PH热榜 | 2026-07-10

Congrats on the launch!

100k builders on an open source tool is genuinely impressive.

BTW, what's the split between solo devs and actual teams using this in production?

Apache 2.0 and soc2 out of the gate is a wild combination for a new open source agent project. usually security teams block this stuff instantly. congrats for launch @emirkarabeg 🙌 qq can i run the entire stack locally via docker-compose and still keep the visual canvas working?

love that files and tables live in the same workspace layout, context switching between three different DBs and vector tools completely breaks agent memory. amzing launch

1,000+ integrations is a lot to keep working. Are those maintained in house or community contributed?

Love the emphasis on replacing unnecessary LLM calls with deterministic code instead of assuming everything needs an agent. Just Curious have you noticed users converging toward a small set of reusable agent patterns over time, or is every team's workflow still highly bespoke? Congrats on the launch! 🚀

Hey Emir! Congrats on the launch. Love open source projects and even more if it's backed by YC. Wish you all the best here

Love that you're supporting multiple integrations instead of reinventing every connector . That should make it much easier to fit into existing tech stacks.

This looks promising! How does Sin handle multi agent collaboration ? Can different agents here context and coordinate on the same workflow?

who do you think gets the most value out of Sim today, developers, technical teams, or can non-technical users get productive quickly as well?

The "chat to build" vs canvas vs code approach is interesting, most tools force you to pick one. Which one do most of your users actually end up sticking with?

@emirkarabeg The per-node flexibility between chat, canvas, and code is what other tools get wrong. Curious, which mode do most of your users end up living in?

Open-source is the interesting bet here, most workflow

tools in this space are closed, which means you're

locked into their roadmap and pricing forever. Curious

how you're thinking about sustainability though, is

this open-core with a paid tier down the line, or

fully community-driven?

1,000+ integrations out of the gate is a lot to

maintain well. How are you handling the ones that

break when a third-party API changes, is that on the

core team or does the community help maintain

specific connectors?

Congrats on shipping this, open-source infrastructure

tools like this are genuinely valuable even when they

never get the attention flashier consumer apps do.

Congrats Emir!
Most agent builders just throw LLM calls at everything and the bill quietly becomes a problem at scale. How does Sim decide which steps get deterministic code versus an LLM call, is that the builder's call or does Sim suggest it

I use Sim every day, it's fantastic

Quick question: when an agent workflow fails in production, how do you actually debug it? Like, is there a trace/replay view, or are you digging through logs across the different integrations it touched? Anyways, congrats on the launch!!

connected my OpenAI agent to Slack and Notion in like 10 minutes, the drag and drop flow editor feels really natural

Open source is the reason I'd actually build a workflow on this instead of renting one that vanishes in a year. When a step in the middle of a run fails, does Sim let me resume from that step or does the whole thing start over? Re-running everything from the top is what made me drop the last couple orchestration tools I tried.

I like that you're treating AI agents less like individual tools and more like a coordinated system. As companies adopt more agents, I think orchestration will matter far more than adding another capable model. The value shifts from what a single agent can do to how reliably they work together.

Congratulations

i like the focus on production ready AI agents instead of simple demos. how does Sim monitor agent health and recover from failed workflow executions?

Played around with it for an hour and honestly the drag-and-drop agent builder feels really snappy, hooked up a Slack-to-Notion workflow in like ten minutes without writing code.

How does the pricing work for an open-source project, especially once you start scaling beyond a handful of agents and hitting those 1,000+ integrations?

@emirkarabeg The per-node flexibility between chat, canvas, and code is what other tools get wrong. Curious, which mode do most of your users end up living in?

The shared memory across agents in the same workspace is cool 👍 can you also set boundaries so certain agents only access certain data, or is it all-or-nothing?

This is so needed - agents working in isolation just aren't as effective!

this is super cool!

Hey PH 👋 Udaya here, founder of PlugThis. We're a team of four in Bangalore.

Till last year we were building AI products for large enterprises. Then the AI builder wave hit, and tools like Lovable, Bolt, and v0 genuinely inspired us. I built web apps, dashboards, and visualisations with them, and they performed incredibly well.

But when I tried building Chrome extensions with them, it just didn't work. Extensions have a completely different architecture (manifests, content scripts, service workers) and these tools weren't designed for that.

I've been building extensions since 2015 in my first startup, so I know this pain personally. Extensions are the most personal software there is 🧩

We spend 7 hours a day in a browser, and everyone of us has some 15 minute repetitive task they wish would disappear. But there was nowhere to go to build a simple extension for ourself.

We searched the market for something like this. Couldn't find it, so we built it.

PlugThis builds Chrome extensions from a plain English prompt. Real Manifest V3 code that you own, and you can wire up a Supabase backend to make it full stack.

One from my own browser: an extension that reads a Twitter profile as I scroll, researches the person, and scores how close they are to my ICP. Took a few prompts ⚡

For launch week, the Builder plan is at the Starter price, 3 days only 🎁

We are here through the day. Try it out, and tell us what you liked and what you wish it did.

And if you build an extension for yourself today, please share it in the comments. That's honestly the whole point of this product 🙌

I've had a few Chorme extension ideas setting in my notes because getting from an idea to working extensions always felt like too much work. i really like that you're focused on one problem instead of trying to build everything for everyone.

I've noiced there are plenty of AI tools for websites but not many focused on broswer extensions. that focus alno makes Plug this stand out to me.

always wanted to experiement with Chorme extension but the setup process kept pushing me toward others projects. this makes the whole process feel much more approcahable.

The versions history caught my attention. i've learned the hard way how valuable it is to roll back when an experiment doesn't work.

@nefer_ai I like that this is built specifically for Chrome extensions instead of trying to cover every kind of app. Sometimes focusing on one problem is what makes a product really instead out.

I appreciate that every change is versioned. It makes experimenting feel much less risky. Knowing I can always go back encourages me to try more ideas.

Firefox or Edge support planned, or Chrome-only for now?

Solid product. I've used this to create a few of my Chrome extensions to make my day to day work easier.

Although people with AI dev tools can create their own, I think everyone (even non-devs) needs to make their own chrome extensions to make their workflows easier, that's where this just works.

With PlugThis sitting under Chrome Extensions, AI Workflow Automation, and AI Agents, how much of the extension-building process are you trying to cover? Is the goal prompt-to-working extension, or more like generating the boilerplate and helping refine pieces such as popup UI, background scripts, and permissions?

