PH热榜 | 2026-05-15

Heya! I'm Steven, founder of TinyHumans.

A few months ago I tried to set up an open-source AI agent for my dad. Three hours later and after wrestling with API keys, YAML and a terminal he had never opened in his life, we both gave up.

That's when I realised that every powerful AI agent today is built for the 0.01% who can spin up their own runtime. The other 99.99% are watching the agent revolution from the sidelines.

So we built OpenHuman.

OpenHuman is a super-intelligent AI agent that anyone can use. Two-minute setup. No config files. A simple GUI you'd hand to your parents and they'd actually figure out. Connect Gmail, Slack, Telegram, Notion, and GitHub in one click and it just works.

A few things I'm proud of:

* It runs locally. Encrypted vault. We never sell your data.

* It never forgets. Real memory across sessions, not session-only.

* It's open-source under GNU.

* It's free to start: no engineer, no GPU, no $6k setup bill.

Early signal has been wild: 8000+ GitHub stars, 5000+ users in the first 7 days, and 150% week-over-week growth.

Today, we're opening it up to all of you. Note that it is still in beta, so you're super early if you find bugs. Feel free to report them to me over at the Discord.

I'll be in the comments all day. You can break it, roast it, tell me what's missing, ask anything. We ship fixes live.

Would love your feedback!

Hi 👋

I'm Ankita, the product marketer at OpenHuman.

I joined this team because Steven and the rest of TinyHumans were the only people I'd met who genuinely wanted to build an AI agent for everyone, not just engineers. The people who don't write code, don't want to wire API keys, don't want to read YAML files. People like my parents, honestly.

Watching the last few months of shipping has been wild. The skills marketplace went from a handful of integrations to 118+. Memory went from session-only to actually remembering you across weeks. And Tiny the mascot has somehow become the most-discussed feature in our internal slack.

If you've ever wanted to use an AI agent but felt the setup wasn't worth the headache, today is your day. It gets way more useful the longer you use it because it actually remembers what you tell it.

Try it free at tinyhumans.ai/openhuman.

Hop into our Discord if you want to chat, ask for a feature, or report a bug. I'm reading everything today.

Thanks for being here 🙏

Ankita

Launching alongside great products like this makes today even more exciting. Really impressed by what you’ve built. Supporting fellow makers today would love to cheer each other on 🚀

One of the weirdest things about using OpenHuman now is how fast normal AI tools start feeling “dumb”.

The moment your AI starts remembering context across chats, understanding your workflow, pulling information across apps, and quietly helping in the background… it’s really hard to go back.

Feels less like using ChatGPT and more like having a second operating system running alongside you.

Also watching non-technical people set this up in like 2 minutes after spending months hearing “AI agents are too complicated” has been pretty wild !

can it use my skills and maybe commands? Tool calls? Is it good for coding?

Amazing product

How did this not exist before?! Love the idea, want to give it a try even though I don't necessarily need an agent in my life.

Something I've seen that feels related: the concept of moving our personal data from the custody of vendors (social, doctors, marketers, employers etc) back to us - the owners. Eg a graph of all your data with granular sharing and privacy policies that you directly control. There's an obvious privacy aspect to this, but it's also natural to imagine how agents can thrive with (controlled) access to it, as an extension of the permanent memory you've built.

Finally an AI that doesnt forget everything after every chat. Sounds actually useful for normal people not just tech guys.

The privacy angle is a big reason to go local-first, but the persistent memory is what actually makes it usable day to day. I I am tired of re-explaining my tech stack and project goals every single time I open a new session.

Since you mentioned that it is in beta and remembers everything, I want to know how you handle context window limits or database bloat over time. Does it start getting sluggish once it knows too much about my work history, or is there some kind of automated cleanup?

The gap between powerful agent and usable by normal people is still massive and most projects only solve the first half.

Congrats on launch!! I like the idea.

How it's handle hallucination thing?

What is the subscription price to use this? Is it open source or need to pay anything?

@itsnotgaf Congratulations. And happy product launch.

Hey Product Hunt 👋

Sergey here, co-founder at HasData.

HasData is the managed web data service for AI agents and pipelines. We handle the messy infrastructure like proxies and anti-bot. We also maintain dozens of ready APIs for Google, Maps, Zillow and e-commerce.

And one thing that sets us apart from every other API in this space: we only bill on success. Failed requests cost nothing. You pay for data that actually arrives, not for retries on a broken proxy.

Today we are launching our AI Agent, an MCP server, a CLI and Agent Skills for Claude Code and OpenClaw. You can now connect HasData straight to your AI stack.

The catalog is pretty powerful if you know exactly what you need. But it gets frustrating if you do not. Picking the right tool, learning the parameters and parsing the output takes time away from actual building.

The new AI Agent fixes that. Describe what you need. The Agent picks the matching API, runs the job, and returns a dataset. Enrich any row from the same chat with contacts, firmographics, or whatever's missing.

The MCP server, CLI, and Agent Skills give the same flow to anyone working from Claude, ChatGPT, the terminal, or Claude Code.

Two things worth knowing:

• We're giving 10,000 free credits during launch week.

• If you're a HasData user already, everything works on your existing account, catalog, and workspace. No separate plan, no migration.

It's live at app.hasdata.com/chat. I'm here all day, so is the team. Drop a comment, break it, tell us what's missing.

Hey Product Hunters!

Nick's here, CPO and Growth Advisor for HasData. I believe this launch is 100% worth your attention, let me tell you why.

Web scraping always used to be a complicated tool for developers to use: if you do it on your own, you have to build and maintain the whole scraping infrastructure by yourself, but that's even not the most complicated part. Parsing your data and extracting relevant values used to be another kind of a nightmare - before LLM and ChatGPT, of course.

Today, thanks to HasData and AI capabilities integrated into this products, I feel like searching and extracting the data you need from web is like building a website with Lovable instead of hiring a freelancer to built it for you with WordPress. You simply pick the required scraper, or even further, prompt your request to HasData AI Agent - then get the relevant data set ready for your CRM enrichment or cold outreach in minutes. Or, what I personally love the most, simply use something like Claude or OpenClaw connector and use the whole power of HasData scraping capabilities in your favourite AI tool - simple as that!

So go ahead and try HasData AI scraping tools today for free and let us know what you think - we are happy to consider your feedback and feature requests ❤️

Your AI is only as smart as the data behind it. And right now most teams I talk to are either flying blind, or they've got somebody on the team who basically does proxies and CAPTCHAs full-time and rewrites parsers every time a site changes its DOM.

I'm Roman, CEO & CTO at HasData. We've spent years on that infrastructure so you don't have to.

Behind one key:

• 40+ ready-made APIs across search (Google SERP), travel (Booking, Airbnb, Google Flights, Google Maps), real estate (Zillow, Redfin), shopping (Amazon, Google Shopping), jobs, and social

• A Web Scraping API for everything else: any URL in, parsed JSON out

• An AI Agent that drives the catalog from natural language if you just want a one-off pull without writing code
  

Use it however you build: HTTP API, CLI, Claude Skill, OpenClaw, or MCP server.

Free to start at hasdata.com. You get 10,000 credits when you sign up. If we save you a week of proxy work, throw us an upvote.

I'll be in the comments today. Happy to get into the APIs, the agent, pricing, what we're working on next, whatever.

This is a really good service. I used it to collect data for Callersmart website. Data quality is perfect, fast response and good price. They have different platforms no-code scrapers:

Recommend!

