NASA Building Local LLM System for Space Missions

NASA's Johnson Space Center is developing the CMO-DA (Crew Medical Officer Digital Assistant) to address communication delays and blackouts that may occur during lunar and Mars exploration. This system adopts a Local-first architecture that operates on local hardware without cloud connectivity.

Key technical features include the following:

• Use of RamaLama: RamaLama, an open source CLI tool that wraps llama.cpp, is used to manage models like container images, providing automatic GPU detection and passthrough capabilities.
• RAG (Retrieval-Augmented Generation) application: Based on spaceflight medical literature, it supports crew members in diagnosing and treating symptoms.
• Edge computing implementation: With support from Red Hat, it is being tested via offline edge deployment based on HPE hardware, and is currently being validated on a ground twin model of the HPE Spaceborne Computer installed on the International Space Station (ISS).

This project aims to treat models not as simple installations but as Portable artifacts, ensuring reproducible and cryptographically verifiable deployment even in space environments where physical access is not possible.