- https://github.com/exlee/pikchr_pl DiagramIDE (diagram workbench amiga-style where you can script Pikchr diagrams using TCL, Prolog, Pikchr or - recently mruby) - it's one of the crates in repo
- https://svg.axk.sh - semi manual SVG fitter so that I can easier vectorize AI-generated pseudovector images
- https://github.com/exlee/rik - this one makes me laugh everytime I use it - it's an AI harness with text editor as an UI (i.e. it reacts to comment strings)
These are not vibe coded but AI made it much easier to slide through major friction points (e.g. for SVG fitter I really didn't want to reinvent fitting algorithms)
It follows a lot of the conventions of Rails which is probably why it has turned out quite well
I've done a lot of little things in Emacs since. Just minor things to improve my workflows or build up Chief of Staff type information flows.
Normal browsers have built-in dev tools - this has built-in design tools. so I can visit my app, open up a surrounding canvas, pull fragments into the canvas, do some design-ish stuff, and merge it back into code. All in the same UI. It was cool enough that I'm going to release it, but for now it's very useful for myself.
It took more than a year. Why? Cloud infrastructure is too expensive.
This was inspired by a karpathy tweet and the prototype created using another tool of mine: The LLM Plugin Generator plugin.
The llm model gateway lets you serve models from the LLM cli as a an openai API. This allows you to use saved consortiums in your various clients as if they where a regular model. Bringing massive parallel reasoning to any workflow.
It occured to me at some time that an collection of parallel running LLMs is not really a consortium. A consortium is a group of organizations. A group of groups. So I rectified disagreement this by adding support for actual consortiums, where each member of a consortium can itself be a consortium of models. e.g.
llm consortium save cns-glm-n3 -m glm-5.1 -n 3 --arbiter mercury-2 llm consortium save cns-k2-n3 -m kimi-k2.6:3 --arbiter mercury-2 llm consortium save cns-meta-glm-k2 -m cns-k2-n3 -m cns-glm-n3 --arbiter cns-k2-n3
Yes, even the arbiter/judge can be comprised of a consortium of models, bringing parallel reasoning to the task of judging parallel reasoning chains.
Consortiums can also now contain groups of specialists. These custom user-defined expert characters address the prompt from a different perspective. And a Westworld style Attribute matrix can be randomized to inject some more entropy into the process.
I think this is similar. Unfinished. https://github.com/mattjoyce/roundtable-consensus
So basically a way to just go on an hour long walk with myself, spit everything from the top of my dome stream of consciousness style, and then have Claude structure whatever I said.
It's nice to have something that structures my thoughts by just thinking out loud.
I vibecoded it (it's approaching 20K lines including tests). It works quite well but there are some bugs, so will have to do some actual engineering. But the UX is working quite well.
Not exactly a tool, but I also made pelohard.com which ranks the most recent Peloton classes by difficulty. Updated twice daily.
- classless CSS library: https://leftium.github.io/nimble.css
- HN client: https://hn.leftium.com
- local realtime streaming transcription prototype: https://rift-transcription.vercel.app
---
These projects were started without AI, but heavily augmented with coding agents:
- https://weather-sense.leftium.com
- console.log replacement: https://github.com/Leftium/gg
- Thin layer over Google forms/sheets: https://veneer.leftium.com
Calibre web UI 2.0 to replace Calibre's mediocre web interface. Used for browsing ebook catalog, searching and cataloging with a simple list feature.
Project Gutenberg local mirror UI to browse my local copy of Project Gutenberg books
A couple of MCP servers for self-hosted services to give access to OpenClaw. Currently working on a daily digest that Claw will generate that includes feeds from these: what news stories were popular in my feed reader, did my baseball team win, etc.
OneNote to markdown/obsidian canvas converter. It did that using interop api to read the actual XML of the onenote files.
Work time tracker as 1px line on edge of monitor. Shows thin line at the edge of the display which fills up based on what i am doing.
