But I don't come to an AI UI generator with a new design. I'm guessing most people don't.
Most of us have trouble with design, have a design we know is bad, but we've already started building it.
We want to improve the design.
Why aren't there AI tools which model editing?
I would compliment the makes of Claudes chat as an alternative - their "Improve" and "Explain" options when you select text are useful for precisely this reason."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["design"],"value":"Ask HN: Where are AIs for editing (e.g. designs?) vs. creating?"}},"_tags":["story","author_authorfly","story_41600215","ask_hn"],"author":"authorfly","created_at":"2024-09-20T09:26:11Z","created_at_i":1726824371,"num_comments":0,"objectID":"41600215","points":1,"story_id":41600215,"story_text":"I get why many AI tools like the magical sheen of a big "What UI do you want to design? Generate" interface entrance.
But I don't come to an AI UI generator with a new design. I'm guessing most people don't.
Most of us have trouble with design, have a design we know is bad, but we've already started building it.
We want to improve the design.
Why aren't there AI tools which model editing?
I would compliment the makes of Claudes chat as an alternative - their "Improve" and "Explain" options when you select text are useful for precisely this reason.","title":"Ask HN: Where are AIs for editing (e.g. designs?) vs. creating?","updated_at":"2024-09-20T17:45:32Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mattkevan"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"\u2022 A social ebook reading app where you can create reading groups and have realtime discussions.
\u2022 A visual moodboard and notes app that uses local models to link and surface content, a bit like an AI powered Memex.
\u2022 A new UI design tool for Mac/iOS with deep support for design systems and AI agents.
\u2022 A CMS and static site generator that runs entirely in the browser. Download the site as a zip or publish directly to GitHub/Netlify.
https://github.com/sparktype-project/sparktype"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: What are you working on? (June 2026)"}},"_tags":["comment","author_mattkevan","story_48528779"],"author":"mattkevan","comment_text":"\u2022 A social ebook reading app where you can create reading groups and have realtime discussions.
\u2022 A visual moodboard and notes app that uses local models to link and surface content, a bit like an AI powered Memex.
\u2022 A new UI design tool for Mac/iOS with deep support for design systems and AI agents.
\u2022 A CMS and static site generator that runs entirely in the browser. Download the site as a zip or publish directly to GitHub/Netlify.
https://github.com/sparktype-project/sparktype","created_at":"2026-06-14T17:29:13Z","created_at_i":1781458153,"objectID":"48530032","parent_id":48528779,"story_id":48528779,"story_title":"Ask HN: What are you working on? (June 2026)","updated_at":"2026-06-14T18:14:58Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"vlugovsky"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hello HN community,
Many of you have probably used drag-and-drop builders to create internal tools or dashboards. They\u2019re great for quick prototypes - but as projects grow, the same problems always appear:
1) Performance issues. D&D builders add a lot of runtime overhead. Once you add more screens, logic, or connected data sources, things start to lag - and you have little control over optimization.\n2) Limited flexibility. You\u2019re restricted to the vendor\u2019s components and configuration options. Anything custom usually turns into a hack.\n3) Design overhead. You still spend a lot of time manually picking components, arranging layouts, and tweaking spacing before even connecting data - and the final result often looks\u2026 meh.\n4) Maintenance pain. Opening an old project (or one handed off to a new teammate) often means digging around to find where things live. There\u2019s no simple way to reason about structure or reuse.
At UI Bakery, we kept asking ourselves how to fix these issues without losing the speed of a low code builder.
First, we built an AI assistant trained on our docs. Then, about 1.5 years ago, we launched the AI App Generator, which could generate UI Bakery apps from text prompts. That helped, but under the hood it was still drag-and-drop - with all its inherent limits.
That\u2019s when we decided to rethink the foundation entirely.\n\u201cWhat\u2019s the most flexible, customizable project model possible?\u201d\nThe answer: code.
So we rebuilt our approach and are now launching UI Bakery AI App Agent - a new way to build secure internal software by chatting.
Key features:
- Data connectors: Connect to SQL, NoSQL, REST APIs, and third-party services.\n- Speed: Start from a plain-text prompt - AI generates a fully functional app in minutes.\n- Code ownership: Access and edit the generated React code directly.\n- Security: RBAC, SSO, SOC 2 compliance, audit logs, and deployment either in our cloud or fully on-prem.\n- Customization: Use any React component you find online or build your own - it\u2019s all just code.
Who is it for?
- Founders/CTOs automating internal processes.\n- Developers tired of rebuilding CRUD/admin UIs from scratch.\n- Enterprise teams needing internal tools with advanced security (on-prem, SSO, RBAC).
We see UI Bakery as the next evolution beyond drag-and-drop builders \u2014 from "fast but limited" to "fast, customizable, and secure."
We\u2019d love to hear what the HN crowd thinks:\nIf you\u2019ve used drag-and-drop builders, what\u2019s been your biggest pain scaling or maintaining them?\nWould you trust AI to generate internal tools that you can later expand like a normal codebase?
Try it here: https://uibakery.io"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","ui","tool"],"value":"Show HN: UI Bakery AI Agent \u2013 build secure internal tools by chatting"}},"_tags":["story","author_vlugovsky","story_45669218","show_hn"],"author":"vlugovsky","created_at":"2025-10-22T13:54:44Z","created_at_i":1761141284,"num_comments":0,"objectID":"45669218","points":3,"story_id":45669218,"story_text":"Hello HN community,
Many of you have probably used drag-and-drop builders to create internal tools or dashboards. They\u2019re great for quick prototypes - but as projects grow, the same problems always appear:
1) Performance issues. D&D builders add a lot of runtime overhead. Once you add more screens, logic, or connected data sources, things start to lag - and you have little control over optimization.\n2) Limited flexibility. You\u2019re restricted to the vendor\u2019s components and configuration options. Anything custom usually turns into a hack.\n3) Design overhead. You still spend a lot of time manually picking components, arranging layouts, and tweaking spacing before even connecting data - and the final result often looks\u2026 meh.\n4) Maintenance pain. Opening an old project (or one handed off to a new teammate) often means digging around to find where things live. There\u2019s no simple way to reason about structure or reuse.
At UI Bakery, we kept asking ourselves how to fix these issues without losing the speed of a low code builder.
First, we built an AI assistant trained on our docs. Then, about 1.5 years ago, we launched the AI App Generator, which could generate UI Bakery apps from text prompts. That helped, but under the hood it was still drag-and-drop - with all its inherent limits.
That\u2019s when we decided to rethink the foundation entirely.\n\u201cWhat\u2019s the most flexible, customizable project model possible?\u201d\nThe answer: code.
So we rebuilt our approach and are now launching UI Bakery AI App Agent - a new way to build secure internal software by chatting.
Key features:
- Data connectors: Connect to SQL, NoSQL, REST APIs, and third-party services.\n- Speed: Start from a plain-text prompt - AI generates a fully functional app in minutes.\n- Code ownership: Access and edit the generated React code directly.\n- Security: RBAC, SSO, SOC 2 compliance, audit logs, and deployment either in our cloud or fully on-prem.\n- Customization: Use any React component you find online or build your own - it\u2019s all just code.
Who is it for?
- Founders/CTOs automating internal processes.\n- Developers tired of rebuilding CRUD/admin UIs from scratch.\n- Enterprise teams needing internal tools with advanced security (on-prem, SSO, RBAC).
We see UI Bakery as the next evolution beyond drag-and-drop builders \u2014 from "fast but limited" to "fast, customizable, and secure."
We\u2019d love to hear what the HN crowd thinks:\nIf you\u2019ve used drag-and-drop builders, what\u2019s been your biggest pain scaling or maintaining them?\nWould you trust AI to generate internal tools that you can later expand like a normal codebase?
Try it here: https://uibakery.io","title":"Show HN: UI Bakery AI Agent \u2013 build secure internal tools by chatting","updated_at":"2026-03-05T22:52:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"AllYourTech"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","ui","design","tool"],"value":"Hello HN! I\u2019m excited to share PixelDojo.ai, an all-in-one generative AI studio for creators. I built PixelDojo to be a one-stop platform where you can generate images, create short videos, and even train custom AI models (LoRA fine-tunes) from your own images. The idea came about because the AI creative process felt fragmented. 10 months ago I set off on a journey to create the ultimate AI playground. A place where cutting-edge image-generation tools are easily accessible.
My goal was to bring all the latest AI creativity tools under one roof so that anyone, even a \u201cnoob\u201d with no prior knowledge could produce great results out-of-the-box.
Core Features
Multiple Image Models: PixelDojo integrates several state-of-the-art image generation models (like Flux and SDXL variants, plus even Google\u2019s Imagen) in one web app.
You can create visuals from text prompts or by uploading an image for variations. The platform offers one-click access to 44+ AI tools spanning upscaling, style transfer, face enhancement, and even character consistency features. (Think Stable Diffusion, but with a choice of tuned models for photorealism, art styles, etc., and an easy UI.)
AI Video Generation: You can turn text or images into short AI-generated videos (think 5\u201310 second clips). Under the hood it uses multiple video models \u2013 for example, a \u201cKling\u201d model for cinematic quality and a \u201cMinimax (Hailuo)\u201d model for lifelike motion.
In practice, this means you can type a prompt (or take an image you made) and animate it into a video within seconds on our cloud GPUs. No video editing skills or GPU hardware needed. It\u2019s still early for AI video tech, but it\u2019s exciting to play with and improve.
Custom Model Training (LoRA): This might be my favorite feature. PixelDojo lets you train your own Stable Diffusion-based model on your images with a few clicks. Upload, say, 5\u201310 photos of a person or product, and it will fine-tune a LoRA (Low-Rank Adaptation) model that captures their likeness/style. We even support single-image training for quick experiments.
Creative Tools & Community: Beyond the big features above, there are a bunch of handy tools: AI upscalers, background removers, pose control, style remixing, etc. The idea is to cover the whole creation workflow. We also have a community gallery and leaderboards so creators can share and discover cool results. The platform is uncensored (within legal limits) to give artists full creative freedom.
And for developers/hackers, we provide an API + Zapier integration (and you can import your PixelDojo models into external apps or pipelines).