Guys, you are amazing. I will try this later because I am building my own (tho manually because I set the goal for myself to learn coding and programming). But this one tool is brilliant! :)

Good idea. It would be even better if your service somehow helped with the promotion and marketing of extensions, because this is much more difficult than development, especially in our time

What happens when Chrome changes a manifest requirement?

BYO API key means full ownership, but no free AI features out of the box. Real cost upfront.

the manifest/content-script/service-worker mismatch is exactly why lovable and bolt hit a wall on extensions. the abstraction they picked works for pages not for browser plumbing.

real q: how does plugthis handle the distribution side? chrome web store trust is the second wall after the code works. asking because building the extension is now the easier half of shipping one.

me thinks version history is valuable but can developers compare changes visually A side by side view would help greatly.

This is a great idea. Does the app publish to chrome store directly post building ?

Congrats on the great launch! QQ - what about permissions requested at runtime through chrome.permissions.request or optional_permissions? Those probably almost never show up in a static call graph, so does the analyser read them as dead and nudge you to drop something you only ask for on demand?

@nefer_ai Tested it today and it's genuinely fun to use. I expected to spend time fixing boilerplate, but it got me to a usable extension much faster than I thought. If you keep improving the generation quality, this could become a go-to tool for anyone building Chrome extensions. Congrats! 👏

+1 to @thenameisarian on the browser-plumbing wall — that manifest/service-worker gap is exactly where the web-app builders stall. @nefer_ai your distribution answer already covers the Web Store side, so I'll ask the two things next to it:

Extensions can't be previewed in an iframe the way a web app can — you load them unpacked and click through real pages. How does PlugThis close the build→see-it-actually-working loop while you're iterating?

And on review: does the generator scope host/API permissions down to just what the feature needs, or is trimming an over-broad manifest still on the human before submitting? Least-privilege is usually what makes or breaks that first review 👌

The focus on one narrow thing is what stands out here, since Lovable and Bolt genuinely do fall apart once you hit manifest files and content scripts. Curious how it handles permission scoping, since that's usually where generated extensions either ask for way more access than they need or break silently when Chrome tightens something. Also, since the code is yours to download, does switching between chat refinement and hand editing the code directly cause any drift, or does it stay in sync both ways.

I like that you picked a narrow but painful category. Chrome extensions look small from the outside, but manifests, content scripts, service workers, permissions, store rules, icons, screenshots, and listing copy make them very different from a normal web app. as a builder, I can immediately think of a few annoying browser workflows that would be perfect as tiny personal extensions.

Curious how PlugThis handles permissions and Chrome Web Store review risk. if someone asks for an extension that reads pages, injects UI, or works across many websites, does it guide them toward safer/minimal permissions automatically?

This looks cool! Does it support complex extensions with multiple pages?

"Complex extensions from plain English" is bold. Where's the actual ceiling here?

I enjoy the idea of creating Chrome extensions without extra setup but how do you test compatibility across different Chrome versions Providing built in testing tools before publishing could help developers catch problems earlier with confidence.

I like the focused approach but how do you handle Chrome permission requests Maybe adding permission explanations could improve user trust.

Congrats on the launch! 🚀

That "15-minute repetitive task" line got me. Mine is copy pasting snippets from a bunch of open tabs into one notes doc, one at a time. Something that just tossed every highlight into a single running note would genuinely save me.

Nicely done!

Does the generated extension actually get reviewed and approved before hitting the store, or am I on the hook for submitting it myself?

I'm Alima, co-founder of ChatCut. I’ve spent years producing ads, documentaries, and branded content, and editing was always the most painful part of the process.

When we started building ChatCut, we thought we were building for editors. But pretty quickly, we realized the pain was much bigger than that. Businesses, YouTubers, TikTokers, teachers, doctors, founders — everyone in the world is making video now, and almost no one enjoys editing.

Editing software is powerful, but it still feels stuck in another era. As video became part of every job, we felt there needed to be a different kind of editor.

We spent a lot of time thinking about where AI actually belongs in video editing. Not as a one-click generator, and not as a gimmick on the side, but as an editing assistant that understands your footage, your intent, and the timeline.

That’s what we’re building with ChatCut: a full video editor with an autonomous AI agent, a professional timeline, built-in stock footage, sound effects, transitions, and built-in generation tools like Seedance 2, GPT Image 2, Nano Banana, ElevenLabs voices.

You shouldn’t need ten different platforms to finish one video. And AI should help where it’s actually useful, while humans keep control of the story and the final cut.

AI won’t replace human creativity, but it will change how we work. We’re building ChatCut to be part of that change.

Every account comes with 20 free credits and we're giving 10% coupon to Product Hunt users: www.chatcut.io

I’m Kaiwen, co-founder of ChatCut. Before building this, I spent years making documentaries, commercials, and short films for VICE, Discovery, and brand clients.

The thing that bothered us was simple:

AI was getting very good at generating images and video.

But editing was still stuck in the old workflow.

You could generate more footage than ever, but turning raw material into a clean, watchable video still meant scrubbing timelines, cutting dead air, finding the good moments, adding captions, making motion graphics, looking for B-roll, fixing music, and repeating the same tiny operations again and again.

And most “AI video editors” were not really editing. They were removing filler words, adding captions, or generating pixels.

That is useful, but it misses the real job.

Editing is not just removing the bad parts. Editing is deciding what the story is.

So we built ChatCut as an AI video editing assistant that works inside a real editor.

You can upload footage and ask it to:

• creatively edit from your directions, prompts, or script

• clean up talking-head videos and turn them into social reels

• generate style-consistent, editable motion graphics

• create B-roll, images, music, and voice

• apply or even generate custom effects and transitions

• make timeline edits you can still manually tweak

The important part: ChatCut does not trap you in a template. It gives you an editable timeline.

Our bet is that the future of video editing is not “press one button and accept whatever AI gives you.”

It is closer to working with a very fast assistant editor: you explain the intention, it does the rough work, and you stay in creative control.

ChatCut is already useful for talking-head creators, founders, educators, marketers, ecommerce teams, TikTokers, internal training teams, and many other people turning raw footage into publishable video.

It is also still imperfect in exactly the ways a real AI editing agent is hard: taste, timing, reliability, and performance with long or heavy video files.

That is why we are launching here.

If you make videos, I’d love for you to try ChatCut on a real piece of footage and tell us where it feels magical, where it still feels dumb, and what would make it part of your weekly workflow.

