It would be very useful to have the option to edit HTML, CSS, and JavaScript separately, in their own tabs or sections, while still keeping the option to write everything in a single file for those who prefer that. This would make larger pages more organized without limiting different workflows.
It would also be excellent to have the AI integrated directly into the editor. That way, users could manually adjust the code and ask the AI for specific changes in the same place, without going back to the workspace.
Another important improvement would be better integration between UI Pages and Embedded Views. When creating an Embed View, there could be an option to select an existing UI Page from the workspace, instead of manually copying and pasting the link.
Version history would also be very important. Since the AI can change a lot of code at once, being able to restore a previous version would make experimentation much safer.
Finally, a Data Sources / Dependencies panel would be very helpful. It could show which databases, fields, and entities the page uses, along with the permissions applied to that data, such as which roles or groups can view or edit those fields. Not to edit permissions there, but to clearly understand what data the page accesses and which access rules are involved.
As an extra for the future, real-time rendering while editing could also help with visual adjustments, but it does not feel like the most important thing right now.
Even as an experimental feature, the potential is already huge.
I think it would be very hard (impossible) to show which entities the view uses, since this is something that is not fixed but is determined at ‘run time’ i.e. when someone visits the view.
Just like any other data view, the entities that show up may depend on the person viewing, the datetime they are viewing and the values of entities themselves.
And we don’t have field-level permissions, so there is nothing to show in that regard
FWIW, we don’t show the permissions that are enforced for databases in existing data views, so I suspect we wouldn’t put time into implementing that for custom pages only.
By the way, the custom pages implement the same access controls that every other data view does, i.e. it should not be possible for a custom page to result in data being readable/editable to users who otherwise would not have those capabilities.
What I meant is not to force a specific style, but rather to offer it as an option by default.
For example:
When the user describes what they want to build, the agent could ask:
Do you want this to follow Fibery’s native UI style?
Or do you prefer a custom design?
If the user chooses the Fibery-style option, the agent could rely on predefined:
design guidelines
UI patterns
styling conventions
so the generated page looks and feels like a native part of Fibery.
If the user prefers a custom style, they can simply describe it and the agent would follow their instructions instead.
So it’s not about restricting flexibility — just providing a native-looking default option, which can be especially useful for users who want seamless integration inside Fibery without extra effort.
I just ran another test, and I’m even more impressed.
I asked it to create a QR Code scanner for entrance control. The idea is to generate QR codes using the entity’s Public ID. When the QR is scanned from a phone, the page reads that Public ID, finds the matching entity, and shows its information in the dashboard.
With this, it becomes possible to manage entrance flow in a very practical way: check whether the person is registered, whether the payment is correct, whether any data is missing, or any other important information. All directly from a phone, just by opening the page and scanning the QR.
The possibilities feel limitless. Creativity is the limit!
Edit: The camera worked perfectly. When I scanned the QR code, it showed all the information from the entity I selected perfectly.
OMG I’m so pumped about this sprint experiment! It reminds me of this thread - Frontend is essential! - where many users [myself included] crave custom UI/UX on top of our Fibery backend data. If you guys can figure out how to enable public portals and reasonably/attractively price licensing for external users, this could soon be the unicorn I’ve been treasure-hunting for years… Let’s ride!!
Surely this will be a sad day to come. It would be nice also, if we can bring our own agent API in case the team expects to make added pricing due to AI. at least to seperate the add-ons for AI and maintain or improve the current non-AI-related/dependent core features.
What I do is I make some examples so I understand how the HTMLs are working (how to get data from Fibery and puts it back), so I can make changes by hand and I need to use less credits. Otherwise you can get a second Fibery account perhaps to get more credits. 12 or 20 euro’s per month is cheap if it really helps you move forward until Fibery’s new AI agent is announced.
This is very powerful, I love this freedom! It’s great that Fibery is not just no code, but also has places where you can actually code. We can do things that the Fibery team never foresaw, our creativity is the limit! This adds to the top of the pyramid of the hierarchy of needs (as well as strengthen some lower layers) described in the book ‘universal principles of design’.
This blog post summarizes it well (and the book itself is also highly recommended):
As well as some other topics I’ve made. Like I said many times now, this ability to make and embed HTMLs within Fibery (and do it with AI) is very powerful and I love it with all my heart
We built a custom page with a large conditional form that distributes input data across 6 related databases (not only linearly, but also in a matrix-like structure), and also pulls data from more than 5 databases to display options inside the form.
