---
id: skops
version: "0.14.0"
license: unclear
license_treatment: unclear
maintenance: active
---

# skops — A set of tools, related to machine learning in production.

License: unclear · Maintenance: active · Downloads: 20.0M/mo

## What it is and what it does

Skops is a Python library that solves the problem of safely sharing and deploying scikit-learn models in production. It provides two main tools: skops.io for secure, pickle-free serialization of estimators, and skops.card for generating model cards that document what a model does and how it should be used. The library is built on top of numpy, scipy, scikit-learn, and packaging, and targets developers and researchers who need to move trained models from development into production environments or share them on platforms like Hugging Face Hub.

The package is actively maintained, supports Python 3.9 through 3.14, and runs on macOS, Windows, and Unix-like systems. It is classified as Beta-stage software and sits in the top 5000 PyPI packages by download volume. The core value proposition is avoiding pickle's security risks while providing a standardized way to document and version models alongside their serialized artifacts.

Use it for:

- Serialize a trained estimator to disk without pickle, then load and deploy it in a web service.
- Generate a model card documenting a model's performance, intended use, and limitations for sharing.
- Store multiple pipeline versions securely in a model registry without trusting pickle deserialization.
- Create reproducible model documentation that includes training data summaries and evaluation metrics.
- Share a trained model with collaborators in a format that is both human-readable and secure.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Skops provides secure persistence and model card generation for scikit-learn estimators, enabling you to save trained models without pickle and document them for production deployment.

Yes, if you work with scikit-learn models in production or need to share them. The low-friction install, active maintenance, and pickle-free persistence address real security and reproducibility concerns. Verify the license before use in proprietary contexts, as it is currently marked unclear in the metadata.

## Install

pip install skops
uv add skops
poetry add skops

## Installing skops

Before you install: Low friction installation with five runtime dependencies (numpy, packaging, prettytable, scikit-learn, scipy). The project is actively maintained with recent commits and a stable release cadence.

License in practice: License treatment is unclear—no SPDX identifier or raw license text is available in the package metadata. Verify the actual license before using in proprietary or restricted contexts.

Quickstart:

pip install skops

import skops.io as sio

model_data = {'trained_model': 'example'}
sio.dump(model_data, 'model.skops')

Requires Python 3.9 or later and scikit-learn as a runtime dependency.

Verify before relying:

- Exact license identifier and terms—metadata shows 'unclear' treatment with no SPDX or raw license field populated.
- Whether model cards generated by skops.card are compatible with all Hugging Face Hub model metadata schemas.
- Performance overhead or compatibility constraints when persisting complex pipelines with custom transformers.

## Package facts

- License: not declared (unclear)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 20.0M/month (top 5,000 on PyPI)
- Known vulnerabilities: none known

## Tags

scikit-learn model persistence, sklearn model serialization, secure model storage, model card generation, ml model documentation, pickle-free model saving, production model deployment, ml-deployment, model-serialization, sklearn-tools

[View on SkillFed](https://skillfed.io/packages/skops) · [View on PyPI](https://pypi.org/project/skops/)