Well done team! Public web data on-demand, payable on-delivery. That's the hook/promise of HasData. :)

I'm a power user of their YouTube Transcript API. And have been a fan of the team's work (and their work effort) since before Roman came onto my podcast.

How can you not be bullish on HasData when they're shipping so much goodness on a monthly basis?

Hey PH! I'm Valentina, technical writer at HasData, and yes, I actually use their tools daily.

The team already covered features, so I just want to add why I personally like this. I tried and used every scraper, every API, every no-code tool, every feature they built. What I really like is how fast you go from start to the result. Even first time using a tool, takes almost no time to understand how it works.

And the project keeps moving forward. They keep building new tools, new scrapers, covering more sources. The latest addition is an AI Agent, so you don't even pick APIs or scrapers yourself, just get the ready data.

If you work with data in any capacity, give it a try. And I also happy to answer any questions in the comments today and further.

Overall dope features but I have one question as well. I have seen while extracting webpage data from Firecrawl, that the extracted text contains all the advertisements links as well, this sorts of corrupt the data and need some cleaning before feeding to LLM. So I want to understand does HasData remove those links in some way ?

Hey there! I just wanted to say qudos for your service. I used it for some enrichment work and honestly you nailed it!!! It took me only about 20 minutes to connect your API to my setup (agent skill for Claude vibe me through), and I was impressed with how smooth the process was.

I had a dataset of 607 rows, and the email enrichment came back with a 74% success rate, which I think is fantastic. After checking for bounces, it settled at 69%. Not bad at all for just 20 minutes of work!

I've added you to my favorites for scraping services because you definitely deserve a spot there.

Good luck with the launch, Bulba Scrapping service )) Fingers crossed for a massive scale-up!

Good luck

I’ve been using HasData for a while now, mainly for Google Maps scraping, and it’s been really easy to use.

What I like is that it removes a lot of the usual friction around collecting structured data. Instead of spending ages working out the best way to scrape, clean, and organise the data, I’ve been able to get useful results pretty quickly.

I’m especially excited to try the new AI agent. Getting data from less common websites is often the painful bit. It can be time-consuming, inconsistent, and a lot of trial and error, especially when every site has a slightly different structure.

If the agent can make that process smoother and more reliable, it’ll be a really useful addition.

Congrats on the launch. Looking forward to seeing how the product develops.

Building a small side project that needs lead data and was about to write a custom scraper. Now I think I'll just use this - will give it a try at least. Congrats on launching today!

Congratulations team!!! All the best in this new launch!

Congrats on the launch and good luck today! Love your landing page design btw, especially the widget that gets you started with a link to scrape or an agent prompt - good part of onboarding process nowadays, would love to know how it works for you

Hey guys, congrats on launch! I've recently seen Firecrawl doing the same (I guess?) AI craping agent - what's the difference between them and you if any?

Scraping for AI agents is a huge bottleneck right now. How does HasData handle complex anti-bot measures like Cloudflare or dynamic JS-heavy sites without constant manual tweaks?

Honestly, Roman mentioning that the AI Agent cross-checks ambiguous pages against a secondary source is totally flying under the radar — it's the most underrated part of this whole launch.

Quick practical question though: how does the Agent actually decide when to escalate to a cross-check? Is it relying on confidence scores (like the extraction model saying "I'm only 60% sure"), strict schema validation ("required field missing"), or some other heuristic?

Asking as an indie dev building AI tools, because getting that "when to escalate" logic right is usually exactly what makes or breaks agentic systems in production.

Congratulations on the launch! Does the agent run on a schedule for recurring tasks, or only on demand?

Congrats on Launch! I hope we could get a LinkedIn extension even though it blocks scraping...

Seems like you guys did a really great job connecting your scraping tools to AI via mcps, cli, etc. Might give it a try today!

Hey guys, congrats on launch!  I run an AI agency and we have quite a lot of projects where scraping is a big part of the stack - market research, price monitoring, lead enrichment, competitor tracking, AI agents with web data. So this looks super relevant for us. Scraping is always one of those parts that sounds easy in the begining, but then you spend a lot of time on proxies, blocks, JS rendering and weird site structures. Just one question how well does it work with websites that have heavy anti-bot protection or dynamic pages with a lot of JS rendering?

Scaling scrapers for AI agents is usually a massive headache because of proxy rotation and constant site updates. Paying only for successful requests feels like a fair model. Do you provide a way to train the AI extraction for specific non standard layouts, or is it purely automated? Also, what system is being used behind it?

How do you deal with reliability there, do you surface a confidence score or require a quick validation step?

Congrats team! Happy to find your product, Google maps no-code scrapers is sth I really need, will try today.

Congrats on launch! Love how the agent picks the matching scraper itself, that part always slowed me down on other tools.

most scraping services charge you whether the data comes back clean or not and you end up paying for broken responses. how does the openclaw skill integration work in practice though, does the agent automatically pick which scraper to use or do you still have to configure each one manually

Congrats on launch, guys! The scraper catalog covers the obvious targets, but what happens when someone needs a niche source not in the library? Does the agent fall back to AI extraction on raw html, or wait for the team to build a dedicated scraper?

Hey, such a nice selection of web scraping apis! But why should I choose you if I'm already on Apify or a similar solution?

Congrats, team! As your current user, I'm happy to see you finally launched on Product Hunt! Congrats and all the best. P.S.: those guys do a great job, take a look if you need to scrape some web data 😻

Congrats on the launch! Used enough unstable scrapers to appreciate when someone takes this problem seriously.

Btw, is the tiktok scraper coming anytime soon?

@rajiv_ayyangar, thank you so much for hunting us!

Hey PH Community 👋

We're Yagiz, a Senior Technical Product Manager at Amazon and an independent researcher and Yigit, co-founder and GP of Vela Partners. Today, we're launching PHBench in collaboration with the University of Oxford (Ben Griffin and Rick Chen) and Vela Partners, the leading quant VC.

And yes, the irony of launching a Product Hunt benchmark on Product Hunt is completely intentional 🙂

Here's the origin story. We kept asking a question nobody had answered: Can you predict which Product Hunt launches will raise Series A funding, based solely on what you see on launch day (votes, rank, team size, category, timing)?

So we built PHBench. We collected 67,292 featured PH launches going back to 2019, matched them to Crunchbase funding records, and identified 528 verified Series A raises within 18 months. Seven years of data. Every featured launch.

Three findings I think this community will find interesting:

→ The signals work. Our model is 4.7x better than random. Statistically significant.

→ The strongest predictor isn't votes alone. It's team size × community engagement together. A large coordinated team achieving high traction is more predictive than either signal alone.

→ B2B categories convert at 3x the baseline rate. API, Payments, Fintech. If you launch a developer tool on a Tuesday with a big team and high engagement, that's a strong signal.

We also tested three frontier Gemini models on the same task. The most capable model performed the worst. Better reasoning doesn't help with pure numbers.

The dataset is available on HuggingFace. The leaderboard is live. The code is public. Can you beat our baseline?

The paper is on arXiv and has been submitted to the NeurIPS 2026 Evaluations & Datasets Track.

Would love your feedback — especially from anyone who's launched on PH and gone on to raise Series A. You're in our dataset :)

Been quietly working on this with Yagiz, Yigit and Rick for a while.

While I mostly focus on using founder profiles to predict raises, PHBench tries the same prediction but from the product side. A similar question but from the other side.

Have a go at the leaderboard if you fancy; the data's on HuggingFace.

Are you using only launch day signals, or do you include post launch traction like follows and comments over the first week?

