OpenMed is local-first healthcare AI software that extracts clinical information and removes personally identifying details from clinical text on hardware controlled by the user. Healthcare developers use its Python runtime, Apple Silicon and mobile SDKs, and browser support for on-device clinical NER and PII de-identification.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add maziyarpanahi/openmed --skill running-openmed-ondevicegit clone --depth 1 https://github.com/maziyarpanahi/openmedWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/maziyarpanahi/openmed/running-openmed-ondevice)<a href="https://agentmods.dev/skills/maziyarpanahi/openmed/running-openmed-ondevice"><img src="https://agentmods.dev/badge/skills/maziyarpanahi/openmed/running-openmed-ondevice/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/maziyarpanahi/openmed/running-openmed-ondevice"><img src="https://agentmods.dev/badge/skills/maziyarpanahi/openmed/running-openmed-ondevice.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00175 | $0.02227 |
| Opus 5 | $0.00088 | $0.01113 |
| Sonnet 5 | $0.00035 | $0.00445 |
| Haiku 4.5 | $0.00017 | $0.00223 |
Grade A, and why
running-openmed-ondevice scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 179 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Running OpenMed on-device
OpenMed runs fully on-device by design. These three backends let you take it further at the edge: MLX (Apple Silicon acceleration), CoreML (iOS/macOS / Neural Engine), and ONNX / WebGPU (cross-platform and in-browser). The flow is the same: convert → (quantize) → run locally. Because inference is local, raw PHI never leaves the device — the strongest privacy posture OpenMed offers.
When to use this skill
When you need OpenMed where there is no server: an iOS/macOS app (CoreML),
fast NER/de-id on an Apple Silicon Mac (MLX), or a portable/browser deployment
(ONNX/WebGPU). For a hosted endpoint use serving-openmed-rest-api; for an
agent tool use deploying-openmed-mcp; for corpora use
batch-processing-clinical-text.
Pick a backend
| Backend | Extra | Best for | Quantization |
|---|---|---|---|
| MLX | openmed[mlx] |
Apple Silicon Macs; fastest local NER/de-id; on-device LLMs | 4-bit / 8-bit weights |
| CoreML | openmed[coreml] |
iOS/iPadOS/macOS apps, Neural Engine | int8 palettization |
| ONNX / WebGPU | openmed[onnx] |
cross-platform runtimes, browser (transformers.js) | fp16 (WebGPU); int8 via ORT |
Quick start — MLX (Apple Silicon)
pip install "openmed[mlx]"
# Convert a HF token-classification model to an OpenMed MLX artifact, 8-bit:
python -m openmed.mlx.convert --model OpenMed/<some-ner-model> --output ./mlx_ner --quantize 8
import openmed
# Run NER/de-id through the normal API — pass the local artifact dir as model_name.
# The loader auto-detects the MLX backend from the artifact (or set backend explicitly).
result = openmed.analyze_text(
"Patient received 75mg clopidogrel for NSTEMI.",
model_name="./mlx_ner", # local MLX artifact directory
output_format="dict",
)
# Force MLX via config if you prefer to be explicit:
from openmed.core.config import OpenMedConfig
cfg = OpenMedConfig(backend="mlx") # None=auto-detect, "mlx", or "hf"
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 179 lines · 175 tokens per session scan A 1e0573b91ea7
running-openmed-ondevice is a skill published in the GitHub repository maziyarpanahi/openmed (5,290 stars, last pushed today), licensed Apache-2.0. It adds 175 tokens to every session and 2,227 once invoked, about $0.0009 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.
Other skills, from other repositories
add-new-model
Use this skill when the user wants to add or port a new model architecture to MLX-VLM — mapping a Hugging Face modeltype to a new file under mlxvlm/models, writing the ModelConfig, matching layer/weight names, reusing a similar existing model, adding a test class, and validating the port. Covers vision-language…
convert-quantize
Use this skill when the user wants to convert a Hugging Face model to MLX or quantize/dequantize one with mlxvlm.convert, including bits and group size, quant modes (affine, mxfp4, nvfp4, mxfp8), RTN vs AWQ, mixed-bit recipes, dtype casts, calibration (text or multimodal), local vs Hub paths, revisions, uploading to…
server-inference
Use this skill when the user wants to run or debug MLX-VLM server inference, including uv run mlxvlm.server, /v1/models, /v1/chat/completions, /v1/responses, streaming, OpenAI-compatible clients, health checks, metrics, model unload/reload, adapters, trust-remote-code, and server request/response failures.
cli-inference
Use this skill when the user wants to run or debug MLX-VLM inference from the command line, including uv run mlxvlm.generate, image/audio/video inputs, local model paths, Hugging Face model IDs, deterministic repro commands, and CLI errors around processors, prompts, model loading, or missing weights.
hf-cache-models
Use this skill when the user wants to list, inspect, or report MLX-VLM model candidates available in the local Hugging Face cache directory, including the server's opt-in hf-cache discovery mode, cache-dir overrides, JSON output, or issue-ready cached model lists.
actual-setup
Set up Actual Computer (actual.inc) inference in Hermes.