Openmed

OpenMed is an open source clinical AI toolkit for detecting and removing protected health information from medical text, running entirely on the user's own hardware. The core is a Python toolkit with APIs for extracting and redacting personal identifiers, backed by encoder transformer models fine tuned for token classification. Because the models classify tokens rather than generate text, there is nothing to hallucinate; each entity comes with a confidence score, and redaction can be done by masking, hashing, date shifting or synthetic replacement. It is aimed at hospitals, clinicians, healthcare organizations and developers who need privacy preserving text processing without sending records to a cloud vendor. OpenMed detects more than 55 PHI types, including all 18 HIPAA Safe Harbor identifiers, across 34 model backed languages, with models ranging from 33 million to 568 million parameters across 13 biomedical domains. OpenMedKit brings the same capability to Swift and Kotlin mobile apps, a WebGPU build performs redaction inside the browser, and REST and gRPC services can be deployed in a private VPC. Signed audit reports and policy aware release gates support compliance reviews. Welna, an iOS app for personal health signal analysis, and OpenMed Agent, a preview for clinical workflow automation, sit alongside the toolkit. There is no charge. OpenMed is released under the Apache 2.0 license with the model weights included, and it requires no license key, API key or subscription.