WikiTech

Turn any podcast into a searchable, timestamped wiki

WikiTech reads a show's RSS feed, transcribes every episode, and writes a page for each person, company, product and topic the hosts talk about, every mention linked to the moment it was said. Run it on your own server, for your own shows.

Browse the wikis Request the code

In the wild

  • Grumpy Old Geeks Official

    wiki.gog.show

    The tech news podcast's own wiki: every episode, searchable transcripts, and the people, companies and products they talk about.

Running on this site

  • Accidental Tech Podcast

    Marco Arment, Casey Liss, John Siracusa

    Three nerds discussing tech, Apple, programming, and loosely related matters.

    8 episodes Latest

  • Don Talk

    Donovan Adkisson

    Politics. The issues facing Americans today. Those are the topics we'll discuss on this show.

    1 episode Latest

  • The Book of Donovan

    Donovan Adkisson

    The Book of Donovan is an introspective journal hosted by Donovan Adkisson, a 30-year technology veteran, community journalist, and South Georgia native. Moving beyond the...

    4 episodes Latest

  • The Thinking Atheist

    The Thinking Atheist

    Seth Andrews, a former Christian broadcaster and believer for 30 years, ultimately escaped the bonds of superstitious thinking to embrace the more satisfying explanations that...

    17 episodes Latest

  • Waveform: The MKBHD Podcast

    MKBHD

    A tech podcast for the gadget lovers and tech heads among us from the mind of Marques Brownlee, better known as MKBHD. MKBHD has made a name for himself on YouTube reviewing...

    10 episodes Latest

Search every show or read how the wikis work.

What every show gets

  • Episode pages

    A summary, a timeline of key moments, the links mentioned on air and the full transcript, with a player that starts at any line you click.

  • A page for everything mentioned

    People, companies, products, media and topics each get a page listing every episode that mentions them, with the quote and the timestamp.

  • Search that jumps to the moment

    Full-text search across summaries, pages and every word said, forgiving of accents and misspelled names. A hit in a transcript opens the episode at that second.

  • A links directory

    Every link from the show notes and the conversation in one place, filterable by website.

  • Spelled right

    A per-show list of names and terms steers both the transcription and the writing, so the hosts' names and the show's in-jokes come out right.

  • The audio stays put

    The player streams from the publisher's own feed. Nothing is re-hosted, so every play still counts for the show.

How it works

  1. Read the feed

    New episodes are picked up from the RSS feed automatically. A back catalogue is only queued when you ask, with a time and cost estimate first.

  2. Get a transcript

    If the feed publishes a transcript it is used as is. Otherwise the audio is transcribed with whisper.cpp, on the server or on a faster machine on your network, or with OpenAI's API.

  3. Write the wiki

    A language model reads the transcript in timestamped chunks and writes the summary, the key moments and the mentions. Choose Anthropic Claude, Google Gemini, OpenAI, or a local model through Ollama.

  4. Link it all up

    Mentions are matched to the pages that already exist, so a name said in episode 12 and episode 300 lands on one page. Lock a page after correcting it and no reprocessing will overwrite it.

Every model call is logged with its token count and cost. A 90-minute episode costs roughly nine cents with Claude Haiku, and nothing at all with a local model.

Run your own

WikiTech is a Laravel application that runs on one small Linux server or container, with no Docker and nothing to subscribe to. Point it at a feed, and it does the rest on a schedule. The admin panel handles shows, episodes, corrections and every setting.

You will need

  • Debian 13, PHP 8.4, PostgreSQL 17 with pgvector, and Redis
  • whisper.cpp and ffmpeg, which the provisioning script builds and installs
  • An API key for Claude, Gemini or OpenAI, or an Ollama server
  • Node and npm on the machine you deploy from, where the assets are built
  • Optional: Cloudflare in front, for edge caching and a tunnel instead of open ports

Three steps

# 1. Install everything on a fresh Debian 13 box
DEPLOY_HOST=root@your-server deploy/provision.sh

# 2. Test, build and ship the app
DEPLOY_HOST=root@your-server deploy/deploy.sh

# 3. Sign in at /admin and add a feed

Want to run it? Email us for the source code and the full setup guide. Licensed under the GNU AGPL v3.