How It Works

The whole design has one goal: make it impossible for a model to put words in the Bible's mouth without you seeing it happen.

1. You state an opinion

Anything with a moral or doctrinal edge: “Divorce is always wrong.” “It's fine to be rich.” “Christians shouldn't drink.” “You have to forgive everyone.” Then choose Both, Case For, or Case Against. Both is the default because it is the honest one.

2. The app finds the passages before any model speaks

Your words are matched against two public-domain topical indexes, Nave's Topical Bible and Torrey's Topical Textbook, which between them file over five thousand subjects. The verse text itself is searched too. That produces a short list of candidate passages, with their text, that the model is handed as its evidence.

3. The model is held to the text

Its instructions are short and strict. Judge only by what the passages say. Never by its own view, never by what most people think today. Explicit commands and direct teaching weigh more than principles; principles weigh more than examples and narrative. For every passage it cites it must say which side of your opinion it lands on, give one sentence on why, and rate its weight from one to three.

It may cite well-known passages beyond the candidates, but only ones it is certain exist, and only by reference. It is forbidden from quoting verse text.

4. Every reference is checked

Each reference the model returns is looked up in the Bible bundled inside the app. If it resolves, you see the real verse text from the translation you chose. If it does not, that citation is removed and the screen tells you how many were dropped. Models do invent references; this is where that stops mattering.

5. The verdict and the bar

On top of every result is one sentence stating what scripture says about the opinion, and a bar with a percentage for each side. The percentages are computed by the app, not the model: the weights of the passages that resolved on each side, summed. A fabricated reference cannot tilt the bar, and the same passages always give the same split. It is labelled as the weight of evidence, because that is what it is.

What it cannot do

• It cannot settle theology. Serious people have disagreed about these passages for two thousand years. The app shows you the texts and how they weigh; it does not end the conversation.
• A verse cited for the wrong side is not caught. The reference check proves a passage exists, not that the model read it correctly. That is why the full text is always shown: so you can judge.
• Translation matters. Three public-domain translations are included and you can switch. None of them is the last word.
• The free on-device model is cautious. Apple's phone-sized model gives thinner answers than a large cloud model and tends to be one-sided. The app says so on the result. A key for Claude or another provider gives a fuller case.

Two ways to answer

On-Device	Apple's model on iPhone 15 Pro or newer with Apple Intelligence on. Free. Nothing leaves the phone. Older phones still get the passages the index found, with a note that explanations need a key.
Your Own Key	Claude, OpenAI, Gemini, OpenRouter, Ollama on your own network, or any server that speaks the OpenAI chat API. The key is stored in the Keychain on your phone. Your question and the candidate passages go to that provider under your account, and nowhere else.

Get The Beta