Snowcell Python SDK

Python SDK for the Snowcell API.

This is a lightweight alternative to the OpenAI SDK that can be integrated into your applications.

Install

pip install snowcell

Quickstart

import os
from snowcell import Snowcell

client = Snowcell(api_token=os.environ["SNOWCELL_API_TOKEN"])

res = client.chat.create(
    model="Meditron3-8B",
    messages=[{"role": "user", "content": "Say hello."}],
    max_tokens=50,
)
print(res.choices[0].message.content)

Streaming

from snowcell import Snowcell

client = Snowcell(api_token="YOUR_API_TOKEN")

for event in client.chat.stream(
    model="Meditron3-8B",
    messages=[{"role": "user", "content": "Write one short sentence."}],
    max_tokens=40,
):
    # Each `event` is a dict parsed from SSE "data:" lines
    print(event)

API

Client

from snowcell import Snowcell

client = Snowcell(
  api_token="...", # or set SNOWCELL_API_TOKEN env var
)

Chat

res = client.chat.create(
    model="Meditron3-8B",
    messages=[{"role": "user", "content": "Explain attention in one line."}],
    max_tokens=80,
)
print(res.choices[0].message.content)

Async:

res = await client.chat.acreate(model="Meditron3-8B", messages=[{"role":"user","content":"Hi"}], max_tokens=80)

Streaming:

for evt in client.chat.stream(model="Meditron3-8B", messages=[{"role":"user","content":"Hi"}], max_tokens=80):
    print(evt)

Completions

res = client.completions.create(
    model="Meditron3-8B",
    prompt="Hello world!",
    max_tokens=32,
)
print(res.choices[0].text)

Async:

res = await client.completions.acreate(model="Meditron3-8B", prompt="Hello world!", max_tokens=32)

Configuration

You can set these environment variables instead of passing args:

• SNOWCELL_API_TOKEN – bearer token for authentication
• SNOWCELL_INFERENCE_BASE_URL – override the inference origin (e.g., http://127.0.0.1:11434 for local testing)

The SDK sends User-Agent: snowcell-python/<version> and uses timeouts tuned for LLMs: connect=5s, write=30s, read=300s.

Errors

The SDK raises SnowcellError for 4xx/5xx responses with server details included. Common cases:

• 401/403 – check your token / project permissions
• 422 – request validation error (inspect the server message)
• 429 – rate limited; retry after the indicated delay
• 5xx – transient server error; try again later

Versioning

• Initial public release: 0.1.0 (chat, completions; sync/async; streaming for chat)
• Backwards-compatible fixes → patch bump (0.1.x)
• New features/endpoints → minor bump (0.x.y)

See CHANGELOG.md.

License

MIT – see LICENSE.