Meta and Booz Allen run Space llama: AI will work autonomously on the ISS

Meta and Booz Allen Hamilton announced the launch of Space Llama – adapted version of the open language model LLAMA 3.2 on board the American National Segment International Space Station (ISS). This is one of the first cases of practical deployment Llm in conditions of complete autonomy and limited communication.

AI without the Internet Unlike land systems, AI is in orbit with strict restrictions: low power, lack of connection to the network and high delay when connected with the Earth. Space Llama is optimized for local work: it responds to crew requests, provides access to documentation and helps in maintaining the station – all offline.

Technical architecture Integration is based on the following components:

• LLAMA 3.2 is the basis of the model adapted for tasks in the compressed environment.
• BOOZ Allen A2E2 ™ is a containerized AI platform for the edge media.
• Spaceborne Computer-2 from HPE is a space computing platform for infusion.
• NVIDIA CUDA GPU – accelerate the processing of requests with limited power.

This stack provides a stable model of the model in the orbital environment, not exceeding the energy and thermal boundaries of the station.

Preparation for autonomous missions Space Llama is a base for future autonomous missions, including a month or Mars, where land feedback can take hours. Such AI will help with diagnosis, operation planning and emergency solutions.

The modular structure A2E2 also allows you to adapt the system to other environments – from polar stations to military bases.

Recall that in April Meta introduced a new series of LLAMA 4.

The gun