Interesting, most people assume raw upvotes are the proxy for quality. So the finding about team size × community engagement being a stronger signal than votes alone is genuinely counterintuitive but very curious. Have you looked at whether solo founders who hit high engagement are penalized by this model? Do they show up as a distinct cluster? Would love to see how the signal degrades for truly first-time founders vs. repeat ones. Incredible dataset, congrats on getting years of data cleaned!

So excited to see this live! This has been a labor of love, collecting data, running +100 experiments, and testing LLMs against good old gradient boosting.

The leaderboard is open. If you can beat us, you're the new champion. Who's in?

Really excited to bring PHBench to you guys! By extending the short-term productivity signals on Product Hunt to predict long-term funding materialization, we help to identify outlier products that are truly valuable in the VC environment. We think it will be greatly beneficial to the Product Hunt community.

Come to beat our baseline and get to the top of the leaderboard!

Really cool idea, Good luck! @ihlamury @yigit

Would be interesting to see a breakdown of false positives: high PH engagement but no Series A. That’s often where the real insight is.

congrats on the launch! This seems very interesting and exciting.

@yigit, congrats on the launch. I will be wondering the result of https://www.producthunt.com/products/vela-terminal launch after the launch ends!

My best product hunt launches were driven by public curiosity and correlated with it. I was using those metric for A/B/C testing and it was way more making sense when you test yourself as founder or an idea of early prediction when same amount of effort is spent.

I will definitely try to benchmark using my past launches and give feedback!

No analysis on hunter impact? 🥴🥴🥴

After today's launch, we all expect to see PHBench's chances of hitting Series A based on its own model. Good luck!

X Corner Random Encounter: A Quick Take on PHBench 🧠

Huge congrats to Yagiz, Yiğit, and the Vela Partners team on this launch!

I just bumped into Yagiz and Yiğit at the X corner an hour ago. I threw a quick challenge at their core assumption, and they gave me some incredible, patient replies. Here is our quick chat. (With help from Gemini to organize my thoughts into this post!)

My Question on Twitter: Many top launches already raised seed money. They use PH as an amplifier, not a starting point. It feels like the high engagement vs. Series A is more of a correlation rather than causality—the "strong signal" might just be a lagging indicator of their pre-existing capital/resources.

Yagiz's Backdoor Insights: Yagiz dropped a total golden nugget. He told me: "88.3% of the 528 launches that raised Series A in our dataset had prior seed funding."

(And oh my god, that means I’m officially part of the 11.7% "naked" bootstrapped builders trying to survive with zero funding, haha! 😅)

Yagiz honestly framed that PHBench is a predictive benchmark, not a causal study. They don't claim causality. But he noted that while a funded team can buy upvotes, consistently landing Rank #1 is harder to manufacture.

Yiğit's Side Note: Yiğit also shared another fascinating data point with me: "Consistent social media posters increase their likelihood of success. However, not posting doesn't decrease your chance. It’s just neutral."

I guess those successful but quiet builders must have massive "invisible" networks that online data simply cannot capture.

My Takeaway: This conversation completely shifted my view. PHBench is essentially a "Funnel Accelerator" for VCs.

It does not discover hidden gems from absolute scratch. Instead, it predicts which already-backed teams have the top-tier GTM execution to dominate the market on launch day. If you already have seed money, PH is the ultimate stress test for your team.

My Personal Note:

To wrap up, I want to say how much I truly appreciate their hard work. Data cleaning and dataset building are brutal, sweating jobs. Really appreciate to see them doing all this heavy lifting and unselfishly open-sourcing the whole thing to the community. Thanks a lot.

Now that this model is public, founders will start optimizing for the signals it tracks - bigger teams on paper, coordinated engagement, category shopping. Does publishing the feature set risk corrupting the signal over time?

Also curious where EdTech lands in the category rankings. Congrats on the launch!

Hey Product Hunt, Claire here, founder of Lensmor.

We built Lensmor because trade show prep is still weirdly manual.

A lot of B2B teams spend real money on booths, travel, sponsorships, and sales time, but still walk into the event with a messy spreadsheet, a generic exhibitor list, and no clear plan for who they should meet.

Lensmor starts with exhibitor data.

You can search 160,000+ global events, find exhibiting companies, identify decision-makers, reveal verified emails, export lists, and plan outreach before the show starts.

The goal is simple: help teams walk into trade shows with the right target list and more meetings already booked.

We’re also building an AI Agent that turns the workflow into a faster pre-show sprint:
find target exhibitors → generate intelligence briefs → draft outreach → help plan meeting booking and follow-up.

We’re launching today as part of Vercel Day, and Lensmor is built on Vercel.

I’d love feedback from anyone who has used trade shows for pipeline:

Where does trade show prep usually break for your team?
- finding the right exhibitors
- figuring out who to contact
- writing pre-show outreach
- booking meetings before the event
- proving ROI after the show

Proving ROI after a show is always the hardest part because the follow-ups are usually generic and late. If Lensmor helps us book meetings before the event starts, that’s already a win. Is there a way to track which booked meetings actually came from Lensmor lists?

Good one for sales people. How does it help with bulk outreach, like someone want to set outreach sequence automatically?

Congrats, Claire! Proving ROI after the show is notoriously hard. Does Lensmor currently track which exported contacts actually turned into meetings or pipeline, or is that still a manual CRM step?

Congrats! On booking meetings before the event does Lensmor integrate with scheduling tools like Calendly or Chili Piper to suggest time slots based on the event agenda or exhibitor availability?

Congrats on the launch. Curious about the AI agent's autonomy - does it draft outreach sequences for human approval before sending, or do you let it send autonomously to verified emails? Asking as someone building agent harnesses on the other side of this category and always interested in where teams draw the trust line.

Trade show ROI is hard to prove when the team starts with an unstructured list. Lensmor’s approach should make the before-and-after much easier to track.

Congratulations 🎊

Generic lead databases rarely understand event intent. Exhibitor data is a better starting point because it already tells you which companies are investing in that market. Love the idea of Lensmor!

The best part of this is the timing. Outreach before the show is where teams can still change the outcome. After the event, most leads are already cold. Congrats on the launch!

Congrats on the launch, Claire! Love the angle of Lensmor.

From my experience, trade shows are expensive, yet most teams still prep with messy spreadsheets and generic lead lists. Turning exhibitor data into a prioritized meeting list before the event is a practical, high-ROI use case.

Converting exhibitor data into meetings is a high-value niche. Does the tool use AI to personalize the outreach based on the specific company's recent news or launches?

Congratulations, team, the trade show ROI piece you mentioned, how does Lensmor help prove that?
Is there built-in tracking for which booths/meetings actually converted, or would that be on us to track separately in our CRM?

Congrats on the launch, and as one who loves going to expos for leads, I really like the idea! I was wondering whether the database could include side events related to each expo. Based on my experience, most real networking happens outside the exhibition space at those events, so I believe it would be helpful to include those information for reference as well.

Congrats on the launch! Nice surprise to see you here – I was curious about your product recently and wanted to wish you good luck. I'm building in the events space too (B2C, a tool that recommends which events to go to), so it's cool to see what other people are doing in this area. Wishing you a great launch day! Funnily enough, I launched my side product on PH today as well 😄

Congrats, Claire! How do you handle duplicate or parent-subsidiary company listings across multiple shows? One team might attend 5+ events per quarter and needs deduplication.

