Vitalik shares local private LLM solution, emphasizing privacy and security first

ME News update, April 2 (UTC+8). Vitalik Buterin posted sharing his local and private LLM deployment方案 as of April 2026. The core goal is to treat privacy, security, and self-sovereign control as prerequisites, minimize the opportunities for personal data to be accessed by remote models and external services, and reduce the risks of data leakage, model jailbreaks, and malicious-content exploitation through local inference, local file storage, and sandbox isolation. In terms of hardware, he tested options including a laptop equipped with an NVIDIA 5090 GPU, an AMD Ryzen AI Max Pro 128 GB unified-memory device, and DGX Spark, and performed local inference using the Qwen3.5 35B and 122B models. Among them, the 5090 laptop reaches about 90 tokens/s with the 35B model, the AMD setup is about 51 tokens/s, and DGX Spark is about 60 tokens/s. Vitalik said he is more inclined to build a local AI environment based on high-performance laptops, while using tools such as llama-server, llama-swap, and NixOS to set up the overall workflow. (Source: ODAILY)