- Category
- AI Tools
- Rank
- No. 823Tools index
Previous survey · No. 807 ·
- Type
- APP
- Builder
- maziyarpanahi
- GitHub
- 5.3k stars
- Latest release
- v2.5.0
- Date
About
Local-first healthcare AI — clinical NER and HIPAA PII de-identification that runs on your machine.
What it does
OpenMed turns clinical notes and supported documents into medical entities, PII findings, and redacted or synthetic text. It loads selected model artifacts, then runs through Python, mobile, browser, batch, or service interfaces. Offline deployments can prefetch artifacts and block outbound sockets during processing.
Why it's ranked here
OpenMed stands out for unusually broad deployment coverage and serious privacy workflow support. It spans scripts, services, mobile apps, browsers, structured health formats, policy profiles, and audit evidence. The breadth brings configuration burden, and the documentation contains conflicting language counts, so production teams must verify their exact path.
What's good
The package separates a small core from optional extras for model runtimes, services, OCR, acceleration, and integrations. It supports CPU, CUDA, Apple MLX, ONNX mobile inference, and browser execution. De-identification offers masking, removal, replacement, hashing, and date shifting, with policy profiles, leakage metrics, audit reports, and offline enforcement.
Tradeoffs
Initial setup may require network access for model artifacts, and model or dataset terms can differ from the SDK license. Runtime support varies by environment and artifact. Some language routes need user-supplied weights, while Russian and Tamil use placeholder routes rather than dedicated bundled models. Documentation claims 33, 34, and 35 supported PII languages in different places.
How to use it well
Use OpenMed when a healthcare engineering team needs controlled extraction or de-identification across notebooks, batch jobs, internal services, mobile apps, or browser tools. Prefetch and validate models with synthetic fixtures, choose the smallest installation extra, and test leakage against local data. It does not replace clinical fitness assessment, expert privacy review, legal advice, or compliance controls.
Technical notes+
pyproject.toml defines a Hatchling-built Python 3.10+ package, Apache-2.0 licensing, console entries openmed and openmed-mcp, a minimal base dependency set, and many opt-in extras including hf, service, mlx, onnx-runtime, multimodal, and mcp. It configures pytest against tests and limits mypy to four annotated modules. README.md documents Python, REST, batch, Swift, Kotlin/Android, ONNX, browser, CLI, and MCP surfaces. docs/faq.md explains cache warming, OPENMED_OFFLINE=1, socket blocking, model reuse, and the compliance boundary. docs/languages.md distinguishes built-in models, placeholders, environment-configured Indic routes, and user-supplied routes. docs/compliance.md explicitly says OpenMed does not self-certify compliance.
Observed
- License
- Apache-2.0 for the SDK source; model and dataset terms may vary.
- Primary package
- Python package requiring Python 3.10 or newer, built with Hatchling and published through PyPI.
- Install surface
- Base package plus optional extras for CLI, MCP, Hugging Face, services, MLX, ONNX, OCR, data frameworks, and integrations.
- Interfaces
- Python library, command-line tool, MCP server, REST and gRPC services, batch processing, Swift, Kotlin/Android, TypeScript, and browser paths.
- Platform support
- Linux, macOS, and Windows for the Python runtime, plus Apple platforms, Android, React Native, and browser deployment paths.
- Testing structure
- Pytest is configured to discover tests under the tests directory, with integration, slow, contract, fuzz, and doctest markers.
Read from README.md, Makefile, pyproject.toml, docs/faq.md, docs/index.md, docs/examples.md, docs/index.hi.md, docs/index.zh.md, docs/languages.md, docs/migration.md, docs/compliance.md.
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.