My co-founder and I are here all day answering questions. Honest feedback wanted.

Hey everyone! I'm Finn, co-founder of ChatCut.

Before this, I built a video editor used by hundreds of millions of people. My co-founders have worked on top-tier commercials and short films.

Still, video editing felt harder than it should be.

A large part of video editing is the execution work between an idea and a finished video: managing and searching through lots of footage, comparing different cuts, and creating all the post-production polish.Traditional editors are powerful, but they are still built around professional workflows and can take a long time to learn.

So we built ChatCut.

ChatCut is an AI video editor. You describe what you want, and ChatCut's agent helps you edit the video. It can understand your footage, find good takes, clean up talking-head videos, add motion graphics, captions, music, transitions, effects, and even generate image or video assets when needed.

So far, we've seen people use ChatCut for:

- Talking-head videos
- App promo videos
- E-commerce product ads
- Explainer videos
- AI short films

These are just examples. You can also explore many other video ideas with it.

We also built ChatCut for different types of creators.
-If you don't know how to edit, you can edit entirely by talking to the agent.
-If you want to fine-tune things yourself, ChatCut also has a full video editor. You can adjust everything manually, or export XML and continue editing in Premiere Pro or DaVinci Resolve.

ChatCut is available on the web and desktop.And if you use Codex, you can turn it into a video editor with one prompt:

"Read chatcut.io/chatgpt to install the ChatCut plugin and set up a new task for me."

It installs the ChatCut plugin and lets Codex help you edit videos too.

We think video editing should feel less like busywork and more like shaping ideas.Just like vibe coding made it easier to build software, we hope chatcutting can make video creation easier too.

We're still early, and we'd be happy to hear your feedback and keep improving from there.

Thanks for taking a look.

A lot of AI video tools feel impressive until you want to change one small thing and realize you are basically stuck with the output. ChatCut sounds much closer to how editing should work with AI: explain the intent, let the assistant do the heavy rough work, then still keep control over the story, timing, captions, B-roll, and final cut.

As someone doing more product/launch content lately, this feels very relevant. recording footage is not the hard part anymore, turning it into something polished and usable is.

Curious how well ChatCut handles taste over time. if I keep adjusting its edits, does it learn my pacing/style preferences, or is each project mostly independent?

Congrats team! I've always felt video editing software has fallen behind the way people actually create today.

AI tools have exploded over the past year, but most don't solve the part I actually dread. I've spent entire nights cutting interviews with hundreds of clips, and the bottleneck was never generating content. It was making good editing decisions without getting buried in manual work.

I like that ChatCut isn't trying to hide the timeline or replace the editor. Also no more learning curve or navigating endless pop ups of dropdown in Adobe.

Curious how everyone thinks about the tradeoff here. Is the biggest opportunity for AI video tools making editing faster, or making the editing process itself more creative by removing the repetitive work?

Congrats on the launch! Really like the idea of combining an AI editing agent with a real editable timeline.

Quick question: how predictable is the credit/token usage in a real editing workflow? Are credits consumed mostly by generation tools like video/image/voiceover, or also by agent actions such as analyzing footage and making timeline edits? Is there a cost estimate before execution?

Really like the philosophy that AI should assist the editing process rather than replace creative decisions. Keeping every AI edit editable on a real timeline feels like the right balance between automation and control. I'm curious what's the most surprising editing task users consistently choose to do themselves even when ChatCut can automate it? Congrats on the launch! 🎬

Super excited to see the ChatCut launch today! I've been a quite early user of ChatCut and I really like the idea of bringing the power of coding agents to content creation.

As a video creator myself I still remember the days I tried hard to learn After Effects 15 years ago. If you're also into old school video editing, the experience with getting over all of the motion graphics stuff by interacting with an LLM agent is just magical. And what's more, if you have tried to build you own agent or maintained production agentic systems, you'd realize that the amount of challenges to make text to code to motion graphics is truly incredible.

The new Codex plugin feature also represents a fascinating direction where future creative tools will be built for humans and agents alike. Congratulations and looking forward to what are yet to come for ChatCut!

This feels like the direction video editing should be going in. Less repetitive work, more time spent on the actual content.

Love the "humans keep control of the story and the final cut" principle. In practice, after the AI does a rough cut, how much freedom do I have to fine-tune it on the timeline — is every action it takes undoable and editable step by step?

The "save your editing workflow as a reusable skill" feature is the most interesting part of this to me. That's basically the same pattern a lot of us are using with Claude Code / Codex for repeatable dev workflows, just applied to editing. Once you save a skill, does it generalize across very different footage types, or does it start breaking down if the new footage doesn't match the original pacing/structure it learned from?

The autonomous agent plus a real timeline is what would pull me over from CapCut. As a solo creator the thing I'd test first: when the agent makes a cut or drops in a transition, does that land as an editable clip I can nudge frame-by-frame on the timeline, or a locked AI block I can only regenerate?

This tool is very efficient.

Congrats on the launch! Excited to see where this goes!

This is one of the cleanest AI video editing concepts I've seen on Product Hunt this year.

Most AI editors try to replace creators. ChatCut feels like it removes the painful parts of editing while keeping creators in control. The fact that it works across ChatGPT, desktop, and web makes the workflow incredibly interesting.

The editable timeline + XML export is what stands out to me — most AI editors hand you a locked render and you're stuck. As a music producer, my one question: when I drop a track in, can it actually cut to the beat / align B-roll transitions to the transients, or is the music just a background layer? That's usually where these fall short for visualizer and promo content.

"Real timeline, XML export when you want to keep working elsewhere" is the detail that stands out to

me. I've spent today wrestling with a different editor that made a bunch of decisions for me I

couldn't easily undo or take elsewhere, so a tool that respects that you might outgrow it, or just want

a second opinion from another editor, is a real, underrated feature. The "understands your footage, intent, and timeline" claim is the ambitious part. Curious how it handles genuinely messy, unplanned footage, like a founder

just talking into their phone with no shot list, versus more structured content, does it need some minimum organization to work well? This would've saved me a few headaches today. Nice work.

The biggest bottleneck in content creation isn't generating ideas anymore—it's editing.

ChatCut is tackling exactly that problem by making professional video editing accessible to everyone, not just editors. The editable timeline and XML export are especially impressive because they fit into real production workflows.