Plaintext bookmark chrome extension that save links to local markdown file, Dynalist, Workflowy, Github Gist and import export between them. Was originally for Dynalist when AI couldn't do much 2-3 years ago. Recently added these other end points. [2]
A heart rate monitor with finger on camera. It's bit crappy though. Had to make it because many trackers, including google fit, couldn't detect 200bpm. https://github.com/SMUsamaShah/heart-rate
[1]: https://github.com/SMUsamaShah/LookupChatGPT/tree/claude/fix...
This started off as a fancy cron with webhook and became a comprehensive runtime. I have been using it for months on several systems.
Hister is a full text indexer for websites and local files which automatically saves all the visited pages rendered by your browser. It provides a flexible web (and terminal) search interface with offline result previews & detailed query language to explore collected content or quickly fall back to traditional search engines.
It can provide a privacy-respecting search experience for serving "recall" type searches where users retrieve previously visited content, but falls short in "discovery" type searches (yet).
Much kudos. I hope more people discover how powerful even a local search index of previously visited content can be. And I hope more people can build large indexes as well, so we're not just relying on Brave & Mojeek & Marginalia (and EUSP) to rescue us for the fallback discovery searches.
- https://blunders.ai : Chess improvement app
- https://fretwork.ai : Freelancer management app (CRM/Billing/etc)
- https://validity.ai : Provide agents the ability to check the UI code it made (w/out needing to run through your full app)
- Save money on groceries + meal planning. This has probably saved hundreds if not $1k+ for our household at this point (some details here: https://x.com/ryanlanciaux/status/2063604299590939042)
- Orchestration / Starter Kit / Chat : Tool to help me manage multiple agent sessions at once. Some details on this one here https://x.com/ryanlanciaux/status/2063976049537417408
It has less features (no OG media or title/story analysis via Bedrock) but it focuses more on the features I like/need from an HN client
* Auto-Birthday - if you have a contact in your android contacts that has both a mobile number and a birthday in their contact info, you can choose to send them an automatic "Happy Birthday" message on their birthday at a specific time. Can do it with hundreds of contacts. Doesn't use hardly any battery or resources.
* Wrecker - stupid simple "throw a ball at a tower of bricks and try to clear the board" game. High score tables. Made in unity. High battery when in use, No battery use when not playing. Will use internet for high score data.
* GeoNote - Create Geo-fences to generate a notification when you enter a location with your custom text in the notification. My wife is always telling me, "Next time we're here, remind me to only order one piece of toast" or something like that, so I make a note, it pops up the next time we're there and we're both happy. Notes are stored locally. No internet access required. Uses Geo-Fencing which is more battery-friendly than always-on GPS access.
All my apps are free, very privacy-focused and as battery-friendly as possible.
No information leaves your device (other than the high score data in Wrecker).
You have to side-load my apps though. I'm not putting them on the Google Play Store. They're so annoying to deal with! OMG
Nothing says “AI enthusiast” more than automating away social interaction.
It probably should message YOU rather than the person who's birthday it is, so you can send something personal.
Also coming up, automated wedding/funeral attendance by your personal humanoid robot designed to look and sound like you.
https://mediaden.ca - iOS app for storing encrypted photos/videos on storage I (the user) exclusively owns, with zero servers, zero telemetry, and a host of other privacy related features.
https://gitlab.com/grepular/foxcage - Runs Firefox inside podman to isolate it from the host. Has some interesting features that I wanted and nothing else gave me.
https://gitlab.com/grepular/claude-sandbox - Yet another Claude sandbox. Runs it inside podman again. Has a pretty powerful proxy system for securing your credentials.
Currently working on a tool for sanitising email. Will be blogging it up at https://www.grepular.com/blog/ when it's ready for others to use. Does things like applying policies to html/svg/calendar/vcard parts to whitelist or blacklist tags/attributes/css/url schemas, clean URLs, fetch remote content at delivery time and attaching to the email to prevent tracking, pgp and smime auto encryption/decryption and a million other features.
