Survival LLM is a free, open-source AI assistant and offline emergency reference library designed to run entirely on local hardware without internet connectivity after initial setup. The application automatically selects and downloads a quantized large language model suited to the host system's available RAM: NVIDIA Nemotron 3.5 Super (80 GB) for Macs with 128 GB+; Qwen3.8 27B (18 GB) for Macs with 32–96 GB; Nemotron 3.5 Lightning (25 GB) for Windows PCs with 32 GB+; and IBM Granite 4.2 8B (6 GB) for systems with 16–31 GB. In blind testing on 22 blackout and first-aid questions, the three larger models achieved 21/22 correct responses, while the 8B model scored 14/22. The 19 GB installed library (9 GB download) includes ~930,000 Wikipedia articles, WikiMed, MedlinePlus, UK Medicines A-Z, WikEM, FEMA Ready.gov guides, CDC travelers' health, U.S. military manuals, wikiHow, iFixit, Appropedia, and Stack Exchange Q&A; a 3 GB variant excludes Wikipedia and iFixit. Installation requires model size plus 5 GB free space, plus ~27 GB temporary during full library install. The app supports Apple Silicon Macs (M1+, macOS 14+) and 64-bit Windows 10/11, with setup taking 20 minutes to 2.5 hours. Features include source citation with highlighted passages, user document ingestion (PDF, Word, ebooks), a reasoning token budget (Off to 16k/unlimited), memory notes for persistent context, and battery draw of ~100 W for 5–40 seconds per query on MacBook Pro. Version 0.3.4 adds reasoning limits, fallback answer generation, and explicit source attribution.