Curious how the intent signal scoring works in practice; is it based on the exhibitor's booth activity at past shows, or are you pulling from outside signals like hiring and funding too? asking because the ICP filter looks solid but intent is usually where these tools get fuzzy

Congrats on the launch! Does the AI Agent generate intelligence briefs that include recent news, job changes, or funding rounds? Or is it more static company/exhibitor data?

Congrats! If a team's biggest bottleneck is 'writing pre-show outreach,' does Lensmor offer A/B testing or open-rate tracking for different message versions? Or is that outside scope for now?

Congrats on the launch, folks 🤩 Good luck!

Congrats, Claire! What's the typical workflow gap after a team exports a list from Lensmor do they usually leave Lensmor to send emails, or are you building native sending + tracking?

The product feels useful because it connects event data to the sales workflow. That bridge is usually handled manually by someone on the team.

I like that Lensmor starts with the event context first. For RevOps teams, a cleaner target list before the show can save a lot of messy follow-up work later.

Congrats on the launch! This is such a practical use of data. It’s not just more dashboards—it’s a clear path from event research straight into real sales conversations.

Hey Product Hunt!

Tech lead at Lokuma here.

The core shift in v2.0: we replaced the fixed "generate-once" pipeline with a real agent loop — the model plans, writes code, inspects output, and self-corrects until it's satisfied. That's what makes the quality jump feel so dramatic.

A few things under the hood we're proud of:

1.Plan-first — the agent drafts a build plan before writing any code. Users approve it before execution starts.

2.Targeted patching — edits use find-and-replace on the specific changed section, not full file rewrites. Faster and more predictable.

3.Live preview — updates as the agent works, not just at the end.

We built this because we were tired of AI builders that look great in demos but break the moment you customize. Lokuma 2.0 is meant to be a real collaborative builder.

Happy to answer questions about the architecture — looking forward to your feedback!

Hi Product Hunt,

I'm Mu, back again.

In March we launched Lokuma Design Agent here - a design intelligence layer that other AI agents call. It took #1 of the day, and that response shaped what came next.

Design Agent is still one of our core. It's evolving fast — the world is moving toward agent-first stacks, and design intelligence for AI is a long road we're committed to.

But there's another half of the picture.

A lot of people aren't building with their own AI agents, yet.
They're sitting down to make a website themselves.
They still need better tools.

So today I'm launching Lokuma Website Builder 2.0.

Design Agent is for AI.

Website Builder is for the people AI builds for.
Same conviction, different side of the screen.

Why now?

Over the past few months, agent architecture has matured. Tool dispatch, persistent memory, self-repair, observability — the runtime is finally ready. And SMBs and creators are entering AI website building in waves. But most of what they get is still a v1.

Most AI website builders generate a great first draft.
Then they leave.

You change the headline, the design breaks.
You add a section, the brand drifts.
The hero photo is wrong, but you can't say "make it more golden-hour" — because the agent doesn't remember what your brand looks like.

That's not a prompt problem.
That's a runtime problem.

So in the month after Design Agent shipped, we went heads-down to build the runtime underneath.

Lokuma 2.0 is a design-aware agent harness for websites.

The same agent that builds your site can edit it next month, restyle it next quarter, swap a palette, fix a broken form, ship the change. It remembers your brand, your structure, your previous iterations, your live source code.

Generation isn't running a website.
A first draft isn't enough.

You need something that stays.

Why us?

I've spent the last decade building tools designers actually use — Readdy and Creatie reach 500,000+ designers and creators today.
Lokuma Website Builder 1.0 quietly shipped in February and showed us where AI-built websites actually break.
Design Agent gave AI a designer's brain.

This one gives that brain a body — and that took the whole team.

We're a small indie crew — designers who write code, AI researchers who care about typography, growth gurus who were running AI-native marketing experiments before that was a category. The roles blur, deliberately. A harness like this lives in the seams between design and runtime, between agent loops and SMB workflows — exactly where mixed-discipline teams move fastest.

Curious how others here see it:

When your AI ships v1, what's the first thing that breaks the second time you try to change it?

— Mu

The real test will be long term edits most tools can generate a good first version but maintaining design integrity after 10–20 changes is where things usually break.

I like lokuma, i tried it and figure it create sites with better design than directly from claude, very nice! good job :)

I like the move toward targeted patching instead of full file rewrites. Whenever I use AI tools available for UI work, they usually break existing CSS while trying to add a single new button. Does this agent actually understand my existing design system and tokens before it starts patching, or is it still guessing a bit?

what if I'm an agency building 5 different client sites?
Can the agent learn from one site's design system and apply lessons to the next one, or is each site starting fresh?

The concept of a 'design agent harness' is intriguing. Does the agent iterate based on high-level feedback (e.g., 'make it more professional'), or does it require specific UI instructions?

I've spent the last few years marketing AI website builders. From inside that seat, the pattern is hard to miss: every new launch promises something different, but the outputs converge within a quarter. Different brands, same skeletons. Different prompts, similar typography. After a while it stops being a tooling problem and starts feeling like a category problem.

The thing that's hard for AI isn't generating a page. It's generating constraint. Design is mostly about what gets left out — the hierarchy that decides what earns attention, the system that decides what's allowed at all. Models are trained on abundance; they want to add. Design wants to subtract. That's the real gap.

Lokuma 2.0 starts there. It builds the design system first — color, type, spacing, hierarchy, the whole grammar — before a single page exists. The site is then composed inside the system, not the other way around. Drop in a reference site and it picks up the design language without copying the layout. Bring an old site across and it gets rebuilt on top of a proper system, content and assets intact.

It's a strange thing to ship from inside a category I've watched commoditize itself. But this is the part that's hardest to fake, and the part that — if it works — holds up. AI is going to keep making generic things faster. Taste is the only layer that compounds.

The "operate" part is what catches my attention. Building a site is table stakes now — maintaining it, updating content, fixing broken stuff without touching code is where 90% of non-technical founders get stuck. Does the agent handle things like SEO meta tags and structured data automatically, or is that still manual?

Congrats on the launch. I tried it out and the onboarding questions stood out to me.

The best part for me was being able to paste in websites for design inspiration. That made the process feel more grounded because I was not just describing a site from nothing. I could give it examples of the style I had in mind and let the builder work from there.

I also liked how descriptive the questions were before generation. For a website builder like this, those first inputs seem really important because the more clearly the agent understands the user’s taste, references, and intent upfront, the less cleanup and back-and-forth the user probably has to do after the first design is generated.

Excited to see where this goes. The product feels like it is trying to make the design process more collaborative instead of just generating a first draft and leaving the user to clean it up.

I've been using agentic website builders like Antigravity, and there are some clear advantages and disadvantages to Agentic Website Builder 2.0 by Lokuma.

On the plus side, it's very easy to use, meaning someone from a non-tech background can easily pick it up. It also handles the design part entirely on its own and honestly does it much better than Antigravity, which requires the user to provide exhaustive details and a hell of a prompt just to make things look aesthetic.

However, it lags behind when it comes to flexibility. Antigravity is basically VS Code on steroids, which devs can leverage much better, but that is not the case here with Agentic Website Builder 2.0.

Honestly, the biggest headache with AI builders is trying to change one small line of text a week later and watching the entire layout drift into absolute chaos. If this actually keeps edits localized without turning the codebase into spaghetti over time, it’s a massive win.

Is the focus on UI only, or does it include UX too? From another comment and reply, if I ask it to "make it more professional" does it understand how the target audience will perceive "professional"?

Congrats, Mu! For agencies building sites for clients – does Lokuma support handoff where the agent can explain its design decisions to a non-technical client during revision rounds?

