PH热榜 | 2026-05-24

The part of this Stitch update that changes the workflow is the DESIGN.md import, and it doesn't get enough attention in the headline features.

Generative design tools have a consistent blind spot: they ignore everything you've already built. You prompt them, get a screen that looks nothing like your product, and spend the next hour reconciling tokens and components. Stitch now reads your existing codebase, Figma file, or live website before it generates anything, extracting your design language via an open standard called DESIGN.md. The output starts from your context, not from scratch.

Paired with the rest of the I/O update:

• Streaming generation with live steering before the screen is finished

• In-place edits for element-level changes without full regeneration

• HTML-native canvas with real animation and interactive state previews

• MCP-based codebase sync to push visual edits back to your code via an agent

• Export to Figma, Netlify, Lovable, and Bolt in one click

Built for developers and product builders using AI coding agents who need a design layer that connects to their codebase rather than creating a parallel one.

Free in Google Labs with generation limits. Give it a go at stitch.withgoogle.com.

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

I like how Stitch MCP works with Antigravity - I can finally get consistent design done.

Did you remove the ability to make animations?

Curious how Hatter handles component consistency when the sketch input is ambiguous — does it infer a design system automatically, or does the user need to define tokens upfront?

I cancelled hiring a web designer and a front-end developer- but I do have my visual identity and brand assets carefully and intentionally designed

It's great to see stitch becoming better. I started my career from design and started using stitch time after time, since it launched.
I really liked the tool until it was on Gemini Pro 3.0 and when it started using 3.1, there were issues and inconsistencies and that's when i moved to claude design.

I'll give it a go again and hope for the best, congrats to the whole team!

tried a few generations and the speed is impressive. I can see this being genuinely useful for rough ideation, especially when you want to explore layouts quickly even for zero UI/UX experience user.

Already used Stitch for one of my websites and honestly had zero UI UX experience going in. The results were good enough that people actually complimented the design without knowing it was AI generated. Version 3.0 with live canvas and in-place edits sounds even better. This is the kind of tool that makes you look like you know what you are doing even when you don’t.

Congrats on the launch! Having one place to manage across Ollama, MLX, and llama.cpp is something I've been doing by hand for too long. Gonna give this a try.

The 'Runs on this Mac' feature for checking if a model can run on the hardware that I have is my favourite part! That's usually the first thing I want to know before downloading some huge model.

Two things I'd love to see: more pre-download details like license, context length, and RAM estimate, and a quick way to open the original Hugging Face model card from inside the app.

Quick one on storage - if I've already pulled Qwen 32B via Ollama and then discover it again in ModelHub, do you dedupe against the existing local file? Or do I end up with two copies eating 20GB? Well done guys overall

Congrats on the launch! 🚀 Having one unified menu bar app to manage models across Ollama, LM Studio, and MLX makes organizing and discovering Hugging Face models way cleaner.

The "Runs on this Mac" hardware checker sounds incredibly useful before committing to a massive download. Can't wait to give this a spin!

Congrats on the launch!

Local LLMs via menu bar is the right UX switching between models shouldn't require a browser tab. Does it handle model downloads itself or do you bring your own? Curious about the memory footprint running models in the background.

Cool! Congrats on the launch

I usually encounter more issues with how to quickly validate whether a model is suitable for your scenario after obtaining it, and whether there is corresponding code that can quickly verify and reduce the time spent trying one by one. I would like to know if your tool has such a function or scenario.

Does ModelHub also track which models are actually being used so you can archive the ones you never run?

Does ModelHub handle quantization format filtering during discovery — like surfacing only Q4_K_M builds based on available VRAM, or is model selection still manual?

I like that this isn’t trying to become another LM Studio or Ollama replacement. Simpler UI to just manage my local LLMs.

Was that an intentional product call from day one? Also, do you see this becoming more recommendation-driven, like “best models for your Mac” based on chip/memory?

Hey folks! We built modelhub to manage your cluttered local models! Do check it out and give us valuable feedback! :D

Hi Product Hunt! 👋 I'm Charles, founder of Freu AI.