Huge congratulations to the team. This deserves attention today! 👏

Love AI tools that let humans be still in charge

I cut a lot of product demos and the part that eats my afternoon is trimming dead air and lining up captions. When I tell ChatCut to "cut the pauses," is it working off the transcript timing or actually detecting silence in the audio? Those two go wrong in really different ways once a clip runs long.

How do you work with 4k / 8k footage streams?

How does the AI handle hour long or longer videos? I want to cut an hour long podcast for example and multi-scene / multi-angle.

Editing video directly inside ChatGPT or via web is a massive time-saver for Shorts/Reels creators. Hard to beat CapCut, but the workflow integration is a huge plus.

Best of luck with the launch, team! The XML export is a great touch for flexibility. Curious, what's the ideal point for someone already using Premiere to bring ChatCut into their workflow?

Congrats on the launch, ChatCut team!

I first saw Kaiwen Li’s product video on Xiaohongshu earlier this year, and the concept immediately felt exciting. I also really liked his video style as a director.

I’ve been looking forward to ChatCut’s official launch and commercialization, so it’s great to see it live on Product Hunt now. Big congrats to the team!

Really like that every AI edit stays editable on the timeline. Congratulations!

how does it actually figure out where to make cuts on its own, and can it handle longer interviews without me having to guide every single edit?

Tried it on a messy interview clip and the auto-cuts actually matched the pacing I wanted, which I did not expect. Liked that the timeline stayed editable after instead of locking me into AI choices.

Congratulations on the Launch @ChatCut ! Super exciting to see such a powerful tool that's also so convenient! Does this mean that I can have ChatGpt do additional contextual research on my videos and provide more depth in the motion graphics?

Is API available?

Hello PH fam!

Long before there was Claude Tag and Victor, there was Scarlett.

We built Scarlett over a year ago as a fun experiment to see what agents could do. To our surprise, she almost immediately started doing probably 90% of our work for us.

We've been reluctant to launch because we knew the competition would be fierce, but our benchmarks show that Scarlett is more than holding her own on both speed & quality.

She also has a host of unique qualities & features:

1) She Just Works. After a year of experimenting with various models, architectures & backends, she just works (and does work).

2) Autopilot. She is trained on 50+ business/growth books, you have the option to use her passively or to set her free to essentially run your whole company from marketing to support.

3) iMessage. A lot of people (solopreneurs) don't like or need slack, so we have an alternative.

4) Use Our Keys. With OpenClaw, Claude, etc you need to buy subscriptions to the things you want to use like HeyGen or XAI, but with Scarlett you can use our keys and we'll pass along the cost.

5) Right Model, Right Job. You can use Opus for chat, Sol for Coding, Fable for design, you pick the right model for the job.

We hope you love her as much as we do!

Cheers,
Dan

Love how the landing page cuts straight to the point with "real colleague" framing instead of leaning on the usual AI buzzwords. The copy feels like it was written by someone who actually understands team workflows, not just tossed into a prompt.

Passing API costs through a single source removes a lot of friction for developers, it keeps the focus on building instead of managing keys and billing. Great work Dan Sutera this looks like a fantastic tool.
any plans for native vector database integrations for long-term semantic memory? congrats👏👏

We’re launching Scarlett today!

Scarlett is your digital assistant inside Slack. The best part is how simple it feels: you just ask Scarlett, and she gets it done. You don't have to switch tabs or jump between Claude, ChatGPT, and other tools. She can generate images, write content, answer questions. It can help your team directly where the work already happens.

Would love to hear what you think!

hey interesting product, how does the co-worker get all the context?

finally tried this for a few days and the handoff to my actual project manager felt smoother than i expected, not robotic at all

the Autopilot capability sounds powerful. for founders who are just getting started, what are the first three workflows you would recommend automating with Scarlett to see value quickly?

Congratulations on the Launch!!

Congratulations on the launch!

Scarlett sounds like a practical AI agent rather than just another chatbot. how does she decide when to act autonomously versus asking the user for approval, especially in business workflows?

I like that you're positioning Scarlett as a co-worker rather than another AI assistant. That's a much higher bar, because teammates aren't judged by how much they can do—they're judged by whether people trust them with responsibility. If AI reaches that point, the relationship users have with these products changes completely.

Congrats on the launch, Dan! 🚀 It’s awesome to see you guys pulling back the curtain on how Scarlett runs. The decision to skip the traditional Vector DB route and stick to an SQL-heavy setup for speed (similar to Hermes) is a really fascinating engineering choice.

Since you mentioned Vector DBs sometimes over-cluttered the context, how do you handle deep semantic search or long-term recall when a user asks about a conversation from months ago? Is it purely structured querying, or do you have a hybrid text-search layer built into the SQL database?

The iMessage angle is underrated for solo operators. A lot of small businesses do not live in Slack, and forcing them into a new cockpit can kill adoption. The piece I would want very visible is the control surface: what Scarlett is allowed to do alone, what needs approval, and what happened while nobody was watching.

This is neat. Does it keep context across different Slack channels or only within threads?

Congrats on the launch @dansutera @yuanhai_shi @cynthiacui

What makes Scarlett different from Claude Code and Viktor?

Congrats on the launch! I'm curious, after a year of using Scarlett internally, what's one workflow your team can no longer imagine doing manually? Those real-world use cases are always the most interesting.

Congrats on the launch! The "real colleague, not a bot" positioning is the hard part to earn. Curious how Scarlett handles uncertainty — does she escalate to a human in-thread when she's not sure, or attempt and flag afterwards? That handoff boundary is usually where trust with an AI coworker is won or lost.

Great product guys, love the fact it comes with 500+ connectors. Let's goo!

congratulations! after a year of testing internally, what was the biggest wow moment that convinced your team she was ready for a public launch?

Played around with Scarlett for a bit and was surprised how naturally it slotted into our slack workflow without feeling like another chatbot. The autopilot features still need some tuning but the teammate framing actually clicked for me.

Congrats! the autopilot feature is fascinating. can users customize how much control Scarlett has over marketing, support or operations based on their comfort level?

congratulations! the she Just works philosophy really stood out. what were the biggest engineering challenges you faced over the past year to make Scarlett reliable enough for daily use?

one thing that caught my attention was the ability to use your managed keys for external services. that's a great onboarding experience. how easy is it for users to switch to their own API keys if they decide to later?

Really interesting direction. How do you think about control vs. autonomy - especially when decisions have real business impact? Congrats on the launch!

