The new UI for Assistant is worse in every measurable way

magiruuvelvet

i'm annoyed at

the chat bubble look. at least give us an IRC-style look please.
information density is far less. there is so much meaningless whitespace everywhere. the old assistant was far more information dense. i do more scrolling than working.
the model selection window is extremely tiny. i can't see the full name of my custom assistants. i waste more time with scrolling than getting work done.
the model selection window resizes when using the search bar. this is bad for accessibility. why doesn't it stay at a static size?
animations are still hardcoded and don't seem to respect prefers-reduced-motion. it is unclear to me what is happening there. maybe make a bug report upstream to whatever library you are using. i improved my animation-api-stub code since then and migrated it into a manifest v2 browser extension that bypasses CSP.
the sidebar scrolls to the bottom each time i open the assistant, which is just weird behavior.
folders are collapsed each time i open the assistant. this one can be fixed with CSS since the elements are in the DOM already with a hidden class.
regression: the html tags <summary> and <details> are no longer escaped and end up in the DOM. i reported this issue months ago for the old assistant where this was fixed. it is unclear to me how this was missed in the new assistant.
the sorting in the model list changed somehow. i carefully prepended my custom assistants with sorting helper characters (Hanzi characters) to create groups. these groups are now all over the place and don't follow Unicode code pointer order anymore for some reason. not sure how this was achieved on a technical level. the sorting makes no sense to me now.

the old assistant was a great product worth paying for.

Luis

Thank you all for the thoughtful feedback. We're working through it at the moment!

(addresssing a comment above)

About the sidebar,
moving to a single sidebar (the old Assistant used a two sidebar approach) involved a few trade-offs that we're still working through. To address your points directly: 2, 3, 5, and 6 will be solved soon.

multi-line chat summaries are gone so now I have no idea what really was in the thread. this is also really bad

What we had previously was just a snippet of the thread's opening text. We still use that layout when you use the sidebar search but it felt very heavy in the single sidebar.

About the thread,

Thread window:
no floating cost/copy pane - I used it all the time

I'm not sure what you're referring to here, could you expand on this?

insufficient feedback when the model is "thinking" after the query has been submitted; what is it doing? three dots are blinking... and blinking...

We're actively working on this. On top of that, we have major improvements coming to the whole answer experience and the current purple boxes will finally be gone. The team is putting a lot of effort into all of this, please keep the thoughtful feedback coming

Pleonasm

Luis The team is putting a lot of effort into all of this....

Parenthetically, it seems to me that the Kagi team is absorbed on supporting a transformation of Kagi into an "AI first" company, rather than concentrating on supporting the core mission of delivering excellent search services.

Hopefully, someday we'll hear Kagi say, "...we have major improvements coming" for the search experience rather than the "answer experience"?

niu3

Luis What we had previously was just a snippet of the thread's opening text. We still use that layout when you use the sidebar search but it felt very heavy in the single sidebar.

Thanks for replying directly and considering the feedback - appreciated.
Perhaps the snippets could be an option or a hidden property we could bring out with custom CSS. I found them more meaningful than the titles themselves. Maybe it's just me.

Luis no floating cost/copy pane - I used it all the time

While looking at a chat thread or a segment of it, there was a little pane on the bottom right which had a copy icon and an "i" icon. It was visible while scrolling and allowed to either copy the answer/thread segment or to see info on the cost, model etc. Now this is stuck at the bottom, like in other chatbots. The previous approach was more convenient.

Thanks again - no doubt there's a lot of work going on. The first assistant was such a refreshing experience vs. the garbage others are serving, and truly felt like the Kagi spirit - focus on what matters. No doubt the team can continue in the same vein.

ax7899

Pleonasm major improvements like the AI kill switch hopefully

PolicyPants

I've had almost universally better experiences with the new Assistant compared with the old. The new UI feels smoother, follow higher refresh rates much better, and just overall feels like a proper Assistant UI like those that mainstream LLM companies have.

This feels like a loud minority type of concern to me. There are obviously problems with it that needed solving (like the UI not respecting reduced animations) and there may be accessibility issues with forcing a switch like this, but things like... the "Chat bubble"? Even Kagi's HTML frontpage has a bubble. It's the search bubble. Just like with the assistant, which now has a bubble as well. To me it seemed like parity with Search.

The one thing I can agree with however, that has been mentioned above is that all of this should be optional. Breaking changes should not be given a 48 hour window to adapt, and even after a 30 day window, this should still be optional for months. And if demand for the old UI is high, why not keep it as an option? Why force everyone to switch so quickly?

NEPPPY1

I am really upset at this new UI. Please bring the old one back.

I am a very new user and fell in love instantly, and now it feels like everything I was trying to get away from. Grossss

Soma2501

"We moved from tags to folders to make the product simpler and easier to reason about"

I just don't think tags are harder to understand than folders especially when they were already functioning as folders.

"...but in practice they were lightly used: only ≈20% of users had any tags at all, and just ≈4% had a thread with >1 tag"

I don't think that's as relevant because even single-tag workflows were broken by this change. Previously:

I could type a tag name to immediately create or select it. Now I must scroll a growing list of folders.
I could search for the tag itself as a navigation mechanism. Now I can only search thread contents within a folder.

I also do not believe the majority of users were specifically asking to replace tags with folders. Why not involve these feedback foruns in these decisions for early feedback instead of forcing a switch that can break existing workflows and usecases? More roadmap and decision rationale transparency would prevent situations where changes isolate a subset of existing users for unclear benefit.

