trnscrb

Product Introduction

1. Definition: trnscrb is a macOS-native menu bar application for local, on-device meeting transcription leveraging Apple Silicon’s Neural Engine. It falls under the technical category of privacy-first productivity tools, using OpenAI’s Whisper model for offline speech-to-text conversion.
2. Core Value Proposition: trnscrb eliminates cloud dependency and subscriptions by offering end-to-end offline transcription for Zoom, Google Meet, Teams, Slack, and FaceTime. Its primary keywords include local transcription macOS, offline meeting notes, and private speech-to-text.

Main Features

1. Auto-Detection Engine: Triggers recording when mic activity exceeds 5 seconds alongside active meeting apps (browser or native). Uses macOS process monitoring for real-time detection without background processes.
2. On-Device Whisper Transcription: Runs the Whisper small model optimized for Apple Silicon Metal acceleration. Processes audio locally—zero data leaves the device—and deletes recordings post-transcription. Supports speaker diarization via HuggingFace tokens.
3. Claude Desktop MCP Integration: Saves transcripts as searchable plaintext in ~/meeting-notes/. Enables semantic search, summarization, and action-item extraction via commands like get_last_transcript or enrich_transcript (Claude API optional).
4. BlackHole Audio Capture: Optional virtual driver for dual-channel recording (mic + system audio). Installs automatically via trnscrb install, enabling full participant audio capture without third-party tools.

Problems Solved

1. Pain Point: Addresses data privacy risks in cloud-based transcription tools (e.g., Otter.ai) by keeping audio/transcripts 100% on-device. Solves subscription fatigue with a one-time purchase model.
2. Target Audience: Ideal for security-conscious professionals (remote engineers, project managers, legal teams) and Apple Silicon Mac users needing compliant meeting documentation.
3. Use Cases: Essential for scrums, client calls, and compliance-sensitive industries (healthcare, finance) where cloud storage of recordings violates data sovereignty laws.

Unique Advantages

1. Differentiation: Unlike cloud-dependent competitors, trnscrb requires no accounts, internet, or ongoing fees. Outperforms manual note-taking with timestamped speaker labels and Claude API-augmented analysis.
2. Key Innovation: Whisper + Metal synergy enables near-real-time transcription on M1/M2/M3 chips. The Claude Desktop MCP integration creates a closed-loop workflow—from recording to AI-powered insights—without data exports.

Frequently Asked Questions (FAQ)

1. Does trnscrb work without internet?
   Yes. Recording, transcription, and transcript searches operate entirely offline. Only trnscrb enrich (optional AI summaries) requires internet to call Claude’s API.
2. How does speaker diarization work in trnscrb?
   Speaker labels (SPEAKER_00, SPEAKER_01) require a free HuggingFace token for PyAnnote-audio modeling. Without it, transcripts still generate but lack speaker separation.
3. Is BlackHole safe to install?
   BlackHole is an open-source macOS virtual audio driver. trnscrb automates its setup for system audio capture, and all code is MIT-licensed for transparency.
4. Can trnscrb transcribe non-English meetings?
   Whisper small supports 99 languages. While optimized for English, it auto-detects languages like Spanish, German, or Japanese during local processing.