The pitch

Record meetings.
Transcribe and summarize them.
All on your Mac.

Lore captures your microphone and the meeting's system audio, runs Whisper locally to transcribe, and uses an on-device Llama model to summarize and answer questions. Nothing you say or hear ever leaves your machine.

What's inside

The feature set

• 🎙 Dual-channel capture Mic + system audio recorded simultaneously, mixed cleanly with neural echo cancellation.
• ✍ On-device transcription Whisper runs locally. Pick tiny / base / small.en for the speed-vs-accuracy tradeoff you want.
• ✨ Summaries Llama 3.2 1B (4-bit) generates structured summaries the moment a transcript lands.
• 💬 Q&A Ask anything about a recorded meeting. Each conversation persists per meeting.
• 🔍 Cross-meeting search ⌘K from anywhere. SQLite FTS5 ranks results by relevance and jumps to the segment.
• 🎵 Word-level karaoke Transcript highlights the current word as audio plays. Click any word to seek there.

The privacy promise

Nothing leaves your Mac.

Audio stays on disk. Transcripts stay on disk. Summaries and Q&A run through a model file on disk. There is no cloud component, no analytics, no telemetry, no API call to anywhere except Hugging Face — once, on first use, to download the models.

Before you install

System requirements

macOS

14.0 (Sonoma) or later. Tested through 14.5 / 15.x.

CPU

Apple Silicon only — M1, M2, M3, M4 of any flavour. Intel Macs not supported (the LLM runtime uses MLX, which is arm64-only).

Disk

App is ~20 MB. First-run model downloads add ~75 MB (Whisper tiny.en) and ~700 MB (Llama 3.2 1B). Plan for ~1 GB free.

RAM

~330 MB while idle, ~1.5 GB while a summary or Q&A is generating. 8 GB Macs are fine.

Step 1 of 3 — install

Install

1. Open Lore.dmgDouble-click the file I sent you. The disk image mounts and a window opens.
2. Drag Lore into ApplicationsDrop it onto the Applications shortcut inside the DMG window.
3. Eject the DMGRight-click the disk in Finder → Eject. (Optional, but cleaner.)

Step 2 of 3 — first launch

The macOS warning is expected.

Lore isn't notarized by Apple yet (that costs $99/yr and I haven't paid it). macOS will block the first launch unless you opt in explicitly. The trick: don't double-click the first time.

1. Open ApplicationsFinder → Applications, find Lore.
2. Right-click Lore → OpenOr two-finger-click on a trackpad. The context menu shows "Open" at the top.
3. Click Open in the dialogThe warning says "developer cannot be verified". Click Open. (The default Move to Trash button is what blocks you.)

Step 3 of 3 — permissions

Two permissions to grant.

Both permissions only fire when you actually start a recording. Lore won't ask until you click ⌘R.

Microphone

Captures your voice. Required. Granted via the standard macOS prompt the first time you record.

Screen & System Audio Recording

Captures the audio playing through your speakers — the other people on the call. Without this, Lore only records you.

Now you're in

What you'll see on first launch

1. Splash screenWordmark fades in, then "Local Recorder" appears below. Click Enter when you're ready (or hit Return).
2. Library is mostly emptyYou'll see six sample meetings tagged with a "Sample" badge. Trash them when you're tired of looking at them — they're just there so the app doesn't feel barren.
3. Hit ⌘R or click "New recording"Permission prompts fire on this first attempt. Grant both, then you're recording.

Try it for real

Record a meeting

For your first test, anything that produces both speech and system audio is great — a Zoom call, a YouTube video you talk over, a podcast you respond to.

1. Hit ⌘RRecording panel opens. The clock starts. Mic + remote levels show.
2. Talk and let stuff play through your speakersBoth tracks are captured separately and combined later — no echo bleed.
3. Hit StopDetail view opens, showing a "processing" loader where the audio dock will be.

What happens next

Post-processing

Five steps run automatically. The progress bar at the top of the detail view shows you which one's active.

• 1. Echo cancellationDTLN-aec on Apple Silicon's Neural Engine. Cleans speaker bleed out of the mic track. ~30 s for a 5 min recording.
• 2. Audio mergeStereo mix.m4a written: your voice on the left, remote on the right. ~10 s.
• 3. TranscriptionWhisper. First run downloads the model (~75 MB). After that, ~1 min for 5 min of audio.
• 4. SummaryLlama 3.2 1B. First run downloads ~700 MB. After that, summaries take ~10 s.
• 5. Search indexFTS5 ingestion. Instant.

Listen back

Audio dock + word-level highlighting

When the meeting hits "Ready", the dock at the bottom loads mix.m4a. Click play. Watch the transcript above — the current word lights up as it's spoken. Click any word in the transcript to jump there.

Ask anything

Q&A

On the Q&A tab, type a question about the meeting and hit Ask. Llama generates the answer locally. Conversation history persists per meeting — close Lore, come back later, the thread is still there.

Things to try: "What were the action items?", "Did we agree on a date?", "Quote what Priya said about the deadline."

Find anything you've recorded

⌘K — cross-meeting search

Hit ⌘K from anywhere in the app. A Spotlight-style palette appears. Type any word or phrase — results stream in from every transcript with the matched word highlighted. Use ↑/↓ to navigate, ↵ to jump to that exact segment in the transcript view.

Things to test

How to help me

• Record a real meetingZoom, Meet, Teams, FaceTime, in-person on speakerphone — whatever you have. Watch for clarity in the cleaned mic track.
• Try a long recording30+ minutes. Memory should stay sane, transcription should finish, summary should still be coherent.
• Try Q&A on a meetingAsk 5+ questions. Notice if answers stay grounded in the transcript or hallucinate.
• Try ⌘K searchSearch for a phrase you remember. Confirm it lands in the right segment.
• Switch themesSettings → General → Theme. Five palettes — see if any feel off.
• Send a screenshot or noteIf anything breaks, looks weird, or just feels wrong. Reply to the message you got the DMG in.