If the decision was driven by resource constraints or prioritizing new-user acquisition over maintaining advanced features, that is something the community should be told directly. What is the broader product strategy here?

mspgrunt

Following the "The new Kagi Assistant is here" post on the June 16th changelog it seems this new layout is here to stay and this new experience will mostly be stuck in stone.

You know, "Forced, Inferior, and Needless UI Redesigns" is something that I have come to expect from Microsoft or Google, but is not what I expect from the Kagi team. Today I have downgraded my subscription to search only and will be looking elsewhere for multimodel access. I'm sure I'm not the only one so I will update this thread later when I've found an alternative that meets my needs.

somerabbit155

I'm also sadly disgruntled by this change. I love kagi. I bought the socks and tell everyone I know about it.

Kagi is the best search experience today - by far. If kagi went away, I might be forced to start my own business to replace it.

And the assistant? Was fine, worked great for me.

I don't care too much about the new ui (though there's a few regressions already mentioned in this thread). But I do care about losing my data, and how kagi is choosing to operate. You've removed my data, with no way to export and no clear reason why. Kagi staff didn't even reply: https://kagifeedback.org/d/10713-bring-back-tags-in-assistant/6

The 4% of users with 2+ tags likely represent some of your most influential and dedicated users. Their impact on your business is outsized in various ways but it doesn't seem like a good idea to upset them. This concerns me because I care about the future of kagi.

Overall, I'm upset that no one at kagi appears to be listening to these concerns or taking them seriously. I genuinely don't understand - if you wanted to make folders the default, with an option for tags buried deeply in the settings ui (and advertsides nowhere), what harm could that possibly cause?

As a business lesson - forced ui overhauls are rarely (practically never) received well. Redesigns typically remove features and reduce stability - I personally ran into a bug today where code output from the code agent is completely malformed because my inputs included a triple backtick code block inside a comment. So unless there is a gripping technical reason to do so, redesigns such as this should not be done, or should be made optional for as long as possible - even without the deletion of user data.

If there is a strong technical reason that the old ui needed to be removed - I don't see how that could possibly be related to the folders/tags change. Maybe an outdated client library? But it would have been ideal to split these changes into two releases.

I don't have any real recourse here - I'm sold on kagi for the foreseeable future. Maybe I'll downgrade to search only. But in my view, kagi's reputation as a users-first organization has been tarnished. And I won't be as vocal about recommending kagi going forward, lets my own reputation be tarnished.

Maybe this was a small enough change that it was truly not known by leadership - but that in itself is imo a red flag for how feedback is being surfaced to the product team. The fact that forced adoption was delayed by a couple weeks tells me this is likely not the case.

In any case, to restore kagi's reputation in my mind, we need clear communication on what led to this change, why it was important to remove our data and existing functionality without optionality, and why there was no response to users given the feedback we surfaced.

As of now, I don't have a lot of confidence to invest in kagi's ecosystem given my investments have not been respected so far.

ax7899

somerabbit155 Yeah I feel the same.

I mean, the two anti-AI threads have gone for over a year as some of the most popular on the forum and Kagi staff have effectively ignored them. Adding a no-AI toggle would be the easiest thing in the world. There's already a CSS that does most of the work. They could've done it in 2024 and we'd be happy, I can only assume the delay and lack of communication is a business decision.

Now the threads about the UI changes are getting the same treatment. There'll be no switching back, they will ignore us until we stop complaining.

I can't help but picture them glancing here and side-eyeing us at this point. I think they've done as any other company and just accepted old users will be alienated. Growth and an ever-expanding ecosystem is more important. Needless to say they've lost my trust and I'm very sad

mspgrunt

somerabbit155 This is the crown summary of the thread.
It seems incredibly arrogant by the Kagi team to kill a feature that 'only' 4% users utilize.
The whole damn product caters to power users. You are alienating your most vocal advocates. Kagi is not growing by freaking ad impressions. Its growing whenever theres a thread complaining about AI / Google and replies start repping Kagi. Youre killing that.

The lack of communication is just insult to injury.

magiruuvelvet

I found an alternative by now. I now use a LLM router service in the EU with proper compliance and budgeting tools and model whitelist features + I have access to more models (including models that were removed from Kagi Assistant) + a TUI (Terminal UI) to control and manage my sessions. I need to manually handle Kagi search integration with MCP now, but I now have absolute control in my new environment, which is a huge improvement. (No more weird word salad injected into my system prompts.) Even compared to the old Kagi Assistant. I don't think I will go back to Kagi Assistant, even if they would theoretically bring back the old UI/UX. I will downgrade to a Professional plan at the end of my billing cycle, since costs are now effectively relocated to the Kagi Search and Extract API.

It was a good run, but time to move on.

v3max

The new Kagi Assistant view is terrible... The previous view was cleaner and more intuitive. Now everything feels stretched out, widened, with a lot of empty space- it's visually painful to look at.

I'm frustrated by the lack of tags, the disproportionate sizing of individual elements, and the strange heading font modeled after the "kagi" typeface- it looks poor.

The interface loads slowly and the whole thing is awful to use...

Kagi- you pride yourselves on listening to your users and building the entire pillar of your company around them. There's no sign that you're actually listening. The new view annoys me so much that I had to give it up. Why change the previous view when it was perfectly fine?

Soma2501

"search or create new folder" is good but its only on the website, not the app

There also seems to be a decent delay before we can interact with the page after going to it...

justadreamer

Why and how was this thread tagged as "done"? The issues described in this thread were not even acknowledged, and are nowhere close to being "done".

« Previous Page