About Thoropod

Thoropod is a Mac app for people who love one podcast and want to know it inside out. Put in the feed of a show, and it downloads the episodes, transcribe them on your Mac, use an AI model to tag each episode along the dimensions you care about. You get a catalogue to browse, filter, and search: every guest, every book, every running bit.

Here's a podcast app that is thorough, get it? And maybe something about dinosaurs, which have nothing to do with podcasts. But dinosaurs are cool and fun.

Thoropod is for superfans

If you have ever wanted to know which episodes a particular guest was on, every time a show talked about a certain film, where a running joke started, or what exactly was said in that hilarious segment three years ago, this is the app for that. It is the desktop, do-it-yourself version of the companion sites that exist for a few well-loved podcasts. But here it is as a generalized tool to work for any show, including private feeds like Patreon or Memberful.

How it works

1. Subscribe. Paste a feed URL or the show's website. Thoropod reads the feed and lists the episodes.
2. Download. All of them, some of them, or none: episodes can also stream from the host when you play them. Audio can be kept as an archive or deleted after transcription.
3. Transcribe. whisper.cpp, built into the app, turns each episode into a transcript on your Mac. Nothing leaves your computer. On Apple silicon an hour of audio takes about a minute.
4. Find the speakers. Optionally, the app works out who speaks when, and once you name a voice it recognizes that person in every other episode. You'll have to do a little bit of listening to match voices to names, but the app has tools to make this pretty easy. You can tag dozens of speakers across hundreds of episodes in short time.
5. Analyze. Each transcript (with the show notes, and the speaker names when known) goes to an AI model with your instructions: list the hosts, the guests, the topics, the films discussed, the recurring segments, whatever suits the show. That is the one step that uses an outside service, unless you point it at a local model, which costs nothing.
6. Browse. One tab per dimension shows every value with its episode count. Click a few to filter, combine them across tabs, search every transcript, and click any line to listen..

Everything lives in a folder per podcast: a small database, the audio, one JSON file per transcript, per analysis, and per speaker map. You can open it in the Finder, back it up, and share the useful parts as a package.

What you can do with it

• Find things. Which episodes had a given guest, covered a given book, or mentioned a phrase. The Transcripts tab searches every word of every episode in an instant.
• Tidy the catalogue. The model spells names inconsistently; rename or merge values in a tab and the fix sticks through later re-analysis.
• Know the voices. Name a host once and every episode gets the name. Guests are named from the analysis's suggestions with a click each.
• Share your work. Export an analyzer package: the dimensions, the prompt, the transcripts, and the analyses, without your keys or a private feed URL. A friend imports it, points it at the feed, and has the whole catalogue without transcribing or paying for analysis.
• Use several Macs. Other Thoropods on your network can take on transcription, speaker-finding, and analysis, so a long archive finishes overnight instead of in a week.

Where to start

Open the Setup Wizard (the wand in the toolbar, or Help › Setup Wizard…) to download a whisper model, optionally the speaker models, and add an AI provider or a local model. Then add a podcast with the + button, download a couple of episodes, transcribe them, and try an analysis on those before you commit a whole archive: a cheap, fast model does surprisingly well, and a few cents an episode adds up over hundreds.

How to use it, the next document, goes through every part of the app.