Sources & licences

Original Arabic text

The Arabic text of the hadiths comes from the collection of hadithapi.com. It is cleaned (control characters, whitespace) but not altered, and checked against Sanadset 650K as a second source.

Translations

All new English, German and Turkish translations are made directly from the Arabic – with Claude (Anthropic), following a fixed glossary and style rules. No third-party translations are copied or used as a template for them.

Hadiths that have not been reworked yet show the texts of our previous collection for now: the English translation from hadithapi.com and German and Turkish machine translations made with Gemini (Google) that have not been reviewed. These pages are clearly labelled and will be replaced step by step.

Every translation is labelled: “AI-translated · not reviewed” until a person has reviewed it, then “reviewed”. Translator’s notes are kept separate from the text. The Arabic original is always authoritative.

Chains of narration

The separation of chain (isnad) and text (matn) and the name lists of the chains use Sanadset 650K as a suggestion. Licence: CC BY 4.0. The data has been checked, supplemented and modified; uncertain matches are marked on the page.

Mghari, M., Bouras, O., El Hibaoui, A. (2022): Sanadset 650K: Data on Hadith narrators. Data in Brief 44, 108540. doi:10.1016/j.dib.2022.108540

Narrators

Names, dates, places, generations, assessments, teachers and students come from the dataset Hadith Narrators (Kaggle, fahd09) (licence CC0 1.0), which is based on muslimscholars.info. Linking the names in the chains to these persons is our own work.

Search

Search by meaning uses the language model multilingual-e5-small (Wang et al., Microsoft, MIT licence) in Xenova’s ONNX version. It runs on our own server; search queries are not sent to any external AI service.

Fonts

Amiri, Source Serif 4 and Inter, each under the SIL Open Font License, self-hosted.