Why use this? For many, it comes down to convenience and cost: Instead of juggling multiple AI notebooks or paying for several separate tools, PixelDojo\u2019s single subscription gives access to everything in one place.
Content creators and small businesses have been using it to spin up marketing graphics, concept art, and social media content extremely fast \u2013 some report cutting design costs by 80% by doing work in-house with these AI tools.
Of course, it\u2019s not a silver bullet and human creativity is still key, but it can handle a lot of tedious production work (and it\u2019s pretty fun to experiment with new models as they come out). The project is bootstrapped and very much a labor of love.
My hope is that PixelDojo can empower fellow creators and builders here to prototype ideas or just make awesome art/videos without barriers.
I\u2019d love to get the community\u2019s feedback, questions, or ideas!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","generator"],"value":"Show HN: PixelDojo \u2013 AI image and video generator with one-click LoRA training"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://pixeldojo.ai"}},"_tags":["story","author_AllYourTech","story_43935614","show_hn"],"author":"AllYourTech","created_at":"2025-05-09T11:21:39Z","created_at_i":1746789699,"num_comments":0,"objectID":"43935614","points":2,"story_id":43935614,"story_text":"Hello HN! I\u2019m excited to share PixelDojo.ai, an all-in-one generative AI studio for creators. I built PixelDojo to be a one-stop platform where you can generate images, create short videos, and even train custom AI models (LoRA fine-tunes) from your own images. The idea came about because the AI creative process felt fragmented. 10 months ago I set off on a journey to create the ultimate AI playground. A place where cutting-edge image-generation tools are easily accessible.
My goal was to bring all the latest AI creativity tools under one roof so that anyone, even a \u201cnoob\u201d with no prior knowledge could produce great results out-of-the-box.
Core Features
Multiple Image Models: PixelDojo integrates several state-of-the-art image generation models (like Flux and SDXL variants, plus even Google\u2019s Imagen) in one web app.
You can create visuals from text prompts or by uploading an image for variations. The platform offers one-click access to 44+ AI tools spanning upscaling, style transfer, face enhancement, and even character consistency features. (Think Stable Diffusion, but with a choice of tuned models for photorealism, art styles, etc., and an easy UI.)
AI Video Generation: You can turn text or images into short AI-generated videos (think 5\u201310 second clips). Under the hood it uses multiple video models \u2013 for example, a \u201cKling\u201d model for cinematic quality and a \u201cMinimax (Hailuo)\u201d model for lifelike motion.
In practice, this means you can type a prompt (or take an image you made) and animate it into a video within seconds on our cloud GPUs. No video editing skills or GPU hardware needed. It\u2019s still early for AI video tech, but it\u2019s exciting to play with and improve.
Custom Model Training (LoRA): This might be my favorite feature. PixelDojo lets you train your own Stable Diffusion-based model on your images with a few clicks. Upload, say, 5\u201310 photos of a person or product, and it will fine-tune a LoRA (Low-Rank Adaptation) model that captures their likeness/style. We even support single-image training for quick experiments.
Creative Tools & Community: Beyond the big features above, there are a bunch of handy tools: AI upscalers, background removers, pose control, style remixing, etc. The idea is to cover the whole creation workflow. We also have a community gallery and leaderboards so creators can share and discover cool results. The platform is uncensored (within legal limits) to give artists full creative freedom.
And for developers/hackers, we provide an API + Zapier integration (and you can import your PixelDojo models into external apps or pipelines).
Why use this? For many, it comes down to convenience and cost: Instead of juggling multiple AI notebooks or paying for several separate tools, PixelDojo\u2019s single subscription gives access to everything in one place.
Content creators and small businesses have been using it to spin up marketing graphics, concept art, and social media content extremely fast \u2013 some report cutting design costs by 80% by doing work in-house with these AI tools.
Of course, it\u2019s not a silver bullet and human creativity is still key, but it can handle a lot of tedious production work (and it\u2019s pretty fun to experiment with new models as they come out). The project is bootstrapped and very much a labor of love.
My hope is that PixelDojo can empower fellow creators and builders here to prototype ideas or just make awesome art/videos without barriers.
I\u2019d love to get the community\u2019s feedback, questions, or ideas!","title":"Show HN: PixelDojo \u2013 AI image and video generator with one-click LoRA training","updated_at":"2025-05-09T12:08:18Z","url":"https://pixeldojo.ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zinedkaloc"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hello, I\u2019m super excited to introduce you to our latest open-source project: AI-Component Generator.
(This tool uses OpenAI's conversational models to let you create and update UI components by chatting with them.)
We believe this will simplify the design and development process in a fun, interactive way.
Check it out here: https://aicomponent.dev
Would love to hear your thoughts!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","generator"],"value":"Show HN: AI Component Generator \u2013 Open-Source Component Generator"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/zinedkaloc/aicomponent.dev"}},"_tags":["story","author_zinedkaloc","story_37622001","show_hn"],"author":"zinedkaloc","created_at":"2023-09-23T10:27:22Z","created_at_i":1695464842,"num_comments":0,"objectID":"37622001","points":1,"story_id":37622001,"story_text":"Hello, I\u2019m super excited to introduce you to our latest open-source project: AI-Component Generator.
(This tool uses OpenAI's conversational models to let you create and update UI components by chatting with them.)
We believe this will simplify the design and development process in a fun, interactive way.
Check it out here: https://aicomponent.dev
Would love to hear your thoughts!","title":"Show HN: AI Component Generator \u2013 Open-Source Component Generator","updated_at":"2024-09-20T15:10:19Z","url":"https://github.com/zinedkaloc/aicomponent.dev"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"gildandstain"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Location: Portland, OR
Remote: Yes! (and hybrid if local)
Willing to relocate: No
Technologies: TypeScript, React, Tailwind, NextJS, Sanity CMS, Node, Python, PyTorch, Netlify, Vercel, Figma, Storybook, Chromatic
R\u00e9sum\u00e9/CV: https://zackrdavis.github.io/resume.pdf
Email: zackrdavis[at]gmail[dot]com
I'm a frontend engineer with lots of experience building fast, fun, accessible user-interfaces. I've worked for startups, enterprise clients, agencies, museums, and artists. I built a robot actor for a theater company and an ML-driven real-time font-generator that you can read about here: https://zackrdavis.github.io/fonting-with-gans-2
I'm looking for close collaboration with designers and product managers, where I can use my design sense and technical chops to their fullest. Specifically, I'd love to help break ground on UI-for-AI puzzles, but I'll be happy anywhere I get to build tools for creating, learning, or understanding data."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Who wants to be hired? (June 2024)"}},"_tags":["comment","author_gildandstain","story_40563280"],"author":"gildandstain","comment_text":"Location: Portland, OR
Remote: Yes! (and hybrid if local)
Willing to relocate: No
Technologies: TypeScript, React, Tailwind, NextJS, Sanity CMS, Node, Python, PyTorch, Netlify, Vercel, Figma, Storybook, Chromatic
R\u00e9sum\u00e9/CV: https://zackrdavis.github.io/resume.pdf
Email: zackrdavis[at]gmail[dot]com
I'm a frontend engineer with lots of experience building fast, fun, accessible user-interfaces. I've worked for startups, enterprise clients, agencies, museums, and artists. I built a robot actor for a theater company and an ML-driven real-time font-generator that you can read about here: https://zackrdavis.github.io/fonting-with-gans-2
I'm looking for close collaboration with designers and product managers, where I can use my design sense and technical chops to their fullest. Specifically, I'd love to help break ground on UI-for-AI puzzles, but I'll be happy anywhere I get to build tools for creating, learning, or understanding data.","created_at":"2024-06-03T16:12:07Z","created_at_i":1717431127,"objectID":"40564134","parent_id":40563280,"story_id":40563280,"story_title":"Ask HN: Who wants to be hired? (June 2024)","updated_at":"2024-09-20T17:13:48Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sminchev"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"I really wanted to see how far I can go. Can I create a meaningful and complex application, big enough, but without knowing the language.
I have 18+ years of experience as software developer. But I have no experience with Kotlin. And to learn Kotlin, to learn the Android libraries, it is not an easy job. I may need at year of active learning and trying things, before having the confidence to start doing something.
So, I asked myself, how far can I go with AI tools? And I went far!
I created https://howareu.app/
The boring statistics:\n 126,000 lines of Kotlin across 398 files\n 45,000 lines of tests across 130 files\n 3000+ unit tests\n 50+ production lessons captured\n 4 months from zero Kotlin experience to production app;\n 0 lines of Kotlin written manually
Below are the main things that I did and helped me do the product!
Good framework\nVibecoding something that big is not an option. You can't write anything in one session. You need a way to keep the context somehow. You need specification. To maintain my specifications I use BMAD framework. It helped me a lot to clear the requirements, prepare the architecture, UI, find any gaps. I used it also a lot for brainstorming session and for marketing. For the latter, it failed
Follow the project rules\nI use Claude Code with Opus 4.6 (yes, the old version) and keep all my important rules saved in CLAUDE.md file. Every rule in that file exists because I violated it once and something broke. The file grows with the project. This is the key insight: CLAUDE.md turns one-time lessons into permanent constraints. The AI never forgets a rule I put there. I forget constantly.
Keep your documentation up to date\nI start my session with a custom command that loads all important documents. Enough context, so that when I start prompting, the AI has the basis to make best decisions possible.\nI end my session with custom command to save all learned, and architectural updates. This way each session is built based on the knowledge from the previous. I have a history, I have an AI that gets smarter with each next session.
Be as descriptive as possible
When you prompt, be as descriptive as possible. I use BMAD to create technical specifications out of my prompts. If the idea is not clear in my head, I do brainstorming session. All possible to minimize the guessing and wrong interpretations during the development phases.
Do you all do code reviews?\nI do. Usually, with a different LLM. This gives a 'different point of view'. Improves greatly the code quality!
What this is not\n It is a no-code, but If you know the language, it is worth checking and correcting. With time, the needed small fixes will become less. It is always good to understand and to make the design decisions yourself.\n It is not effortless. The workflow took months to build. The documentation is extensive.\n It is not magic. The AI makes mistakes. The difference is that mistakes are caught by the process (tests, reviews, rules, audits) instead of by users;
The takeaway
AI coding tools are not magic code generators. They are force multipliers for engineering process. If your process is "open chat, type prompt, hope for the best," you will be disappointed.
