major webui rewrite IS DONE! feel free to submit feature requests or bug reports against the webui again

[Rose22]

as you know, most of openlumara is manually coded. but there is one part that is mostly ai generated - the webUI. and it has gotten far too difficult to maintain!

bugs spring up seemingly out of nowhere, i have difficulties modifying it manually (without AI help) to reflect changes i do to the core, and when i touch one part of it, everything else breaks... it's total spaghetti code with one function pointing to another to another to another and it's just, really bad and unclean, and obviously slop. it was never meant to become the main way to interact with openlumara

it's been taking up most of my dev time, leaving other parts of the framework with unresolved bugs and outdated code, because i keep getting bug reports about the webUI, and i'm constantly busy fixing them.

it's frankly been a thorn in my side lately and it's a good lesson on why (partial) vibecoding is so bad, even if it's "just the frontend". for reference, this doesn't happen with the core, because i wrote it manually

the way i made the webUI is not exactly vibecoding, but it approaches it. i started with a oneshot by a powerful model, then started to refine it through thousands of prompts requesting small tweaks and features, over the course of many months. so it's mostly AI generated, but not exactly vibecoding. however... the tech debt is still very much a thing

i'm going to do it properly this time. in a good framework, with clean code, in my own style. it's going to take time, so please bear with me, and don't submit PR's against the webui

[Rose22]

Screen.Recording.2026-07-18.at.10.19.10.mp4

the rewrite is already well underway! it's still missing a bunch of features from the current webui, but this is going faster than i expected. you can see the progress at https://github.com/Rose22/openlumara/tree/webui-rewrite. the codebase is MUCH cleaner!

[PraxisOG]

I love this project, it helps me a ton with everyday executive function. If you're rewriting the webui, could you change a behavior to make it more consistent with the other channels? Sending slash commands in the webui while the api is down gives an error 503 and doesn't send the message through, as currently it gives the connection error which halts the message send. In every other channel I can send a slash command to a custom module to change or load local models through llama.cpp, but it errors out on the webui unless a model is already loaded. Figured I'd ask here instead of sending a bug report

[Rose22]

I love this project, it helps me a ton with everyday executive function. If you're rewriting the webui, could you change a behavior to make it more consistent with the other channels? Sending slash commands in the webui while the api is down gives an error 503 and doesn't send the message through, as currently it gives the connection error which halts the message send. In every other channel I can send a slash command to a custom module to change or load local models through llama.cpp, but it errors out on the webui unless a model is already loaded. Figured I'd ask here instead of sending a bug report

yup, thats one of the things i tried to fix in the old webui! unfortunately that then broke EVERYTHING in ways i only noticed weeks later, making me have to revert about 3 weeks of work. so the rewrite is very much needed...

right now i'm trying to just get it back to the standards of the old webUI, but after that i want to fix a bunch of bugs like those, as well as make it even more user friendly. oh and support for sending all kinds of multimedia, not just images.

the new webUI is much closer to the core and how all other channels work. i'm also going to rewrite the telegram channel by the way, as that too is mostly ai generated.

[Rose22]

by the way, progress is happening rapidly. i got the settings dialog working:

(and as you can see, themes are working again too)

and toolcall streaming is working (although still a bit janky due to how hard it is to convert partial streamed json to a nice view)

Screen.Recording.2026-07-19.at.11.47.55.mp4

i've also taken the opportunity to make parts of the webui modular, just like the rest of openlumara:

and i'm just getting started when it comes to that. i'll make anything that makes sense modular, so that you can turn off exactly the parts of the webUI you don't want. as with the rest of openlumara, turning off any parts of the webui also stops the code from ever being loaded (such as the javascript and css files)

EDIT: oh btw did i mention that for the webui to work properly, i had to insert a bunch of ai generated code into the core? so this will allow me to finally get rid of all that and get the core back to a pure state (without any ai generated code)

[Rose22]

the webui rewrite is DONE! :D

[Rose22]

i documented a lot of the progress on discord, but i'll repost it here so its not locked behind discord. i think it might be fun for people to see what went into this:

16th of July, 2026

this is how i started lol, just trying out alpine.js

then that same day i got a chat list

17th of July, 2026

the next day i got this

then this

then this

and at the end of the day, i already had this going!:

Screen.Recording.2026-07-17.at.11.34.45.mov

Screen.Recording.2026-07-17.at.15.08.08.mov

Screen.Recording.2026-07-17.at.15.39.04.mov

18th of July, 2026

progress was happening really fast

Screen.Recording.2026-07-18.at.10.19.10.mp4

i already had basic streaming going, toolcalls were just not being formatted nicely yet!

already started to get the structure of the settings dialog

19th of July, 2026

got theme support back in

and added a bunch of settings that make the webui backend literally never load certain parts of html

i also managed to get toolcalls to look okay again

and by this point i was using the webui-rewrite branch of openlumara to help work on the code, rather than the old webui

Screen.Recording.2026-07-19.at.10.41.22.mp4

i reduced the height of the thinking section so that it would no longer take up all the space, inspired by some other UI's that did this

Screen.Recording.2026-07-19.at.17.09.00.mov

i was able to add fancy animations across the UI in one go by just using a single x-animate attribute coming from an alpine.js plugin. have i mentioned how nice alpine.js is to work with?

