November 17, 2026 · 7 min read

Multiverse Computing brings offline AI to military equipment maintenance, deploying FieldGuide on purpose-built HPE ProLiant Compute

FieldGuide draws on the customer’s own S1000D technical data to guide repairs at the edge, running fully offline on CPUs alone.

Multiverse Computing

When a military vehicle breaks down in the field, the technician on site may not be an expert in that particular system. Finding the right repair procedure can mean searching through extensive technical manuals or consulting a specialist who may be unreachable when communications are limited or unavailable. In the meantime, critical equipment remains out of service. What technicians need is a way to access reliable, equipment-specific guidance directly at the point of repair, even without a network connection.

To address this challenge, Multiverse developed an AI-powered maintenance assistant, powered by HPE and Intel, that gives operators access to approved technical procedures wherever repairs happen. Powered by a language model optimized using Multiverse Computing's CompactifAI technology, FieldGuide runs fully offline purpose-built, rugged hardware that can be deployed as CPU-only or GPU accelerated, with or without connectivity.

Operators ask FieldGuide questions in plain language and receive step-by-step instructions drawn only from approved technical data. Every answer cites its source data module, so the operator can open it and check. A less experienced technician gets guided help outside their specialty, and an experienced one gets a direct answer.

Customers stay in control of their data. Ingestion is done by the customer directly from its S1000D common source database, so Multiverse Computing never needs access to confidential documentation. The model, the index and the data stay on the server, no data is transmitted to external servers, and the assistant keeps working with no network link.

In practice, that means a technician facing an unfamiliar fault code, a hydraulic issue outside their specialty, or a torque spec buried in a multi-volume manual gets a direct, sourced answer on the spot — instead of a radio call to a reachback expert who may be hours away or unreachable. That keeps vehicles moving through the repair bay faster, takes routine questions off the queue for scarce subject-matter experts, and shortens the time it takes a newer technician to work confidently outside their core specialty. Equipment readiness stops being a function of who’s available on the other end of a radio, and becomes a function of the technical data already on hand.

Advanced models meet rugged hardware

CompactifAI reduces model size by 80 to 95% with a 2 to 4% accuracy loss, which lets a capable model run from server memory on the CPU and take advantage of the AI acceleration built into every Intel Xeon 6 core. With no accelerator card, platforms have one less high-power component to carry, which can run at different temperature ranges and keep in spares, which matters in vehicles and shelters where size, weight and power are tight.

The reference configuration pairs FieldGuide with a single EL220 Gen12 server in the HPE ProLiant Compute EL2000 chassis, a 2U platform designed for harsh and remote environments, with a 32-core Intel Xeon 6 processor, 128 GB of DDR5 memory and 3.84 TB of NVMe storage.

The launch of FieldGuide builds on the collaboration between Multiverse Computing, HPE and Intel. In September 2026, the three companies ran an optimized speech-to-text model on Intel Xeon 6 processors inside an HPE ProLiant server with no discrete accelerators, following their July milestone in which CompactifAI models were shown running on Intel Xeon 6.

Your data, your environment

Multiverse Computing has delivered or is delivering 25 RAG engagements across cloud, on-premise, edge and embedded environments in 7 countries, 14 of them as full deliveries rather than proofs of concept. FieldGuide is one example, starting with a pilot on one platform’s S1000D data set on the EL2000 reference configuration. Accuracy is then validated against a query set agreed with the customer’s maintenance team, and the solution is rolled out to units with ongoing support as technical data changes. Latency and throughput are validated on the customer’s own data during the pilot.

FieldGuide illustrates how smaller, more efficient AI models can bring advanced capabilities into environments where connectivity, computing resources and access to specialized expertise are limited.

To discuss a pilot on your own S1000D data, contact business@multiversecomputing.com.

Want to know more?

Reach out to us at business@multiversecomputing.com