Privacy Policy

Who we are

VerseFlame is church-presentation software, operated from Canada. If you have any questions about your privacy or this policy, email us at privacy@verseflame.com.

What we store

Your account. Your email address and a securely hashed password, plus your name if you provide one.

Your church’s content.The church and campus records you create, the songs and lyrics your church enters, your service plans, the images and videos you upload, links to media you use (such as YouTube clips), and the current state of what’s showing on your displays. Your church owns this content. Uploaded images and videos are served from unlisted links — that is how a projector shows them without signing in — so anyone given a link could open it. When you delete an image or video, it is permanently removed after 30 days.

Not your payment details. When you subscribe, your card information goes directly to Stripe, our payment processor. VerseFlame never sees or stores your card number.

If you join the waitlist.Your email address, which button you used, and your church’s name if you add it. We use it only to email you about VerseFlame, and we delete it whenever you ask — email privacy@verseflame.com.

How voice recognition works

While the microphone is on, the speech recognition happens in your own web browser, using its built-in speech feature. In Chrome, that means the audio is sent to Google’s speech service; in Microsoft Edge, to Microsoft’s. Either way it is handled under your browser’s terms. VerseFlame receives only the resulting text — the words the browser transcribed — not the audio itself. We never record the room.

That text does not always stay with us. Depending on which features your church is using, short passages of it — and during live translation, all of it — are sent on to Anthropic. See What we send to Anthropic below, which says exactly what goes, when, and what happens to it there.

If you joined on your phone during a service

Some churches use VerseFlame to ask the room a question — you scan a code or enter four digits, type a short answer, and it may appear on the screen at the front.

We do not know who you are.There is no account, no sign-in, no email, and no phone number (a trivia game can give you a nickname — see below). Your device is given a random identifier so the same person can’t answer one question twice, and so the person running the service can see how many phones have joined; it is not linked to you and we cannot use it to identify you.

Joining is counted, even if you never type anything. When you open the page, we record that a device joined — the random identifier and the time, nothing else. It lets the person at the front see that people are getting in, which is the difference between waiting and assuming the code is broken. It is a count of phones, not of people, and it is deleted on the same clock as everything below.

What you type — and the fact that you joined — is deleted within 48 hours of the session ending— automatically, whether or not it was shown on screen, and including anything the person running the service chose to remove. After that, only the church’s own question remains, and that is deleted within 90 days.

If you play Bible trivia.Some trivia games give each phone a nickname from a list (“Brave Otter”), which you can change to one word of your own — please not your real name. A name you type is checked before it can appear. The screen at the front shows only the top five, and you can choose not to appear there at all. Your score is worked out from your answers and is not kept, and your nickname is deleted when the session ends.

Diagnostic information

Off by default.At a church’s request, we can turn on a per-campus diagnostic setting that records what the system did— for example, “heard a reference and showed this verse” — to help us improve accuracy. This records structured events about the app’s own actions. It does not record the words of a sermon, prayer, or conversation.

That is a promise about what we store. It is not a promise that nothing was sent anywhere — see Being stored and being sent are different things below.

A partnership, for founding partners only. Founding partner churches — shown by a “Founder” badge in the app — have agreed to share more than the setting above collects, so that VerseFlame’s voice recognition gets better for everyone. It is a real bargain with two halves, and both are worth stating plainly.

What you share. On a founding-partner campus we keep, as text — never as audio:

• Everything the microphone transcribes while it is on — not only the commands. The browser hands us every finished phrase and we keep it, whether or not it was addressed to VerseFlame.
• The half-finished guesses. Before settling on a phrase, the browser offers rough interim versions — often wrong, sometimes nonsense. We keep some of those too, including words the browser heard and then left out of the finished phrase.
• During read-along, the words being read.
• During live translation, both languages — the English the microphone heard and the translation that was shown, side by side.

In plain terms: a service preached with the microphone on is kept as text. Not a summary of it — the words.