If your process is "document the architecture, define the rules, automate the lifecycle, capture every lesson, review everything critically", the AI becomes unreasonably effective. The investment is not in better prompts. It is in better engineering.
Honesty first
Why am I writing all this?! Because I am desperate! I made this app to help people. Make something good. I want to be sure that I cover as many cases as possible, and that it will really help people. The moment I share in the proper channels that such app exists, I get treated as an intruder, who sells evil!
If somebody would like to help, please, install it, and give me a honest feedback!\nThank you!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"I created a 126K line Android app with AI \u2013 the workflow that worked for me"}},"_tags":["story","author_sminchev","story_48184549","ask_hn"],"author":"sminchev","children":[48330433,48330528],"created_at":"2026-05-18T19:44:03Z","created_at_i":1779133443,"num_comments":2,"objectID":"48184549","points":2,"story_id":48184549,"story_text":"I really wanted to see how far I can go. Can I create a meaningful and complex application, big enough, but without knowing the language.
I have 18+ years of experience as software developer. But I have no experience with Kotlin. And to learn Kotlin, to learn the Android libraries, it is not an easy job. I may need at year of active learning and trying things, before having the confidence to start doing something.
So, I asked myself, how far can I go with AI tools? And I went far!
I created https://howareu.app/
The boring statistics:\n 126,000 lines of Kotlin across 398 files\n 45,000 lines of tests across 130 files\n 3000+ unit tests\n 50+ production lessons captured\n 4 months from zero Kotlin experience to production app;\n 0 lines of Kotlin written manually
Below are the main things that I did and helped me do the product!
Good framework\nVibecoding something that big is not an option. You can't write anything in one session. You need a way to keep the context somehow. You need specification. To maintain my specifications I use BMAD framework. It helped me a lot to clear the requirements, prepare the architecture, UI, find any gaps. I used it also a lot for brainstorming session and for marketing. For the latter, it failed
Follow the project rules\nI use Claude Code with Opus 4.6 (yes, the old version) and keep all my important rules saved in CLAUDE.md file. Every rule in that file exists because I violated it once and something broke. The file grows with the project. This is the key insight: CLAUDE.md turns one-time lessons into permanent constraints. The AI never forgets a rule I put there. I forget constantly.
Keep your documentation up to date\nI start my session with a custom command that loads all important documents. Enough context, so that when I start prompting, the AI has the basis to make best decisions possible.\nI end my session with custom command to save all learned, and architectural updates. This way each session is built based on the knowledge from the previous. I have a history, I have an AI that gets smarter with each next session.
Be as descriptive as possible
When you prompt, be as descriptive as possible. I use BMAD to create technical specifications out of my prompts. If the idea is not clear in my head, I do brainstorming session. All possible to minimize the guessing and wrong interpretations during the development phases.
Do you all do code reviews?\nI do. Usually, with a different LLM. This gives a 'different point of view'. Improves greatly the code quality!
What this is not\n It is a no-code, but If you know the language, it is worth checking and correcting. With time, the needed small fixes will become less. It is always good to understand and to make the design decisions yourself.\n It is not effortless. The workflow took months to build. The documentation is extensive.\n It is not magic. The AI makes mistakes. The difference is that mistakes are caught by the process (tests, reviews, rules, audits) instead of by users;
The takeaway
AI coding tools are not magic code generators. They are force multipliers for engineering process. If your process is "open chat, type prompt, hope for the best," you will be disappointed.
If your process is "document the architecture, define the rules, automate the lifecycle, capture every lesson, review everything critically", the AI becomes unreasonably effective. The investment is not in better prompts. It is in better engineering.
Honesty first
Why am I writing all this?! Because I am desperate! I made this app to help people. Make something good. I want to be sure that I cover as many cases as possible, and that it will really help people. The moment I share in the proper channels that such app exists, I get treated as an intruder, who sells evil!
If somebody would like to help, please, install it, and give me a honest feedback!\nThank you!","title":"I created a 126K line Android app with AI \u2013 the workflow that worked for me","updated_at":"2026-05-29T23:08:07Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"IuliiaShnai"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ui","design","generator","tool"],"value":"Wanna share a new micro-tool I built: pitch deck generator
How to works? \n1. Add startup details\n2. Generate your deck\n2. Edit slides\n3. Turn it instantly in a link on Papermark\n4. Share and track views ( on each page)
Design inspired by Midday deck which has over 17k views and amazing UI."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","generator"],"value":"Show HN: AI Pitch Deck Generator"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://deck.papermark.io"}},"_tags":["story","author_IuliiaShnai","story_40718539","show_hn"],"author":"IuliiaShnai","created_at":"2024-06-18T14:56:08Z","created_at_i":1718722568,"num_comments":0,"objectID":"40718539","points":1,"story_id":40718539,"story_text":"Wanna share a new micro-tool I built: pitch deck generator
How to works? \n1. Add startup details\n2. Generate your deck\n2. Edit slides\n3. Turn it instantly in a link on Papermark\n4. Share and track views ( on each page)
Design inspired by Midday deck which has over 17k views and amazing UI.","title":"Show HN: AI Pitch Deck Generator","updated_at":"2024-09-20T17:13:16Z","url":"https://deck.papermark.io"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hfriedrich"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"https://youtu.be/6OZxm7ZEVU0
I spent a while building N.E.O. (Native Executable Orchestrator) \u2013 an AI-powered tool that turns natural language prompts into compiled, live, running .NET desktop applications. Think of it like ChatGPT Canvas or Claude Artifacts, but for native Windows apps. Not mockups or web previews \u2014 actual (optionally sandboxed) binaries.
My core philosophy for this was frictionless setup. The tool requires absolutely no SDKs. All it needs is the standard .NET runtime. No .NET SDK, no Node.js, no manual environment configurations. If your prompt requires Python, N.E.O. automatically downloads and sets up the necessary environment under the hood. From a user perspective: download the installer, set your API keys, type a prompt, and it works.
I'm a teacher, not a Silicon Valley startup founder \u2014 I started this as a fun side project. At some point, I tried to make this a commercial product, but I suck at marketing and it didn't work out. :) But before I let the software collect dust, I wanted to ask HN: is this worth polishing and open-sourcing?
What N.E.O. does:\n- Instant Native UI: You type a prompt like "build me a dashboard with live charts" and the tool generates a running native desktop app (WPF or Avalonia) right before your eyes.\n- Natural Language Iteration: Describe changes in plain language. The AI applies diffs, recompiles (without an SDK!), and the app updates instantly.\n- Hybrid Stack Capabilities: You can mix native (C#), web (React), and Python in a single app. For example: "Build a stock analyzer with a React-based interactive chart, and use Python for the statistical calculations" \u2014 N.E.O. will embed the web components and Python environment behind a native UI.
Some technical details that might interest you:\n- SDK-less Compilation: Compiles C# on the fly without the .NET SDK. NuGet and Python dependencies are fetched automatically on demand.\n- Self-Healing: Errors are fed back to the AI automatically. Even runtime crashes trigger automatic repair loops.\n- Visual Click-to-Edit: Click any control in the running app (WPF only), and a basic property editor appears.\n- Security & Sandboxing: Dual-layer security (14 heuristic rules + AI semantic risk assessment) before applying generated code. Apps can also be run in an optional Windows AppContainer sandbox to restrict network and file access.\n- Branching Undo/Redo: The history is not a linear stack, but a tree. Explore different design directions and easily revert to any branch.\n- Cross-Platform Export: Export native WPF apps for Windows, or generate Avalonia UIs that can be exported for Windows, Linux, and macOS.\n- Apps Can Call AI: The resulting applications can themselves use Claude/GPT/Gemini/Ollama/ LM Studio at runtime via an automatically exposed API.
What I'm honestly wondering:
Desktop app builders feel like an abandoned category. Everyone is building web-first AI code generators. But aren't there still domains where desktop wins \u2014 enterprise internal tools, offline apps, hardware access, kiosk systems? I teach C# and WPF for a living, and one of my main motivations for building N.E.O. was to stress-test the entire C# toolchain: where exactly does an AI-driven native app generator hit a wall because essential APIs, tooling, or framework support is missing?
Is this worth polishing and open-sourcing as a community project, or should I just move on and try to perfect my brownie recipe?
The code is messy and far from perfect, but the core architecture works. Happy to answer any questions about the implementation!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","generator"],"value":"Show HN: Neo \u2013 AI-powered native .NET desktop app generator"}},"_tags":["story","author_hfriedrich","story_47262437","show_hn"],"author":"hfriedrich","created_at":"2026-03-05T15:10:01Z","created_at_i":1772723401,"num_comments":0,"objectID":"47262437","points":1,"story_id":47262437,"story_text":"https://youtu.be/6OZxm7ZEVU0
I spent a while building N.E.O. (Native Executable Orchestrator) \u2013 an AI-powered tool that turns natural language prompts into compiled, live, running .NET desktop applications. Think of it like ChatGPT Canvas or Claude Artifacts, but for native Windows apps. Not mockups or web previews \u2014 actual (optionally sandboxed) binaries.
My core philosophy for this was frictionless setup. The tool requires absolutely no SDKs. All it needs is the standard .NET runtime. No .NET SDK, no Node.js, no manual environment configurations. If your prompt requires Python, N.E.O. automatically downloads and sets up the necessary environment under the hood. From a user perspective: download the installer, set your API keys, type a prompt, and it works.
I'm a teacher, not a Silicon Valley startup founder \u2014 I started this as a fun side project. At some point, I tried to make this a commercial product, but I suck at marketing and it didn't work out. :) But before I let the software collect dust, I wanted to ask HN: is this worth polishing and open-sourcing?
What N.E.O. does:\n- Instant Native UI: You type a prompt like "build me a dashboard with live charts" and the tool generates a running native desktop app (WPF or Avalonia) right before your eyes.\n- Natural Language Iteration: Describe changes in plain language. The AI applies diffs, recompiles (without an SDK!), and the app updates instantly.\n- Hybrid Stack Capabilities: You can mix native (C#), web (React), and Python in a single app. For example: "Build a stock analyzer with a React-based interactive chart, and use Python for the statistical calculations" \u2014 N.E.O. will embed the web components and Python environment behind a native UI.
