OpenMed 1.0.0 ships

OpenMed 1.0.0 is out, and it adds a public Apple stack for running medical text scanning directly inside iPhone and Mac apps. (github.com) (sourceforge.net) The April 14 release bundles four pieces into one version: the Python package, Apple Silicon support through Machine Learning eXchange, a public Swift package called OpenMedKit, and updated Apple-platform docs. (sourceforge.net) OpenMedKit is distributed through Swift Package Manager from version 1.0.0 of the GitHub repository, and its docs list iOS 16 or later, macOS 13 or later, and Xcode 15 or later as requirements. (openmed.life) This software is built for named-entity recognition, which is a way of teaching a model to tag spans of text such as names, dates of birth, or Social Security numbers. OpenMed’s Swift example shows an app detecting “John Doe” and “123-45-6789” from a sentence on-device. (openmed.life) For hospitals, insurers, and health app developers, that changes where scanning can happen. Instead of sending text to a remote server first, developers can bundle a converted Core Machine Learning model and a label file inside the app package. (openmed.life 1) (openmed.life 2) Machine Learning eXchange, or MLX, is Apple’s machine-learning framework for Apple silicon, and Apple says it is optimized for the unified memory architecture used in those chips. Apple also publishes Swift bindings, which lets developers use the same family of tools from native Swift apps. (opensource.apple.com) (mlx-framework.org) OpenMed says its MLX backend can automatically convert supported Hugging Face token-classification models on first use and cache them locally, while keeping the same entity output format as its standard Hugging Face backend. (openmed.life) The 1.0.0 release notes also draw a limit around the launch. OpenMed says Apple packaging is “still being expanded,” and that some model families are not yet part of the MLX rollout. (sourceforge.net) The project’s broader pitch is open-source clinical natural-language processing under the Apache 2.0 license, with model discovery, de-identification tools, and application programming interfaces for batch and service use. The repository says the package supports all 18 Health Insurance Portability and Accountability Act Safe Harbor identifiers in its personally identifiable information workflow. (github.com) (openmed.life) (github.com) In practice, OpenMed 1.0.0 turns a privacy promise into a shipping package: developers can now pull a Swift dependency, compile a model, and run medical text redaction inside the app instead of routing that text through the cloud first. (openmed.life) (sourceforge.net)