A while back, we teased that we were working on extending our browser automation tech to the entire operating system. Today, we are officially launching Freu AI for Mac—an AI agent that automates any desktop software across your OS using natural language.

The Problem: Vision Agents are Too Expensive & RPA is Too Brittle
We hit a massive wall with current GUI automation. Traditional RPA (AppleScript, rigid X/Y coordinate clickers) breaks the moment you resize a window or an app updates its UI. On the flip side, modern multimodal agents (sending screenshots to cloud LLMs) scale terribly for repetitive tasks.

Right now, most desktop agents operate like interpreters. Every time you ask it to "Extract data from this local PDF and enter it into Excel," it takes a screenshot, sends it to the cloud, reasons about the visual layout, and clicks.

The Traditional Cost: ~10k tokens (Image context) × 5 steps × 10 runs a day = ~500k tokens/day just to navigate the exact same desktop UI, not to mention the unbearable latency.

The Solution: AOT Compilation + Semantic UI (SUI)
Freu AI changes this by introducing Ahead-of-Time (AOT) compilation for OS-level tasks. Instead of the agent analyzing the screen from scratch every single time, you show it the cross-app workflow once.

Freu AI uses a cloud vision-based model to "compile" that session into a deterministic, reusable DSL.

The Freu Cost: You pay the cloud "AI reasoning" token cost once when the agent watches and learns your workflow. But for future runs? The agent simply invokes the pre-compiled DSL command locally. This drops your recurring execution costs to zero and reduces latency from minutes to seconds.

How it works under the hood:
When you record a desktop workflow, our engine doesn't just save a dumb macro. It uses Semantic UI (SUI) to understand the screen:

Perceive: It recognizes buttons, text fields, and icons across any app.

Resolve: It anchors to the semantic meaning of the UI, not rigid coordinates. If Spotify moves their "Play" button, Freu AI still finds it.

Execute: It binds these visual anchor points into our DSL and executes them deterministically.

🎁 The Open-Source Bonus:
While the Mac desktop app is our core product, we are open-sourcing freu-cli today—our underlying DOM-based browser automation engine. You can drop it into your own agents to give them instant "muscle memory" for web tasks. Repo here: https://github.com/freu-ai/freu-cli

🔮 What’s Next: The Local Vision Execution Engine
We are relentlessly upgrading our stack. Very soon, we will launch a capability to run the execution phase using a lightweight, SUI-optimized vision model running entirely locally on your hardware. While we will always rely on powerful cloud LLMs to understand your complex intent during the initial "learning" phase, this upcoming local engine means your day-to-day repetitive executions will cost exactly zero API tokens and keep your real-time screen data 100% private.

We’d love for you to try Freu AI for Mac. I’d love to hear your feedback on our AOT approach or how you're currently handling repetitive cross-app tasks. My co-founders and I will be hanging out in the comments all day to answer your questions! 🚀

Quick question, can it handle when things go wrong mid-workflow? Like if a dialog pops up unexpectedly, does it know how to recover or does it just brick?
All the best team

What kinds of apps does it work best with right now, native macOS apps or web apps in the browser?

Compiling workflows once via a deterministic DSL and skipping LLM calls at runtime is a smart tradeoff. We've hit exactly this problem at RetainSure: brittle selectors break on every UI update and token costs add up fast. This architectural choice solves both at once. How does freu-cli handle mid-execution interrupts? If a modal pops up, does it replan via LLM or does the DSL have recovery logic baked in?

Nice product

As someone who spends 2 hours a day moving data between apps, if this actually works I'm installing it right now. The zero-cost execution is the game changer.

The best part for me is not having the agent re-read the same UI every time. If a workflow is repeated daily, teaching it once and running it locally sounds way more practical and I'd mainly want to see how it handles small layout changes after app updates.

Pretty simple but a very cool approach, also awaiting local vision engine!

Hi everyone!

Love small practical tools with a very clear pain point.

Every USB-C cable looks the same, but some charge fast, some are slow, some handle displays, and some quietly break your setup.

WhatCable makes that visible in plain English. It basically tells you whether the cable, charger, device, or Mac port is the problem, so you don’t have to guess why your SSD is slow, your dock is acting weird, or your display refuses to work.