It creates its own copy of your workdir for the agent to play in, and then you pull changes out ala git diffs or commits.
It's a MASSIVE time saver, and I use it as my daily driver.
This one generates maps from OpenStreetMap data + some custom curated info in YAML: https://github.com/c0nsumer/trailmaps.app-map-generator
This one converts a basic chunk of OpenStreetMap data to an SVG so I can mark it up (by hand) in Adobe Illustrator to make specifically-styled print/PDF maps, such as what get installed at trailheads: https://github.com/c0nsumer/osm_to_ai
This one takes GPS recorded rides and builds custom/personal heatmaps serving up the map tiles so I can use them in map editing software: https://github.com/c0nsumer/local-heatmap-tile-server
And all of this has been put together to make the custom, local, specific-use-case maps that are at https://trailmaps.app (which, via local curation, are overall better mobile/online maps than many of the bigger auto-generated systems such as Trailforks, Gaia, RideWithGPS, etc, for visualizing local systems).
It's neat stuff where I understand all the inputs, outputs, and how most of it works, but AI tooling (Claude, mostly) has allowed me to bolt it together much faster than I would have writing it myself.
I also built this site for educating potential customers and other privacy professionals about the increasing tempo of CCPA enforcement actions driving compliance: https://ccpa.world/enforcement
I could have probably coded this from scratch quicker considering that it took me two weeks to remove all of the hallucinated imaginary enforcement actions against real companies and also the citations to non-existent California law that the models kept injecting into my enforcement summaries.
1. A dashboard that tracks my personal metrics (github, strava, todo completion, flossing)
2. A eink display for that dashboard
3. A realtime node graph that shows a codebase (and/or its diffs) in a way that I can visualize what functions call which, and under what conditions
4. A agent that automatically fills out government forms and creates invoices for my friends brewery based on the delivery notes in their google calendar.
imping - PingPlotter-like app. They didn't have a Linux version and I'm a paying customer, so I vibe coded this one: https://github.com/zenakuten/ImPing
utcolor - text colorizer for Unreal Tournament 2004 https://github.com/zenakuten/utcolor
utquery - Unreal Tournament 2004 Game Browser tool https://github.com/zenakuten/utquery
utstatsdb - This is an old project that did not work anymore with modern php+mysql. I had claude fix it. https://github.com/zenakuten/utstatsdb
- small club website: https://www.kolibrinkpg.com
- ticketing system with Stripe payments and QR scanning at the door
- Instagram/media ingestion for the club site
- genealogy tool with GEDCOM import
- scripts for downloading/archiving public-domain film material
- playlist/library tooling for DJ use
- music collaboration/sync tool for Ableton projects
- normal work stuff in a much larger existing codebase
I have become a lot more strict about process after being burned a few times. Mostly: make the change small, be clear about what it is supposed to do, check the assumptions before coding, use tests/logging/manual checks as evidence, and don’t merge anything I can’t review and explain myself.
1) Use chatgpt pro from codex cli, opencode, claude etc as you can't get it via API. This has been the biggest boost in productivity for me as I don't have to copy and paste.
https://github.com/agentify-sh/desktop
2) A small gate to make sure any agent cannot run destructive rm -rf or git reset --hard commands, it has saved me many many times
https://github.com/agentify-sh/safeexec
3) For mac users, summarizes and speaks out loud after codex finishes a turn
Art search for magic cards
Code: https://github.com/VadimKey/xorpingtonian
Catalog (in Russian): https://vadimkey.github.io/xorpingtonian/
During vibe coding I found that emojis are not that simple as I thought about them.
https://bedtimebookhelper.com/
After coming back from paternity leave, I found that my team had really leaned in to AI driven development. This project was half catching up and half attempting to solve the burn out from the repeated books my wife and I were experiencing.
And the inverse as well, of course.
Runs on a raspberry pi that I was otherwise using to take backups periodically… has been working pretty good honestly.
