AI assistant for emergencies released

A free and open-source AI assistant called Survival LLM has gained attention on Hacker News. Designed for emergency situations, the tool aims to provide reliable information without needing an internet connection.

AI operates offline with pre-downloaded data

Version 0.3.4 of Survival LLM was published on October 27, 2026. It runs directly on a user's computer after installation, using a library of pre-downloaded sources for its answers.

Sources include Wikipedia, medical guides, and more

The AI draws from about 930,000 of the most visited Wikipedia articles, medical encyclopedias like WikiMed and MedlinePlus, emergency guides from FEMA and the CDC, military survival manuals, wikiHow tutorials, and iFixit repair guides.

AI cites sources for each answer

Every response from the AI includes citations of the articles it used. This allows users to check the source material and verify the information provided.

Users can add personal documents for customization

Users can upload their own documents, such as a generator manual or a family emergency plan. The AI treats these documents the same way it treats its pre-loaded sources.

AI adapts to different computer specifications

The application adjusts to the machine it is installed on, checking available RAM and downloading the appropriate model. For example, a Mac with 128 GB of memory receives a larger model from NVIDIA, while a computer with 16 to 31 GB of RAM gets a smaller model from IBM.

Library size varies based on user needs

The full version of the library requires 9 GB of download, but a lighter version without Wikipedia or iFixit only needs 1.3 GB. This allows users to choose based on their storage capacity.

AI performance varies with model strength

The developers note that the AI can make mistakes. Tests show that the most powerful models answered correctly in 21 out of 22 cases, but the model used on computers with 16 to 31 GB of RAM succeeded only 14 out of 22 times.

AI use is energy-intensive

Using Survival LLM is energy-intensive, consuming about 100 watts during the 5 to 40 seconds it takes to generate each response on a MacBook Pro.