For 90 days. After that the words are stripped out automatically, by a job that runs every night. The row they came from stays — the counts, the timings, what the system did — because that is what shows us whether the software is improving over a year rather than over a Sunday. The words themselves are gone for good: there is no archive and no undo.

What it buys. This is the only record that shows us where the system was wrong. Counts and timings tell us a verse appeared; only the words tell us it was the wrong verse, or that a reference was spoken clearly and missed entirely. Much of what we have fixed came out of exactly this data — a book name misheard the same way five services running, a reference said twice and dropped both times, and a translation that repeated the previous sentence back to the reader. None of those were visible in the numbers. Your services are why they are fixed, for every church, including the ones that share nothing.

It is yours to leave. It applies only to founding partners, can be switched on only by us and never self-serve, and is never enabled for any other church — for every other church this spoken text is stripped out before anything is stored. If you would rather it was off, email privacy@verseflame.com and we will switch it off, and you keep the Founder badge and everything else that came with it.

VerseFlame’s own demonstration accounts keep everything, like a founding partner’s. If you try VerseFlame on one of ours, what you say into it is kept on the same terms.

We may access diagnostic data to identify and fix problems and to improve VerseFlame.

Service providers we work with

We use a small set of trusted providers to run VerseFlame. Each handles data only to do its job:

• Supabase— stores your church’s data and handles sign-in.
• Stripe — processes payments and calculates tax. Your card details go to Stripe directly; we never see them.
• Resend — sends our emails (sign-up confirmation, password reset, team invitations, and billing notices).
• Anthropic — runs the Claude AI behind several VerseFlame features. Because this is the one service we send what people say out loud to, and what they type into a word cloud, it has its own section below: What we send to Anthropic.
• API.Bible— provides the verse text for the modern translations we license. Their licence requires us to report each time one of those verses is shown, so your display screens load API.Bible’s own reporting script, which tells them which verse was shown, together with an identifier for your church’s campus — never a person.
• Cloudflare — stores and plays the videos you upload.
• YouTube— when you show a YouTube clip, your display plays it from YouTube, under YouTube’s terms.
• Vercel — hosts and serves the application.

What we send to Anthropic

These VerseFlame features work by sending text to Anthropic’s Claude. In every one we send text, never audio. Most send the words your browser transcribed; one sends what people type; one sends only Bible text.

• Find a verse by description.The description you type or speak, and anything you say to narrow it down (“the Old Testament one”), is sent to Claude to suggest a reference. This runs only when someone asks for it.
• Verse narrowing. When a chapter is already on the screen, short excerpts of what the speaker is saying are sent to Claude to work out which verse is being discussed, so the screen can follow along. This is on by default.
• Live translation. Where your church has it switched on, everything the microphone transcribes during the service is sent to Claude, continuously, for as long as translation is running — so it can be read in another language on a phone.
• Checking the translation. While live translation is running, each translated line is sent to Claude a second time, together with the English it came from, to check that it says what was said. It is part of live translation and runs only where translation does.
• Grouping word-cloud answers.When your church puts a word cloud on the screen, the answers people type are sent to Claude while the cloud is still collecting, so answers that mean the same thing (“grace”, “gracious”, “so much grace”) show as one. Only the answers are sent — never who typed them, and never anything that could identify them. Whoever runs the cloud can switch this off for any question. This is on by default.
• Writing Bible trivia. To write trivia questions, passages from the public-domain Bible translations built into VerseFlame are sent to Claude. Nothing anyone says or types is sent — only the Bible text the questions come from.

If you are reading a translation on your phone, it is held for 45 minutes. The translated lines, and the English they came from, have to sit somewhere between being spoken and reaching your phone, so we keep them in a short-lived buffer and delete them automatically after 45 minutes. It is there so that if you lock your phone during a song and come back, you can catch up on what you missed instead of finding a hole in the middle of the sermon.