Some technical details that might interest you:\n- SDK-less Compilation: Compiles C# on the fly without the .NET SDK. NuGet and Python dependencies are fetched automatically on demand.\n- Self-Healing: Errors are fed back to the AI automatically. Even runtime crashes trigger automatic repair loops.\n- Visual Click-to-Edit: Click any control in the running app (WPF only), and a basic property editor appears.\n- Security & Sandboxing: Dual-layer security (14 heuristic rules + AI semantic risk assessment) before applying generated code. Apps can also be run in an optional Windows AppContainer sandbox to restrict network and file access.\n- Branching Undo/Redo: The history is not a linear stack, but a tree. Explore different design directions and easily revert to any branch.\n- Cross-Platform Export: Export native WPF apps for Windows, or generate Avalonia UIs that can be exported for Windows, Linux, and macOS.\n- Apps Can Call AI: The resulting applications can themselves use Claude/GPT/Gemini/Ollama/ LM Studio at runtime via an automatically exposed API.
What I'm honestly wondering:
Desktop app builders feel like an abandoned category. Everyone is building web-first AI code generators. But aren't there still domains where desktop wins \u2014 enterprise internal tools, offline apps, hardware access, kiosk systems? I teach C# and WPF for a living, and one of my main motivations for building N.E.O. was to stress-test the entire C# toolchain: where exactly does an AI-driven native app generator hit a wall because essential APIs, tooling, or framework support is missing?
Is this worth polishing and open-sourcing as a community project, or should I just move on and try to perfect my brownie recipe?
The code is messy and far from perfect, but the core architecture works. Happy to answer any questions about the implementation!","title":"Show HN: Neo \u2013 AI-powered native .NET desktop app generator","updated_at":"2026-03-05T23:42:13Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"alexdanilowicz"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Alex and Teddy here. We\u2019re launching Magic Patterns (https://www.magicpatterns.com), an AI prototyping tool that helps PMs and designers create functional, interactive designs and websites. There\u2019s a demo video at https://www.youtube.com/watch?v=SK8C_tQBwIU, as well as video walkthroughs of specific examples at https://www.magicpatterns.com/docs/documentation/tutorials/v...
While other tools help with \u201cAI-assisted coding,\u201d we have been quietly focused on \u201cAI-assisted designing.\u201d With Magic Patterns you can visually communicate your idea, get hands on feedback from customers, and test new features.
Teddy and I are best friends and former frontend engineers turned founders. We arrived at Magic Patterns after several pivots\u2014always in the design tooling space, but different products that all struggled to get usage. We started working on Magic Patterns after an internal hackathon. Teddy built a UI library catalog and I messed around with GPT 3.5. We thought it\u2019d be fun to combine the two: an AI component generator. Describe whatever you want, and get back a React component!
That started to take off and we gained users, but it wasn\u2019t developers using the tool. Instead, it was PMs, designers, and leadership who could finally communicate their ideas. They use it to test new ideas quickly, get feedback from customers, and improve communication with internal teams. Also, hobbyists (and programmers who aren\u2019t designers) use us to create designs and UIs that they wouldn\u2019t be able to otherwise.
We use Sonnet 3.5 and 3.7, and leverage a fine-tuned model for fast-applying edits. The most challenging part is determining the most relevant context to feed to the LLM. We attempt to solve this with our click to update feature and by letting users define a brand preset, or default prompt.
Unlike other tools in this space, we\u2019re specifically focused on (1) product teams\u2014we're realtime and collaborative; and (2) frontend only\u2014we don't spin up a database or backend because we aren't solving "idea to fullstack app."
A common workflow is a product manager building an interactive prototype and then passing it off to a designer for more polish or directly to engineers. Many teams are even skipping Figma entirely now, telling us that it feels like an unnecessary middleman. Teams are instead generating clickable prototypes, collaborating directly with stakeholders, and using that as the mockup.
With Magic Patterns, you can: - Collaborate with your team on our infinite canvas; - Match your existing designs by creating reusable components directly; - Brainstorm features and flows. (The latter is what we use it for internally.)
We started as a way to build small, custom components, but now people are one-shotting entire\nwebsites and hosting them with us, or building dashboards that they share internally or in customer demos. People have sold $10k/mo contracts with Magic Patterns designs!
Small business owners\u2014everyone from fishermen to driving instructors to hotel managers\u2014are using us to build their websites and then hosting them with us. Example sites built by Magic Patterns include https://getdealflow.ai/ and https://joinringo.com/. It\u2019s amazing how people who couldn\u2019t have done that before are now able to, and super gratifying to us to be empowering people in this way.
You can get started with our docs here: https://www.magicpatterns.com/docs/documentation/get-started..., and you can try the actual product. Simply go to https://www.magicpatterns.com and prompt for any UI you want.
Today no login is required, just click \u201cComing from Hackernews?\u201d and you\u2019ll get 5 messages free to try. Once you hit the limit, you\u2019ll then be prompted to login. Plans start at $19/mo for another 100 messages a month (https://www.magicpatterns.com/pricing).
We\u2019re stoked to be sharing with HN today and are open to all feedback!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","design"],"value":"Launch HN: Magic Patterns (YC W23) \u2013 AI Design and Prototyping for Product Teams"}},"_tags":["story","author_alexdanilowicz","story_43752176","launch_hn"],"author":"alexdanilowicz","children":[43752574,43752593,43752637,43752860,43752900,43752916,43753215,43753615,43753616,43754344,43754523,43755276,43755483,43755754,43755795,43755895,43756451,43756663,43756680,43757466,43757543,43757565,43757695,43758671,43758909,43759515,43759589,43759806,43760141,43760297,43760703,43760745,43760926,43762430,43763550,43763610],"created_at":"2025-04-21T14:07:03Z","created_at_i":1745244423,"num_comments":114,"objectID":"43752176","points":185,"story_id":43752176,"story_text":"Alex and Teddy here. We\u2019re launching Magic Patterns (https://www.magicpatterns.com), an AI prototyping tool that helps PMs and designers create functional, interactive designs and websites. There\u2019s a demo video at https://www.youtube.com/watch?v=SK8C_tQBwIU, as well as video walkthroughs of specific examples at https://www.magicpatterns.com/docs/documentation/tutorials/v...
While other tools help with \u201cAI-assisted coding,\u201d we have been quietly focused on \u201cAI-assisted designing.\u201d With Magic Patterns you can visually communicate your idea, get hands on feedback from customers, and test new features.
Teddy and I are best friends and former frontend engineers turned founders. We arrived at Magic Patterns after several pivots\u2014always in the design tooling space, but different products that all struggled to get usage. We started working on Magic Patterns after an internal hackathon. Teddy built a UI library catalog and I messed around with GPT 3.5. We thought it\u2019d be fun to combine the two: an AI component generator. Describe whatever you want, and get back a React component!
That started to take off and we gained users, but it wasn\u2019t developers using the tool. Instead, it was PMs, designers, and leadership who could finally communicate their ideas. They use it to test new ideas quickly, get feedback from customers, and improve communication with internal teams. Also, hobbyists (and programmers who aren\u2019t designers) use us to create designs and UIs that they wouldn\u2019t be able to otherwise.
We use Sonnet 3.5 and 3.7, and leverage a fine-tuned model for fast-applying edits. The most challenging part is determining the most relevant context to feed to the LLM. We attempt to solve this with our click to update feature and by letting users define a brand preset, or default prompt.
Unlike other tools in this space, we\u2019re specifically focused on (1) product teams\u2014we're realtime and collaborative; and (2) frontend only\u2014we don't spin up a database or backend because we aren't solving "idea to fullstack app."
A common workflow is a product manager building an interactive prototype and then passing it off to a designer for more polish or directly to engineers. Many teams are even skipping Figma entirely now, telling us that it feels like an unnecessary middleman. Teams are instead generating clickable prototypes, collaborating directly with stakeholders, and using that as the mockup.
With Magic Patterns, you can: - Collaborate with your team on our infinite canvas; - Match your existing designs by creating reusable components directly; - Brainstorm features and flows. (The latter is what we use it for internally.)
We started as a way to build small, custom components, but now people are one-shotting entire\nwebsites and hosting them with us, or building dashboards that they share internally or in customer demos. People have sold $10k/mo contracts with Magic Patterns designs!
Small business owners\u2014everyone from fishermen to driving instructors to hotel managers\u2014are using us to build their websites and then hosting them with us. Example sites built by Magic Patterns include https://getdealflow.ai/ and https://joinringo.com/. It\u2019s amazing how people who couldn\u2019t have done that before are now able to, and super gratifying to us to be empowering people in this way.
You can get started with our docs here: https://www.magicpatterns.com/docs/documentation/get-started..., and you can try the actual product. Simply go to https://www.magicpatterns.com and prompt for any UI you want.
Today no login is required, just click \u201cComing from Hackernews?\u201d and you\u2019ll get 5 messages free to try. Once you hit the limit, you\u2019ll then be prompted to login. Plans start at $19/mo for another 100 messages a month (https://www.magicpatterns.com/pricing).
We\u2019re stoked to be sharing with HN today and are open to all feedback!","title":"Launch HN: Magic Patterns (YC W23) \u2013 AI Design and Prototyping for Product Teams","updated_at":"2026-05-02T02:11:41Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sarreph"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hey HN! We\u2019re Rory, Chris, and Felix from Kenobi (https://kenobi.ai). Kenobi lets you add AI-based content personalization to any website. As a site owner, you install our personalization widget with a script tag, just like you would for e.g. a chatbot integration. As a visitor, you interact with the widget (right now by providing a company name) and Kenobi changes the site content to suit you.
We\u2019ve built a demo that anyone can try here: https://kenobi.ai/start
We believe that large parts of the web are about to go from being static to dynamic because of how adept LLMs are at transforming rendered HTML. And right now we\u2019re focussing on B2B landing page content (as opposed to application UIs) because there is a lot of commercial opportunity for increasing top-of-funnel inbound conversions.
