Available now
AyahOnDevice
AyahOnDevice v2.0 Stable is here.
Identify any Ayah with fully local, on-device AI — from the Holy Quran. Exact phrase discovery, local Confirm & Learn, complete English/বাংলা UI, and a Quran reader whose displayed text always comes from bundled Quran data.
A NivareQ™ project
Meet AyahOnDevice
AyahOnDevice is a local-first Quran ayah identification app and reader. Speak a short Quran fragment or type/paste Arabic to search locally, then open the matching canonical ayah or passage in the bundled Quran reader.
Two NivareQ firsts
- Our first open-source app. AyahOnDevice’s original source is public on GitHub under the Apache License 2.0. Third-party Quran text, translations, fonts, model/runtime components and other bundled resources retain their respective upstream terms.
- Our first app with on-device AI. Speech recognition runs locally on the device using WebGPU when available, with a CPU/WASM fallback. The first model preparation may require Internet access; normal recognition after preparation stays on-device.
Stronger local identification
- Exact Match Context. Exact normalized Arabic phrase matches are shown separately from fuzzy ranking, including all canonical occurrences when a phrase appears in more than one ayah.
- Confirm & Learn. Optionally confirm a result so similar future searches can be re-ranked locally. Raw matcher similarity is preserved, learning is bounded and reversible, and Undo and Reset controls stay on-device.
- Repeated ayat stay honest. Identical full ayat and shared openings are handled without forcing a false single winner; canonical occurrences remain discoverable under the same interpretation.
- What local AI heard stays visible. The recognized transcript is shown as an approximate search clue. AI output never becomes displayed Quran text.
Quran reading and context
- Complete English/বাংলা interface. Switch the full app interface between English and বাংলা independently of the selected Quran translation.
- Uthmani and IndoPak display. Read the complete Quran locally with Uthmani or IndoPak text and choose from five English and two Bengali translations.
- Familiar/context labels. Fifteen selected ayat and passages include curated English/Bengali familiar or contextual labels that never affect Quran text, matching or learning.
- Saved, notes and History. Keep Saved ayat, optional titles and notes, local History, confirmations and corrections together on the device.
Private and local by design
Normal voice recognition is fully local and on-device: audio is not sent to a cloud speech service for transcription. Search History, Saved data, notes and learned recognition remain in local browser storage unless you explicitly export a backup file.
Personal Backup v2 can export and restore Saved items, History, learned recognition and portable preferences. Restore validation and atomic storage handling protect local continuity.
Availability
AyahOnDevice v2.0 Stable is available at ayahondevice.nivareq.app. The first preparation of the default voice model may require an Internet connection; after the app shell, Quran data and model are cached, Settings can verify offline readiness.