Congratulations on the launch!

i really like the idea of using your API keys instead of requiring users to manage multiple subscriptions. how do you handle usage limits and cost optimization behind the scenes?

@dansutera This one's personal. I've been duct-taping AI tools together at Ctruh to cover exactly this gap, and an AI teammate that actually holds context and owns the ops glue is the dream. Congrats team, taking it for a spin this week. 🔥

Hey Product Hunt 👋,

I am S.S. Rahman, co-founder of ConnectMachine. We're back: v2.0 is live today on iOS and Android.

What ConnectMachine is: A private AI agent for the people you meet. Scan a business card or a LinkedIn QR, add a quick note, and it remembers who you met, where, and what was said. Later, ask it in plain English ("who did I meet at Web Summit who works in fintech?") and it answers, by text or voice. In the background it keeps contacts organized and enriched. When the timing matters, it surfaces the right person and quietly handles the follow-up.

There's no feed and no noise. Nothing is public. It's a self-curated personal network you actually control.

A quick why: Since our first launch we've had thousands of users and hundreds of conversations, here on Product Hunt and elsewhere. One idea kept circling: "not just a card reader." Nobody wants another card scanner. What people want is for the handshake to lead somewhere. Instead, contacts end up scattered across apps and the context evaporates the minute a meeting ends. Social networking apps don't help. They reward broadcasting, not remembering. ConnectMachine v2.0 is built around fixing that.

What's new in v2.0:

• Ground-up redesign: cleaner, faster, and your branding stays consistent across My CM Page, Card, and Wallet. Everything to manage your network in one place.

• AI Note Taker: record a meeting, assign speakers, get a transcript and summary in one tap. Works in any language.

• Smarter AI Concierge: handles multi-step prompts (ask for three things at once, it sorts them out) and works in your language. Private, trained on your network, visible only to you.

• Smart Event Detection: automatically groups contacts around the events where you met them.

• Team Plan: branded team cards, custom domain, and admin controls. Update bios, websites, and calendar links across your whole team from one place.

• More ways to capture: LinkedIn QR, WhatsApp snap-to-contact, and searchable scanned cards. The original card image is preserved for every scan.

• Cleaner contacts: notes at save time, one-tap duplicate merge, and incremental import that skips unchanged contacts.

• ShareBack form: when you share your card, the other person can share theirs back instantly. Networking that's actually mutual.

• Follow-up reminders: email and SMS nudges per contact. Never forget to circle back.

• Full branding control: brand colors, fonts, logo, banner, custom avatar, drag-and-drop field reorder, and a branded QR code that replaces our logo with yours.

• Multi-language: the full app in English, Spanish, French, Italian, and Russian, with auto-detect on first launch.

• Contacts Webhook: push contact events straight to Salesforce, HubSpot, Pipedrive, or any CRM via Zapier, Make, or n8n. No manual entry.

Worth noting: the preserved card image, the webhook, and the language support all began as user requests in our reviews. People asked, we shipped. Most of v2.0 was built that way.

Who it's for: Founders, operators, investors, consultants, and now teams.

Anywhere you do live networking: conferences, expos, client meetings, any room where you meet people faster than you can remember them.

No spreadsheets, no lost cards, no "who was that guy at booth 14?"

Try it: iOS: https://apps.apple.com/us/app/connectmachine/id6751988305
Android: https://play.google.com/store/apps/details?id=com.connect.machine

ConnectMachine started as a digital card tool. You told us you wanted real memory behind every connection, so that's what we built. The review that stuck with me most said, simply, that we kept our promise. v2.0 is us keeping it again.

I'll be in the comments all day. If something annoys you, tell me that too. That kind of feedback moves the product more than compliments do.

Thanks,
S.S. Rahman

Hi everyone Vinod here , Co-founder at ConnectMachine.
Rahman covered the big picture, so I want to zoom in on the part I'm most proud of: the memory layer.
The hardest problem in networking isn't collecting contacts it's remembering context. That's what our AI now solves:

• AI NoteTaker records your meeting, separates speakers, and hands you a clean transcript + summary. What was said becomes part of the relationship, not something you scramble to recall a week later.

• Agentic recall  ask "who did I meet at the fintech summit?" or "what did Sarah and I discuss last month?" in plain language, and the agent answers with real continuity.

• A smarter Concierge that now handles multi-step prompts throw three tasks at it at once and it untangles them.

• Smart Event Detection automatically clusters contacts around the events where you met them.

The goal: every person you meet ends up in your network with context  location, event, and notes not just a name and a title. Would genuinely love to hear how the NoteTaker performs in your real meetings.

scanned a stack of old paper cards from a conference and it actually pulled out the right context notes for follow-ups, that part impressed me. love that nothing shows up in any public feed.

Congrats on v2.0. Something I noticed reading through: every privacy guarantee here protects the person holding the app. Nothing is public, visible only to you, a network you control. But the person whose card got scanned, and whose voice ended up in a transcript, doesn't have an account. Is there anything on their side of it, or does consent for the Note Taker rest entirely on the user remembering to ask the room?

Hi Everyone! Neha here, Marketing Director at ConnectMachine.😊

My favorite part of v2.0 is how much better it feels and how well it represents your brand.

• Ground-up redesign based on real user feedback. It's cleaner, faster, and more consistent across My CM, Card, and Wallet.🎨

• Branded cards in minutes. Match your card to your brand with custom layouts, colors, and logos. Create multiple identities and share only what you choose.🃏

• Real-time card updates. Change your details once, and your contacts are notified automatically. No more outdated numbers floating around.🔁

• Multi-language support with automatic language detection, making the experience feel native wherever you are.🔠

• Apple Wallet priority keeps your card ready for instant sharing whenever you need it. 🈸

For founders, consultants, and anyone whose personal brand is part of their business, your identity now stays sharp and consistent everywhere.🙌

I'd love to hear what you think about the new design and which feature stands out to you the most!❤️

Try it -
iOS: https://apps.apple.com/us/app/connectmachine/id6751988305
Android: https://play.google.com/store/apps/details?id=com.connect.machine

Congrats on the launch @rohanrecommends ! Great hunt!

Does the webhook support real-time updates when a contact or note is changed?

congrats on the launch!

The QR scanning was noticeably faster than other apps I have tried, and the fact that everything stays encrypted locally is a real plus. Wish the CRM sync took a beat less setup, but overall it feels thoughtfully built.

