serwin-sdk 0.2.1

SDK for Serwin cloud platform worker scripts

serwin_sdk

SDK for writing worker scripts on the Serwin cloud platform. It handles moving data in and out of your job — pulling input from a bucket, publishing results, and deploying trained models — so your script can focus on the actual logic.

Install

pip install serwin_sdk

Quick start

import os
from serwin_sdk import WorkersContext

ctx = WorkersContext(api_key=os.environ.get("API_KEY"))

# pull everything from a bucket into a local folder
ctx.input(bucket="my-input-bucket", dest_folder="input")

# ... your script's own logic goes here ...

# push everything in a local folder back up to a bucket
ctx.publish(dest_folder="output", bucketname="my-output-bucket")

Configuration

WorkersContext reads its settings from environment variables set on the worker VM at provision time. You can also pass any of them explicitly to override the environment:

Env var	Purpose
API_KEY	Authenticates requests to the control plane
CONTROL_PLANE_URL	Base URL of the control plane API
JOB_ID	Identifies the current run
WORK_DIR	Local scratch directory (default /tmp/work)

ctx = WorkersContext(
    control_plane_url=os.environ.get("CONTROL_PLANE_URL"),
    job_id=os.environ.get("JOB_ID"),
    work_dir=os.environ.get("WORK_DIR"),
    api_key=os.environ.get("API_KEY"),
)

API

ctx.input(bucket, dest_folder, filenames=None)

Pulls data from a bucket into dest_folder. Omit filenames to pull the whole bucket; pass a list to pull only those specific files.

ctx.input(bucket="checko19", dest_folder="input")
ctx.input(bucket="checko19", dest_folder="input", filenames=["cars.csv"])

ctx.output(local_path, url)

Uploads a single local file to a presigned upload URL.

ctx.publish(dest_folder, bucketname)

Uploads everything under dest_folder to bucketname, preserving folder structure.

ctx.publish(dest_folder="output", bucketname="checko20")

ctx.deploy(model_path, profile, name=None)

Uploads a trained model artifact (file or folder) and registers it with the control plane as a deployable endpoint.

ctx.deploy(model_path="model", profile="linear-regression", name="cars-price-model")

profile must be one of the platform's known pipeline profiles (e.g. linear-regression, llm-train, llm-infer).

ctx.run_requirements(filepath="requirements.txt")

Installs a requirements.txt next to your script before it runs, if one exists. No-ops if the file isn't found.

License

MIT