Survival LLM packs an open model and 1.5M articles into a free offline emergency assistant
Survival LLM, an MIT-licensed app, runs open-weight models and an offline library of medical and emergency guides on a Mac or PC, with sources for every answer.
By Gürkan E.Published
Tech AI LatestSurvival LLM is a free, MIT-licensed app that runs an open-weight language model and an offline library of about 1.5 million articles on a Mac or Windows PC. It answers emergency and first-aid questions without internet and shows the passages it used. Developer Ronny Fernandez released it on 24 October 2026 and shipped versions 0.3.0 to 0.3.4 on 26 and 27 October. His own tests found the smallest model gave harmful answers, so it is not offered.
Key facts
- First release: 24 October 2026; versions 0.3.0 to 0.3.4 on 26 and 27 October
- Licence: MIT; runs on Apple Silicon Macs and 64-bit Windows
- Models by memory: Nemotron 3.5 Super, Qwen3.8 27B, Nemotron 3.5 Lightning, Granite 4.2 8B
- Library: about 1.5 million articles, including about 930,000 from Wikipedia
- Developer's test: best model got the main point right on 21 of 22 questions
- Answers show cited passages; the app replaces sources the model never opened
What Survival LLM is
Survival LLM is a desktop app that bundles a local language model with an offline reference library, built for blackouts and other emergencies. Its project page says users set it up while online, after which it needs no internet and nothing they ask leaves the computer. Every answer lists the articles it read, and clicking one opens the source with the cited passage highlighted.
The developer, Ronny Fernandez, published the code on GitHub under the MIT licence. Version 0.1.0 appeared on 24 October 2026. Versions 0.3.0 to 0.3.4 followed on 26 and 27 October, adding a memory feature, a visible reasoning view and a reasoning budget setting, several of them crediting fixes from users.
Which open models it runs
The app checks the computer's memory and downloads a model to fit. Macs with 128 GB or more get NVIDIA's Nemotron 3.5 Super, an 80 GB download. Macs with 32 to 96 GB get Alibaba's Qwen3.8 27B. Windows PCs with 32 GB or more get Nemotron 3.5 Lightning, and machines with 16 to 31 GB get IBM's Granite 4.2 8B. Models run through Ollama. Setup takes from 20 minutes to about two and a half hours.
The library includes about 930,000 of Wikipedia's most-read articles, the WikiMed encyclopedia, MedlinePlus, FEMA and CDC guidance, U.S. military medicine and survival manuals, wikiHow and iFixit guides. The full download is about 9 GB, or 1.3 GB without Wikipedia and iFixit. Each source keeps its own licence, and some allow only non-commercial use.
What the developer's tests found
The README reports a test of 22 blackout and first-aid questions, graded blind by AI judges against official guidance. Nemotron 3.5 Super scored 86% accuracy and got the main point right on 21 of 22 questions. Qwen3.8 27B also got 21 right, and Granite 4.2 8B got 14. Granite 4.2 3B gave five harmful answers, including telling users to stay 3 feet from a downed power line when guidance says 35, so it is not offered. The developer notes the tests ran on an earlier build of the library.
A separate test found that models named sources they had never opened. The app now replaces any such line with the list of articles the model actually read.
Why it matters
Survival LLM shows open-weight models doing a job cloud assistants cannot: working with no connection at all. It fits the wider push toward local models, as in our Nemotron 3.5 Bolt story. Its own tests also show the limit, since small models still make dangerous mistakes. More open source news is on the Open Source AI hub.
What changed
| Before | Offline emergency references and local AI chat apps existed separately, and most local assistants did not cite their sources |
|---|---|
| After | A free MIT-licensed app combines a sized-to-fit open model with a 1.5-million-article offline library and cited answers |
| Who it affects | People preparing for outages or working without connectivity; developers building local retrieval apps on open models |
Questions people ask
- Does Survival LLM work without internet?
- Yes, after setup. The model and library are downloaded first; then it answers questions offline and nothing leaves the computer.
- Which AI models does Survival LLM use?
- It picks by memory: NVIDIA Nemotron 3.5 Super, Qwen3.8 27B, NVIDIA Nemotron 3.5 Lightning or IBM Granite 4.2 8B, run through Ollama.
- Is Survival LLM safe for medical advice?
- The developer warns that AI models make mistakes, especially small ones, and every answer shows its sources so users can check them.
Sources
- Survival LLM (Survival LLM project, 27 Oct 2026)
- anonnerdfrenzy/survival-llm repository and releases (GitHub, 27 Oct 2026)
Reference pages
- NVIDIA company
- IBM company
- Alibaba company
- Survival LLM tool
Spotted an error? See the corrections policy and tell the editor. Facts in this story come from the sources listed above; vendor claims are attributed.
Comments
Loading comments.
