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

Survival LLM is a free, open-source AI assistant and emergency library that operates locally on your personal computer. This innovative tool can provide answers to a wide range of queries, such as food safety during power outages, treating burns, and water purification, by drawing upon trusted sources and citing their exact locations within each response.

To set up Survival LLM, all you need is an internet connection initially. Once installed, the software requires no further connection, running entirely on your device. The app remains consistent across all platforms, checking your computer's memory upon opening and downloading the appropriate AI model. The size of the model directly impacts the intelligence of the assistant, with larger models generally performing better.

In rigorous testing, Survival LLM achieved a remarkable 22 out of 22 correct answers to blackout and first-aid questions, when graded blind against official guidance. Smaller models, while more compact, tend to make more errors, which is why the app consistently displays the source material for cross-referencing. The extensive library, which comprises around 9 GB of data, can be downloaded entirely or a more modest 1.3 GB version (3 GB once installed) can be chosen, omitting Wikipedia and iFixit.

The installation process necessitates a certain amount of disk space, specifically, the size of the chosen model plus an additional 5 GB of free space, and approximately 27 GB while the entire library is being installed (10 GB for the smaller model). Compatibility extends to Apple Silicon Macs running macOS 14 Sonoma or newer, as well as 64-bit versions of Windows 10 or 11. The setup process itself can range from 20 minutes to approximately 2½ hours, depending on your computer's performance and internet speed.

Finally, the material used in Survival LLM is sourced from openly licensed content (with some exceptions for non-commercial use only) and public-domain references. These sources are individually licensed, indicated alongside each article, and the entire library can be searched in a matter of seconds.