Changelog

Improved

• Performance improvements. A round of optimisations across the live detection pipeline for a smoother service on lower-end machines.

New

• Apple Speech (on-device) provider. Macs can now transcribe a full service entirely on the device using Apple's built-in Speech framework — no API key, no cloud round-trip, and nothing leaves your machine. Pick it under Settings → Transcription → Provider → Apple Speech. First use prompts for Speech Recognition permission; if you've never used Dictation in the chosen language, macOS will download the on-device model the first time.
• Sermon-point templates. Each detected sermon point is now classified into one of five overlay templates — Sermon point, Numbered, Key thought, Scripture, or Group — and the broadcast header reflects the template choice. Bible references in a point automatically route to Scripture; speaker numbering ("point one", "secondly") routes to Numbered; quotable standalone lines route to Key thought. Every row in the Sermon Points panel shows a small template chip you can tap to swap the template when the AI guessed wrong; if the point is already on air, the overlay re-renders immediately.
• Grouped sermon points. When two to four points belong together — a numbered series the pastor laid out, paired contrasts, sub-points of a parent thought, or an end-of-section recap — the AI now emits a single Group row that stacks all of them on one overlay card. Each child gets its own row with its own number, its own body, and an optional inline Bible reference rendered in a smaller, lighter style next to the body. An optional group headline sits above the children. The originals stay in the outline too, so the operator can show the group or any individual point.

Improved

• Faster live verse detection on Apple Speech. Apple's on-device recognizer keeps refining a partial transcript and only commits it to a final result on long pauses or session rollover — which delayed bible verse detection until the operator stopped speaking for many seconds. Kairos now commits a partial as a final after about a second of stable text, so verses surface in the feed as soon as the phrase that contains them lands. Detection on Deepgram, Gladia, OpenAI, and Kairos managed STT is unchanged.
• No more vanishing partials. Long in-progress phrases no longer silently disappear from the live transcript when Apple's recognizer rolls to a new utterance — the prior phrase is now committed as a permanent transcript line.
• Sermon-point overlays stay readable. Each template enforces its own character and line caps before anything reaches the broadcast canvas, so a long-winded model response no longer overflows the text box mid-service. Oversize copy is rewritten or, as a last resort, gracefully truncated rather than spilling off-screen.
• Stable rows in the Sermon Points panel. Points are now diff'd by a content hash across extract polls, so rows no longer re-mount and reset their open dropdown every 30 seconds when the model echoes the same outline back.

New

• Sermon Points. As your pastor speaks, Kairos listens in real time, identifies the key points, and turns them into a clear bullet outline in the new Sermon Points panel. Each bullet is one click away from the broadcast — tap to display, tap again to hide. Updates roughly every 30 seconds from the live transcript. Turn it on under Settings → Sermon points.
• Sermon Notes. When the service ends, Kairos compiles every verse cited, the key points, and a full sermon summary into polished notes from the saved session. Each note can be downloaded as a PDF or published to a shareable link (kairos.live/n/<id>) you can send to your congregation as the closing message. Find it under Sessions → the session you want → Notes.
• Session audio recording. Recording now captures the room audio too, not just the transcript. Every saved session shows a full-width waveform player above the transcript tab — click anywhere to scrub. Audio lives next to the session and is removed automatically when the session is deleted.
• Shared media controls. A single play / pause / scrub bar now sits between the canvases and the Up Next grid whenever a video is staged. It drives the video on both Preview and Program — they stay in lock- step, no more split native controls and no audio doubling. The bar hides itself automatically when the active media is an image or a lyric is on screen.
• Songs in the queue. Songs can be queued like verses and media. Hitting the play icon on a queued song stages it and flips Program on air in one click — no second trip to the Lyrics panel.
• OpenLyrics XML import. Settings → Lyrics → Import now accepts a single OpenLyrics .xml file or a folder of them, alongside the existing EasyWorship and OpenLP importers. Duplicate titles are skipped so re-running an import never overwrites manual edits.

Improved

• Start recording mid-service. The live timer pill is now clickable while transcription is running — one tap starts the recording even if you closed the "Record this session?" prompt earlier. Click it again to stop and save.
• Send to live, every time. The PREVIEW Send to live button now enables for any stageable content — verse, lyric, or media — not just Bible verses, so a queued video or a manual lyric goes on air with the same single action.
• Overlay order matches what you see. When you send a lyric on air while a video is staged, the lyric now fully replaces the video on both canvases (and the playback bar steps aside) instead of being painted over. Hide the lyric to bring the video back.