How does the AI actually figure out who to remember and what mattered, does it run on-device only or hit your servers in any way?

This is useful. Does the AI note taker work offline for conferences with spotty internet?

I started using their app this year, love it!

What happens if two people scan each other's cards at the same event? does it merge into one memory or stay separate?

Hey! Great update. I was wondering, what data stays on my device, and what data is sent to your servers or shared with third parties?

@syed_shayanur_rahman Congratulations. And happy product launch.

Finally ditched the stack of bent paper cards in my wallet after trying this - the scan is genuinely fast, even with crumpled receipts from meetups last week.

The value here is very "in-the-moment," scan and note right away or the context is gone. Do you know how often people actually do it in the moment versus catching up later? Congrats on the launch!

How do you handle corrections when the system remembers something incomplete or wrong? I’m curious whether users can see where a remembered detail came from and update it explicitly.

Congratulations on the Launch!!

The "attach a voice note to a scanned contact" piece is the real differentiator, not the card itself. Most networking apps solve storage, not recall

Curious how the AI Concierge decides what's worth surfacing later, does it prompt you to add context right after a scan?

Early feedback has been really exciting!! GPT-5.6 is setting a new bar for performance while also reducing token usage and latency, particularly in coding, complex agentic workflows, and tool-heavy tasks.

Is the solar system naming convention around to stay? Or just for the 5.6 release?

two months ago, @sama posted on X:

we still get looksmaxxed on frontend a little but we IQmog hard now

nailed it 👏👏

The API documentation is genuinely some of the best I've worked with, clear examples that actually run and sensible defaults that save hours of trial and error.

Does it automatically perform semantic compression during preprocessing to reduce input token usage?

How does it compare to Fable 5?

Programmatic Tool Calling plus explicit prompt caching is the real headline here for anyone running agentic workflows at volume. That's a meaningful cost and latency lever, not just a spec bump. Curious how Sol/Terra/Luna routing works in practice: automatic based on task complexity, or manual per-call?

The “more smarts per token” positioning is interesting. In practice, where have you seen the biggest efficiency gains—long reasoning tasks, coding, or everyday chat?

Curious how the pricing scales for smaller teams just starting out with API access, especially compared to self-hosting open-weight models.

Tried it for a quick prototype last week and was surprised how clean the API docs are now. Getting a basic text generation setup running took maybe ten minutes.

The API docs are genuinely some of the cleanest I've read, with code samples that actually work on the first try. Really impressive that a tool this complex still feels approachable.

the docs are genuinely well organized, and getting a working prototype running over the weekend was way faster than i expected.

GPT is backk

this just makes sense! apple should have done this, but very glad Juicy exists!

I like the focus on staying native, lightweight, and local instead of adding another resource-heavy background app. Has building Juicy uncovered any surprising patterns in how different Mac models or Apple Silicon generations age from a battery health perspective? Congrats on the launch! 🔋

Replaced my usual stack of AlDente and coconutBattery with Juicy this week and the custom alerts at 50% actually feel polished instead of janky, plus the per-app energy insights showed me a Chrome tab was secretly murdering my battery.

Love the design :)

i've been using it for months now and it keeps getting better. looks sleek in my menu bar and i love the battery information it gives me. so much easier than digging for this info in the system preferences.

The "impossible to ignore" framing is the part I like, and it's worth naming why it works: most apps fighting to interrupt you are serving themselves, but a battery alert interrupts you for your own goal — not losing work — so the aggression is aligned instead of extractive. That's a different category from a nag, and it's why the 10%/5% system alerts fail: not urgent enough at the moment that actually matters to you. My real question is the charge-limit side. Third-party charge limiting on the Mac has historically meant going below the public API (SMC-level), which is exactly the thing that breaks on an OS point-release or behaves differently across Apple Silicon generations. Is Juicy using something Apple now exposes officially, or maintaining the lower-level path — and how much of your support load is "it stopped limiting after I updated macOS"?

I think this is a great product! Dominik has been responsive to questions and issues that have come up when going to the beta version of new Mac OS. He has been helpful in pointing me in the right direction. The product is comprehensive and well laid out, and it works! That's the big thing! I will certainly recommend this to all the people I know using Apple products. Juicy gets my vote for Product Hunt.

Finally a battery app that doesn't feel like a workaround. The custom alerts at 50% saved me from a dead phone on a hike, and I love seeing the energy hog culprit apps right in the menu bar.

the screen glow and sound choice on custom alerts feels really considered, like someone actually stress-tested the default 20% warning and thought "no, this needs to be harder to miss." nice craft.

Love that the alerts use screen glow instead of just a banner, makes the 20% warning feel intentional rather than punishing. Nice move bundling all those tools into one native Swift app instead of the usual stack of utilities.

I've been using the app for a while now, and I'm very happy with it. It would be great to add widgets to the home screen. Is this planned for the future?

I’ve been using Juicy from the first day and I love it. Continue the great work!

I’ve really enjoyed using Juicy! It’s rare to find a utility app that’s both beautiful and genuinely useful. The custom battery alerts, charge limiting, and battery health features have all become part of my daily routine. It feels like a feature Apple could have built themselves. Wishing you an amazing Product Hunt launch! 🙌

This is one of those apps that everyone should install on their Macs by default 👌

Been using this product for a while now and really love it! Especially the last update, such a beautiful app and love the new little icon in my toolbar.

Congrats on the launch! 🚀

The 50-80% charge limiting got me, I always mean to protect my battery and never actually do it. Love that it's native and fully local too. I should give it a try!

I love the way how it notifies to charge the computer

Very good idea + very good execution. If only I had a Mac lol Does it also deal with unused applications still running in the background that might impact battery life?

appreciate Nim! yeah luckily they have not done it (yet) hehe more space for Juicy to grow!

How difficult is migration from existing desktop projects and could step by step examples make adoption much smoother?

The "no browser, no WebView, no compromise" positioning is very clear.

Desktop apps are in a weird place where a lot of tools feel like websites wrapped in a window, which is fine until performance, memory, offline behavior, or native feel actually matter. A toolkit that gives declarative UI and a predictable state model while still rendering natively sounds like a strong direction.

Curious how close Native SDK gets to platform-native behavior out of the box. Does it handle things like keyboard shortcuts, menus, window controls, accessibility, and OS-specific patterns automatically, or do developers need to wire a lot of that themselves?

i appreciate the message based state model because predictable updates can reduce debugging time. have you tested performance with very large applications and would publishing benchmark results help developers understand where this toolkit performs best?

