PH热榜 | 2026-05-23

RANT ALERT

The whole GEMINI 3.5 (trash) is SO Bloody DISAPPOINTING, I used to be such a MASSIVE Antigravity lover.

I was honestly excited leading up to I/O to see how they take their solid foundation and build on top of it. But look where we're now.

It's honestly a Joke that with all these resources, they CHOSE to make the Fastest Slop Generating model.
Don't get me started on Antigrav CLI, it used to be good as a Open-source project and i had faith, until they close sourced it and made it astronomically shittier and broke most of the developers hearts.

With the way the Input box dances around when generating, makes me think they used 3.5 Flash for it :')

They didn't leave AntiGravity IDE Alone either, and chose to take a fat shit on it too. WHY DO I NEED 2 APPS? One Codex clone (But worse) and an IDE Which is just mid.

What can we expect from sellouts like windsurf founders, who dropped being one of the only innovative competitors to cursor, to JOIN the DARK SIDE OF TRUST ME BRO BENCHMARKS and STAGED DEMOS (like running doom) [IF they're spear-heading this they're the ones to take the blame for it]

ON TOP OF THIS SHIT SHOW, They have the Audacity to LIE ABOUT 3x Limits in the IDE, they removed the PERFECTLY FUNCTIONAL Browser use which gave video walkthroughs and Debugged (GOATED FEATURE)
and replaced it with /browser which needs to be configured separately and STILL doesn't work.

And wait, THE SLIMIEST THING YET, this boils my Blood as a Die hard Anti Grav user. They CLAIM 3X limits, and STILL CLAUDE Doesn't last a SINGLE PROMPT and being on the pro plan doesn't really mean much.

Also, Can someone from @Google Antigravity tell me WHY Does the GPT Model and other ancillary models offered by AntiGravity, run out when i've never used it? (Eg: I never touch sonnet, but after one prompt to opus; GPT sonnet and G3.1 Pro run out)

IF YOU HAVE SUCH A SURPLUS OF TPUs AND MONEY TO DISH OUT TOKENS on SLOP WHICH MOST SERIOUS USERS NEVER ASKED FOR. Rather allocate more tokens to developers / creators / entrepreneurs who actually use your tools, INSTEAD OF BURNING THEM ON A F*CKING BLACK HOLE SIMULATION and Information agents which is going to serve me ADS and not REAL information...

Such a Disgrace and waste of resources and a borderline act of bad faith to its users...

One more thing i cant understand is HOW DOES A TRILLION DOLLAR COMPANY BOASTING ABOUT A QUADRILLION TOKENS SPENT doesn't make good use of it and Blow every competitor out of the water? (Total BS number btw cuz its all being force fed to people with AI MODE which is a Hallucination and machine)
Almost makes me think that its a choice, so they don't get summoned by the Senate for having a Monopoly on this Market too (then its a 5D Chess move)

Most people are gonna be out here glazing google, and i might get banned or flagged for being honest and real. My only question is, Who's to be blamed for this?

AI psychosis is real, and the state of the frontiers is a GREAT EXAMPLE OF IT.

and yes i'm calm, and this is not hate, but frustration and pain from a REAL USER.
Thank you for your attention to this matter @rohanrecommends @joshtwoodward @logankilpatrick @sundar_pichai

Antigravity CLI brings full agentic coding capability to the command line, running on the same core engine as the Antigravity 2.0 desktop platform.

Agentic coding tools have mostly been designed around visual interfaces. That works until you're on a remote server or running SSH-heavy workflows where a GUI simply isn't in the picture. The CLI closes that gap without asking you to change how you work.

• Multi-file editing and multi-step reasoning directly in your terminal, handling complex tasks without a graphical layer

• Shared agent engine with Antigravity 2.0, so improvements to the platform apply automatically to the CLI too

• Keyboard-first and optimised for remote sessions, with minimal resource overhead

• Bidirectional settings sync and session export, letting you pick up terminal work in the desktop app when needed

• Auth via system keyring with SSH-aware fallback built in

• macOS, Linux, and Windows support

Built for developers who primarily work in the terminal and want agentic capability without switching surfaces. Give it a try and drop your experience in the comments below.

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

Developer workflows are moving from commands to intentions. Interesting glimpse of where terminals are heading.