What we’ve run into so far:
The Fibery AI chat runs out of context quite quickly.
Our current workaround is:
use Fibery AI to generate the initial structure
then continue development in ChatGPT, where we pass the base logic and relationships
so it understands how not to break the underlying data structure.
A small thing, but in the Edit HTML Code page there is no search field — Ctrl+F doesn’t seem to work properly.
On the design side — we’re trying to make it look like a native Fibery View, but it’s not perfect yet (though getting close).
It would be ideal if the Fibery AI could generate UI using native Fibery components, for example: dropdowns, checkboxes - if that’s possible .
This way we could build custom pages, but still reuse native Fibery UI elements, making them feel fully integrated.
But now it works stably and perfectly - thank you very much!
Thank you so much for the feedback!
From the next week we will work on a new prototype that will solve 1 and 2.
We also already working on 3.
As for 4, it is hard to say, since Fibery UI was build without AI in mind and we are not sure how easy it will be to refactor these components and make then AI-ready. Maybe it will be easier to restyle some existing library we have on the market
That’s great to hear — thank you so much to you and your team!
We’d also like to highlight that even in its current state, the ability to build custom pages connected to databases already solves a large number of real use cases and works really well for us. Looking forward to seeing the next iteration !
Also some feedback below. I know its experimental so its a bit of a brain dump to give you better context
When agent asks clarifying questions there is no easy way to give responses to multiple questions and the only CTA is “Go Ahead”. I believe Claude is best experience at this (multiple choice with option to type “something else”)
The ability to see the streaming reasoning from the agent is helpful to figure out where it goes off track so I can improve follow up prompts/context or stop it. You can see it after the final generation but not during and no code diffs.
The ability to paste screenshots would be so helpful to provide context for UI
Its great that follow up tweaks maintain the same URL. It’s also scary that follow up tweaks maintain the same URL
The QA agent does a good job of catching errors and outputs almost always work. But wonder what is the main model?
The designs are not great everything is a “card” (all the models are better/worse at this). I tried to influence the design with a simple “design.md” file including principles but it made no real difference to the output. I then simply included a prompt “Make sure you use shadcn” and it made some improvement (although still not great). If it got closer to Fibery design as a default would be cool.
To get better designs I used Pencil (pencil.dev) and its phenomenal. It seems perfectly matched to this type of design process. I downloaded the initial Fibery designed HTML, prompted and tweaked with Pencil, then exported new HTML and uploaded back to Fibery. It worked perfectly, all the code remained in tact and far better design. It took me about 15 minutes total. But some type of “git push” to get the file back into Fibery or testing would really help
I hit “page agent encountered an error: Chat is too long. Please, start a new conversation.” It sounds like you’re already on to fix this.
Also now that I’m able to create a “control surface” it did highlight the fact that data from Fibery connections is stale.
Fibery can be a “single pane of glass” between multiple systems. But if data is stale then it makes it very difficult to take action. I can deep link a URL to the source system and switch context to make updates. But if I make an update in a source system I still have to wait up to an hour (or more depending on the schedule) before I can act upon it in Fibery.
Loving working with this and can see the power of what it can create!!
To follow on with our experience, I’ve tried it 3 times now when I had a few minutes to try and build a dashboard like I’ve wanted to build for a long time. The dashboard layout it generates is very workable; however, it seems to just ignore the Page Agent altogether and create a static list of entities, so it ends up being a really nice mockup. No matter how I prompt it explicitly to generate a dynamic / queried / whatever list and not static html, it still generates a static html.
That said, I’m still very excited about this feature and hope it continues to be improved, because the ability to generate custom pages is a huge unlock and takes away one of the main reasons why I was afraid we may have to exit Fibery in the distant future.
I tried to create a Dashboard for my finances. It seems to work ok, and it will be amazing when it works Although I still did not manage. Since I had to fix a lot of mistakes and clear out some wrong things, after a while, I always get, “page agent encountered an error: Chat is too long. Please, start a new conversation.” And then everything is gone, and I have to start all over again.
That happened 3 times, and I sent more than half of my tokens
What I meant is not to force a specific style, but rather to offer it as an option by default.
.
!
Although I still did not manage. Since I had to fix a lot of mistakes and clear out some wrong things, after a while, I always get, “page agent encountered an error: Chat is too long. Please, start a new conversation.” And then everything is gone, and I have to start all over again.