I also built a program that fills/submits my time reports, and does the same for all of my subordinates - then signs them off… Saves everyone like 5 minutes if we remember to do it, or 12 minutes and frustration if we’re reminded by HR about it (which happens because who the hell cares about time reports?)
I've also built a release notes app for my QA teams, a DORA metrics app, a thing to map UX journeys with Playwright, and a ton of games and stuff. AI got me back into enjoying building things again.
It has "levels" of compilation, with EASY mode being about as easy as Ruby, and the compiler can present you with options to get that as strict & performant as Rust/Tokio.
I'm going to need at least a month to finish all the documentation, though.
But no honestly, unfortunately most tools I did for myself are not for hobbies but something that I needed for work... like this one (https://github.com/crmkit/crmkit) most recently.
Already have $180 ARR prebooked (the money that I used to pay for youtube music), looking forward for more.
if anyone has links for open-source self-hosted spotify/yt music replacement, I would gladly appreciate links
Github: https://github.com/jantznick/youtube-spotify
It essentially uses youtube as the music source, I think I heard somewhere that playing through embedded videos skips ads but I'm not really sure, in all my time testing it I never noticed ads, but I'm also on premium so that may have been why.
by all means critique, I don't know that I have a ton of time left for it and I'm sure there's bugs here and there. I was having issues getting it to autoplay on desktop when the window itself wasn't the active tab. I never really tried it on mobile.
I was trying to get some DB of artist/song info but doing that was proving to be complicated.
I collect song metadata from various places (genre, instruments, track credits, rating). I also scrape charts by year, genre etc.
Then I run an ETL job on the json data I have downloaded, pre-building queries for extremely fast lookup tables. This gets saved to Duckdb, which is used by my go web ui/api.
It's very early days, and I only spend one or two hours a week on it, but right now it's amazingly useful. It had roughly 80k song metadata. To preview the suggested songs I ended up building a very cut-down YouTube music player, except that the playing song has all the metadata right there, and everything is a link that can take you to the artist, composer, instrument, genre, album etc. It's a great way to "wander through your collection".
Unfortunately this is only useful to me, because I targeted the music I listen to.
Next step is to download lyrics and extract song meaning, keywords etc. Then use MusiCNN, (or CLAP,OpenL3, HTSAT) to extract embeddings. Finally train my own model for nearest-neighbor retrieval based on a mix of metadata, giving the user the ability to tune it on the fly.
Can be used to resize images, but the main purpose was pixelation for a game I was making.
https://charleswiltgen.github.io/TagLib-Wasm/ – Also built for myself initially, I FOSS'd it because there was nothing like Mutagen for TypeScript/JavaScript runtimes. (I don't dislike Python, but think it's a bit of a mess.) This was my first serious project to leverage LLMs for coding.
https://pwascore.com/ – Built because I wanted to quantify how bad Safari was at PWAs. Learned that, objectively, Safari is as bad as PWAs as Firefox (which is to say, not terrible, and not to blame for why PWAs continue to be mostly-irrelevant).
When you install the CLI, it (with your permission) asks to update your base agent prompt files (e.g. `~/.codex/AGENTS.md`, or `~/.Claude/CLAUDE.md`) with info about how to use the tool.
This means all your agent chats know about SDocs, and it’s nearly always your agent which invokes the tool: “Hey Claude, sdoc me a list of all my open MRs”, etc.
I did a ShowHN about it here: https://news.ycombinator.com/item?id=47777633
basically trying to see what a vertically integrated agent looks like, where the agent has deep access inside a framework and it operates from within a framework, so like, instead of reading files, opening processes etc - it gets a bunch of framework specific runtime tools(logs are the easiest example)
Tuber[0] - this is my favorite, use it multiple times a week. It's just a little CLI wrapper around yt-dlp for my most common use cases - downloading the video, or the audio, or the subs. And then, if you've got the Claude CLI installed, it can also shoot the subs through Claude for a summary. I use it all the time, it's a great little thing!
