NASA tests offline medical AI assistant for astronauts

NASA's Johnson Space Centre has developed and is testing an offline medical AI assistant with Red Hat to support astronauts when communication with Earth is unavailable.

Known as the Crew Medical Officer Digital Assistant, the tool is designed as a clinical decision support system for deep-space missions. It aims to help astronauts diagnose and treat symptoms without relying on a live link to Earth-based doctors, a constraint that becomes more acute as missions move farther from the planet.

At the centre of the project is RamaLama, an open-source software project backed by Red Hat. Researchers are using it to run AI models locally instead of through a cloud connection, marking a shift from an earlier prototype that depended on remote computing.

That change addresses one of the practical limits of using AI in space. Many current AI systems depend on constant network access, but spacecraft operating far from Earth can face long delays or complete communication loss, making autonomous tools more important for crew health and safety.

The system is being tested on HPE hardware described as the terrestrial twin of the Spaceborne Computer aboard the International Space Station. The setup gives researchers an environment to refine the software before any further operational evaluation.

Offline design

NASA's medical assistant began as a proof of concept built on spaceflight medical literature. The latest work has focused on turning it into a fully disconnected edge system that can operate independently in remote conditions.

RamaLama packages AI models as containers, allowing them to run consistently across different devices. Here, the software supports both language models for medical reasoning and vision-language models for image-based symptom analysis.

This multimodal approach lets the assistant handle text and visual inputs on the same local system. It also avoids the need for a large supporting infrastructure, an important factor where computing resources, power and connectivity are limited.

Researchers say container-based deployment also makes the system more auditable and reproducible. Those features are especially relevant in medical and mission-critical settings, where developers and operators need to understand exactly how a model was packaged and run.

Space constraints

The medical challenge is straightforward: astronauts on future lunar and Mars missions may not be able to consult Flight Surgeons in real time. Signal delay increases with distance, and in some situations communication could be interrupted altogether.

That leaves crews needing more self-sufficient tools, especially if no trained medical specialist is on board. A digital assistant of this kind is intended to support a Crew Medical Officer by offering structured guidance when immediate external support is not possible.

Red Hat describes RamaLama as a way to simplify how developers run, pull and serve AI models. The software uses Open Container Initiative-compliant containers, making it possible to move models between systems while keeping deployment consistent.

For NASA, that portability matters because systems developed on the ground must behave predictably when moved into far harsher, less forgiving environments. Space operations leave little room for uncertainty in software behaviour, particularly in medical scenarios.

Wider uses

The work also points to possible uses beyond human spaceflight. A medical AI system that can function without a cloud connection could have applications in remote settings on Earth, where reliable communications and specialist care are harder to access.

Those parallels reflect a broader industry push to move AI workloads closer to where data is generated, rather than routing every request through centralised data centres. In medicine, local processing can also help when speed, resilience or network availability matters more than access to remote computing capacity.

The project remains in testing rather than operational deployment. The system will be refined through terrestrial testing before being demonstrated to NASA leadership for evaluation of further use.

The next iteration is expected to use Red Hat Enterprise Linux AI, replacing the current setup with another software base for managing the containerised applications behind the assistant. The aim remains the same: to provide medical guidance when astronauts are too far from Earth to rely on immediate outside help.

By shifting from a cloud-dependent model to a local system, the Crew Medical Officer Digital Assistant can continue operating even when the link to Earth is severed.