Open comments for this post

@yf on Astilbe · about 3 hours ago

7h 14m 22s logged

Added toolkit skeleton, and added chat branching + regeneration!

Open comments for this post

@yf on Astilbe · about 13 hours ago

7h 30m 22s logged

Theme chooser and auto-titler implemented! Auto-titler automatically uses the loaded model to generate a title for your conversation (I did implement using smallest model, but the current model is shown to be much faster due to cache anyway) in the backend. I also implemented a little orientation tour. This time it’s all light mode, to make up for my silly mistake last devlog of posting a darkmode with my light mode update.

Open comments for this post

@yf on Astilbe · 1 day ago

2h 14m 41s logged

Added a light theme and some icons (svg)!

Open comments for this post

@yf on Astilbe · 1 day ago

9h 20m 9s logged

Streamlined the onboarding experience, now onboarding preferences do save properly! Model recommendations also dynamically updated. I started a git to prepare for major feature additions but it is not yet uploaded to GitHub. Stay tuned!

Open comments for this post

@yf on Astilbe · 2 days ago

2h 27m 20s logged

Devlog X?

Hi all, Astilbe is approaching YAGNI full basic functionality. I’m really excited!

Anyhow I apologise for taking this long to write a proper devlog. However, I will accordingly update all my previous devlogs with the detail we all deserve very soon as I finish the polish.

This devlog will be a summary of the previous devlogs combined:

CSS is standardised. I have to concur that it is substantially, the new styling, is AI-generated; however, a lot of details were manually tweaked. I also ensured standardisation of styling and tested edge cases personally. (Animations were also comprehensively standardised and implemented)
Settings is now separated into basic and advanced settings. Advanced settings is enable-able with a toggle. If you ask why, it’s because Astilbe is designed to be grandma-friendly and useful for people who are not necessarily computer savvy.
Model Library is now fully functional. Basically, you can now download any model from the library, and you can adjust your tiers to the models that you like. (Tiers is just the models set for the “fast”, “balanced” and “advanced” modes.)
Error handling is unified now. Catch-alls are implemented more comprehensively and I have been doing manual edge-case testing to ensure polish for every single function added.
In this devlog, I added smart conversation ranking and infinite scroll, in order to save compute from being wasted by displaying EVERY SINGLE old conversation in the scrollbar, similar to how Claude does it. Conversations are now also ranked by the last_used time.

Anyhow, that’s all for now. Thank you for reading!