MakinaRocks has deployed an on-device “Equipment Operation AI Staff” on the ROKS Pohang, a Republic of Korea Navy vessel restricted by strict maritime security protocols from accessing external internet networks. The deployment marks a critical operational test for bringing generative AI tools directly into closed-net naval environments.
The Bottom Line
- The Deployment: Physical AI firm MakinaRocks installed its proprietary AI agent framework aboard the ROKS Pohang of the Navy’s 1st Fleet Command.
- The Constraint: Because military ships operate at sea under strict security, standard cloud-based generative AI is blocked. The system instead runs locally via lightweight on-device architecture.
- The Function: The tool assists sailors in rapidly querying naval gun technical manuals and maintenance documents rather than autonomously operating ship equipment.
On-Device Architecture for Closed-Net Naval Operations
Naval vessels pack hundreds of complex mechanical systems into tight quarters, requiring crews to consult vast technical manuals and maintenance guidelines under high-pressure conditions. Standard commercial generative AI applications rely heavily on cloud infrastructure and continuous internet access. That model breaks down entirely at sea, where military security protocols strictly limit external network connectivity. To solve this bottleneck, MakinaRocks built a lightweight on-device operating environment designed to function entirely within a closed shipboard network.
The system operates using the company’s proprietary AI operating system, known as Runway. The technology functions as an operational reference assistant. Sailors input questions regarding equipment upkeep or troubleshooting, and the AI agent framework parses the query intent to fetch precise documentation from stored technical manuals.
Inside the AI Agent Framework Aboard the ROKS Pohang
The deployed system utilizes a hierarchical agent structure designed to split complex administrative and technical search tasks. At the center sits a primary representative agent that analyzes the exact intent behind a sailor’s query. Once the intent is categorized, the system routes the request to specialized sub-agents focused specifically on information retrieval and technical data analysis.
This division of labor ensures that answers are pulled accurately from verified naval gun technical manuals and maintenance documentation. Following its physical installation on the ROKS Pohang, the system underwent initial operational checks and was officially demonstrated to key naval leadership, including the Commander of the ROK Navy’s 1st Fleet.
Future Field Testing and Defense Expansion Plans
The current phase extends far beyond a simple single-day demonstration. MakinaRocks and the ROK Navy intend to maintain active, continuous operation of the equipment-operation assistant aboard the vessel to measure long-term field applicability. Feedback gathered directly from sailors interacting with the system at sea will be channeled straight back into system updates and interface refinements.
According to statements provided by MakinaRocks CEO Yoon Seong-ho, the Pohang-class test serves as a proving ground for applying physical AI technology directly to closed-network military environments. The ROK Navy is actively reviewing the real-world performance metrics from this initial pilot to determine whether the equipment-operation assistant should be expanded across additional fleet vessels and broader defense sector applications.
| Project Parameter | Operational Detail |
|---|---|
| Technology Provider | MakinaRocks (CEO Yoon Seong-ho) |
| Vessel & Command | ROKS Pohang, ROK Navy 1st Fleet Command |
| Core Infrastructure | Runway AI Operating System & On-Device Framework |
| Grant & Oversight | Korea Research Institute for Defense Technology (KRIT) |
