Dark Fault
Offline · Open source · Apple Silicon

Your own library and AI, with the internet unplugged.

Dark Fault turns a Mac into a private encyclopaedia and assistant. Read Wikipedia, search it, and ask questions that are answered from it, with sources. Nothing leaves your Mac.

A grid fault is the short circuit that trips the lights out. Born as Project D.A.R.K. S.K.Y.: Durable Archive of Reference Knowledge, on Silicon, Keyless, Yours.

Three things, done well

One small program for your Mac. No Docker, no admin password, no account.

Read

An offline library

Wikipedia, Wiktionary, Wikivoyage, medical encyclopaedias and more, in the open ZIM format from Kiwix. Drop any .zim file in and it appears.

Search

Find it as you type

Titles suggest as you type. A one-time index adds full-text search, and a local embedding model can re-rank results by meaning.

Ask

Answers with sources

Chat with a local model through Ollama. Turn on "Use the library" and each answer lists the articles it drew on, with links you can check.

Keyless means keyless

Dark Fault was built to run on a Mac you also use for real work. These are not settings. They are how it is made.

Install

You need an Apple Silicon Mac (M1 or later) and Go to build it. For chat, install Ollama. The library and search work without it.

Disk space depends on what you download: about 1 GB for Wikivoyage, 13 GB for Wikipedia introductions, 119 GB for all of Wikipedia with pictures.

To start Dark Fault at login, the repository includes a launch agent that runs as you. No sudo.

# build and install (no sudo)
$ git clone <source repository, coming soon> dark-fault
$ cd dark-fault && make install

# get a pack, index it, start
$ darkfault fetch wikivoyage-en
$ darkfault index --all
$ darkfault serve
# open http://127.0.0.1:8740

Models it can use

Dark Fault refuses any model that is not in its catalogue, even if Ollama has it. Every catalogue model has an Apache 2.0 licence.

ModelUseMaker
granite4.2:3b · granite4.2:30bChatIBM
mistral-small3.2:24b · devstral-small-2:24bChatMistral AI
gpt-oss:20bChatOpenAI
granite-embedding:278mSearch re-rankingIBM