Pulpit AI listens for a scripture reference and puts it on the projector — no typing, no slides to build the night before. Runs on the church laptop, works without internet.
No slide decks, no scrambling to find a verse mid-sermon. Just speak — or type — and it's up.
Say “John 3:16” or even a loose paraphrase and the verse is found and projected. Hold-to-talk, or hands-free continuous listening.
KJV, WEB and ASV ship with the app; hymns project stanza-by-stanza. Arrow keys or “next” walk through verses live.
Choose a sanctuary gradient, a calm starfield, a living aurora — or drop in your own image. The congregation screen never has to be plain black.
Compose announcements, giving details with a QR code, or a quick notice — and put them on screen with one click. Bookmark the ones you reuse.
Save a whole service order and step through it. Enable extras like a live clock or slide templates — or write your own handler plugin.
Off by default. Bring your own key to route transcription through Groq or resolve tricky paraphrases with DeepSeek — the verse always comes from the local Bible.
The operator stays in control — nothing reaches the congregation screen until it's meant to.
Hold the mic and say the reference, leave it listening through the sermon, or type it into quick reference. Live transcription shows what it's hearing.
A confident match goes straight up. A borderline one waits as a card you tap to accept — so a mis-hear never surprises the congregation.
The verse, hymn or slide fills the second screen over your chosen backdrop. Step forward and back, or blank it, without touching a slide deck.
The details that matter when you're the one trusted with the screen.
Speech runs on-device by default (transformers.js), so audio never leaves the laptop. A speech-band filter, echo cancellation and adaptive voice detection lift the preacher's voice out of music and crowd noise, and a neural voice-activity filter drops non-speech before it's ever transcribed.
Live interim text appears while someone is still speaking, so you're never talking to a ghost. For the loudest rooms, an optional cloud transcription tier (Groq Whisper) is one toggle away.
Bundled public-domain translations (KJV, WEB, ASV) work fully offline. Say a range like Romans 8:38-39, then step verse-by-verse with the arrow keys or “next”. Hymns project one stanza at a time, wrapped as sung lines rather than prose.
A dedicated congregation window scales type to the screen. Pick from built-in backdrops — Sanctuary, Midnight, Dawn, Aurora, Starlight — or use your own picture, with a dimmer so verse text stays readable. A live/preview toggle lets you line up the next thing without the congregation seeing you work.
Compose announcements and giving slides (with a QR image), save reusable ones as bookmarks, and lay out a full service as a preset you step through. Extras ship as toggleable plugins — a digital clock, slide templates — and you can import your church's own hymnal or Bible as a content pack.
Offline is the default and the app is fully functional that way. Cloud tiers are opt-in, use your own API key, and fall back silently to on-device if a call fails. Even with the smart-matching tier on, the model only ever returns a reference — the scripture text itself is always read from the local dataset, never generated.
Free, and built fresh for each release by CI on every platform. Pick your operating system.
First launch downloads the on-device speech model once (~50–80 MB), then runs offline.
Pulpit AI ships with the King James Version, World English Bible, and American Standard Version built in (all public domain). Need something else? A translation is just a content pack: one file, installed from inside the app (Plugins → "Import a hymnal or Bible…"). No restart, no reinstalling the app.
More translations, including how to add ones still under copyright, are on the wiki.
Pulpit AI runs fully offline out of the box. In a noisy room, or when the preacher paraphrases a quote instead of citing chapter and verse, two optional cloud tiers can help: Groq for faster, more accurate transcription, and DeepSeek for matching a loose paraphrase to its verse. Both stay off until you turn them on in Settings, and each uses your own API key, billed by that provider directly, not by us.
gsk_).sk-).No. Everything core — voice recognition, scripture, hymns, projection — runs on the laptop with no connection. You only need internet for the one-time speech-model download and for the optional cloud accuracy tiers.
Uncertain matches don't auto-project — they appear as a confirmation card the operator taps to accept or dismiss. You can also type any reference directly in quick reference, and blank the screen instantly at any time.
Windows 10 or newer, macOS 11 or newer, and 64-bit Linux (AppImage or .deb). It runs on modest hardware — a smaller/faster speech model is available for older laptops, and a GPU is used automatically when one is present.
Yes — compose announcement and giving slides in-app and bookmark them, and lay out a full service as a preset. Developers can add capabilities through the plugin system documented in the wiki.
Pulpit AI itself is free. The optional cloud tiers use your own API keys and are billed by those providers only if you choose to enable them.
Install it on the booth laptop this week and never fumble for a verse again.
Download Pulpit AI