Rafiq does not answer from memory. It searches a library of the Quran, the hadith, tafsir and the works of scholars, then writes the reply from what that search found, with the sources shown under it.
The library holds about 1.17 million passages:
First Rafiq works out what is being asked. It reads past typos, follows on from earlier turns in the chat, and notices when a question is outside what it can answer.
Then it searches the library two ways, by meaning and by the words used, and looks to the scholars' rulings first. The passages it finds are ranked against the question, and only those close enough to it are kept.
A language model then writes the reply from those passages, and the sources it used are listed under the answer, so each one can be opened and read.
If the library holds nothing close enough to the question, Rafiq says so. It does not fill the gap with a guess.
Rafiq follows Ahl al-Sunnah wal-Jama'ah. Where the scholars differ on a ruling, it gives the Hanbali and Athari position first, then notes the other positions briefly and says whose they are. That leaning is stated here so that it is never hidden.
Some questions are not only questions of fiqh. When one touches a medical emergency, a person in danger at home, or a punishment that only a court may carry out, the answer opens with what to do first, such as seeing a doctor or getting to safety, before the ruling.
An answer shows what the scholars said about a question. It is not a fatwa, and it is not a ruling on your own case. A question that turns on your own circumstances belongs with a qualified scholar who can hear them.
Every source is linked under the answer. Read it, especially before acting on a ruling. If an answer is wrong, the flag under it sends a report saying what was wrong, and reports are read to correct Rafiq.
The reply is written by DeepSeek, a language model. It is given the passages the search found and instructed to write from them.