Super handy stuff!

Cool project - need this for Android!

USB-C cable confusion is one of those problems that sounds trivial until you're trying to figure out why your display isn't working. Every developer has a drawer of mystery cables. Does it work on iOS too or Mac only?

That is really cool, Please tell me I can use this for my iPhone and iPad as well, by any chance @daz1uk?

This looks promising. I always have to call my husband to check the cable and figure out stuff. He knows all the USB cables and how fast it charges. To me, they look all the same. haha

Definitely going to try it out!

I've got a mess of cables to sort through, with some that are suspect. Now I can find out for sure!

Love it! As one of the folks who keeps "known good cables" separately from the rest of the flock.

Hey friends

Sacha here, founder of @Edgee .

Two weeks ago Anthropic announced that starting June 15, your programmatic Claude usage gets capped at a $20-$200 monthly credit pool. For heavy Claude Code users, that's roughly a 25 to 40x cut in effective inference.
Same with Copilot that is moving to usage-based pricing June 1st.

A lot of people are angry about it. I get it. But we're builders, and the right answer to a market change is to ship better tools, not to complain.

We started building Fallback Models the week before Anthropic's announcement, after one too many Anthropic outages. The timing is now coincidentally perfect.

Here's what our Fallback Models feature does:

→ Anthropic down? Route to Kimi K2.6, GLM, Qwen, Gemma, or others.

→ Plan limit hit? Same thing, automatically.

→ Want to route always? Pick your model.

You can also fall back to your own Bedrock, Vertex, or Azure account in one click. Same Claude Code on top, your cloud underneath, zero code changes.

And it works the same with Copilot, Codex...

How it fits with our other features:

- Compression: use fewer tokens

- Teams: see who uses tokens and on what

- Fallback Models: keep working when your primary model can't

Fallback Models ships with our Team plan. The compression engine that powers all of it is free to try, no credit card.

Two questions for you:

- Which fallback models would you actually want to use?

- What other failure modes should your coding assistant handle?

Will be in comments all day 🙏

edgee.ai/fallback-models

smart approach to a real pain point. the rate limiting on Claude Code during peak hours has killed my flow more times than id like to admit. curious how the token compression affects output quality though — does it handle long context windows well or is there a tradeoff with the 50% savings?

The auto-fallback when rate limits kick in is the part I always end up wiring by hand. Good luck with the launch!

Congrats on the launch!! This solves a real issue for developers who can’t afford downtime when Claude is rate limited or down. Keeping coding running with simple fallback models will make workflow feel more stable.

The transparent proxy approach here is clever. Intercepting at the API layer means zero client changes, and that matters. We've burned time at RetainSure debugging failures partway through a session when Claude's rate limits kicked in at the worst moments. How do you normalize tool_use schemas across models? Claude's format doesn't map cleanly to Qwen or Gemma, and that mismatch can quietly degrade agent output.

Can we set the sequense of fallbacks? See, I'd love to give you a sequence of the LLMs I don't pay for and then, last resort, OpenAI and Grok can squeeze the last of my life blood out of me. Thanks.

The fallback angle is practical for agent workflows, especially when a coding session is mid-task and the provider limit hits. I’d be curious how you surface model switches in logs, since silent fallbacks can make debugging output differences harder.

Fallback models for Claude Code is exactly what's needed hitting a rate limit mid-task and losing context is painful. Does it maintain the full context when switching models or does the fallback start fresh?

Runway Agent is a chat-based AI inside the Runway web app that handles ideation, editing, sound design, and final output as a single conversation.

You describe what you need, the agent builds toward it, and the output is a complete, publish-ready file rather than raw footage requiring further work.

What makes it different: Most AI video tools are prompt-in, clip-out. Runway Agent operates at the project level, covering generation, editing, and audio in one session. The "finished" framing in Runway's own copy is the operative word here. The output includes sound design, not just visuals.

Key features:

• Conversational interface for creative direction and execution

• Full video output with editing and sound design included

• Targets ads, shorts, and social content formats

• Available on Runway web

Who it's for: Content teams, solo creators, and marketers producing short-form video for social, ads, or branded content who want to go from idea to finished output without context-switching between tools.