Our journey to Kenobi today is a long and snaking one. You may notice from the post title that we did YC\u2019s Winter 2022 batch (I know that 4 years is practically ancient in YC-dog-years). Kenobi is a hard pivot from our original idea that we got accepted into YC with \u2014 a company called Verdn which did trackable environmental donations via an API. Since the summer, we\u2019ve been hacking on different ideas\u2026 We started with personalized UI screenshots for outbound campaigns, but then people told us they wanted transformations to their actual site[0] \u2014 so we built an agentic workflow to research a visitor-company and \u201cpre-render\u201d changes to a landing site for them. Ultimately, there was too much friction in getting people to incorporate personalized URLs into their cold outbound campaigns[1]. Besides, people kept asking for us to do this for their inbound traffic, and so our current iteration was born.
Right now with Kenobi you pick a page that you\u2019d like to make customizable, and choose [text] elements that you\u2019d like to make dynamic. You can define custom prompting instructions for these elements, and when someone visits your page, our agentic workflow researches their company, and presents the updated content as quickly as possible, usually within a few seconds.[2] You also get a ping in Slack every time this happens so you know who is using your site.
We\u2019ve been experimenting with features such as generating custom imagery that actually looks good and native to the page design, and pulling in company data sources so that e.g. the right case study can be presented based on a visitor\u2019s industry and ICP profile. Our most requested feature is deanonymizing traffic so that Kenobi\u2019s personalization can happen automatically as visitors land on your page \u2014 this is coming very soon, as right now you have to specify where you\u2019re coming from.
It\u2019s surprised us just how much business value we\u2019ve gotten from knowing who (most probably) is on the page and asking for a personalized experience. We\u2019ve seen response rates 3x of what we would normally from following people up from companies we know visited our site.
There are many players in this space already, and everyone seems to have their own angle. We are keen to hear thoughts on what people think the future of the personalized internet looks like!
Cheers from London!
P.S. - there's also a video that Chris recorded showing the end-to-end Kenobi experience right now https://www.loom.com/share/bc0a82a2f2fd40f695315bae80e8f5d8
[0] - Many of them had tried AI \u201cmicrosite\u201d generators but found the maintenance of managing a separate website(s) just for closing deals to be burdensome and inefficient.
[1] - Despite having a CSV export and Clay integration option for our pre-generated website changes, getting people to weave the URLs into their email sequences (everyone uses different tools) seemed almost insurmountable without building what would ostensibly be our own sequencing software.
[2] - We use light foundation models with grounded search for the research step, and translate these into markup changes via another light LLM pass and our own DSL which is optimized for speed."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Kenobi \u2013 AI personalized website content for every visitor"}},"_tags":["story","author_sarreph","story_46277709","show_hn"],"author":"sarreph","children":[46277795],"created_at":"2025-12-15T17:41:57Z","created_at_i":1765820517,"num_comments":2,"objectID":"46277709","points":7,"story_id":46277709,"story_text":"Hey HN! We\u2019re Rory, Chris, and Felix from Kenobi (https://kenobi.ai). Kenobi lets you add AI-based content personalization to any website. As a site owner, you install our personalization widget with a script tag, just like you would for e.g. a chatbot integration. As a visitor, you interact with the widget (right now by providing a company name) and Kenobi changes the site content to suit you.
We\u2019ve built a demo that anyone can try here: https://kenobi.ai/start
We believe that large parts of the web are about to go from being static to dynamic because of how adept LLMs are at transforming rendered HTML. And right now we\u2019re focussing on B2B landing page content (as opposed to application UIs) because there is a lot of commercial opportunity for increasing top-of-funnel inbound conversions.
Our journey to Kenobi today is a long and snaking one. You may notice from the post title that we did YC\u2019s Winter 2022 batch (I know that 4 years is practically ancient in YC-dog-years). Kenobi is a hard pivot from our original idea that we got accepted into YC with \u2014 a company called Verdn which did trackable environmental donations via an API. Since the summer, we\u2019ve been hacking on different ideas\u2026 We started with personalized UI screenshots for outbound campaigns, but then people told us they wanted transformations to their actual site[0] \u2014 so we built an agentic workflow to research a visitor-company and \u201cpre-render\u201d changes to a landing site for them. Ultimately, there was too much friction in getting people to incorporate personalized URLs into their cold outbound campaigns[1]. Besides, people kept asking for us to do this for their inbound traffic, and so our current iteration was born.
Right now with Kenobi you pick a page that you\u2019d like to make customizable, and choose [text] elements that you\u2019d like to make dynamic. You can define custom prompting instructions for these elements, and when someone visits your page, our agentic workflow researches their company, and presents the updated content as quickly as possible, usually within a few seconds.[2] You also get a ping in Slack every time this happens so you know who is using your site.
We\u2019ve been experimenting with features such as generating custom imagery that actually looks good and native to the page design, and pulling in company data sources so that e.g. the right case study can be presented based on a visitor\u2019s industry and ICP profile. Our most requested feature is deanonymizing traffic so that Kenobi\u2019s personalization can happen automatically as visitors land on your page \u2014 this is coming very soon, as right now you have to specify where you\u2019re coming from.
It\u2019s surprised us just how much business value we\u2019ve gotten from knowing who (most probably) is on the page and asking for a personalized experience. We\u2019ve seen response rates 3x of what we would normally from following people up from companies we know visited our site.
There are many players in this space already, and everyone seems to have their own angle. We are keen to hear thoughts on what people think the future of the personalized internet looks like!
Cheers from London!
P.S. - there's also a video that Chris recorded showing the end-to-end Kenobi experience right now https://www.loom.com/share/bc0a82a2f2fd40f695315bae80e8f5d8
[0] - Many of them had tried AI \u201cmicrosite\u201d generators but found the maintenance of managing a separate website(s) just for closing deals to be burdensome and inefficient.
[1] - Despite having a CSV export and Clay integration option for our pre-generated website changes, getting people to weave the URLs into their email sequences (everyone uses different tools) seemed almost insurmountable without building what would ostensibly be our own sequencing software.
[2] - We use light foundation models with grounded search for the research step, and translate these into markup changes via another light LLM pass and our own DSL which is optimized for speed.","title":"Show HN: Kenobi \u2013 AI personalized website content for every visitor","updated_at":"2026-03-05T23:12:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"pancsta"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hello HN!
Intro: I'm pleased to announce AI-gent Workflows [0], a new AI Agents platform with a local reasoning layer. It's based on a purpose-built state machine [1] (Actor Model) and comes with various devtools, observability, basic tools (eg search, scraping), and ubiquitous UIs. It can work natively on mobile devices, and UI sessions are more akin to remote desktop connections (easy sharing without any syncing).
Context: There was a brief HN discussion [2] while still in the stealth mode, as it addresses some design mistakes of other frameworks. In secai (the project's codename), everything is an "organic workflow" from the bottom layer up, including tools. It's called a "stateful flow graph" and it decides about the control flow. Each graph transition can be debugged to a very granular decision-making step. That leads us to the most interesting part - reasoning.
Reasoning: Turns out LLMs are great at translating procedures to that state machine, which happens via simple relations between states. Combined with Inversion of Control (IoC), all the actions are performed and reasoned about within the state machine itself (not an FSM). This isn't a low-code solution, but there is a schema layer, which may be given to a codeless person to easily adjust an agent. Because agents are state machines and agents can easily define state machines, they can effectively create themselves (to a degree).
Prompts: Everything is defined in a schema and prompts are mostly data structures. Each state can have a prompt attached to itself, and each prompt can have documents (context). Above states, there's an abstraction of Stories with actors (state machines). All the prompts are nicely archived in an SQLite file.
Memory: There are 3 memory layers - long term (SQL), short term (dynamic state machines), and the transition log. The last one is actually a stream of ML-ready binary vectors. Thanks to the relational nature of the "stateful flow graph", combined with logical clocks (each state has one), it can be used to mine useful data insights.
Devtools: This is probably the broadest topic, but I'll keep it short - distributed debugger, REPL, SVG diagrams, code generators, grafana dash generators, deep-linked Otel traces, and more.
Performance: Before starting the work on the AI layer, I've ported libp2p-pubsub to this state machine to see how much slower it would get in their own load tests - resulting in +15-20% CPU, with the same memory ceiling.
It took 18 months and >30k CLoC of Golang to make it.
Enjoy!
[2] https://news.ycombinator.com/item?id=43703196"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: AI-gent Workflows \u2013 locally reasoning AI Agents"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://ai-gents.work"}},"_tags":["story","author_pancsta","story_44386314","show_hn"],"author":"pancsta","children":[44386449],"created_at":"2025-06-26T11:27:52Z","created_at_i":1750937272,"num_comments":1,"objectID":"44386314","points":3,"story_id":44386314,"story_text":"Hello HN!
Intro: I'm pleased to announce AI-gent Workflows [0], a new AI Agents platform with a local reasoning layer. It's based on a purpose-built state machine [1] (Actor Model) and comes with various devtools, observability, basic tools (eg search, scraping), and ubiquitous UIs. It can work natively on mobile devices, and UI sessions are more akin to remote desktop connections (easy sharing without any syncing).
Context: There was a brief HN discussion [2] while still in the stealth mode, as it addresses some design mistakes of other frameworks. In secai (the project's codename), everything is an "organic workflow" from the bottom layer up, including tools. It's called a "stateful flow graph" and it decides about the control flow. Each graph transition can be debugged to a very granular decision-making step. That leads us to the most interesting part - reasoning.
Reasoning: Turns out LLMs are great at translating procedures to that state machine, which happens via simple relations between states. Combined with Inversion of Control (IoC), all the actions are performed and reasoned about within the state machine itself (not an FSM). This isn't a low-code solution, but there is a schema layer, which may be given to a codeless person to easily adjust an agent. Because agents are state machines and agents can easily define state machines, they can effectively create themselves (to a degree).
Prompts: Everything is defined in a schema and prompts are mostly data structures. Each state can have a prompt attached to itself, and each prompt can have documents (context). Above states, there's an abstraction of Stories with actors (state machines). All the prompts are nicely archived in an SQLite file.
Memory: There are 3 memory layers - long term (SQL), short term (dynamic state machines), and the transition log. The last one is actually a stream of ML-ready binary vectors. Thanks to the relational nature of the "stateful flow graph", combined with logical clocks (each state has one), it can be used to mine useful data insights.
Devtools: This is probably the broadest topic, but I'll keep it short - distributed debugger, REPL, SVG diagrams, code generators, grafana dash generators, deep-linked Otel traces, and more.
