maziyarpanahi/openmed

Local-first healthcare AI: clinical NER & HIPAA PII de-identification that runs 100% on-device. 2,200+ medical models, 21 languages, Apple MLX + Python, no cloud, no patient data leaving your network. Apache-2.0

What it solves

OpenMed addresses the privacy and cost challenges of processing sensitive medical data in the cloud. It allows users to extract biomedical entities and remove personally identifiable information (PII) without ever uploading patient data to an external server.

How it works

The project provides a suite of specialized medical models (over 2,000 available) that run locally on the user's hardware. It supports various backends including Apple Silicon (via MLX), NVIDIA GPUs (via CUDA), CPUs, and even web browsers (via Transformers.js). It can perform tasks like Named Entity Recognition (NER) for diseases, drugs, and anatomy, as well as HIPAA-aware de-identification through masking, replacement, or hashing.

Who it’s for

  • Healthcare developers building mobile apps (iOS/Android) that require offline clinical intelligence.
  • Data scientists needing to de-identify large clinical datasets while maintaining data residency.
  • Researchers working with sensitive medical text in air-gapped environments.

Highlights

  • 100% On-device: No data leaves the local device or network.
  • Cross-platform support: Runs on iOS, iPadOS, macOS, Android, and in the browser.
  • Specialized Medical Models: Includes models for disease detection, pharmacology, anatomy, and genes.
  • HIPAA-aware: Capable of identifying and redacting all 18 Safe Harbor identifiers.
  • High Performance: Optimized for Apple Silicon using MLX, offering significant speedups over CPU-only execution.