Quran dataset (~338 pairs)
Mood-based ayah guidance, Tawheed, Salah, Sabr, hope, anxiety, death, and afterlife, using Sahih International and Pickthall translations only.
إيمانVibes
First Islamic AI Model by ImanVibes
Coming soon
Chat in ImanVibes
Coming soon
API Access
An open-source Islamic model built for calm, grounded, citation-aware answers.
Alif-1.0 is an open-source Islamic AI model fine-tuned on top of Qwen3.5-2B by Alibaba, trained on 1,042 carefully curated instruction-response pairs covering Quran, Hadith, and Islamic history.
The name Alif is the first letter of the Arabic alphabet, the first letter of Allah, and the first letter of Al-Quran.
A compact fine-tune focused on Islamic source discipline rather than scale for the sake of scale.
1,042
instruction-response pairs
783
training steps across 3 epochs
0.30
final loss, down from 1.84
~3 hrs
free T4 training time
Mood-based ayah guidance, Tawheed, Salah, Sabr, hope, anxiety, death, and afterlife, using Sahih International and Pickthall translations only.
Built from Sahih Bukhari, Sahih Muslim, Sunan Abu Dawud, Jami at-Tirmidhi, and more, with collection name, book number, and hadith number attached.
Drawn from Ibn Kathir, Ibn Khaldun, Al-Tabari, and other classical scholars, spanning the life of the Prophet PBUH, Sahabah, Karbala, the Crusades from the Muslim perspective, the Ottoman and Mughal empires, and modern Islamic geopolitics.
Qwen3.5-2B by Alibaba, released under Apache 2.0.
LoRA using the Unsloth library.
Google Colab on a free T4 GPU plus Kaggle on free T4 x2 GPUs.
Dataset research support came from Zai by Zhipu AI and Claude by Anthropic.
The model was tuned to stay cautious, cite responsibly, and defer when scholarship is needed.
Never fabricates ayahs or hadiths.
Built-in responses against Islamophobia and extremism.
Cites Quran 5:32 against any violence-related questions.
Respects all religions and does not speak negatively about other faiths.
Every response includes a beta disclaimer.
Directs complex fiqh questions to qualified Islamic scholars.
Alif-1.0 is an experimental beta model fine-tuned on a small dataset of 1,042 pairs. Responses may be inaccurate or incomplete. Always verify important Islamic matters with a qualified Islamic scholar. ImanVibes and its creators accept no responsibility for any decisions made based on this model.
The model page is live, while product integrations are still on the way.
Alif-1.0 is open source under Apache 2.0 license