Performance: Before starting the work on the AI layer, I've ported libp2p-pubsub to this state machine to see how much slower it would get in their own load tests - resulting in +15-20% CPU, with the same memory ceiling.
It took 18 months and >30k CLoC of Golang to make it.
Enjoy!
[2] https://news.ycombinator.com/item?id=43703196","title":"Show HN: AI-gent Workflows \u2013 locally reasoning AI Agents","updated_at":"2025-06-26T12:30:45Z","url":"https://ai-gents.work"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ynnickw"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hi HN,
I built AppLaunchFlow to help indie developers and small teams create App Store marketing assets faster.
When launching an app, you usually need a lot of materials:\n\u2022 App Store screenshots\n\u2022 promo graphics\n\u2022 social media visuals\n\u2022 marketing copy
Most tools only help with designing screenshots, but they don\u2019t help with the actual marketing workflow.
So I built a tool that starts with your app screenshots and uses AI to extract the important information from them \u2014 features, UI structure, and value propositions \u2014 and then helps generate all the assets needed for a launch.
Main features:
- AI screenshot analysis \u2013 extracts features and messaging from your UI
- App Store screenshot generator \u2013 optimized layouts for iPhone & iPad
- Promo graphics generator \u2013 social media and marketing visuals
- Figma-style editor \u2013 fully customizable layouts, not locked templates
- App launch asset generator \u2013 quickly produce everything needed for a release
The goal is to become a complete launch toolkit for app developers.
Instead of spending hours in design tools, you upload screenshots and iterate quickly until you have polished assets ready for:
- App Store listing\n- Product Hunt launch\n- social media marketing\n- promo graphics
This tool is mainly built for indie developers, small teams, and app agencies who want to ship fast without hiring a designer.
I\u2019d love feedback from the HN community.
Website:\nhttps://applaunchflow.com
Curious about what assets you usually create when launching an app
Thanks!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","tool"],"value":"Show HN: AppLaunchFlow: AI tool that turns app screenshots into marketing assets"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.applaunchflow.com"}},"_tags":["story","author_ynnickw","story_47259123","show_hn"],"author":"ynnickw","created_at":"2026-03-05T08:30:37Z","created_at_i":1772699437,"num_comments":0,"objectID":"47259123","points":1,"story_id":47259123,"story_text":"Hi HN,
I built AppLaunchFlow to help indie developers and small teams create App Store marketing assets faster.
When launching an app, you usually need a lot of materials:\n\u2022 App Store screenshots\n\u2022 promo graphics\n\u2022 social media visuals\n\u2022 marketing copy
Most tools only help with designing screenshots, but they don\u2019t help with the actual marketing workflow.
So I built a tool that starts with your app screenshots and uses AI to extract the important information from them \u2014 features, UI structure, and value propositions \u2014 and then helps generate all the assets needed for a launch.
Main features:
- AI screenshot analysis \u2013 extracts features and messaging from your UI
- App Store screenshot generator \u2013 optimized layouts for iPhone & iPad
- Promo graphics generator \u2013 social media and marketing visuals
- Figma-style editor \u2013 fully customizable layouts, not locked templates
- App launch asset generator \u2013 quickly produce everything needed for a release
The goal is to become a complete launch toolkit for app developers.
Instead of spending hours in design tools, you upload screenshots and iterate quickly until you have polished assets ready for:
- App Store listing\n- Product Hunt launch\n- social media marketing\n- promo graphics
This tool is mainly built for indie developers, small teams, and app agencies who want to ship fast without hiring a designer.
I\u2019d love feedback from the HN community.
Website:\nhttps://applaunchflow.com
Curious about what assets you usually create when launching an app
Thanks!","title":"Show HN: AppLaunchFlow: AI tool that turns app screenshots into marketing assets","updated_at":"2026-03-05T23:42:03Z","url":"https://www.applaunchflow.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mazahermuraj"},"story_text":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","ui","design","tool"],"value":"Hey everyone
A bit of background:
A colleague of mine (UI/UX) said that she uses ChatGPT to help her generate a colour palette of tints and shades based on a base hex colour.
She then copied the suggestions given by ChatGPT into Figma and created a colour palette that suited the design she was working on.
But she had a few issues:
ChatGPT was inconsistent or did not produce what she wanted\nChatGPT didn't always show her the colour of the hex, so she used it in combination with a tool like Cooler\nChatGPT took away from her the eye for design when looking through the colours. She wanted to make the choice, not let AI do it for her.\nSo I built a small tool that tries to simplify that process:
\u2022 Input a base colour (or generate one randomly)
\u2022 Auto-generate tints (5% steps) and shades (10% steps). But you can adjust this
\u2022 Two modes: generate a palette or create a custom one
\u2022 Copy the output and paste it into Figma or whatever design tool you want to use
There\u2019s absolutely no promotion or paid plans at the moment. Just a free tool you can use without signing up. But if you want to save a palette to your library, you can sign up for free too.
I'd love honest feedback on:
- Usefulness of the workflow
- UX/usability thoughts
- Any missing features you\u2019d want
Here's the link for context (no signup): https://colorpal-sage.vercel.app/
Thanks in advance\u2014really appreciate your time & feedback!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["generator"],"value":"Show HN: Built a simple colour palette generator \u2013 would love feedback"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://colorpal-sage.vercel.app/"}},"_tags":["story","author_mazahermuraj","story_44428730","show_hn"],"author":"mazahermuraj","created_at":"2025-06-30T22:45:55Z","created_at_i":1751323555,"num_comments":0,"objectID":"44428730","points":1,"story_id":44428730,"story_text":"Hey everyone
A bit of background:
A colleague of mine (UI/UX) said that she uses ChatGPT to help her generate a colour palette of tints and shades based on a base hex colour.
She then copied the suggestions given by ChatGPT into Figma and created a colour palette that suited the design she was working on.
But she had a few issues:
ChatGPT was inconsistent or did not produce what she wanted\nChatGPT didn't always show her the colour of the hex, so she used it in combination with a tool like Cooler\nChatGPT took away from her the eye for design when looking through the colours. She wanted to make the choice, not let AI do it for her.\nSo I built a small tool that tries to simplify that process:
\u2022 Input a base colour (or generate one randomly)
\u2022 Auto-generate tints (5% steps) and shades (10% steps). But you can adjust this
\u2022 Two modes: generate a palette or create a custom one
\u2022 Copy the output and paste it into Figma or whatever design tool you want to use
There\u2019s absolutely no promotion or paid plans at the moment. Just a free tool you can use without signing up. But if you want to save a palette to your library, you can sign up for free too.
I'd love honest feedback on:
- Usefulness of the workflow
- UX/usability thoughts
- Any missing features you\u2019d want
Here's the link for context (no signup): https://colorpal-sage.vercel.app/
Thanks in advance\u2014really appreciate your time & feedback!","title":"Show HN: Built a simple colour palette generator \u2013 would love feedback","updated_at":"2025-06-30T22:50:01Z","url":"https://colorpal-sage.vercel.app/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"abelsm"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hey HN! It\u2019s Abel and Alex here to share what we\u2019ve been working on for just over a year: FlutterFlow (https://flutterflow.io). It\u2019s like WebFlow, but for Flutter.
Flutter is an open source framework for building cross platform applications. FlutterFlow combines a UI builder with pre-built templates and Firebase/API integrations, generates clean Flutter code, and allows you to deploy to app stores directly from your browser. This enables extremely fast iteration, from product idea concepts and designs to working Flutter apps. As an example of what\u2019s possible, we built an internal app for playing trivia games by using the jservice.io API and Firebase all in under 2 hours:
Time lapse of building the app: https://youtu.be/Fm4jjpuKM1E
Link to live version of the app: https://app.flutterflow.io/run/ByqwG33rw80qyN39ooNq
Exported source code: https://github.com/FlutterFlow/ff-trivia/tree/flutterflow
Alex and I, along with a friend of ours from Google, quit our jobs in 2019 to work on a cross-platform mobile app that ultimately failed. It was a learning opportunity, and it also led us to feel the pain of the slow iteration process every time we wanted to roll out a new experience. We were able to experiment with various landing pages within hours, but building new screens and app experiences took weeks or even months. For over a year now, we\u2019ve been tirelessly working on fixing this problem.
I first fell in love with coding by pure luck as a kid in Ethiopia. My father, who at the time owned an internet cafe, decided to start taking night classes in CS in the late 90s. Ultimately he didn\u2019t use his degree professionally, but I ended up with learning materials and a compiler, Turbo C++ 3.0. As I grew older, eventually ending up as an engineer at Google, I started to appreciate that as engineers we were often tasked with solving problems even when the solution didn\u2019t necessarily involve writing code. Alex comes from a physics background, doing his undergrad at Stanford, and transitioned to study CS and AI there as well. In 2016 he joined the team I was on, a small ML group within Google Maps. He\u2019ll often admit he had underestimated the amount of skill involved in building beautiful, fast and functional apps. And he certainly didn\u2019t expect to love building with Flutter as much as he does, having been entrenched in ML for most of his career. Yet here we are.
There has recently been a healthy amount of skepticism towards no-code tools, mainly due to concerns of extensibility and scalability. This is definitely the case for some apps - a good example is a tool such as FlutterFlow itself. It would be very difficult to build all of FlutterFlow recursively. We do use it internally for many of our pages, but using a visual builder to implement our code generator seems far fetched. This doesn\u2019t imply however that there isn\u2019t a middle ground that enables fast iteration in a visual builder, coupled with the ability to write code that seamlessly integrates with the overall experience. We\u2019re not quite there yet, but we believe this is the right direction.
Finally, we believe Flutter is going to be the catalyst that drives this movement. It\u2019s composability, the fact that it\u2019s super cross-platform (Android/iOS/Web/Desktop/Embedded), and the vibrant and passionate community it fosters give it a unique advantage. Whether we do it or someone else, the application builder of the future will be built on Flutter. Huge thanks to our users, the Flutter team and the Flutter community.