How does pricing work once you go beyond the hobby tier — is it purely based on bandwidth and build minutes, or do serverless invocations factor in too? Trying to estimate what a mid-sized Next.js app would realistically run.

I thought this kind of approach was already outdated. Does it still make sense to have this type of hybrid platform in the world of AI assistants, where building natively is no longer a big challenge? And you no longer need to rely on third parties to use recent native APIs or to avoid surprises when your packages and app break.

I am excited about this - having an alternative to the standard Electron/Tauri and having explored Flutter and React Native, this sounds very compelling and the sample apps - especially the note-taker - look quite smooth. I am wondering if the fact that our platform is a monorepo with React/Next/Tailwind/TypeScript will be a multiplier in being able to use this? We've built heavily on Vercel SDKs (we use the AI Gateway, the AI SDK, Nextjs) and we're big V0 users as well! Love this, @Vercel!

How does pricing actually work for mid-sized projects, and are there any hidden limits on bandwidth or build minutes I should know about before committing?

Usually going native like that means giving up easy stuff like embedding web content or styling with CSS. Is the tradeoff worth it in practice, and what's it actually like to build with something closer to React, or something new to learn? Excited to try it out!

I've been driving native desktop apps with automation and half my breakage comes from each OS handling windows and menus differently. Does Native SDK give you one API across Mac and Windows, or are you still writing platform-specific code under the hood for the fiddly stuff like tray icons and file dialogs? That gap is usually where cross-platform toolkits quietly leak.

Deployed a Next.js site in minutes and the preview URLs just work every time. The DX feels polished without being overengineered, which is rare.

A built-in cost dashboard showing real-time bandwidth and function invocation spend per project would be huge. Right now I have to dig through usage logs and try to estimate monthly bills myself, which makes budgeting for client work way harder than it needs to be.

How does pricing scale once you start getting serious traffic — is there a clear breaking point where it jumps from a flat tier to something unexpectedly higher?

I like the focus on native applications instead of browser based solutions. Could adding more sample projects for different industries help developers discover practical ways to use the toolkit?

me enjoying native development wants offline documentation. would detailed guides make onboarding faster for first time users and contributors alike?

How does memory usage compare with existing desktop frameworks during larger projects?

Notion is growing up.

Product work already lives across feedback, specs, decisions, tasks, docs, and handoffs, so having agents help with the routing/triage/summarizing part makes a lot of sense. the key detail is that humans still own the judgment calls, because shipping software is rarely just moving tickets from one column to another :)

Curious how Ship OS handles messy feedback that is half bug report, half feature request, half customer emotion. does the agent simply summarize and route it, or can it also connect it to existing projects, PRs, and past product decisions?

How does Ship OS know when a task actually needs a human judgment call versus one it can just push through on its own?

This is interesting. When agents triage feedback, how do they handle contradictory signals from different customer segments?

Is this an actual new launch, or just an ad for notion? As of right now, don’t see this “ship os” product in their changelog.

Curious how the offline mode actually holds up when you jump between wifi networks mid-project?

Been using it for a few weeks and the database setup is surprisingly flexible, way more than I expected. Took a little to figure out how I wanted everything organized but now it just clicks.

The way databases, docs, and calendars all snap together inside one clean canvas is honestly really impressive, makes everything feel like it belongs in the same workspace.

Publishing directly from Claude sounds like a real time-saver. How do you handle AI-generated content review before it goes live? Is there an approval step built into the workflow?

How does the multi-channel seeding actually handle formatting differences across platforms like LinkedIn versus a blog CMS?

Congrats on the launch. The interesting shift here is from AI generating text to AI participating in the whole campaign workflow. When Claude has access to brand context, analytics, and publishing channels, what is the review checkpoint you see teams using most: approval before scheduling, approval before publishing, or review after the first performance signals?

Finally, something that doesn't make me dread cross-posting. The multi-channel seeding actually works as advertised, saved me 20 minutes on a single article.

looks great, congrats on the launch!

Valeri, the constant hopping between tools to get one post out is what always drains me before I even publish. Keeping the whole thing in one calm place is honestly the dream for a small team.

When Claude schedules and publishes a whole campaign through StoryChief, is there a checkpoint before it actually goes live or does it just ship?

Been around since 2018 which is rare for a marketing tool at this point. Curious how the AI Canvas and the William agent from earlier this year changed the actual workflow versus just being new features bolted onto the same core product. Also wondering if the AI Power Mode reviewers keep mentioning is generating drafts from scratch or mostly cleaning up and matching tone on stuff people already wrote.

Congrats on the launch!

The editor's distraction-free layout with inline collaboration cues is a really thoughtful touch. It feels like the team actually understands the messy back-and-forth of editorial work instead of just bolting on features.

Congrats for you! Curious to see how you frame the collaboration side versus the scheduling side in the product.

Excited to hunt Muse Spark 1.1! 🚀

Some highlights:

• 🧠 New multimodal reasoning model for agentic AI

• 💻 Major improvements in coding, tool use, and computer use

• 🖼️ Stronger multimodal understanding

• ⚡ Faster execution with multi-agent orchestration

• 📚 Massive 1M-token context window

• 🔌 Public preview of the Meta Model API for developers

• 🤖 Available now in Thinking mode on Meta AI and meta.ai

If you're building AI agents or developer tools, this one is worth checking out. Curious to hear what everyone thinks!

The focus on agentic tasks caught my attention. What kinds of real-world workflows have shown the biggest improvement with Muse Spark 1.1 compared to traditional chat models?

Gave it a quick spin on a small coding task and it nailed the tool use on the first try, which honestly caught me off guard.

Tried it on a quick coding task and the tool calling was noticeably sharper than the last version, picked the right API without me hand-holding. Solid upgrade.

The tool use improvements are genuinely noticeable, it handled a multi-step browser task without me hand-holding it. Coding feels tighter too. Curious to see how it stacks up against the reasoning models dropping this week.

Tool use feels noticeably snappier than the previous version, and the multimodal understanding caught a subtle detail in a chart I threw at it.

How does the local agent actually decide when to call a tool versus just answering, and is that planning step visible somewhere so I can audit what it's about to do?

How does the local agent actually handle tool approvals on macOS without the cloud, does it prompt every single time or can you whitelist actions for a session?