I dont get it. How is this different from just normal antigravity? i download it i get antigravity. is there a curl to instlal the antigravity cli or smth? Found it. its bellow normal antigravity download area. anyways here is the curl: curl -fsSL https://antigravity.google/cli/install.sh | bash

Fixed half a duplication bug then this: "Individual quota reached. Contact your administrator to enable overages. Resets in 167h39m7s." 😅

I just woke up and opened Product Hunt to see what’s launching today. Memdex caught my eye so I went through the homepage, and the way you described the context detection flow felt clean. Great work! By the way I lose my best AI chats all the time. Scrolling through the history is a nightmare. So this idea... I like it. I have a question though. Since everything stays on my device, how does the matching actually work? Is it just looking for the same words, or does it understand the meaning behind them? Either way, cool product. Rooting for you.

The idea of memory is extremely helpful. I use many AIs, yet heavy on Claude. instead of copying and pasting prompts or ask it to summarize it then paste it in a different AI, phew!!! this alone is brilliant "you open a fresh Claude window and it already knows your project".

"Reusable local memory" is the part most AI tools paper over with "your data is private" wording without actually solving the locality problem. Curious which decision was harder — the storage shape or deciding what NOT to capture. That second one usually defines how a memory tool ages.

Hey Andrew! Sounds obvious but it actually isn't. Also jumped between all of them and know how hard is to set again the context. I'm sure ti's gonna change the game. Wish you all the best!

Reusable conversational memory is one of the biggest unlocks for practical AI workflows.

Good idea and implementation! Congrats on this launch!

First of all congrats on your launch!
I just checked your website and landing page storytelling is great which is usually first thing that caught my eye in any new tool and something i spent good time to deliver it in my own tool as it is the first impression for any new user.
I use Claude a lot and many times it misses the past convos even though i ask it to update the memory but in fact it still repeats the same old information. Does Memdex helps solving this issue?

The Grammarly-for-context framing is strong. The place I’d be careful is separating durable memory from stale task state.

Some things really should follow me everywhere: product constraints, preferred wording, customer language, recurring project facts. Other things are only useful for this week’s thread and become confusing later. I’d love a lightweight way to mark injected context as “keep reusing this” vs “only for this task/session,” plus a small preview of why Memdex thinks a prior chat is relevant before it inserts it.

Writers and researchers don’t want more features. They want less friction between thinking and writing.

Hey @andreaigner This is really an intuitive product. I’ve actually struggled with this a lot myself, so I’d love to try it out. Any plans for a Windows version?

feels like a very real pain point for anyone doing deep research. The work gets fragmented across PDFs, notes, citations, browser tabs, and scattered thoughts. The local AI angle is especially interesting because research needs trust and grounding, not just fast summaries.how notemd handles evidence quality when finding a source for a claim, does it distinguish between strong support, weak support, contradictory findings, and loosely related papers?

Congrats on the launch!! This feels clean and useful for people who want seamless and free writing. A local first setup also makes it more trustworthy for focused work and research.

There’s real craft in making nostalgia feel modern instead of gimmicky

Love this idea. I can't live without music, you hit the nail!

This scratches a weirdly nostalgic itch 😄 Curious, what was the hardest retro effect to recreate authentically without hurting performance too much?

I’d turn this on just for the atmosphere while working!

Hey Product Hunt! 👋

I'm Alex, the maker behind Bulkmark.

Like most of you, I'm a chronic bookmarker on X. Every interesting thread, every useful tip, every "I'll read this later" tweet… saved. And then? Never opened again.

My bookmarks folder became a graveyard of good intentions.

I tried Notion exports, read-it-later apps, even spreadsheets. Nothing stuck. The friction was always too high.

So I built Bulkmark, the tool I wish existed.

Here's what it does:

1. 🔄 One-click sync + auto-tagging

   Connect your X account and Bulkmark imports every bookmark you've ever saved, then tags each one by topic automatically. No manual sorting. Ever.

2. 💬 Chat with your bookmarks

   Ask anything. "What did I save about pricing?" "Show me everything on AI agents." Your bookmarks finally answer back, with the original tweets cited.

3. 📋 Curated lists you can share (or not)

   Group your best finds into public lists. A reading playlist, a founder collection, a launch playbook. Build it, name it, share the link.

4. 📩 Friday digest in your inbox

   Every Friday, a clean email with the best of your week worth revisiting. No app to open. No feed to scroll.

---

Why I built it this way:

I didn't want another dashboard begging for attention. Bulkmark is designed to feel like a quiet reading room. Warm, focused, calm. It works in the background, then delivers value when you actually have time to read.

