Survival LLM: AI assistant and emergency library that runs on your own computer

The library holds about 930,000 Wikipedia articles plus WikiMed, MedlinePlus, FEMA and CDC guidance, military manuals, wikiHow, iFixit and Stack Exchange answers. On first launch, the app checks available memory and downloads a model sized for the machine, from IBM Granite 4.2 8B on 16 to 31 GB systems to NVIDIA Nemotron 3.5 Super on Macs with 128 GB or more. In an earlier build's test, Nemotron answered 21 of 22 questions correctly and Granite answered 14. Setup runs 20 minutes to 2.5 hours, and the full library is a 9 GB download (19 GB installed). Most assistants fail without a connection, which is when emergency questions tend to come up. The project's answers point to the exact passage used, so a reader can check a small model's output against a fixed source instead of trusting its memory. Users can also add their own PDFs, Word files and ebooks to the search index. The code is open source and the app ships for Mac and Windows. For founders, the useful signal is the design: retrieval over a fixed local corpus with citations lets a modest model handle a narrow domain offline, and the model choice tracks hardware closely. The caveats matter as much. The creator warns that smaller models make more mistakes, and inference draws substantial battery power while the model works. Anyone shipping a local assistant should plan for hardware tiers, verifiable sourcing and power budgets.