Article
AI in Healthcare AI News Dell Technologies Hardware & Electronics

Dell and Nvidia unveil six-agent AI healthcare system that fits under a desk

The system keeps patient data on hospital hardware rather than sending it to external servers, and can reportedly be deployed in about an hour

by TechDefused Newsroom
A healthcare professional is shown sitting at a desk in an office setting, engaging with a patient via video call on a laptop. The scene includes elements of a professional workspace, such as medical supplies and plants. — Credit: Photo by Vitaly Gariev on Unsplash c Photo by Vitaly Gariev on Unsplash

Dell Technologies and Nvidia have published the Local Healthcare Agent playbook, demonstrating a six-agent clinical AI architecture designed to run entirely on a deskside system.

The reference architecture is designed to keep protected health information inside hospital infrastructure, rather than routing it through external cloud services.

Patient records are handled where they are created, an approach the companies say reduces both compliance risk and processing delay.

What powers the system

The configuration is built around a Dell Pro Max with GB300, paired with Nvidia's Grace Blackwell Ultra GB300 Superchip.

The playbook says the hardware delivers 20,000 TFLOPS of FP4 compute, a measure of processing speed for AI workloads, and can host models with up to one trillion parameters.

At inference, the system runs containerised models including Nvidia Nemotron 3 Super, a 120 billion parameter mixture-of-experts model, alongside specialist inference tools for protein structure and visualisation.

Six agents doing different jobs

The playbook lays out six specialised AI agents working together.

One acts as a coordinator, while the remaining five focus on patient data, labs and vitals, medications, clinical analysis and molecular visualisation respectively.

The system supports fully air-gapped deployments, meaning it can run with only read-only outbound connections and no direct link to external networks.

Rules that update without retraining

Rather than embedding clinical thresholds directly into the underlying model, the agents read human-readable skill files at the moment a query is made.

That design means updated clinical guidelines can be applied immediately, without needing to retrain the underlying AI models.

Part of a broader on-premises push

The blueprint extends a recent push by Dell and Nvidia to package agentic AI as on-premises infrastructure for regulated industries such as healthcare.

The companies say the approach is intended to reduce cloud computing costs and speed up deployment for organisations bound by strict data rules.

The Local Healthcare Agent playbook is available now, with the companies saying organisations can deploy the full six-agent system in approximately an hour.

by TechDefused Newsroom