Congrats, Mu! For SMBs who are not technical how do they tell Lokuma what their brand looks like? Do they upload a style guide, point to an existing site, or describe it conversationally?

Really thoughtful launch congrats! What's the most unexpected thing that broke during your own testing when editing an AI generated site a month later? That story would help us trust the runtime.

Congrats on the launch, Mu! You said most AI website builders generate a great first draft, then leave when your AI ships v1 what's the first thing that breaks the second time you try to change it? for us it's always the hero section layout

Congrats on this launch!

Hey Product Hunt 👋

Joy here, back for round two.

In March, we launched Lokuma Design Agent — a design intelligence layer built for other AI agents to call. It hit #1 Product of the Day, and what we heard from that community changed everything about what we did next.

Design Agent is still core to what we're building. It's moving fast, agent-first stacks are becoming the default, and we're in it for the long game on design intelligence for AI.

But there's a side of this we hadn't fully addressed.

As a branding & marketing agency owner, I've seen how SME owners struggle.

You're wearing ten hats. You're good at your craft — whether that's running a restaurant, a studio, a clinic, a shop. But building a website? That's a different skill set entirely, and hiring someone to do it right isn't always an option.

I watched businesses around me struggle with this for years. Not because they lacked ambition, but because the tools assumed too much, too much time, too much budget, too much technical fluency. They'd settle for something that looked cheap. Or they'd spend money they didn't have on something they couldn't maintain.

That gap is what Lokuma is really about. Not just a better AI tool, but an honest attempt to give small businesses something they've never quite had: a website that looks like it was made for them, and actually stays that way.

What Lokuma 2.0 actually is:

Most AI builders are one-and-done. They generate, then vanish. What you're left with is a site you can't confidently touch — because the thing that built it isn't there anymore.

Lokuma 2.0 is different. It's a design-aware agent harness: the same agent that builds your site sticks around. It comes back next month when you need to update the headline. Next quarter when you want a full restyle. When a form breaks, when the palette needs a refresh, when your offering changes and the whole page needs to catch up.

It knows your brand — not as a setting you configured once, but as something it's held onto: your visual language, your structure, your previous iterations, your live source code. Every edit is informed by everything that came before.

Because here's the thing most builders miss: generating a website isn't the same as running one. A great first draft is just the beginning. The real test is what happens the second time you need to change something — and the third, and the tenth.

You don't just need something that launches. You need something that lasts.

So join us and explore Lokuma 2.0 together.

Great one team, BTW, if someone eventually wants to hand this off to a developer or export it cleanly, what's the code quality like?

Hey PH. Building Lokuma 2.0 these months — with the team, with AI as a collaborator — keeps returning me to one thought: the era already shifted. The question isn't what AI can do. It's what each of us chooses to become next. — DK

Hi PH! Full-stack engineer at Lokuma. I tune guitars on weekends. Tuning agents on weekdays turns out to be the same craft — pluck, listen, find the half-step off, adjust one thing, pluck again. Most of Lokuma 2.0 was built by agents I spent two months tuning. Different strings, same instinct.

Gradient Bang is a massively multiplayer, completely LLM-driven game. Come play Gradient Bang with us. See if you can catch me on the leaderboard.

This whole thing started because I wanted to explore a bunch of things I’m currently obsessed with, in an application of non-trivial size, that felt both new and old at the same time.

So … a retro-style space trading game built entirely around interacting with and managing multiple LLMs. Factorio, but instead of clicking, you talk to your ship AI and figure out how to make money, make friends, and make havoc for your enemies.

Some of the things we’ve been thinking about as we hack on Gradient Bang:

- Sub-agent orchestration
- Managing very, very, very long LLM contexts, including episodic memory across user sessions
- World events and large volumes of structured data input as part of human/agent conversations
- Dynamic user interfaces, driven/created on the fly by LLMs
- And, of course, voice as primary input

If you’ve been building coding harnesses, or writing Open Claw agents, or doing pretty much anything that pushes the boundaries of AI-native development these days, you’re probably thinking about these things too!

The game is entirely open source. So if you want to see how we built it, you can clone the repo and start asking Claude/Codex about the code. If you want to add a feature, submit a PR.

New today, design your own corporation ship agents, run them in a Vercel Sandbox, and bring them into the game. Think you can make your pair trading loops faster? That's going to give you a pretty big advantage in the game. Want to run with unlimited corp ship compute using open source models? You can do that, now!

See the Vercel Sandbox subagents starter repo here: https://github.com/pipecat-ai/gradient-bang/tree/main/deployment/vercel

Gradient Bang is one of those products that immediately makes you want to keep exploring. Super fun, witty and packed with interesting ideas around agents, voice, and dynamic UIs. Glad I got to contribute to it in a small way, awesome work by the team!

Was super interesting to work with the @Daily.co team on this. The use case goes way beyond gaming into complex enterprise workflows. 🚀

Would love to read a deep dive into the subtopics you tackled - managing long-term context, dynamic UIs, and sub-agent orchestration. Will hit Claude on the repo logic in the meantime! 🤖💻

Gradient Bang is built on Pipecat, the leading open-source Python framework for building real-time voice and multimodal agents. We're hosting a hackathon on May 30th at YC along with our friends at Cekura, NVIDIA, AWS, and Twilio. Come join us!
https://events.ycombinator.com/HW0opxy78

Takes me back to the 90s!!

Having worked with agents and particularly voice agents for the last 2 years, the craft behind this game is quite amazing. We ran 100s of simulations to test these agents and I have only 1 thing to say - If you haven't tried it yet, do now period

Launching together today makes Product Hunt even more exciting. Love the product it really caught my attention.

Cheering for fellow makers today, and would love to support each other. Wishing you a fantastic launch 🚀

love this! Do you have future plans of launching more such games?

looks cool! Love that players can ship their own subagents into sandboxes. Curious how you keep the game balanced when some players can write tighter loops or throw way more compute at their corp ships than others?

A multiplayer game driven by LLM prompts sounds like absolute chaos in the best way. How do you handle the latency issues that usually come with real-time LLM interactions?

Hey PH! 👋

Most people don't know this yet: when you apply for a job, an AI tool scans your LinkedIn before any human sees your name. Platforms like Juicebox, HireEZ, and Gem are filtering candidates at scale and most profiles are invisible to them.

The candidate might truly be qualified, but because their profile isn't written for how AI reads it, they never get shown to a recruiter.

Findable fixes that. Paste your LinkedIn URL and we score your profile in 30 seconds, showing you exactly how findable you are to AI recruiting tools and what's hurting your visibility.

Here's what you get:

📊 AI readiness score out of 100: see exactly where you stand across role clarity, skills, and completeness

✍️ Rewritten headline: optimized for how AI recruiting tools actually surface candidates

🔧 Copy-paste fixes: specific changes you can make on LinkedIn right now

📋 Resume scoring: upload your resume and we'll find gaps and check alignment with your LinkedIn

🎯 Role-specific match score: enter any target role and see how well your profile matches it today

🚀 Long-term goal mode: reframe your entire action plan around where you want to be, not just where you are

📋 Action plan: specific things to build and publish that make you a credible candidate beyond just fixing your profile

🏢 Companies matched to your trajectory: aligned, realistic, and stretch companies with a personalised explanation of why you fit

📩 Who to reach out to: specific people at those companies with pre-written outreach messages ready to copy

A few ways people are using it:

🎯 Active job seekers: know exactly what AI recruiting tools see when they scan your profile and fix it before your next application.