We\u2019d love for you to give it a try and share your thoughts. What do you think the future of application development is going to be?
p.s. we were on HN when we announced our launch back in May: https://news.ycombinator.com/item?id=27238381\nWe\u2019ve made a lot of progress since then, enabling app store deployment, payments, ability to add custom code and much more: https://community.flutterflow.io/c/whats-new-in-flutterflow"},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: FlutterFlow (YC W21) \u2013 Build Apps Visually"}},"_tags":["story","author_abelsm","story_29628716","launch_hn"],"author":"abelsm","children":[29628972,29629070,29629291,29629566,29629604,29629767,29629865,29629992,29630050,29630145,29630510,29630977,29631383,29631636,29632327,29632784,29633030,29634823,29635345,29635810,29639783],"created_at":"2021-12-20T19:07:58Z","created_at_i":1640027278,"num_comments":62,"objectID":"29628716","points":190,"story_id":29628716,"story_text":"Hey HN! It\u2019s Abel and Alex here to share what we\u2019ve been working on for just over a year: FlutterFlow (https://flutterflow.io). It\u2019s like WebFlow, but for Flutter.
Flutter is an open source framework for building cross platform applications. FlutterFlow combines a UI builder with pre-built templates and Firebase/API integrations, generates clean Flutter code, and allows you to deploy to app stores directly from your browser. This enables extremely fast iteration, from product idea concepts and designs to working Flutter apps. As an example of what\u2019s possible, we built an internal app for playing trivia games by using the jservice.io API and Firebase all in under 2 hours:
Time lapse of building the app: https://youtu.be/Fm4jjpuKM1E
Link to live version of the app: https://app.flutterflow.io/run/ByqwG33rw80qyN39ooNq
Exported source code: https://github.com/FlutterFlow/ff-trivia/tree/flutterflow
Alex and I, along with a friend of ours from Google, quit our jobs in 2019 to work on a cross-platform mobile app that ultimately failed. It was a learning opportunity, and it also led us to feel the pain of the slow iteration process every time we wanted to roll out a new experience. We were able to experiment with various landing pages within hours, but building new screens and app experiences took weeks or even months. For over a year now, we\u2019ve been tirelessly working on fixing this problem.
I first fell in love with coding by pure luck as a kid in Ethiopia. My father, who at the time owned an internet cafe, decided to start taking night classes in CS in the late 90s. Ultimately he didn\u2019t use his degree professionally, but I ended up with learning materials and a compiler, Turbo C++ 3.0. As I grew older, eventually ending up as an engineer at Google, I started to appreciate that as engineers we were often tasked with solving problems even when the solution didn\u2019t necessarily involve writing code. Alex comes from a physics background, doing his undergrad at Stanford, and transitioned to study CS and AI there as well. In 2016 he joined the team I was on, a small ML group within Google Maps. He\u2019ll often admit he had underestimated the amount of skill involved in building beautiful, fast and functional apps. And he certainly didn\u2019t expect to love building with Flutter as much as he does, having been entrenched in ML for most of his career. Yet here we are.
There has recently been a healthy amount of skepticism towards no-code tools, mainly due to concerns of extensibility and scalability. This is definitely the case for some apps - a good example is a tool such as FlutterFlow itself. It would be very difficult to build all of FlutterFlow recursively. We do use it internally for many of our pages, but using a visual builder to implement our code generator seems far fetched. This doesn\u2019t imply however that there isn\u2019t a middle ground that enables fast iteration in a visual builder, coupled with the ability to write code that seamlessly integrates with the overall experience. We\u2019re not quite there yet, but we believe this is the right direction.
Finally, we believe Flutter is going to be the catalyst that drives this movement. It\u2019s composability, the fact that it\u2019s super cross-platform (Android/iOS/Web/Desktop/Embedded), and the vibrant and passionate community it fosters give it a unique advantage. Whether we do it or someone else, the application builder of the future will be built on Flutter. Huge thanks to our users, the Flutter team and the Flutter community.
We\u2019d love for you to give it a try and share your thoughts. What do you think the future of application development is going to be?
p.s. we were on HN when we announced our launch back in May: https://news.ycombinator.com/item?id=27238381\nWe\u2019ve made a lot of progress since then, enabling app store deployment, payments, ability to add custom code and much more: https://community.flutterflow.io/c/whats-new-in-flutterflow","title":"Launch HN: FlutterFlow (YC W21) \u2013 Build Apps Visually","updated_at":"2024-09-20T10:00:53Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sarreph"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hey HN! We\u2019re Rory, Chris, and Felix from Kenobi (https://kenobi.ai). Kenobi lets you add AI-based content personalization to any website. As a site owner, you install our personalization widget with a script tag, just like you would for e.g. a chatbot integration. As a visitor, you interact with the widget (right now by providing a company name) and Kenobi changes the site content to suit you.
We\u2019ve built a demo that anyone can try here: https://kenobi.ai/start
We believe that large parts of the web are about to go from being static to dynamic because of how adept LLMs are at transforming rendered HTML. And right now we\u2019re focussing on B2B landing page content (as opposed to application UIs) because there is a lot of commercial opportunity for increasing top-of-funnel inbound conversions.
Our journey to Kenobi today is a long and snaking one. You may notice from the post title that we did YC\u2019s Winter 2022 batch (I know that 4 years is practically ancient in YC-dog-years). Kenobi is a hard pivot from our original idea that we got accepted into YC with \u2014 a company called Verdn which did trackable environmental donations via an API. Since the summer, we\u2019ve been hacking on different ideas\u2026 We started with personalized UI screenshots for outbound campaigns, but then people told us they wanted transformations to their actual site[0] \u2014 so we built an agentic workflow to research a visitor-company and \u201cpre-render\u201d changes to a landing site for them. Ultimately, there was too much friction in getting people to incorporate personalized URLs into their cold outbound campaigns[1]. Besides, people kept asking for us to do this for their inbound traffic, and so our current iteration was born.
Right now with Kenobi you pick a page that you\u2019d like to make customizable, and choose [text] elements that you\u2019d like to make dynamic. You can define custom prompting instructions for these elements, and when someone visits your page, our agentic workflow researches their company, and presents the updated content as quickly as possible, usually within a few seconds.[2] You also get a ping in Slack every time this happens so you know who is using your site.
We\u2019ve been experimenting with features such as generating custom imagery that actually looks good and native to the page design, and pulling in company data sources so that e.g. the right case study can be presented based on a visitor\u2019s industry and ICP profile. Our most requested feature is deanonymizing traffic so that Kenobi\u2019s personalization can happen automatically as visitors land on your page \u2014 this is coming very soon, as right now you have to specify where you\u2019re coming from.
It\u2019s surprised us just how much business value we\u2019ve gotten from knowing who (most probably) is on the page and asking for a personalized experience. We\u2019ve seen response rates 3x of what we would normally from following people up from companies we know visited our site.
There are many players in this space already, and everyone seems to have their own angle. We are keen to hear thoughts on what people think the future of the personalized internet looks like!
Cheers from London!
P.S. - there's also a video that Chris recorded showing the end-to-end Kenobi experience right now https://www.loom.com/share/bc0a82a2f2fd40f695315bae80e8f5d8
[0] - Many of them had tried AI \u201cmicrosite\u201d generators but found the maintenance of managing a separate website(s) just for closing deals to be burdensome and inefficient.
[1] - Despite having a CSV export and Clay integration option for our pre-generated website changes, getting people to weave the URLs into their email sequences (everyone uses different tools) seemed almost insurmountable without building what would ostensibly be our own sequencing software.
[2] - We use light foundation models with grounded search for the research step, and translate these into markup changes via another light LLM pass and our own DSL which is optimized for speed."},"title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: Kenobi (YC W22) \u2013 Personalize your website for every visitor"}},"_tags":["story","author_sarreph","story_46301865","launch_hn"],"author":"sarreph","children":[46302403,46302428,46302442,46302474,46302525,46302784,46303054,46303205,46303247,46303251,46303700,46304048,46304073,46305299,46305468,46306597,46307087,46307122,46307305,46307974,46309120,46309376,46331714],"created_at":"2025-12-17T16:44:13Z","created_at_i":1765989853,"num_comments":56,"objectID":"46301865","points":46,"story_id":46301865,"story_text":"Hey HN! We\u2019re Rory, Chris, and Felix from Kenobi (https://kenobi.ai). Kenobi lets you add AI-based content personalization to any website. As a site owner, you install our personalization widget with a script tag, just like you would for e.g. a chatbot integration. As a visitor, you interact with the widget (right now by providing a company name) and Kenobi changes the site content to suit you.
We\u2019ve built a demo that anyone can try here: https://kenobi.ai/start
We believe that large parts of the web are about to go from being static to dynamic because of how adept LLMs are at transforming rendered HTML. And right now we\u2019re focussing on B2B landing page content (as opposed to application UIs) because there is a lot of commercial opportunity for increasing top-of-funnel inbound conversions.
Our journey to Kenobi today is a long and snaking one. You may notice from the post title that we did YC\u2019s Winter 2022 batch (I know that 4 years is practically ancient in YC-dog-years). Kenobi is a hard pivot from our original idea that we got accepted into YC with \u2014 a company called Verdn which did trackable environmental donations via an API. Since the summer, we\u2019ve been hacking on different ideas\u2026 We started with personalized UI screenshots for outbound campaigns, but then people told us they wanted transformations to their actual site[0] \u2014 so we built an agentic workflow to research a visitor-company and \u201cpre-render\u201d changes to a landing site for them. Ultimately, there was too much friction in getting people to incorporate personalized URLs into their cold outbound campaigns[1]. Besides, people kept asking for us to do this for their inbound traffic, and so our current iteration was born.
Right now with Kenobi you pick a page that you\u2019d like to make customizable, and choose [text] elements that you\u2019d like to make dynamic. You can define custom prompting instructions for these elements, and when someone visits your page, our agentic workflow researches their company, and presents the updated content as quickly as possible, usually within a few seconds.[2] You also get a ping in Slack every time this happens so you know who is using your site.
We\u2019ve been experimenting with features such as generating custom imagery that actually looks good and native to the page design, and pulling in company data sources so that e.g. the right case study can be presented based on a visitor\u2019s industry and ICP profile. Our most requested feature is deanonymizing traffic so that Kenobi\u2019s personalization can happen automatically as visitors land on your page \u2014 this is coming very soon, as right now you have to specify where you\u2019re coming from.