An agent that can hold a creative brief and execute it end-to-end is a different product category from a generation tool. Whether the execution holds up in practice will be the real test, but the direction is the right one.

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

Chat-based video editing could be a game changer for solo content creators the editing part is always the bottleneck. Does it handle vertical format for Reels and TikTok or mostly landscape output?

I stopped using Runway when the clips didn't go beyond 10 sec and there was no voice over consistency across clips. Hoping 'project level' editing can help with solving for the challenge of consistency and character integrity.

This feels like one of those tools people will either love immediately or ignore completely. The idea is strong though, especially if it cuts down the usual video-editing pain.

This actually looks pretty clean. Honestly, this is the kind of small Mac tweak that makes daily use feel nicer.

The preset model is cleaner than it looks. It's not just icon positions, it's app lifecycle management across contexts, which is where the real complexity lives. At RetainSure I switch between customer-facing work and engineering several times a day and the context tax adds up. How do you handle apps with unsaved state? Does DockFlow warn before closing or delegate that to each app?

When you switch presets, does it also restore window positions or just open and close apps?

Nice work on the hotkeys, context-switching between dock presets is the kind of friction I keep wanting gone. Good luck with the launch!

Hey Product Hunt 👋

I'm Samet. Building Podio with @armanaksoy34

Quick context on why this exists: we kept opening news apps, scrolling for ten minutes, and walking away more anxious without actually knowing more. The feeds aren't broken. They're just doing a different job than the one we wanted them to do.

So we tried audio. You tell Podio what you follow (your industry, a few stocks, the team you support, whatever it is) and it builds a live news station around that. Hit play, you're on air.

It works like a real radio, not a podcast app. There's no episode list to manage. You press play and there's news on what you picked.

Inside the app right now:

🎙 Custom stations you build yourself

🔥 Hot Topics: ready-made stations on whatever's moving today

📅 Daily and weekly channels that show up without you asking

We've been at this about a year and shipped to early users a few months back. The product changes most weeks based on what people tell us, which is why I'm here.

Try it: 📱 iOS: apps.apple.com/us/app/podio-news-podcast-maker/id6747738443 🌐 Web: listenpodio.com

What I'd really value: spend a few minutes with it, then tell me what you'd change. If it clicks for you, an upvote means a lot to a two-person team. If it doesn't, please tell me why. That's the more useful comment anyway.

Huge thanks to @armanaksoy34 for being a relentless co-founder and to everyone on PH for taking a look ❤️

Point about scrolling news apps for ten minutes and just walking away anxious is so incredibly accurate. Endless doomscrolling is a terrible way to start the morning. Replacing that with a passive, hands-free radio format is a brilliant concept. @samet_sezer

Cool! It reminds me one guy who built something like podcast from tweets. This is also an interesting idea :)

I really like this concept. Finally an app that builds me a live station with exactly the topics I care about - no endless scrolling through feeds.

Quick question: How personalized does the content get over time? Does it learn from what I skip or listen to longer?

Came across this today and the concept is interesting. I like the idea of not having to scroll through feeds but I am a bit unsure how well it will actually match my specific interests without feeling too generic. Will give it a try and see if it actually feels personal or not. Btw good luck with the launch.

Tired of doomscrolling news feeds and wasting time?

We've built Podio so you can just hit play and stay informed hands-free.Podio turns your interests into a live, personalized audio stream:

Perfect for commuting, workouts, cooking, or deep work.
No screens, no algorithms fighting for your attention. It's early, and I'd genuinely love your honest feedback (what's working, what's missing, content quality, etc.).

Many thanks to my co-founder @samet_sezer who made this possible.
Can't wait for the feedback and comments today!

Try it free on the App Store: https://apps.apple.com/us/app/podio-news-podcast-maker/id6747738443

Hey Product Hunt 👋

I built Monkey Morse because I wanted to learn morse code but every existing tool felt like it was made in 2010 — clunky UIs, walls of dots and dashes, no sense of progress.

