Survival LLM: A Free Offline AI for Emergencies and Outages

Survival LLM, a free and open-source AI assistant, runs entirely on a user's own computer and answers survival and first-aid questions without any internet connection. Version 0.3.4 was released on October 27, 2026, and the project drew considerable attention on Hacker News this week.

Its author describes it as "designed for emergency situations". Typical questions include how to keep food safe during a power outage, treat a burn or purify water.

Answers backed by a local library

Once installed, the model sends no requests to the outside world. It draws on a library downloaded in advance. According to the project's GitHub repository, that library includes about 930,000 of Wikipedia's most-viewed articles, medical encyclopedias such as WikiMed and MedlinePlus, emergency guides from FEMA and the CDC, military survival manuals, wikiHow tutorials and iFixit repair guides.

Every answer lists the articles it relies on. Clicking one opens the source with the relevant passage highlighted. Users can also add their own documents, such as a generator manual or a family emergency plan, which the assistant consults like the rest.

Built to run on modest machines

The app adapts to the hardware. At first launch it checks available memory and downloads a matching model. A Mac with at least 128 GB of memory gets NVIDIA's Nemotron 3.5 Super, an 80 GB model, while a computer with 16 to 31 GB uses IBM's Granite 4.2 8B, which weighs 6 GB. The full library is a 9 GB download, but a version without Wikipedia and iFixit takes up 1.3 GB.

The developers acknowledge the tool can be wrong. In their test of 22 safety and first-aid questions, the most powerful models answered 21 correctly, but the model for 16 to 31 GB machines got only 14 right, more than one error in three.

Power use is another caveat. On a MacBook Pro, the assistant draws about 100 watts for the 5 to 40 seconds each answer takes, which matters when electricity is scarce.