July 20th, 2026

i tried out this design but it was too much lol

finally made sure API errors properly show up from the backend

and made the stop button just send /stop to the backend, so that it handles it properly :) instead of the old webui's weird workarounds that involved AI-generated websocket trickery, which made it inconsistent with other channels

i made an initial draft of a catppuccin theme and as you can see i was using AI to learn how to code CSS and javascript instead of having it write it for me

also to learn better ways to do things in python

then i added message buttons

July 21, 2026

i started to work on one of my main gripes with the webUI... the way you set up your API connection. the new one automatically tells you whether a connection worked or failed, and why

then i started to work on system logs

July 22, 2026

i showed how much nicer the structure of the webui codebase is

and started work on the mobile webUI... oh god look at that abomination LOL

at this point the turn grouping logic was still client-side but i was going to move it to the backend/server real soon

July 23, 2026

and here you see the first iteration of the server-side turn grouping logic

Screen.Recording.2026-07-23.at.16.02.23.mp4

as you can see the turn grouping logic (the logic that makes everything the AI does appear in one message bubble) was now completely server-side. this is going be amazing for making a new TUI btw, stay tuned

July 24, 2026

added a way to see the system prompt without using commands, by popular demand

July 25, 2026

kept using my local AI (qwen3.6 35b) to find bugs in the codebase so i could go fix them (manually)

added message editing

moved chat exporting to the core, so that you can export from any channel using /export or in the webui using the button

July 26, 2026

fixed the weird timestamps that always say "2 hours ago" even if you literally just said hi to your AI

July 27 -> July 30

i took a break.... its not healthy to work non stop for this long

July 31, 2026

i sent my poor AI off to read the same file over and over until the end of time (well, until i stopped it) to test if i finally fixed the toolcall chains getting cancelled due to context overflow. and yeah, i did!

August 1, 2026

thanks to the new webUI being so much easier to code for, i was finally able to make key/value tables inside settings panels and release my new Replacer module with openlumara

and this, this is one of the big new features, one of the main reasons i wanted to rewrite the webui so badly. the ability to add field descriptions to fields that come from core settings that don't belong to modules, so, i could now massively improve UX by explaining what each setting does :)

this was especially important for the developer role, which confused everyone

i went wild with theme creation for a moment (by asking my openlumara to read the catpuccin color scheme guide and to then generate themes based on it, with a theme for each variation)

...but then i came to my senses and just made the 4 variants

August 2, 2026

i decided to replace the theme grid with a dropdown because there were just too many themes now lol

and added mode switching buttons next to the dropdown

made sure to test it on mobile

aaaaand i started to rewrite the coder module from scratch because it was bothering me that too much of it was AI generated and that it was messy and hard to maintain

i added one of the most requested features to that coder rewrite... the ability to point it at multiple folders

i considered adding this but decided to scrap it for now. it wouldve been a bit too complex to implement considering i wanted to release the webui rewrite soon

did the same thing for template paths!

and made sure the system prompt of the new coder was nice, small, concise, and tidy

August 3, 2026

this was the day i decided to finally release it!

...with a big scary warning about your chats having to be migrated cuz i completely reworked the chat system so that it loads chats from individual files rather than one big database json file

August 5, 2026

spent lots of time cleaning up the code

---

and that's what went into making this new webUI! hope you enjoyed this peek into the process

[abc-nix]

Is there a list somewhere to compare the feature parity of both old and new webui? I can see there are new thing in the new webui, but there are also things left behind in the old webui. I am sure some things you also don't want to keep or have implemented differently.

[Rose22]

Is there a list somewhere to compare the feature parity of both old and new webui? I can see there are new thing in the new webui, but there are also things left behind in the old webui. I am sure some things you also don't want to keep or have implemented differently.

kinda. i have a todo of the missing stuff:

• command autocomplete (which pops up when you start typing a command)
• the ability to search inside a single chat
• keyboard shortcuts (this time i'll make them customizable)
• the Storage Editor (i think i'll just make that a seperate channel considering how complex that is. doesn't make sense to make it part of the main webui)
• tag filters

it's not that much anymore, but it was enough to delay the release for another few weeks, and i didnt wanna wait anymore. and right now i'm polishing up whats already there and fixing any bugs that get reported before i start adding these features back in