🔄 Passive candidates: not looking right now but want to stay findable. Run your score, make the fixes, and let opportunities come to you.

👩‍💼 Recruiters: send it to every candidate before they go to interview. Takes 30 seconds and makes your submissions stronger.

🎓 Students and early-career folks: your LinkedIn is your resume now. Make sure AI can actually find you before you start applying.

Free. No login. 30 seconds.

Try it at: https://tools.crustdata.com/crustimate

Nice tool! Definitely useful for candidates applying. Tried it out and got some good recs for improving my score.

Best of luck team with your launch.

Quick question, have you tracked whether people who actually implement these fixes get more recruiter outreach? Like, real before/after data?

Curious if you personalizing recommendations by role (PM vs engineer vs marketing), or if it’s a universal scoring system.

This feels especially useful for teams without dedicated product ops or research synthesis layers. Smaller teams will likely get the most immediate value.

As a developer, I usually spend half my day clarifying what a PM actually wants. If this can bridge that gap by translating spec-speak into something actionable for my dev environment, that’s a win for me. Does it export directly to Linear or Jira, or integrate with any other tools?

Great idea. I’m still waiting for someone to build an AI service that analyzes user behavior on a website, understands their goals and doubts based on the entry point, visited pages, mouse movement, etc., and then takes actions that increase the chances of conversion. This could be a popup with relevant information, an email, or even a dynamically changing interface tailored to the user’s intent. That would be gold for product managers!

@sevil_kubilay nice, congrats, btw, what do you mean by "customer signals" exactly? Curious to know more how it works and delivers the value and btw, great work :)

'Cursor for PMs' is a bold tagline! Can mia help with drafting technical specs and PRDs by referencing an existing codebase, or is it focused on the ideation stage?

The hard part has always been deciding what counts as signal. One loud enterprise can drown fifty quiet churners, and most teams never see the second group.

Hope you'll solve it!

The "Cursor for PMs" framing really clicked for me, it immediately communicates that this isn't another dashboard, it's a tool that executes.

Sevil, curious: of the 500 customers you mentioned, what's the workflow change they report most often after using mia for the first 30 days?

Love the direction. I can see how it works when a solid growth team actually sit next to Product.

As a PM in a legacy org, I'd be happy if I even got actionable insights based on multimodal inputs - I'll write the PRD, hell I'll even write the code - but please help me make sense of the hairy data. Something I'm gonna try solving for myself.

Well done. Will you be able to trace a shipped feature back to the original customer signal that triggered it? That traceability layer would be huge for making the case internally that PM-led discovery actually moves the needle

The gap between "raw feedback" and "something an AI can actually build from" is real and painful.
Curious...does mia handle conflicting signals from different customer segments, or does it surface them and leave the call to the PM?

Hey team. Getting an error at the signup "Access blocked: This app’s request is invalid"

Congrats on the launch! Been following closely this space as it's probably a big gap in between Coding Agents and all the data produced by the business. The only thing that is really integrated is the ticketing (e.g linear).
Whats is your take in the right UX for the long term of the product? More of a self-service insight platform, or more integrated directly into Slack and Coding agents. I'm asking because I think the problem is very genuine but the UX is hard to nail.

Hi there PH. I'm back with another banger launch and this time an open-sourcing project.

A bit of context: I came across TrustClaw on Twitter, before I joined Composio. Every now and then I'd run into a founder who tells me they were using it as their daily driver.

Today Sarah(MTS at Composio and creator of TrustClaw) made it open source on GitHub. It is MIT-licensed, with a CLI that gets you from zero to a working instance in two minutes. The inner founder in me has been twitching about this all week and there's a real category of people who want their own always-on agent, on their own infra, with their own data, and not a hosted SaaS sitting between them and their accounts. TrustClaw is the cleanest version of that I've seen.

Huge shoutout to Sarah who shipped in less than 4hrs and turned this from an idea to what it is today.

Deploy a 24/7 agent via Vercel that talks to you on Telegram, remembers things, and plugs into ~1000 apps over OAuth via Composio, but didn't want to ship your tokens to a third party. Clone the repo and build your agent in < 2 mins. Btw OpenClaw but more secure.

Sarah is here to answer any questions. Let us know what you think. All feedback welcome :)

Hey, creator of TrustClaw here!

3 months ago OpenClaw had taken the internet by storm and I was immediately inspired to build some way to make it easier for non-technical people (and lazy people like me) to spin up their own personal agent without having to deploy their own docker container or set up plugins, skills, MCPs, CLIs for all their apps. So I built my own version and launched it and it went viral on X. After months of convincing, I am so excitid to announce today that it is open-source!

The craziest part is you can deploy all this in a single command!

npx @ComposioHQ/trustclaw deploy

Or, without touching any code at all via our Vercel template.

When preparing this project for the open-source community, easy deployment was a MUST. Vercel's stores (database + redis), cron jobs, AI Gateway, AI SDK covered absolutely everything this project needed. Their ecosystem is incredibly powerful and made this project so easy to get up an running.

Being able to self host is a big deal for personal agents that have access to my Telegram or Gmail. Having clear audit logs to see exactly what actions were taken while I was “asleep” would really help build trust in the system.

Is there a way I could run this on my laptop? Because I would try a safer version of OpenClaw. I haven’t even tried OpenClaw yet because I think it’s not secure

One click to deploy is insane, no more excuses to not ship. Can't wait to see what people build on top of this 👀

The combination of scheduled actions, self hosting, and access to real apps through OAuth makes it easier for people actually building workflows they rely on daily, especially with the added control and privacy that comes from being able to run it yourself. Good job guys.

vercel hobby caps functions at 5 minutes and cron jobs only run once a day. for anything beyond basic scheduled tasks that seems like it could hit a wall pretty fast. are most users running this on hobby or does it realistically need vercel pro to be useful day to day

The 'doing things while you sleep' promise is bold! What kind of security guardrails are in place to ensure the AI doesn't perform unauthorized actions in a production environment?

When the agent performs scheduled actions while you're asleep, what's the rollback mechanism if it makes a mistake? Can you audit what it did and undo bulk operations, or is it fire-and-forget once the cron triggers?

Setup was super quick & easy using Vercel Single Click deploy but when I started using it I couldn't get it working and pages were loading blank (tools and no responses). What did I do wrong!? :(

I've spent entire weekends trying to set up self-hosted agent stuff and it always falls apart somewhere in the auth flow . the OAuth-through-Composio approach makes sense too, like the agent never actually sees my raw credentials which is the part that always made me nervous. one thing I'm genuinely unsure about though , if I'm on Vercel hobby plan, how long can a single agent task actually run before it just cuts off?

The Game Boy as ambient indicator that an agent is running async is a fun design call. Curious about the specific signal: is it binary (something is running, nothing is running), or is the screen actually communicating which agent, which task, how far along?

The visualization choice seems load-bearing to me, because "your agent is doing something" is much less useful in practice than "your agent is on step 3 of the migration and waiting for your token approval." Which side did you land on, and why?

This is a fun way to solve the silent agent problem. Sometimes I am not sure whether my script is stuck or just thinking. Having an interactive agent that shows its state or mood makes it much clearer whether it is actually working or not. Beyond the aesthetic side, does it provide any real debugging feedback if the agent hits a loop, a rate limit, or a code issue that it is unable to solve?

This is such a clever visual cue for 'deep work.' Does the Standboy character have different animations based on the type of task the agent is currently performing?