It\u2019s surprised us just how much business value we\u2019ve gotten from knowing who (most probably) is on the page and asking for a personalized experience. We\u2019ve seen response rates 3x of what we would normally from following people up from companies we know visited our site.
There are many players in this space already, and everyone seems to have their own angle. We are keen to hear thoughts on what people think the future of the personalized internet looks like!
Cheers from London!
P.S. - there's also a video that Chris recorded showing the end-to-end Kenobi experience right now https://www.loom.com/share/bc0a82a2f2fd40f695315bae80e8f5d8
[0] - Many of them had tried AI \u201cmicrosite\u201d generators but found the maintenance of managing a separate website(s) just for closing deals to be burdensome and inefficient.
[1] - Despite having a CSV export and Clay integration option for our pre-generated website changes, getting people to weave the URLs into their email sequences (everyone uses different tools) seemed almost insurmountable without building what would ostensibly be our own sequencing software.
[2] - We use light foundation models with grounded search for the research step, and translate these into markup changes via another light LLM pass and our own DSL which is optimized for speed.","title":"Launch HN: Kenobi (YC W22) \u2013 Personalize your website for every visitor","updated_at":"2026-06-21T08:59:20Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dallen97"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"Hi HN \u2014 I built Seedream4, a browser-based AI image generator & editor. You can generate from text or upload an image and do precise edits (background removal, in/out-painting, style transfer, upscaling) and export PNG/JPG/WebP. There are example prompts you can run right away, and uploads (PNG/JPG/WEBP up to 10MB) support higher-res results including 4K.
Why\nI got tired of manual masking/retouch steps for marketing and product visuals. I wanted something that combines text-to-image with instruction-based edits so you can keep lighting/perspective intact and iterate fast.
What\u2019s different\n \u2022 Instruction-based edits with local brushes/masks, without wrecking the rest of the scene.\n \u2022 Multi-reference consistency to keep brand colors/character identity coherent across shots.\n \u2022 Export-ready workflow (PNG/JPG/WebP, transparent backgrounds; 1:1, 16:9, 9:16, etc.).\n \u2022 No-signup demo: click an example or upload and try.
How it works (high level)\nThe web app sends your prompt/images to third-party model APIs (e.g. modern image-creation models) via my backend. I provide the UI, masking tools, queues, and credit metering. Per the privacy policy: prompts/images are processed by model providers and not retained longer than needed to deliver results.
Pricing\nThere\u2019s a free tier to try core generation/editing. Paid plans add more credits, 4K export, priority processing; commercial usage rights are included in Pro.
Known limitations / looking for feedback\n \u2022 Text & tiny details still fail on some compositions.\n \u2022 Complex multi-object edits sometimes need an extra mask pass.\n \u2022 Throughput can vary under load.\nI\u2019d love feedback on UX, failure cases you hit, performance expectations, API/CLI needs, and whether the pricing/credit model feels fair.
Happy to answer anything about the design choices, queues, and the editor\u2019s internals. Thanks for checking it out."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Seedream4 \u2013 Multi-reference, in/out-painting, 4K export"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://seedream4-ai.com"}},"_tags":["story","author_dallen97","story_45258728","show_hn"],"author":"dallen97","created_at":"2025-09-16T06:29:50Z","created_at_i":1758004190,"num_comments":0,"objectID":"45258728","points":1,"story_id":45258728,"story_text":"Hi HN \u2014 I built Seedream4, a browser-based AI image generator & editor. You can generate from text or upload an image and do precise edits (background removal, in/out-painting, style transfer, upscaling) and export PNG/JPG/WebP. There are example prompts you can run right away, and uploads (PNG/JPG/WEBP up to 10MB) support higher-res results including 4K.
Why\nI got tired of manual masking/retouch steps for marketing and product visuals. I wanted something that combines text-to-image with instruction-based edits so you can keep lighting/perspective intact and iterate fast.
What\u2019s different\n \u2022 Instruction-based edits with local brushes/masks, without wrecking the rest of the scene.\n \u2022 Multi-reference consistency to keep brand colors/character identity coherent across shots.\n \u2022 Export-ready workflow (PNG/JPG/WebP, transparent backgrounds; 1:1, 16:9, 9:16, etc.).\n \u2022 No-signup demo: click an example or upload and try.
How it works (high level)\nThe web app sends your prompt/images to third-party model APIs (e.g. modern image-creation models) via my backend. I provide the UI, masking tools, queues, and credit metering. Per the privacy policy: prompts/images are processed by model providers and not retained longer than needed to deliver results.
Pricing\nThere\u2019s a free tier to try core generation/editing. Paid plans add more credits, 4K export, priority processing; commercial usage rights are included in Pro.
Known limitations / looking for feedback\n \u2022 Text & tiny details still fail on some compositions.\n \u2022 Complex multi-object edits sometimes need an extra mask pass.\n \u2022 Throughput can vary under load.\nI\u2019d love feedback on UX, failure cases you hit, performance expectations, API/CLI needs, and whether the pricing/credit model feels fair.
Happy to answer anything about the design choices, queues, and the editor\u2019s internals. Thanks for checking it out.","title":"Show HN: Seedream4 \u2013 Multi-reference, in/out-painting, 4K export","updated_at":"2026-03-05T22:42:06Z","url":"https://seedream4-ai.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thealexanderlee"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ai","ui","design","generator","tool"],"value":"I built an agent that helps you build Rollercoaster Tycoon-influenced theme parks. You can prompt something like \u201cBuild me a cool theme park\u201d and it\u2019ll build a cohesive theme park with multiple worlds and connected with paths and rides.
The weird part is that I built it using the same ideas we use to make AI-generated websites follow a company\u2019s design system.
I work as an engineer at Magic Patterns, focused on building our Design System Agent, aimed to use your existing brand, components, and conventions instead of producing something that looks like generic AI-generated UI.
By default, models tend to converge on similar-looking designs. This is part of that \u201cvibe coded slop\u201d feeling: the same typography treatments, shadows, cards, icons, animations, etc\u2026 My job is to figure out the right context and guardrails so that when you prompt \u201cBuild me a dashboard,\u201d the result actually looks like your existing product.
At some point I realized those same ideas could be applied to RCT.
While in web, you might have rules about which typography, colors, spacing, and components should be used together.
In a theme park, you need rules like: rollercoasters need complete tracks, rides need entrances connected to paths, paths need to connect different areas of the park, and a pirate-themed world should actually use pirate-themed scenery.
It was interesting seeing how closely the problem of building a coherent theme park resembled the problem of building a coherent product UI. Similarly with web design, simply giving the model the right components wasn't enough.
I ended up building an eval loop where Magic Patterns would generate a park, another agent would grade it against a rubric, and then the agent would update its rules and skill files before trying again.
The rubric checked things like whether rollercoasters formed valid tracks with at least one drop, whether rides were accessible by paths, whether each world used the appropriate themed scenery, and whether the park worked as a whole.
After a lot of iterations, you can now prompt something as simple as \u201cBuild me a cool theme park.\u201d and you should get back a fully functioning park with guests, rides, paths, rollercoasters, and themed worlds.
Best yet, you can also follow the guests, listen to their thoughts, and watch them ride the rides in RCT fashion.
Happy to answer any questions, but would love to see the parks you all create!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Build your own theme park"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.magicpatterns.com/theme-park"}},"_tags":["story","author_thealexanderlee","story_49452037","show_hn"],"author":"thealexanderlee","children":[49456446,49456510,49456568,49456678,49456809,49456860,49457028,49457840,49459216,49459672,49464689,49465159,49467429,49475852,49485820],"created_at":"2026-08-26T16:40:01Z","created_at_i":1787762401,"num_comments":25,"objectID":"49452037","points":55,"story_id":49452037,"story_text":"I built an agent that helps you build Rollercoaster Tycoon-influenced theme parks. You can prompt something like \u201cBuild me a cool theme park\u201d and it\u2019ll build a cohesive theme park with multiple worlds and connected with paths and rides.
The weird part is that I built it using the same ideas we use to make AI-generated websites follow a company\u2019s design system.
I work as an engineer at Magic Patterns, focused on building our Design System Agent, aimed to use your existing brand, components, and conventions instead of producing something that looks like generic AI-generated UI.
By default, models tend to converge on similar-looking designs. This is part of that \u201cvibe coded slop\u201d feeling: the same typography treatments, shadows, cards, icons, animations, etc\u2026 My job is to figure out the right context and guardrails so that when you prompt \u201cBuild me a dashboard,\u201d the result actually looks like your existing product.
At some point I realized those same ideas could be applied to RCT.
While in web, you might have rules about which typography, colors, spacing, and components should be used together.
In a theme park, you need rules like: rollercoasters need complete tracks, rides need entrances connected to paths, paths need to connect different areas of the park, and a pirate-themed world should actually use pirate-themed scenery.
It was interesting seeing how closely the problem of building a coherent theme park resembled the problem of building a coherent product UI. Similarly with web design, simply giving the model the right components wasn't enough.
I ended up building an eval loop where Magic Patterns would generate a park, another agent would grade it against a rubric, and then the agent would update its rules and skill files before trying again.
The rubric checked things like whether rollercoasters formed valid tracks with at least one drop, whether rides were accessible by paths, whether each world used the appropriate themed scenery, and whether the park worked as a whole.
After a lot of iterations, you can now prompt something as simple as \u201cBuild me a cool theme park.\u201d and you should get back a fully functioning park with guests, rides, paths, rollercoasters, and themed worlds.
Best yet, you can also follow the guests, listen to their thoughts, and watch them ride the rides in RCT fashion.
Happy to answer any questions, but would love to see the parks you all create!","title":"Show HN: Build your own theme park","updated_at":"2026-08-29T17:49:33Z","url":"https://www.magicpatterns.com/theme-park"}],"hitsPerPage":20,"nbHits":18,"nbPages":1,"page":0,"params":"query=AI+UI+design+generator+tool&advancedSyntax=true&analyticsTags=backend","processingTimeMS":23,"processingTimingsMS":{"_request":{"roundTrip":16},"afterFetch":{"format":{"highlighting":3,"total":3},"total":1},"fetch":{"query":8,"scanning":13,"total":22},"total":23},"query":"AI UI design generator tool","serverTimeMS":27}