⚡ 100% Local macOS Intelligence

Screen recording that actually transcribes and writes meeting notes.

Native ScreenCaptureKit engine. Lossless system audio + microphone capture. Local Whisper transcription. Realtime live assist HUD. Zero cloud leaks.

$ npm install -g localscribe
macOS 13+ Sonoma / Sequoia Apple Silicon & Intel Bun + Swift + FFmpeg 100% Test Coverage
terminal — 80x24
v0.1.0

~ $ localscribe start --name "Q4 Architecture Sync" --live --notes

building capture helper... done

recording -> 2026-10-21_14-30-00_q4-architecture-sync.mp4

stop with: localscribe stop

live assist: opened in foreground HUD [TUI active]


[14:30:05] sysaudio attached

[14:30:06] microphone channel 0 linked

[14:30:08] whisper live segmenter: ggml-base.en active

⚡ Status: Recording 2026-10-21_14-30-00_q4-architecture-sync (0m15s)

SCREENCAPTUREKIT ✦ ZERO VIRTUAL AUDIO DRIVERS ✦ LOSSLESS SYSTEM AUDIO ✦ LOCAL WHISPER TRANSCRIPTION ✦ HERMES AI MEETING NOTES ✦ BANGLA / MULTILINGUAL WHISPER ✦ LIVE TUI HUD ✦ SCREEN ANNOTATION ✦ NATIVE SWIFT MENU-BAR APP ✦ SCREENCAPTUREKIT ✦ ZERO VIRTUAL AUDIO DRIVERS ✦ LOSSLESS SYSTEM AUDIO ✦ LOCAL WHISPER TRANSCRIPTION ✦ HERMES AI MEETING NOTES ✦ BANGLA / MULTILINGUAL WHISPER ✦ LIVE TUI HUD ✦ SCREEN ANNOTATION ✦ NATIVE SWIFT MENU-BAR APP ✦

Engineered for Deep Technical Work

01

ScreenCaptureKit Engine

Native macOS ScreenCaptureKit capture. Record full displays, individual windows by title or app, or interactively drag-to-select regions on screen.

02

Lossless System Audio + Mic

No BlackHole. No Soundflower. No kernel extensions. Localscribe taps the ScreenCaptureKit system audio stream directly and isolates mic audio on separate tracks.

03

Local Whisper Transcription

Powered by local whisper.cpp. Automatically parses language or runs Bangla with large-v3-turbo, Hindi, Urdu, Spanish, Arabic, or English base.

04

Live Assist TUI HUD

A dual-speaker realtime terminal feed while recording. Distinguishes who spoke (me vs them). Ask Hermes questions in real time with ?question.

05

Timestamped AI Notes

Hermes turns messy conversation transcripts into structured notes: Summary, Key Points, Decisions, and Action Items stamped with transcript clock [mm:ss].

06

Screen Annotation Overlay

Draw directly over the screen during presentations or reviews. Native transparent Metal-backed canvas spawns as a sidecar process.

CLI Command Arsenal

Command Options Description
localscribe start --name, --out, --live, --notes, --transcribe, --draw, --lang Starts recording screen + system sound + mic. Opens live TUI or detaches with -d.
localscribe stop (none) Stops the active recording session, runs stop hooks (Whisper transcription & Hermes notes).
localscribe status --json Checks whether recording is active, returns file path, start timestamp, and elapsed time.
localscribe live --role <role>, --context <file> Re-attaches the realtime interactive TUI HUD to a detached background recording.
localscribe captions (none) Streams live speaker-attributed transcription text lines to standard output.
localscribe notes <file> --lang <language>, --name <name> Regenerates structured AI meeting notes from an existing recording or transcript.
localscribe devices --json Lists all connected displays, active macOS windows, and available microphones.
localscribe list (none) Lists all finalized video recordings in ~/Movies/localscribe sorted newest first.

Zero-Surprise Architecture

┌────────────────────────────────────────────────────────────────────────┐
│                          LOCALSCRIBE CLI                               │
│              (TypeScript + Bun Runtime / Process Runner)               │
└────────────────────────────────────┬───────────────────────────────────┘
                                     │
           ┌─────────────────────────┼─────────────────────────┐
           ▼                         ▼                         ▼
┌─────────────────────┐   ┌─────────────────────┐   ┌─────────────────────┐
│  SCREEN CAPTURE     │   │    AUDIO ROUTING    │   │  WHISPER LIVE ENGINE│
│  Native Swift (SCK) │   │  System Audio (SCK) │   │  ffmpeg 16k mono wav│
│  Displays / Windows │   │  Mic Audio (ffmpeg) │   │  whisper.cpp worker │
└──────────┬──────────┘   └──────────┬──────────┘   └──────────┬──────────┘
           │                         │                         │
           └─────────────────────────┼─────────────────────────┘
                                     ▼
                      ┌─────────────────────────────┐
                      │    RECORDING COMPLETE       │
                      │  ffmpeg AAC + H.264 Muxer   │
                      └──────────────┬──────────────┘
                                     ▼
                      ┌─────────────────────────────┐
                      │    HERMES NOTES ENGINE      │
                      │  Hardened temp sandbox      │
                      │  Delimited untrusted block  │
                      │  Summary + Actions [mm:ss]  │
                      └─────────────────────────────┘