New

• 2 free hours of managed transcription. Every new account is seeded with 120 minutes that drain before anything else. Enough to run a full Sunday service before deciding whether to top up the wallet or subscribe to Pro. No card required to start.
• Trial balance on the desktop and dashboard. The Account tab and Dashboard → Billing both show how many free minutes are left while the grant is active, and hide it once you've moved to wallet or Pro.
• Manual lyrics mode. The Lyrics panel now opens with two buttons — Pick song and Auto-detect. Operators who'd rather drive lyrics themselves can choose a song from the catalog, step through sections and lines with Prev / Next, and send each line to NDI Live on demand. The AI matcher only runs when explicitly turned on.
• EasyWorship & OpenLP import. Settings → Lyrics → Import bulk- loads songs from an EasyWorship plain-text export folder or an OpenLP songs.sqlite. Duplicate titles are skipped so re-running an import never clobbers manual edits.

Improved

• One unified organization model. Personal and church accounts are now the same thing — every signup gets one organization, named after the church you enter at signup, and that organization can invite branches, set per-branch hour caps, and manage team members immediately. Existing personal accounts were automatically upgraded; no "create your church" step anymore.
• Drain order is now trial → Pro → wallet. A Pro subscriber who's still inside their trial no longer accidentally burns their monthly 60-hour allowance before the free minutes; the trial bucket always empties first.

New

• Cloud (Kairos managed). A new STT option under Settings → Speech. Sign in once on the desktop and Kairos handles the connection to the transcription provider for you — no API key to copy, no card to attach to a third party. Minutes are metered server-side and the desktop shows your wallet balance live as the service runs.
• Pricing built for Nigerian churches. Top up your wallet from ₦10,000 upward, or subscribe to Pro (₦ equivalent of $200/year) for 60 hours of managed STT every month with wallet overage at ₦ equivalent of $0.40/hr. Live USD→NGN rate, charged through Paystack.
• Multi-branch churches. A head church can invite branch operators by email, set a monthly hour cap per branch, and see usage roll up across every campus from one dashboard.
• Dedicated Account tab. Sign in to Kairos, see this month's session usage, and sign out — all from one place in Settings → Account. The Speech tab now just points here when you pick Cloud.
• Billing history. Every top-up, session debit, refund, and adjustment shows up in your dashboard with a running wallet balance, scoped to whichever church you're viewing.
• Devices page. See every desktop signed in to your account and revoke any one of them remotely if a laptop walks off.

Improved

• Secure desktop sign-in. Replaced the old "paste a token" flow with a one-time device code: the desktop shows KAIROS-XXXX, you confirm it in the browser, and a short-lived access token is minted to a rotating refresh token in the OS keychain. JS never sees the refresh token.
• Stuck "connecting" recovers automatically. If a managed session fails to start (network blip, expired key), the Start button comes back without needing to click Stop first.
• English forced on managed Cloud sessions. Nova-3's auto-detect was occasionally tagging Nigerian-accented English as French and surfacing the French Bible. Cloud sessions now lock to English; if you need another language, use BYOK with a provider of your choice.
• Tightened voice command triggers so bare topic mentions like "in french" no longer flip the active translation mid-sermon.

Fixed

• The active translation no longer silently switches to a French Bible when the speaker quotes a verse in mixed-language preaching.
• Sessions that ended mid-minute previously rounded down; final usage now rounds the trailing partial up to the next billable minute so the wallet balance always reconciles.

Improved

• Downloads moved to a dedicated CDN. Release binaries are now served from Google Cloud Storage instead of GitHub Releases, so you no longer need a GitHub account to install Kairos.

Fixed

• Installer would silently fail on a clean machine when a previous version had left orphaned config files. Installs are now fully self-contained.

New

• Real-time verse detection. Kairos listens to your sermon feed, transcribes speech as it happens, and surfaces the Bible reference the moment your pastor cites or quotes it.
• Broadcast-ready NDI output. Send detected verses straight to vMix, OBS, Tricaster, or any NDI-compatible switcher as a lower-third overlay.
• Music detection. Worship segments are automatically skipped so the overlay never fires on song lyrics.
• Next/previous verse navigation. Once a passage is on screen, advance or step back through the chapter with a keystroke or remote command.
• Voice control. Producers can hold a chosen translation, jump to a passage, or hide the overlay using spoken commands during the service.
• Guided onboarding. First-run setup walks new churches through audio routing, translation choice, and broadcast output in under five minutes.

Supported translations

KJV, ESV, NIV, NKJV, NLT, and CSB are bundled at install time. No internet connection required after setup.