OpenMed Agent in 15 seconds.
Codex. Claude. Hugging Face. Open source. The full stack.
Bring your own LLM. The medical layer is built around it.
Preview: https://t.co/6BI3qnzSDQ
🚨 Today: OpenMed Agent ships in preview.
Built on @huggingface:
→ HF endpoints power clinical extraction + terminology
→ MCP for your own services
→ Every tool call, every plan, fully visible
1,000+ OpenMed medical models on HF. Preview today.
This guy just made a revolution!
Frontier medical AI running on your iPhone, pulls your Apple Health data + scans reports, redacts everything personal on-device, then reasons with GPT-5.5 through your ChatGPT account.
Zero sensitive data leaves your phone.
Privacy-first health tech done right.
Proud of you Maziyar!
@MaziyarPanahi@Bougey@MorpheusAIs For HIPAA / PHI, use AWS Bedrock/SageMaker, GCP Vertex AI, Azure AI Foundry/ML, Databricks, or HF Enterprise dedicated endpoints. Everything else is not HIPAA compliant, private or not. Even as a hospital or clinic, you can’t spin a B200 and host a model privately. Compliance…
Frontier medical reasoning on your iPhone, with GPT-5.5, and zero identifiers ever leaving the device.
OpenMed Agent reads your Apple Health, wearable trends, and any medical report you scan, redacts on-device, then asks GPT-5.5 via your own ChatGPT account.
App Store next. 👇
I printed a patient's discharge summary, packed with names, IDs, phone, email, and address, then scanned it with my iPhone.
31 pieces of PHI masked on-device. Then it pulled the diagnoses, meds, and doses out as clean data.
No server. No Python. Nothing leaves the phone. 👇
OpenMed Agent, small Sunday update.
→ queue to stack tasks while the agent is busy
→ steer to redirect the agent mid-plan
→ planning view now collapses
→ theme to swap themes mid-session
Quality-of-life wins for builders in the terminal.
We observe improvements of Opus 4.8 on long horizon agentic healthcare tasks but the 1/3 is far from perfect. Major failure mode is still rushed to submit PA when lacking evidence, approving a case when evidence is lacking, i.e, refusal and policy retrieval & adherence
Biology is the next agentic frontier after coding. Anthropic is aggressively improving their models on routine data analysis with careful attention to nuances of different assay types. Opus 4.8 is noticeably better at single cell / spatial analysis. We have already rolled it out to customers across pharma and academia. Cool to see our benchmarks on the system card.
We’re taking steps to accelerate defensive progress in biology:
- Launching Rosalind Biodefense to help trusted builders develop new biodefense and pandemic preparedness capabilities.
- Expanding trusted access to GPT-Rosalind for select U.S. government and allied partners supporting public health and biodefense missions.
Advances in biology can strengthen our ability to prevent, detect, and respond to biological threats. Our goal is to help build a more robust ecosystem – giving trusted defenders frontier AI to develop and operate new defenses for public health and biodefense.
https://t.co/CT4muTSOOZ
Refusing to fabricate should be the first thing you implement in your clinical AI.
Anthropic's newest model "Claude Opus 4.8" is now available in OpenMed Agent.
Opus 4.8 just did the most important thing in clinical AI: it said no.
Asked to reconcile 3 guideline bodies on aspirin, OpenMed Agent searched, found the guidelines weren't in its sources, and labeled 8 gaps instead of inventing them.
Refusing to fabricate is the feature.
@FullStackTaco@nxpatel@EvidenceOpen That’s a fair way to loom at it. The OpenMed Agent has access to PubMed and Medline journals and has citations mandatory. It always provides evidence and human-in-the-loop is a first principle. We’re in preview, working with design partners
20 new open-source PII redaction models today: Vietnamese, Bengali, Korean, Chinese. 20 MLX variants for on-device.
Built on NVIDIA's Nemotron-PII data and OpenAI's privacy-filter weights. ~1.5B more people covered. Apache 2.0.
4 new languages for the OpenMed open-source PII filter family: Vietnamese, Bengali, Korean, Chinese.
20 new models, 20 MLX variants for on-device. ~1.5 billion more people covered.
NVIDIA × OpenAI lineage. Apache 2.0 on @huggingface. Built by one person.
@alexrives This is the day a lot of us have been waiting on. ESM's language-model bet on folding has aged beautifully. Wrote a landscape piece a while back that put ESM in AlphaFold-era context. Link for the backstory:
https://t.co/WAxL7yNZ7p
CARBON: 8B open DNA model, 65K-token context, whole human genome on a single GPU in <2 days.
Clinical-language counterpart: OpenMed, 1,000+ open medical models on @huggingface, eval sets published with the weights. Apache 2.0 on both. 🤗
@MaziyarPanahi@huggingface everything you have ever needed to build is on @huggingface! give this to your coding agent and let’s build!
https://t.co/XojrVX4ttq
5.14M downloads on PyPI. 1.2K followers on the OpenMed @huggingface org. openmed 1.5.2 ships today.
The PII family driving most of that traffic = NVIDIA's Nemotron-PII data + OpenAI's privacy-filter weights, combined and Apache 2.0.
Beautiful when big labs ship open.