What's next:

- Notion export

- Resurfacing (bookmarks you saved 6 months ago, brought back at the right moment)

- More integrations (LinkedIn saves, Reddit saves...)

---

I'd love your feedback. 🙏

Specifically:

• What's missing for you to ditch your current bookmark workflow?

• Would you pay for resurfacing? (re-reading saved content on a smart schedule)

• Any integrations you're dying to see?

Trying Bulkmark is free for 7 days. No credit card upfront. If you find a bug, ping me directly, I'll fix it the same day.

Thanks for checking it out.

Let's clean up those bookmarks together.

— Alex

The Friday digest feels like the most important part to me, because it changes bookmarks from storage into a review habit.

For resurfacing, I’d pay more for “why now?” than for pure scheduling. If Bulkmark can say “you saved three posts about pricing, one about packaging, and this older thread contradicts the new one,” that’s much more useful than reminding me a bookmark exists. A small “use this for…” note when saving could also help the AI sort inspiration, reference, competitor intel, and action items differently.

Hi everyone!

Apache 2.0 on Command A+ is a big deal!

This is @Cohere’s fastest and most powerful model yet. It uses a 218B MoE architecture with 25B active parameters, built for enterprise agents that need reasoning, tool use, vision, long context, and multilingual workflows.

Command A+ can run on as little as two H100 GPUs with W4A4 quantization, which is pretty impressive considering its overall performance across agentic, multimodal, and multilingual tasks.

Love how they put it:

We want to give developers direct access to enterprise-grade agentic capabilities from experimentation to production.

Kudos!

Open source AI workspaces feel like an important counterbalance in the ecosystem.

how does Cohere perform on long-running agent workflows, not just isolated benchmarks. When an enterprise agent has large context, tool calls, retries, vision inputs, and multilingual reasoning in the same run, where do you see the biggest strength, reasoning quality, latency, cost efficiency, or reliability?

We built Forsy after seeing how quickly AI agents are becoming part of serious everyday work.

But most of the data created inside that work is never captured as a useful asset. People are still trying to sell prompts, when the real value is in the workflow data behind the run.

Forsy captures workflow data from the agents you already use (OpenClaw, Claude, Codex, Hermes, etc.) and turns it into sellable structured data. It creates a marketplace for authentic, high-fidelity workflow data with licensing and privacy built in.

We want to help people get more value from the agents they already use, while building the infra for a new agent data economy where authentic, high-fidelity workflow data can be discovered, licensed, and sold.

Interesting concept. The one thing I'm trying to wrap my head around, who's actually buying this workflow data? The big AI labs training their models, or regular companies? Curious who's on the other side of the marketplace.

The "sell your workflow data" framing is interesting because the bottleneck has never been "is the data valuable" — it's "what's the consent layer that doesn't feel like 17 dark-pattern checkboxes." Curious how you landed on the buyer-side flow.

Actually useful. Is it contextual? Like, does it know when you're working on project 1 to use MCP 1 and on project 2 MCP 2?

nice! does it work with windows? I mainly use mac for work, but still sometimes switch to windows for creative work, and connecting mac as a second monitor was always difficult with third party software.

feels like a very honest builder story, you needed a calm, private outliner that worked the same way across Mac and iPhone, so you made one.I like the no-account/no-server approach. Outlines often hold unfinished thinking, not just polished notes.what was harder to get right: performance with large image-heavy outlines, or keeping the experience simple across both Mac and iOS?

Very timely launch. Every mac is half open now.

Where are the leads coming from, and how do you decide the score is actually a fit for that niche?

Generating keyword and metadata variants inside the same panel is super interesting. Does Vibegrowing pull real-time App Store search volume and competition metrics to suggest these keywords, or is it purely using semantic LLM generation based on our app description? great work @melek_kurnazarslan

Wow looks great. It chose the keywords very well. Also designed the logo and screenshots. It really deserves the name ✅.

Heyy PH! I'm Melek.

I'm an indie iOS developer who has launched multiple apps on the App Store Gluten Scanner, CosmicSynastry, Breath, and others.

Every launch taught me how painful and time-consuming ASO is when you're doing it alone.

We didn't hire a team to build vibegrowing we built it myself using AI-native development workflows.

That means I ship fast, stay lean, and deeply understand the pain I'm solving. I've lived this problem on every single launch.

Quick question for the indie devs here:
After finishing your app, how long does it actually take you to get it live on the App Store? (Counting metadata, screenshots, landing page, legal all of it.)

Curious how much time vibegrowing could save your next launch.

