ekorbia

I'm working on Ekorbia: https://github.com/ekorbia/ekorbia-desktop. It's a lightweight, MIT licensed, Rust/Tauri desktop app for private local LLMs with a prompt library, chat overlay, ephemeral chat, model comparisons, and folder/RSS/URL watches.

I"m also still working on my crossword puzzle pipeline, https://github.com/ekorbia/xword-pipeline, and my daily word game site, https://wordfuzz.com/, which has a daily mini crossword and a few other daily word games.

And finally I'm working on a browser based Civ5/Unciv clone which is in its early stages.

Yes, I tried a while back to use LLMs for both the grid fill and the clue writing but was not able to generate dense puzzles. This pipeline uses a Rust fill engine that fills hundreds of candidate grids using a scored word list.

Thanks for the feedback! I will recheck the clues for that puzzle.

I'm working on xword-pipeline: https://github.com/ekorbia/xword-pipeline

It's a pipeline for generating dense crossword puzzles with a Rust grid fill engine and optional Claude based clue writing and QA processes.

I use it to generate the crossword puzzles on https://wordfuzz.com/. The quality is already good for smaller (5x5 and 9x9) puzzles. I'm working on improvements for larger 15x15 NYT style puzzles.

I'm also working on Ekorbia: https://github.com/ekorbia/ekorbia-desktop. It's a lightweight, MIT licensed, Rust/Tauri desktop app for private local LLMs with a prompt library, chat overlay, ephemeral chat, model comparisons, and folder/RSS/URL watches.

The builds are unsigned but you can build it yourself to avoid the warnings. It has a bundled engine on macOS and requires an OpenAI-compatible server on Windows/Linux. Any feedback would be appreciated.

I'm a big fan of word games and I gave this a quick try. At a glance some of the games are a bit confusing and could use a pop-up explaining game play for first time users.

What are you storing in Supabase? I've also AI coded a couple of small word games but I explicitly stayed away from any back-end requirements so I could deploy easily to CloudFlare pages.

Looks cool! I tried this on Chrome and Firefox on my MacBook but couldn't get past the 'Connect your audio interface' screen. Firefox hung after clicking the start button and Chrome gave me a permission denied error.

Is the latency truly comparable to desktop plugins like ToneX or the Neural DSP plugins? I think many guitarists are very latency sensitive.

Yeah, it seems the general consensus is that LLMs can't generate large newspaper style crosswords.

For now I will just use Phil to build mini puzzles and try using an LLM to help generate clues. Thanks.

Thanks. I'll give that a try. I also found https://crosshare.org/ which is open source and has a crossword constructor.

Thanks. I wasn't aware of XWords. It works well for generating puzzles with sequences of intersecting words. It doesn't seem to produce denser puzzles as you add more clues and answers. It just builds a larger puzzle.

Thanks for the response. I agree that specificity is important.

I provided a few specific clues and answers along with the grid layout, and a theme as context. Ideally this would generate a filled crossword that I could then verify and edit.

Claude would churn, restarting from scratch over and over trying to fill the layout. If I remove the grid layout from the context, I can generate sparse sequences like you mention but these are not very fun to play.

I also found myself trying to drag the bars. Is dragging supposed to work on desktop? If not, I would clarify that in the "How to play" instructions.

What dictionary are you using? I tried a couple of words like 'acing' and 'awing' that should work?