Scrapio[1] - this is really specific but I was so pleased with how it turned out. You give it a list of "hacks" ("mods" for Super Mario World) and it goes out to SMWCentral, grabs each of the patch files, and patches a clean ROM. I think I only used it twice but it was just a nice way to chew through a list of hacks and get a few ROMs ready, made quick work of something that would have otherwise taken a bunch of annoying schlep work.
Lotus Eater[2] - calling this a tool is a bit undersell, but I'm still really pleased with it. It's a fan site for jamtronica greats Lotus that scrapes Nugs.net for setlist data and lets you do some mildly interesting analysis on things like song frequency and co-occurrence. Also has a per-user "shows I've attended" thing, Setlist Bingo. It's been fun to hack on.
Lastly, less a tool, more just a toy: last week Google released their Magenta model for doing live music generation. I thought it was really neat, and it's open source, so I opened it up with Claude, and after a few passes and some extremely annoying toolchain issues, I was able to add a spectrograph which does key / chord analysis to the "Collider" app, so you get a live readout of "what the band is playing" and you can pull out your guitar or whatever and join the jam with some info at your disposal. It's the kind of thing that would have taken way too much effort to be worthwhile in the past, but with AI, it's a really neat result of a fun night of weekend hacking. See the README I added in my fork for a screenshot:[3]
[0]: https://github.com/epiccoleman/tuber
[1]: https://github.com/epiccoleman/scrapio
[2]: lotuseater.epiccoleman.com
[3]: https://github.com/epiccoleman/magenta-realtime/tree/eric-mo...
But I dont use "AI" to make them
I use a code generator
I like to use the smallest possible "toolchain", using the least possible resources, to build software tools
Ideally I want the tools to compile quickly on underpowered hardware
- gcode sender and generator https://mycnc.app
- CNC simulator https://sim.mycnc.app
- Cabinet design with door/drawer designer https://cabinet.mycnc.app
Static site generator for my blog, or at least bits of it.
https://github.com/mretallack/OpenEPaperCliTool
----- 3d printer pipeline, so its can print stuff directly without having to use the computer to set it up.
https://github.com/mretallack/3dprinter
----- Experiment with creating a Abdroid Auto app for phones that cannot run real AA. (WIP)
https://github.com/mretallack/AndroidAuto
----- A android 3d clay modeler to create models for 3d printer, with stl export.
https://github.com/mretallack/ClayModeller
----- Uk Fuel finder python lib and Home Assistant intergration for showing fuel stations from UK gov api.
https://github.com/mretallack/ukfuelfinder https://github.com/mretallack/ukfuelfinder-ha
---- Reverse engineer cheep drone video feed, from drone found in charity shop
https://github.com/mretallack/DroneCamera
---- App to send voice to camera using mqtt.
https://github.com/mretallack/CameraSpeaker
---- Added ONVIF to an oss rtsp android app.
https://github.com/mretallack/cams
---- Added Home Assistant to Dicio Assistant.
https://github.com/mretallack/dicio-android
---- Added telegram bot interface to kiro, with group support.
Just a more helpful discord chat generally. It also gaslights you too!
Here is the tool: https://git.sr.ht/~asibahi/hoopoe
Would have made them without UI with a bit more elbow grease invested in web-searching for some examples, maybe even a StackOverflow question.
Generally, I'm not a fan of LLMs and their social effects.
Currently working on a markdown search and wiki backed by Typesense, also has good web search, fetch, crawl. This will power my personal knowledge base system as an important step towards more leverage and better outcomes.
It's a user daemon that runs on my machine and exposes a unix socket, and then a bunch of hooks in claude, zsh, vim, etc, that report directory and commands I've run and all that, pipes it to claude Haiku for summary, and then stores context in sqlite. It also exposes that data as MCP so I can use claude to say "hey what was I doing yesterday," or any arbitrary time range.
I find that in the age of using AI agents, "Wtf was I working on yesterday" is an even harder thing to remember for me, so this helps me kind of track everything with a database that a) has AI summaries already and b) can be accessed by AI as well as a CLI.