The video showcases the Asustor Lockerstor 4 Gen 3 NAS as a powerful local AI platform capable of storing and running large language models like Llama.cpp and OpenClaw efficiently, while also providing robust storage management and security features. It demonstrates the NAS’s ability to perform AI inference tasks, including image analysis and real-time information retrieval, highlighting its versatility as both a secure storage solution and an AI compute device.
The video explores the capabilities of the Asustor Lockerstor 4 Generation 3 NAS, a powerful network-attached storage device designed to handle AI workloads locally. This NAS boasts impressive hardware features including dual 10GB ports, two 5GB ports, 40 Gbps USB 4 ports, 16GB of RAM (upgradeable to 64GB), and internal SSD slots, providing up to 30GB of network bandwidth. The presenter demonstrates how this device can not only store large AI models but also run them locally, showcasing its potential as both a storage and AI inference platform.
The operating system of the NAS offers a rich ecosystem of applications accessible through an app store, including Docker and Portainer, which the presenter uses to manage AI models and agents via a graphical interface. Additionally, a web-based SSH tool called “shell in a box” allows remote command-line access without needing a separate terminal application. This setup facilitates running and managing AI models like Llama.cpp and OpenClaw directly on the NAS in a secure, sandboxed Docker environment.
Performance tests with various large language models (LLMs) reveal the NAS’s capabilities and limitations. The presenter runs models such as Gemma 4 (E4B), Gemma 12B, and GPT OSS 20B, noting token generation speeds ranging from about 4 to 11 tokens per second depending on model size and complexity. The Gemma 4 E4B model stands out for its ability to perform image inference, demonstrated by analyzing a training loss graph image to detect overfitting. Larger models like the 12B and 20B parameter versions run slower but still perform well, with the NAS maintaining responsiveness for other tasks during inference.
The video also highlights the NAS’s robust storage management features, including RAID configuration, BTRFS file system with snapshotting for data protection, and scheduled SMART scans to monitor drive health. Security considerations are addressed by disabling online access by default to reduce hacking risks, though firewall options allow region-based access control. The presenter shows seamless file sharing over the network and easy folder management, emphasizing the NAS’s dual role as a secure storage hub and AI compute device.
Finally, the presenter demonstrates OpenClaw running on the NAS, connected to a Mac for additional processing, using the new Quen 3.6 27B model with MTP layers. This setup supports advanced AI tasks like tool calls for real-time information retrieval, exemplified by querying weather data and researching a recent FIFA controversy. The video concludes by inviting viewers to express interest in further tutorials, underscoring the Asustor Lockerstor 4 Gen 3 as a versatile and capable platform for local AI development and deployment.