Gotta love the solution. Brings back the memories of the good old days. 😅

Tamagotchi that you have to keep alive next? :D

this is the kind of thing that sounds dumb for 5 seconds and then you realize people will genuinely keep it installed for months

also feels like a very “current era” product. humans becoming supervisors for agents means idle time inside editors is suddenly a real UX problem and this weirdly addresses it better than productivity tools do

The saves live in a folder you control part is what got me — no cloud dependency

Really cool to see us launching on the same day. Loved the product and the execution looks great. Sending support from one maker to another would love to support each other 🚀

Cline just open-sourced the agent runtime that powers one of the most-used coding agents in the world.

What it is: @cline/sdk is a TypeScript agent runtime with a plugin architecture, designed to be embedded in any surface VS Code extensions, CLIs, desktop apps, Slack bots, or whatever you're building.

The problem with most agent frameworks is that the loop gets glued to the product surface. Cline had this same issue: the agent grew inside the VS Code extension, and eventually the runtime and the IDE wrapper were inseparable. The SDK is the result of pulling those two things apart: a stateless, reusable loop at the core, with a stateful runtime layer on top of it.

What makes it different: most coding agent runtimes aren't actually portable. They're designed to run inside one product. Cline SDK is explicitly built so sessions can move across surfaces, long-running work survives UI restarts, and the agent loop stays stateless regardless of where it's running. The plugin layer lets teams add domain-specific tools and observe lifecycle events without forking the runtime.

Key features:

• Plugin architecture for registering tools, lifecycle hooks, and agent rules

• Native subagents and multi-agent teams with built-in handoff and orchestration

• Checkpointing, scheduled cron jobs, web fetch, and MCP connectors out of the box

• Provider-agnostic: Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and OpenAI-compatible endpoints

• Example apps included: Slack bot, VS Code extension, desktop app

Benefits:

• Embed a production-grade agent loop without building the harness yourself

• Swap models and providers via config, not code rewrites

• Multi-agent workflows without a separate orchestration layer

• Start small with individual packages, add the stateful runtime only when you need it

Who it's for: TypeScript developers building coding agents, internal dev tooling, or AI-powered CLI applications who want a stable, extensible runtime rather than a prompt wrapper.

The benchmark numbers on Terminal-Bench 2.0 will get attention, but what actually matters here is that Cline is releasing the same runtime they use internally as an open package. That's a different kind of credibility than a framework built specifically for public release.

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends

Honestly Kudos to the Cline team. I have been switching many internal agents over to the new SDK and it has been great! 🫡👏🏻🚀

Plugin-based runtimes live or die on fault isolation. If a buggy tool crashes mid-session, does it kill the whole agent or is each plugin sandboxed? That's were most SDKs get sloppy.

The pain is real — every new Claude/GPT/Cursor session starts from zero and you spend the first 5 minutes re-explaining your stack. A shared context layer that just follows you across tools is the right fix. What happens when context gets stale or contradictory — does Relay surface that, or does it trust whatever was saved last?

Goated idea, real problem i've been suffering from myself. Building in the same space, tried their product and LOVED it. Congrats on the launch guys! Good luck

This is painfully relatable. I run a 3 person startup and we use Claude, Claude Code, and ChatGPT across the team. the context re-pasting is real. Solo founder from Kazakhstan building this is impressive too. Good luck today!

@alimkhan_y Congrats, every day I have this challenge, but I solve it by keeping a clean set of md files in my repos and always instructing the agents to keep them updated (e.g. build plan, architecture, progress, audit, summary, etc.). I was curious to know how it would be different from that and what would be the added value/benefit here?

Btw, very cool product and love the website!

The 'stop repeating yourself' pitch hits home. Does Relay sync your context across different LLM interfaces (like ChatGPT and Claude) simultaneously?

Sleek Analytics v3 is live 🚀

We've been obsessed with one problem: how to make analytics so simple, you actually use it.

Today we're shipping things that change how people track their business:

Real-time global visitor insights, our new iOS app, and daily digests in Telegram—all built on Vercel. We're using Vercel's Edge Functions for our webhook infrastructure, cron jobs for scheduled digests, and deploying globally so data reaches your dashboard instantly, no matter where you are.

Built entirely on Vercel's stack, which made it possible to ship globally, at scale, without managing infrastructure.

Ask us anything—we'll be here all day 💚

Real-time Telegram pings for traffic spikes is a great feature for small teams. Can you set custom thresholds for these alerts to avoid notification fatigue?

Love the UX/UI of your app. Just wishing you all the best after the launch date.

The dashboard looks very easy to understand, a huge upside for small non-technical teams. It looks like you can identify users with an email address (or other identifier). Does it show non-logged-in visitors?

Hi everyone!

Kimi WebBridge is a practical bridge between AI agents and the browser.

Install the extension, connect it to your local agent, and the agent can use your existing Chrome or Edge session to handle web tasks like opening pages, filling forms, collecting information, and moving through websites for you.

The nice part is that CC/Codex, @Cursor, Hermes, and @OpenClaw can use it too.

A lot of daily work still happens in the browser, and WebBridge gives agents a simple way to actually operate there.

The local first approach for browser control is a smart move for security. I have stayed away from most browser agents because I don't want to hand over my session cookies to a third-party server. How do you handle sites that are heavy on shadow DOM or complex anti-bot triggers?

I dogfood Kimi WebBridge and honestly, I use it every single day!

Just install the extension, link it to your agent, and it will surf Chrome/Edge for you — filling forms, grabbing info, clicking around… basically handling all the boring stuff.

It works with Claude Code, Codex, Kimi CLI, Cursor, Hermes, OpenClaw.

Multi-file analysis is the feature I'd actually open this for. Half my week is reading through ten PDFs from a customer to figure out what they're really asking for.

I know how tough interacting with a live browser can be as I've been frequently using Python and Selenium recently. Giving terminal agents a clean way to bridge that gap is a massive step up. I'm really curious how the extension actually passes the page data back to the LLM... does it clean everything up into structured JSON or a lightweight DOM snippet first, or is it just dumping raw HTML? How do you manage the token count?

Giving terminal agents a clean way to interact with a real browser is a huge help. I'm really curious how the extension actually passes the page data to the LLM. Does it clean everything up into structured JSON or a lightweight DOM snippet first, or is it just dumping the raw HTML? Managing the token count while keeping the page context is always the trickiest part of building web agents.

How do you handle sensitive actions like form submissions, is there a confirmation step before it clicks buy or send?

How do you handle user data privacy when bridging AI agents to the live web? Especially for users on sites with sensitive content (banking, health portals)?

Connecting agents to the 'live web' is still a major hurdle. Does the bridge provide a structured data output (JSON) for the agent, or does it just pass raw HTML?

Hey Product Hunt 👋
Launching Wowable on Vercel Day 🚀

We built Wowable because most AI website builders still generate generic content.

The problem: even when the design looks good, every website ends up sounding the same. You still have to rewrite everything manually because every page says “quality service”, “customer-first”, or “best in town”.

Our approach: businesses already have real content online. Customer reviews, social profiles, photos, business listings, and screenshots already describe what makes a business unique far better than placeholder AI copy.

Instead of starting from a blank page, Wowable builds websites from content businesses already have online.

Paste a Google Maps link, Instagram profile, TripAdvisor page, LinkedIn profile, or even a screenshot, and Wowable builds the website from that content automatically.

No prompts. No writing. No complicated setup.

🚀 Product Hunt Launch Offer:

To celebrate our launch, we’re giving the first 50 users 50% off Wowable Pro for the first year.