It is a buffer, not a record. Nothing is archived, nothing is exported, and no transcript of the service is kept from it — a deletion job runs every five minutes and removes anything older than 45 minutes, whether anyone read it or not. It was ten minutes until September 2026, which turned out to be shorter than the time people actually put their phones in their pockets. We do not keep it longer than this because your phone only ever shows the most recent part anyway: past about 45 minutes we would be storing what you spoke, with no one able to read it.

When a Bible verse card appears on your phone, we note which verse it was, your language, and how long after the verse came up it reached you — never the verse text and nothing that identifies you — so we can tell whether verses reach readers in time.

If your church offers read-aloud, the voice is your phone’s own. When you press Listen, your phone speaks the translated lines using the voice built into it. We do not send anything new to do this, and we do not record anything. Your phone’s maker may run that voice on its own servers on some phones, under their privacy terms, not ours. We note that you pressed Listen, your language, and how long you listened, with nothing that identifies you, so your church can tell whether anyone uses it.

If your church offers the premium voice,the translated lines you listen to are sent to ElevenLabs to be turned into speech, under ElevenLabs’ own privacy terms. Only the translated words are sent — never what the microphone heard in English, and nothing about you. The audio is kept on our servers for up to 45 minutes, so other listeners in your language hear the same line without it being made twice, and is then deleted. Your church pays for it; you are never asked to.

Anthropic does not use any of it to train their models. Their commercial terms say so, and we have not joined any program that would allow it. We checked our Anthropic account settings on 5 September 2026 to confirm.

Anthropic deletes what we send within 30 days.There is one exception and we would rather you heard it from us: if Anthropic’s automated safety systems flag a passage, they may keep it for up to two years. That is their process, not ours, and we cannot switch it off. We mention it because a prayer request is exactly the kind of thing that could be flagged.

Being stored and being sent are different things

This is the part most likely to be misread, so we want to be plain about it.

The diagnostic settings above control what VerseFlame keeps. With them off, we keep nothing of what was said in our records — the words are stripped out before anything is written down, and what is left is counts and timings. The one exception is live translation, where it is switched on: each line it translates, and the English it came from, sit in the 45-minute buffer described above so phones can show them.

They do not control what leaves the building. If one of the features above is running, the text goes to Anthropic whether or not we keep a copy. Turning diagnostics off means we are not storing your sermon. It does not mean the words stayed in the room.

If you would rather the automatic ones were off for your church, email us at privacy@verseflame.com and we will switch off verse narrowing, live translation, or both. Find a verse by description only runs when someone asks for it — by typing a description, or by saying something like “find the verse about…” while the microphone is on.

How long we keep your data

We keep your church’s data until you ask us to delete it. Email privacy@verseflame.comand we’ll remove it.

Some things expire on their own, without you asking. Anything a congregant types on their phone during a service is deleted within 48 hours of the session ending, and a trivia nickname when the session ends. Translated lines for phones are deleted after 45 minutes. An image or video you delete is permanently removed after 30 days. And wherever we hold the text of something someone said, those words are stripped out after 90 days by a job that runs every night — the surrounding record stays, the words do not.

Your choices

• Ask us to turn the diagnostic setting on or off at any time, and we’ll do it — email privacy@verseflame.com.
• The founding-partner voice-capture tier is off unless you are a founding partner who has arranged it with us directly.
• Ask us to delete your data, and we will.

Where we operate

VerseFlame is operated from Canada and follows Canadian privacy law (PIPEDA). Wherever you are, we aim to honor reasonable requests to access or delete your information.

Children

VerseFlame is used by churches and operated by adults — staff and volunteers. The founding-partner voice-capture tier could incidentally include the voices of minors present during a service. We are reviewing the specific obligations here with legal counsel before we open to the public.

Changes to this policy

If we change how we handle your data, we’ll update this page and the “last updated” date above.

Contact

Questions? Email privacy@verseflame.com.