Paragraphos

Local podcast + YouTube → whisper.cpp transcription pipeline for macOS. Everything runs on your Mac. No cloud APIs, no telemetry, no account.

macOS · Apple Silicon whisper.cpp large-v3-turbo MIT licensed

Download for macOS Source on GitHub

Unsigned build — first launch needs the three-click Gatekeeper walkthrough. Or build from source.

Queue tab — live transcribe dashboard with progress ring, per-row Audio / Whisper / Finish columns

What it does

🎧 Podcasts

Find feeds by name (iTunes Search) or paste an RSS URL. Daily check; resumable downloads.

📺 YouTube

Subscribe to channels. Captions-first (requested lang → en → any), whisper fallback when none usable.

📥 Anything else

Drop audio / video files, paste a URL (yt-dlp generic), or watch a folder. Bulk-import a directory.

📝 Local transcribe

whisper.cpp on Metal — large-v3-turbo by default. Parallelised across N workers. Audio never leaves the machine.

🩺 Self-diagnosing

Failed feeds bucketed (DNS / TLS / 404 / 5xx / NAT64 / …) with one-click Retry now.

🤖 Headless / agent-controllable

23-command CLI with --json inspection so an LLM agent can drive the whole app.

Install

Download the latest .dmg from the releases page, drag Paragraphos.app to /Applications, then walk through the first-launch Gatekeeper steps. The first-run wizard handles Homebrew, whisper-cpp, ffmpeg, and the ~1.5 GB Whisper model.

Build from source instead:

git clone https://github.com/madevmuc/paragraphos.git
cd paragraphos
python3.12 -m venv .venv
.venv/bin/pip install -r requirements.txt -r dev-requirements.txt
PYTHONPATH=. .venv/bin/python app.py

CLI

23 commands with --json inspection — full GUI parity for headless / agent control. See the CLI section in the README for the table + an example task chain.