Use code: WOWABLE50

Would genuinely love your feedback 👇

Can the owner edit and lock certain sections so the AI doesn’t overwrite key details later?

@josh_bennett1 I just tried generating a website using a random TripAdvisor link and honestly didn’t expect it to turn out this clean. The layout, photos, reviews, and overall structure actually looked like a proper modern business website instead of one of those obvious AI-generated pages. Even the mobile version looked polished. Curious how you guys are handling the scraping and content structuring behind the scenes because the data mapping looked surprisingly accurate.

Hey everyone, Chris here, one of the makers behind Wowable.

Over the last 10+ years running a website and digital agency, I noticed the same thing over and over again: most small businesses know they should have a website, but the process is still surprisingly painful. Hiring designers/developers is expensive, DIY builders take time, and many business owners simply never get around to launching something properly.

At the same time, people are searching online more than ever, not just on Google anymore, but increasingly through AI tools like ChatGPT and other assistants that rely heavily on web content and business information. Businesses without a proper web presence risk becoming almost invisible online.

That’s why we built Wowable. The idea was simple: what if you could paste a Google Maps link, Instagram profile, Facebook page, TripAdvisor listing, or even screenshots/brochures… and instantly get a clean, professional website in under a minute?

The goal isn’t just to “generate a website with AI.” It’s to help businesses get online faster, look credible, and start getting discovered by more customers.

Would genuinely love to hear your thoughts, feedback, criticisms, or ideas on where we should take this next 🙌

@moh_codokiai @josh_bennett1 i tested this just while ago and its really good and by the way the name wowable is very good name.

The site generated from a simple brochure into a polished website is wow. Good luck and would like to know which model using for this or scraping? Also how the seo works for this?

One thing we realized while building Wowable is that most small businesses already have enough content online for a website, they just don’t have time to organize it.

A restaurant might already have:

• hundreds of customer reviews

• photos

• menus

• Instagram posts

• business listings

but still no website that properly reflects the business.

That became a huge part of the idea behind Wowable.

I like that this doesn’t stop at just generating a homepage. The ability to edit themes, manage SEO, connect domains, and view analytics makes it feel more like an actual business product rather than a quick AI demo. Technical question though, how are you handling businesses that have content spread across multiple platforms instead of one source?

@moh_codokiai one site generated personal website from my linkedin and looking great. Loving this to generate more and more.

I tried it, and it is a good idea. However, you need to make it more advanced by using AI to detect logos and themes, as competitors are currently creating sites with similar designs and themes.

I just tried generating a website using a random TripAdvisor link and honestly didn’t expect it to turn out this clean. The layout, photos, reviews, and overall structure actually looked like a proper modern business website instead of one of those obvious AI-generated pages. Even the mobile version looked polished. Curious how you guys are handling the scraping and content structuring behind the scenes because the data mapping looked surprisingly accurate.

Congrats on the launch! Local business owners are too busy running the shop to write web copy. The whole game is meeting them at the content they already have online.

The backtesting part is what makes this interesting. Most trading bots usually just follow trends. How much control do I actually have over the backtesting parameters? For example, can I force it to test against specific historical volatility events like flash crashes, or do I need to specify something else for that?

Hey PH! I'm Bhargava, one of the founders of Mobius.

One thing I kept running into as a retail trader was that the idea was never the hard part. Turning it into something that actually runs in the market was the real issue.

But with Mobius, You describe your strategy in plain English, and we handle the rest. Backtesting, screening, researching, and live trading Mobius handles the whole loop, in one place, in minutes.

We're live today. Excited to hear what you'd build first.

Finally, someone making the Pelosi-tracking actually actionable. I’ve been trying to scrape those filings manually, but the lag usually kills the trade. If this can automate the entry the second the data hits, it’s a massive win. @dheeraj_t

Retail traders fail at the gap between "this looks good in a sheet" and "this is running on real money." Anything that closes that gap fast is going to print.

Knows nothing about trading, but maybe it would feels better if you can track what user have inputed in the landing page in localStorage, and if user decide to register, it will carries to the dashboard as sketch.

Launching together today makes Product Hunt even more fun. Really loved what you built it instantly caught my attention.

Always happy to support fellow makers, and would love to support each other. Wishing you an amazing launch 🚀

The plain English strategy input is the right UX call, most backtesting tools make you learn their DSL before you can even test if your idea is worth anything. Curious how you're handling strategy overfitting when users describe vague conditions — does Mobius add any guardrails there?"

this should ship bundled with macOS for cat owners. two of mine attached, both certified keyboard interns :)

Damn bro, I don't have a cat, but I love your idea so f***g much! This is absolutely brilliant, love it!

Haha, are you making a system to destroy my MAC by letting the cat walk on it? Haha. Good thinking btw.

This is the most 'Product Hunt' product I've seen today. Beyond just locking the keys, does it have a 'cat mode' that displays something fun on the screen while the keyboard is disabled?

Hey! I'm following all AI Video Editors here, because I love to create videos, but editing is a nightmare for me. So, I'm asking the same question all AI Video Editors launching on PH: your video uploaded here(it's super cool btw!) edited using Picsart?

Having 140+ models in one place is a game changer for creators who spend half their day switching between tools. Quick question — is there batch processing? For YouTube thumbnails I usually need 5-10 variations per video and doing them one by one kills the workflow. Would love to test this for thumbnail A/B testing.

Real .pptx export is the killer feature here. Most AI builders give you a proprietary link or a static PDF, which is useless if you need to make a quick tweak in the boardroom. Does it export as editable shapes and text, or just flat images?

Do charts and layouts stay editable once you open it in PowerPoint, or do they flatten into images?

Exporting directly to PowerPoint is a killer feature for corporate workflows. Can Riffly follow a specific brand kit or template provided by the user?

Hey everyone — welcome to the launch! 👋

I built Riffly because I genuinely hated making presentations. Still do. So I built something that does it for me.

This is day 1 of putting it in front of the world and I couldn't be more grateful for everyone who showed up.

Ask me anything. Tell me what's broken. Tell me what you love. Tell me what you'd build differently. I can take it — honest feedback is the only way this gets better.

A few things I'm already working on based on early feedback: → Brand kit support (apply your own colors, fonts, and logo) → Fully editable charts in PowerPoint export → Larger deck mode for 20+ slide presentations

This is a real product built by one person who wanted to solve a real problem. Every comment here shapes what gets built next.

Thank you for being here on day 1. Let's build something great together. 🚀

The thing I keep coming back to with MCP Connectors is that the highest-leverage work in any company already lives across three or four systems, and until now there hasn't been a clean way to let an agent operate on that whole surface from one place. Reading from your warehouse and then acting in Linear, HubSpot, or Slack in the same chat session is a meaningfully different shape of tool than the dashboards most of us grew up with. The team has been deep in this for months and it feels great to finally ship. Happy to dig into anything in the thread today.

The approval gate design is the detail that stands out most — "needs approval" by default before flipping to "always allow" is exactly the right trust model for agents that can actually write to production systems. A lot of agentic tools skip this and it's why people don't trust them with real workflows.

The cross-system chaining is where this gets interesting. Curious how it handles failures mid-chain — if the DB read succeeds but the HubSpot write fails, does it roll back, notify, or just log and move on? That failure handling is usually where these workflows quietly break in practice.

Also wondering if MCP servers you self-host get the same approval UX or if that's only for the pre-built connectors.

My non-engineer team would slack me for a number twice a week and I'd write the query, screenshot it, move on. Tools like this kill that loop, which is what I actually want.