drop-in replacement for the OpenAI SDK. Change the base URL and your key; keep the rest of your code.

---

Install

pip install silvol

With optional framework integrations:

pip install silvol[langchain]   # LangChain
pip install silvol[crewai]      # CrewAI
pip install silvol[all]         # both

---

Quickstart

from silvol import Silvol

client = Silvol(api_key="sk-svl-...")          # or set the SILVOL_API_KEY env var

resp = client.chat.completions.create(
    model="DeepSeek-R1-Distill-Qwen-7B",
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

Async:

import asyncio
from silvol import AsyncSilvol

async def main():
    client = AsyncSilvol(api_key="sk-svl-...")
    resp = await client.chat.completions.create(
        model="DeepSeek-R1-Distill-Qwen-7B",
        messages=[{"role": "user", "content": "Hello"}],
        stream=True,
    )
    async for chunk in resp:
        print(chunk.choices[0].delta.content or "", end="", flush=True)

asyncio.run(main())

---

LangChain

from silvol.integrations.langchain import SilvolChat

llm = SilvolChat(api_key="sk-svl-...")
result = llm.invoke("Summarise the Silvol architecture in one sentence.")
print(result.content)

---

CrewAI

from silvol.integrations.crewai import SilvolLLM
from crewai import Agent, Task, Crew

llm = SilvolLLM(api_key="sk-svl-...")

researcher = Agent(
    role="Senior Researcher",
    goal="Uncover groundbreaking technologies in AI",
    backstory="...",
    llm=llm,
)

---

Models

Model ID	Context	Notes
DeepSeek-R1-Distill-Qwen-7B	32k	Always-on — call directly
meta-llama/Llama-3.1-70B-Instruct	128k	On-demand — deploy first
Qwen/Qwen2.5-Coder-32B-Instruct	32k	On-demand — deploy first

DeepSeek-R1-Distill-Qwen-7B is the always-on model returned by GET https://api.silvol.ai/v1/models — call it directly. Other models are served on demand: provision a dedicated GPU deployment first (from the dashboard or the deployments API), then call it by the HuggingFace model ID you deployed.

---

Authentication

Get your API key from the Silvol Dashboard. Keys are prefixed sk-svl-. Pass it as api_key=, or set the SILVOL_API_KEY environment variable (the SDK reads it automatically). OPENAI_API_KEY also works as a fallback.

---

Links

• Docs: silvol.ai/docs
• Dashboard: silvol.ai/dashboard
• PyPI: pypi.org/project/silvol
• GitHub: github.com/optimuscodexprimus/silvol-python