Sometime last year I moved all my notes to local markdown files, managed with Obsidian. I didn’t fully appreciate how useful this was going to be until AI coding tools got good enough to work with them directly.
One thing that was missing was meeting transcription. Tools like Granola are great, but they’re a whole separate ecosystem, and I wanted transcripts that would land directly in my markdown notes.
So I had AI build transcribe-md — a macOS CLI and skill that records your mic and system audio, transcribes with a local model, and writes clean markdown.
curl -fsSL https://raw.githubusercontent.com/hrescak/transcribe-md/main/install.sh | bashThis installs transcribe-md and registers it as a skill in both Claude Code and Cursor. Type /transcribe-md meeting.md and it starts recording. Here are the main features:
- Two audio channels — Captures your microphone and system audio separately, so you get speaker-labeled transcripts from any video call
- Echo deduplication — Detects duplicate audio when using speakers instead of headphones, and removes the transcription “echo” automatically
- Fully local model — whisper.cpp runs the
base.enmodel with Metal GPU acceleration. Nothing leaves your machine - Real-time output — Transcription is appended to your markdown file as each chunk completes, so you can watch it build up
- Configurable — Set duration limits, use mic-only mode, choose specific input devices
Example output
## Transcript — 2026-03-19 14:30
**[14:30:05] You:** So the main issue is the authentication flow breaks on mobile.
**[14:30:12] Them:** Right, I think the redirect URI isn't being handled correctly by the webview.
**[14:30:21] You:** Can we use the universal links approach instead?
**[14:30:28] Them:** Yeah, that should work. Let me check if the backend already supports it.
Meeting notes workflow
The skill really shines when composed with more skills. I have one that periodically checks the calendar, preps a meeting and starts a transcript. Another one takes the raw transcript, combines it with any notes, and produces a summary with action items and decisions. I’m really enjoying the flexibility of local markdown and coding agents for building systems.
transcribe-md is available on GitHub.