So I borrowed the bit that makes monkeytype addictive (clean typing test, live stats, instant restart).
Would love feedback from anyone who's tried to learn morse before, what made you bounce off?

looks awesome dude

Does Nexpend connect to bank transactions to detect subscriptions automatically, or is it manual tracking only?

congrats on the launch can i link it with my bank for example mercury and it detects subs from there ?

Quick breakdown of what you can actually do right now:

🛒 Browse the marketplace (no login needed) — jellynet.net/marketplace
→ 40 providers across 12 categories: LLMs, image gen, speech, embeddings, vision, crypto, market data

🔑 As a buyer:
→ Sign up, get a universal jn_xxx key
→ Swap your provider URL to api.jellynet.net — that's it, no SDK
→ Smart routing picks the cheapest available key. If it 429s, silent retry on the next one.

💰 As a supplier:
→ Paste your idle API keys
→ Your own calls run free first
→ Surplus earns you 60% per call served
→ 8-hour payouts in USDC or fiat 🤖 For AI agents:
→ x402 micropayments on Solana — no signup, no API key, just pay-per-call with USDC

$5 free credits on signup to try it out.

We broke down the full story in a thread here: https://x.com/jellynet_/status/2058452265095278629?s=20

Would love honest feedback — what's missing, what's confusing, what would make you actually use this. We're building in public and every comment shapes what ships next 🙏

This is a pretty unusual take on the LLM API problem. How do you handle reliability and trust when requests are routed through supplier contributed API capacity?

This looks amazing!! But also kinda overwhelming lol 😅 I'm a complete beginner, never deployed a server in my life, and I just have a blog idea. Is this too advanced for someone like me or can I actually use it?

Upvoted! Rooting for indie makers like you 🚀 One thing that would make me pull the trigger today, do you have any kind of demo or video walkthrough? I learn better seeing it in action before I connect my own server.

Tried to deploy a server and it failed twice. No clear error message, just spun and died. Not a great first impression for something that's supposed to be "simple." 👎

Tried your product, it's very easy to develop any blog/business site using it, and it's very fast with changes deployed to production.

@SlateHut.com - AI website builder made the following website in 1 shot. Check more demo here https://slatehut.com/showcase

We’re two friends who play live poker and work in tech.

Every big series, we had the same problem. Too many schedules, too many tabs, and eventually a spreadsheet full of buy-ins, bullets, Day 2s, backups, and conflicts.

So we built Poker Planner for ourselves.

It helps you plan a poker trip around your dates, bankroll, game preferences, and schedule conflicts.

For launch, new signups get one Vegas 2026 trip unlocked for free before July 1.

Would love feedback from anyone who has ever planned a live poker trip the hard way

Can you export the final schedule to Google Calendar with reminders?

I like the no-bullshit attitude here. Finally a water app that forces real accountability instead of fake streaks and cute animations.

You actually have to drink on camera when prompted.

Quick question: How do you handle privacy? Is the footage stored locally or deleted immediately after verification?

Like most of us, I had a folder full of habit apps I'd stopped opening and reminders I'd trained myself to dismiss without thinking. The notification would fire. I'd close it. Nothing would change.

That's the loop I kept finding myself in, and why I built Drink Your Water.
Passive reminders don't change behavior. Accountability loops do. So instead of another tracker, I built something that actually closes the loop:

→ You get a notification turned into a countdown
→ You go get water
→ You drink it on camera
→ The app confirms it happened

That's it. That's the whole thing.

The front-facing camera verification was the key design decision. Just enough friction to make the action real without making it annoying:

You can't fake it
You can't dismiss it
You either did it or you didn't

Start on a normal mode first. Then try Unhinged and report back.

Happy to answer anything about the behavioral design, camera verification, how onboarding works, or why your kidneys deserve better.

Ask away.

I've deleted more habit apps than I've ever used consistently.

The pattern is always the same → reminder fires → I snooze → I forget → I feel bad for a day → I download a new app.

Drink Your Water breaks that loop in the dumbest-smartest way possible.

You have to drink on camera. That's it. There's no way to fake it, no way to dismiss it and move on.

I tried the normal mode first. Then Unhinged.

Unhinged got me.

If you've ever ignored 47 hydration reminders in a single day (guilty), this one is actually different. Give it a shot.