Has anyone created a "Local LLM Survival Kit"?

Here's what I'm thinking about: A USB thumb drive that you can plug into any PC or laptop, and immediately get a usable knowledge base powered by an LLM, without requiring an Internet connection. I believe the technology for this should be ready. Rough architecture: llama.cpp binaries for CPU-only inference, for Windows, macOS, and Linux (each for all major architectures, a few hundred MB total) Qwen3.5 35B-A3B @ Q4_K_M (22 GB, for systems with >= 32 GB RAM) Gemma 4 E4B @ Q4_K_M (5 GB, for systems with < 32 GB RAM, and for audio/video processing on larger systems) A compressed SQLite database containing: An English Wikipedia dump (120 GB raw, around 30 GB with sqlite-zstd after pruning) Freely licensed books on important topics like medicine, engineering, etc. A simply server with a browser-based chat frontend that hooks the model up to a tool allowing it to search the database All of this should (just about) fit on a 64 GB thumb drive, which retails for well below 10 USD. On almost any PC or laptop from the past 15 years, you should get 5-20 tokens/s with zero setup and no GPU required, regardless of the operating system. Chat sessions are saved back to the drive and come with you wherever you take it next. Does anything like this exist?