Oracle chat

1. Your birth details load a deterministic saju profile — pillars, traits, and timing reads.
2. When you ask something about your chart, Oracle has to fetch the answer: your natal reading, your glyph, a day, a month, a year, a decade, a friend pairing, or a public figure pairing. Each one is a separate request to the engine, and the engine is the only place those numbers exist.
3. Your glyph card is the one part we can promise outright: the app stamps it onto the answer after the fact, so the portrait, number, and archetype are read from your chart rather than written by the language layer.
4. Personal memory (when signed in) carries preferences you have mentioned across chats. It shapes wording; it is never a source of chart facts.
5. Daily luck like/dislike and timeline resonance tags can steer tone — again framing only, not new substance.

Everything above is how the app is built, not a guarantee we can check sentence by sentence. Oracle is asked to answer only from what the engine returned, and it is given no other way to reach your chart — but no automatic check reads a finished answer back and confirms every number in it. If something Oracle says does not match what you see on the Chart screen, the Chart screen is the one to trust.

Asking your own questions is part of Plus. Without it the Oracle screen still opens and still reads your chart, but you pick from a fixed set of questions instead of writing your own, and nothing is saved between visits.

Chat asks your permission before the first question, because writing the answer means sending your question and the engine's findings to Anthropic. Decline and every other part of the app works exactly as before — the fixed-question screen makes no such request.

With Plus you can ask 322 questions a month. The composer starts telling you how many are left once you are down to ten, so the wall is never a surprise, and the count resets on your own subscription date — subscribe on the 20th and the questions come back on the 20th, not on the 1st. It is a monthly allowance rather than a daily one on purpose: asking four questions in one sitting and none for a week is how people actually read, and a daily cap punished exactly that.

Multi-turn threads are saved to your account and listed in the left drawer. Each chat is separate; you can delete the one you have open from the Oracle screen, or all of them at once from the account sheet at the foot of the drawer. Clearing chats does not clear what the Oracle remembers about you — that is a second control in the same sheet, Clear Oracle memory, which empties the summary on your account and on this device while leaving your saved chats alone. The Profile screen has the same control. Neither one touches your resonance feedback, and clearing the memory does not stop the Oracle building a new summary from your next conversation.

Oracle can read a pairing two ways, and they are not the same on consent. Friends are people who accepted your invite or whose invite you accepted, so the pairing rests on something they did. Public figures are read from our own catalog of well-known people, who have not agreed to anything — the same as comparing yourself to a celebrity on the Match screen. Oracle cannot reach anyone else: it has no way to read a person you have only described to it, and it does not guess a chart from a name.