Pre-launch grind is the real killer for solo devs ..you're solving the right bottleneck. How deep is the ASO keyword data? Pulling from App Store directly, or LLM + third-party?

Four years ago I started building Krater alone, on the side, as a copywriting app with templates powered by an early GPT model. That was it. Output quality was rough, but it was the first time AI felt useful to me, so I kept going.

I'm Malte, the founder. As the technology got better, the product grew with it. I added text to speech, then image generation, then video, then more models from every major provider as they came out. Every step was driven by user feedback and what was suddenly possible that month. You can actually see the journey in the older Krater launches below this one, going all the way back to 2023. Since the start of 2026 there have been four of us full time, and we've spent the year rebuilding everything around one realisation:

The problem with AI isn't the models. It's the gap between trying AI and actually using it. People sign up to ChatGPT, get impressed, and a week later they're back to doing things the old way, because using AI well means juggling five tools that don't talk to each other.

Krater is still the place where every AI model from every provider lives in one app. What's new is the agent on top of them. It uses every model and your connected apps to actually finish the task, and it can run on a schedule so the work keeps happening without you. Research goes into your Notion every Monday morning. Images land in a website it just published. A daily summary of your inbox waits for you at 8am. Same chat box, completely different product.

70,000+ people use it today. The description and images above covers what it does. What I'd actually love from you: tell me the task you keep avoiding because AI almost helps but not quite. That's what we're building toward next.

Malte

We are so happy to be launching today and seeing everyone's support. Thank you so much!

Hey guys, thank you so much for checking out our launch! It means the world to us.

I have been using Krater for a little over 3 months now and i love it. It feels very intuitive even for non AI users with it’s simple / commands. But also has features like API, Teams and Tasks for when you start learning more. So there is always something to unfold.

As of now i use krater daily. Mainly just for chats but also to get my news and email briefs with my morning coffee!

Amazing product

Thank you all for checking out our launch!

Do you let people pick and pin their preferred models per task, or does the agent auto choose everything?

Four years building one problem — that's how you hit 70K users. What kills retention most — habit pulling people back to ChatGPT, or missing features?

Had I known there is no iOS app for Krater on my iPad (like there is for Perplexity, Grok, and others), I would not have bothered with the one year subscription. In fact, if there's an option to cancel/refund my subscription (it has been active for 2 days), I would do so.

Curious about silent failure. If a Monday research run quietly drops a garbage summary into Notion four weeks running before anyone checks, what catches it? All in all I think the scheduled-agent piece is the wedge here. Congrats on launching a decent product!

Damn guys! This is some amazing work, seems like something sorely needed in a market of cheap AI rip-offs. Something thoughtfully put together with a real purpose beyond just gobbling up people's money!

What are the current limitations of your services, and is there any such one that's harder to improve than others?

I'm not all too technically gifted but I would love to know more. I'll be giving Krater a good run to learn more about how it works, looking forward to using your services!

This is such a neat idea, Shadow Mode especially. Really lowers the barrier to just trying it out.

One thing I'm curious about though, how does it handle queries that are semantically close but mean the opposite? Like "which foods are good for high blood pressure" vs "which foods should I avoid for high blood pressure" these would probably sit pretty close in embedding space but serve completely different answers. Does the validator catch that, or is this a known edge case you're still working on?

I kept hearing the same thing from realtor friends: their CRM was a glorified spreadsheet. Lead comes in, sits in a tab, gets cold, gets lost. So I built the opposite: a CRM where the AI does the work. A new Zillow lead hits Gmail, Agentype parses it within a minute, drafts a personalized follow-up, finds matching listings, suggests viewing times. The agent reviews and approves while they're out showing properties.

The hardest part wasn't the AI. It was the boring stuff; parsing 23 different lead-notification email formats, surviving Gmail rate limits, and earning enough trust to send on someone's behalf. So every automation starts in draft mode. You flip auto-send on once you trust it.

Also wired up an OpenAPI spec so you can manage everything from ChatGPT or Claude. Ask "who are my hottest leads this week?" and it answers from your live pipeline.
Would love feedback from realtors and anyone building AI on top of messy real-world workflows.

This feels more useful than a “better CRM” pitch. Do agents mostly manage the pipeline through chat, or is the dashboard still the main workspace?

Hey, I would request anyone who are seeing this launch, just give one try we hope you will like our product if not just tell us how can we improve, Thank you for your time

We've been working on for months on Tross and now let's go! Try and share your reviews please.