Frank, as in Rosenblatt. And Frankenstein.
Small open source apps I built with AI agents to solve my own problems. Each one is useful to me, and most of them only to people with the same very specific need.
Why they are all called Frank
Frank Rosenblatt built the Perceptron in 1957, the original neural network and the ancestor of every LLM I use today. He also promised machines smarter than people "very soon", the newspapers ran with it, and the first AI winter followed. These apps were written with AI, so the name is a small tribute to him.
The other Frank is Frankenstein. These are experiments, stitched together to see what agents can build, and I make no promise that they are robust. Each one scratches a very niche itch of mine. They were also my test beds: building them is how I worked out the agent workflow I use every day.
Six of them worth a look
My main ones, all still maintained. Each card links the repository and the posts where I wrote about it.
FrankMD
Rails 8 · Docker
A self-hosted Markdown editor and notes organizer. No database: notes are plain files on disk. It starts with one Docker command, and the fed command opens any folder in it.
The itch: an editor built around the way I write blog posts.
Frank Sherlock
Rust · Tauri · Ollama
Point it at a folder of photos, PDFs and videos, and a local vision model describes every file, reads the text with OCR and indexes it all. Then search for "receipt from market of $42.00". It also finds duplicates and groups faces. Nothing leaves the machine.
The itch: finding one image among thousands on my NAS without sending them anywhere.
- I built a smart image indexer with AI in two days (Frank Sherlock)
- Software is never done: four projects and life after deploy
Frank Yomik
Python · Ollama · Flutter · Chromium
Read Japanese and Korean comics in the original. It finds the speech bubbles, reads them with OCR, and translates them with a local LLM or adds furigana. Hold the pointer over the page and a magnifier shows the result. There is an Android and Linux app and a Chromium extension.
The itch: reading manga in Japanese and looking up only the parts I get stuck on.
- The complete manga setup: Frank Manga+, Frank Yomik and Prettify
- My first vibe coding failure and how I fixed it (Frank Yomik)
Frank Manga+
Rust · Tauri · SvelteKit
A desktop reader for MANGA Plus by Shueisha, which is otherwise a phone app. Library, full catalog search and a page-fit reader. The free tier works as is, and a subscriber secret opens the premium chapters. Linux, macOS and Windows.
The itch: reading MANGA Plus on a real screen instead of the phone.
FrankMega
Rails 8.1 · SQLite
File sharing for a home server. Upload a file and get a link that expires after a set time or number of downloads. Invite-only, with passkeys and two-factor login, and no external services.
The itch: sending big files to family and friends without a third-party service.
Frank Scanlation
Tauri · Linux, macOS, Windows
A Kindle-like desktop reader for scanlation sites. Paste a site's address and it becomes a library card with cover art, chapter tracking and a badge when new chapters land. Linux, macOS and Windows.
The itch: scanlation sites are bright, full of ads and one long scroll.
Smaller itches
Five more, each built for one job I needed done.
Frank GO
Sabaki · KataGo
A beginner-friendly Go trainer built on Sabaki. More than 4,700 tsumego served at your level, historical games to study, and KataGo as an opponent. Overlays show territory and score, the part beginners cannot see yet.
The itch: strong Go software is made for strong players.
Frank Type
Rails 8 · Docker
A typing trainer with public-domain prose instead of random word lists, in the MonkeyType style. No account: history and charts stay in the browser. A digraph heat map shows which key pairs slow you down. Runs in Docker.
The itch: MonkeyType serves random words, and TypeRacer makes me wait for other players.
Frank Karaoke
Flutter · Android
An Android app that wraps YouTube and scores your singing live, from the video's audio and your microphone, with no song files. Four scoring modes and pitch shift. The scoring is still experimental.
The itch: a score for any karaoke video on YouTube.
Frank Lyrics
Chrome MV3 · LRCLIB
A Chrome extension that adds line-by-line controls to YouTube lyric videos. It pulls synced timestamps from LRCLIB, or lets you mark lines by hand, then repeats the exact line you are memorizing. Experimental.
The itch: hunting the YouTube timeline to repeat the same line of a song.
Frank Geary
Vala · GTK
My fork of the GNOME Geary email client, with two fixes as native patches: the accounts sidebar collapses, and the recipient field suggests the people who already wrote to you. I rebase it on upstream often.
The itch: Geary was almost perfect, but its three fixed columns wasted my vertical monitor and the recipient field never suggested anyone.
I have 14 public Frank repositories on GitHub as of September 24, 2026, a few more than the ones on this page.
See every Frank repositoryEverything here is open source
The tools, the skills and the benchmark are public repositories. Fork them and build the version that fits the way you work.