Fully Local Meeting Transcription Skill

Fully Local Meeting Transcription Skill, also referred to as transcribe-md, is a macOS command-line tool designed to transcribe meetings directly into markdown files using a local speech-to-text model. It addresses the need for seamless integration of meeting transcripts into local markdown-based note-taking workflows, particularly for users managing their notes with applications like Obsidian. The tool enables users to record both microphone and system audio, generating speaker-labeled transcripts that reflect conversations from any video call.

Key features include the capture of two separate audio channels—microphone and system audio—allowing for clear attribution of speakers in the resulting transcript. The tool incorporates echo deduplication, which automatically detects and removes duplicate audio caused by using speakers instead of headphones, minimizing transcription errors. en model, with Metal GPU acceleration, ensuring that no audio data leaves the device.

Transcribe-md provides real-time transcription output, appending text to the designated markdown file as each segment is processed, so users can observe the transcript as it is created. The tool offers configurability, allowing users to set duration limits for recordings, operate in microphone-only mode, and select specific input devices according to their needs. It can be installed via a shell script and is registered as a skill in platforms such as Claude Code and Cursor, where it can be invoked with a command to begin recording and transcription.

The tool is suitable for individuals who value privacy and want direct control over their meeting transcripts, such as those who prefer local note-taking systems or wish to automate parts of their meeting documentation workflow.