NemoClaw Knowledge Wiki

LM Studio LM Link Remote LLM Access for Portable Devices

3 min read

Clip title: Private AI on the go… a new trick Author / channel: Alex Ziskind URL: https://www.youtube.com/watch?v=PqBrnip-ZLw

Summary

The video explores the evolving landscape of running large AI models, contrasting the capabilities of high-end MacBooks with more portable devices and the role of innovative tools in bridging this gap. Initially, the presenter demonstrates successfully running several large language models, including GPT-OSS 120B (60GB) and Meta Llama 70B (70GB), on a MacBook Pro with 128GB of RAM. However, this powerful machine, capable of handling huge models, is then symbolically set aside to highlight the challenge of running such demanding AI on a more accessible, portable device like a MacBook Air with only 16GB of memory, which quickly struggles even with a smaller 4B model like Gemma 3 due to intensive prompt processing.

The core solution introduced is LM Studio’s new “LM Link” feature, powered by Tailscale, which allows a less powerful client device to securely and effortlessly connect to remote local or cloud-based machines hosting powerful AI models. This creates a virtual private network, enabling users to leverage high-performance hardware without physically carrying it around or managing complex network configurations. The presenter illustrates this by connecting his MacBook Air to a robust Mac Studio (with 512GB RAM) and an external GPU rig (NVIDIA RTX Pro 6000 with 96GB VRAM), demonstrating the ability to run massive models like Qwen3 Coder 480B (251GB) and Qwen3 Next 80B (45GB) with a 50,000-token context window and impressive speed.

Further emphasizing the scalability and flexibility, the video showcases the MacBook Air connecting to a high-end cloud server from Cirrascale, which features eight NVIDIA B200 GPUs providing over 1TB of VRAM. This setup allows the presenter to run the Kimi K2.5 model (a 1.8TB model) privately and securely, generating detailed architectural and database designs with remarkable speed and quality. This remote local approach ensures privacy and security for sensitive data, as the processing occurs on a user-controlled machine or a dedicated, secure cloud instance rather than public cloud AI services.

Beyond software innovation, the video highlights the importance of reliable power for mobile AI workloads with the Jackery Solar Generator 5000 Plus. This portable power station, offering substantial capacity (up to 60kWh) and output (7200-14400W) with UPS backup, ensures uninterrupted work even in remote locations or during power outages. The overall takeaway is that LM Studio’s LM Link, combined with robust local or remote infrastructure and reliable portable power, democratizes access to large, high-quality AI models. It simplifies the setup and usage of powerful AI tools, enabling users to maintain privacy, security, and efficiency from any device, whether for chat, coding, or complex professional scenarios.

Graph View

Backlinks