On-device is the underrated headline here. In voice work the two things that actually block adoption are latency and privacy — a local model that never ships audio to a server kills both at once, and for anything with sensitive data (legal, medical, insurance) that's the difference between "can't touch it" and "ship it." Curious how accuracy holds up on proper nouns and numbers locally vs the big cloud models — that's usually where on-device stumbles. Nice to see transcription + a real agent in one place. Congrats on the launch 🚀

The "entirely on your Mac, no telemetry" bit is exactly why I'd try this over cloud dictation — but I want to understand the agent step: when Mispher "goes and does it," is that agent running fully local against on-device models, or does the agentic action call out to an API? And is the rewrite/translate the same local model as transcription, or a separate one I'd need to pull down?

the rewrite-in-place feature is the one that'd actually sell me - highlight a sentence anywhere, say make this shorter, and it swaps in place. how does that hold up in apps that don't expose a normal text API, like electron apps or browser textareas buried under heavy JS overlays? does it degrade gracefully, like falling back to a clipboard paste, or does it just fail to find the selection in those cases?

How does the on-device agent handle tool approvals when the LFM2.5 model needs to call something sensitive like Apple Notes, and can I pause approvals globally for trusted actions?

How does the local agent actually handle tool calls that need internet access, like fetching a webpage or hitting an API, while still keeping everything on-device?

love that the pose detection runs entirely on-device, no sketchy cloud uploads just to count my push-ups. the gamified ranks and streaks are a clever way to keep me showing up every day.

Does the camera count reps accurately even when the lighting is bad or you're at a weird angle in a crowded gym?

The on-device pose detection is a smart call, both for privacy and for the kind of low-friction logging that actually keeps people coming back. Love that you resisted the wearable route.

Does the camera actually work well in a dim home gym setup, or do you need solid lighting to get accurate rep counts?

The on-device pose detection is a really thoughtful call, feels great knowing nothing leaves the phone while still getting accurate rep counts.

Real-time rep counting sounds useful for solo training too. How does the model handle exercises where body parts are partially occluded or viewed from different camera angles?

Building on Emrah's reliability question with the part nobody's hit yet: on-device pose tracking for a full session is a thermal budget problem as much as an ML one. A phone running the camera + a pose model continuously for 30 minutes heats up and throttles, and that's exactly when inference gets flaky — so "accurate" in a 2-minute demo and "accurate at minute 25 of a sweaty circuit" can be different apps. Do you duty-cycle the model between reps or downsample frames to stay under the thermal ceiling, or is it running full-rate the whole set? And when the phone's propped against a water bottle at a bad angle, does it miscount silently or tell the user to reposition? The silent miscount is the one thing that'd make me stop trusting the number.

Tried the squat counter in my living room and it actually caught my rep count without me touching the screen. The XP and streak thing feels a bit cheesy but it's weirdly motivating to keep showing up.

the on-device pose counting is such a smart move, especially for privacy. one thing i'd love to see is a quick audio cue or short rest timer that kicks in between sets automatically when the camera detects you've stopped, so i'm not constantly staring at the screen to check when to go again.

How does the camera actually handle things like messy form or switching between exercises quickly? I get that on-device processing is great for privacy, but curious how reliable the rep counting ends up feeling in a real sweaty home workout.

I like that it run on the user's own Claude subscription. That feels like a smart way to give people more control over privacy and costs . Nice touch.

Really interesting concept .How do you handle long term memory across channels? Can teams choose what Yasmine remembers or forgets?

Curious how the per-channel memory works in practice — is it truly isolated so a coworker in #finance can't surface anything from #marketing, or is it one workspace memory scoped per channel at query time? And since she runs on my own Anthropic account, does the per-tool allow/ask/block policy live in Yasmine or ride on my Claude setup, and is it set per channel or globally?

How does the memory actually work per channel — does it carry context across DMs and other channels too, or is it strictly scoped to where Yasmine was set up?

The part that stuck with me is the idea of it remembering each corner of our workspace on its own. That is usually where these things fall apart for me, so I am curious to watch it hold up, @badr7 .

This is cool. Can a Yasmine in one channel query another channel's Yasmine for context?

The idea of each Slack channel becoming its own AI teammate with separate memory is genuinely useful. Setup felt pretty smooth and it pulled the right context without me having to over-explain.

The per-channel memory setup is clever, feels like each teammate actually has context rather than starting fresh every time. Love that it runs on your own Claude subscription too.

Finally tried Yasmine in our team Slack and what stood out was the per-channel memory. It actually remembered context from yesterday's thread without me re-explaining. Genuinely useful.

Finally an AI that just sits in my Slack and actually does the thing. Loved how each channel having its own memory means no more re-explaining context every time.

The per-channel memory idea is genuinely useful, way more flexible than a single bot. Hooked it up to my docs and email in a couple minutes and it actually pulled the right context when I asked for a recap.

Hello hello Product Hunters!

Here's a cool free app that let's you add yourself to ANY photo or video:

Works 100% on your phone so super private

Best way to use it (examples: https://x.com/ThePeterMick/status/2075203667314000161, https://x.com/ThePeterMick/status/2075304624743911747):

1. Open the app

2. Tap "Camera Roll" or "Use Camera" (for a new live pic)

3. Take the photo of your startup's landing page with you pointing at the domain name with your finger

That's it! Post it on socials and tell your startup's story :-)

You can also select an existing video and add yourself to it (like this: https://x.com/thepetermick/status/2075335285781434627)

An easy way to help you pump out more content for your social media

Tag me on X for a repost if you take a pic with it - go WOW your audience today

Cheers,

Peter

how well does the cutout hold up with messy hair or when you move around a lot in the clip?

Tried it on a photo of my dog and the cutout was surprisingly clean around his fluffy ears. The no-login thing is a nice touch, felt like I could just play around without friction.

How well does the cutout hold up with messy hair or when you're moving around a lot in the clip? Curious if it handles motion cleanly or needs a pretty still pose.

How well does the background removal hold up with messy lighting or busy backgrounds, like a crowded room? That’s usually where these tools start showing seams.

The on-device processing is a nice touch, super quick to drop myself into a clip without uploading anything. The cutout around my hair was solid for a first try, though fine edges like glasses still look a bit fuzzy.

Congrats on your launch Peter!

Cool concept! Already support or planning to add custom templates?

how well does the cutout handle things like messy hair or fast movement in the video, or is it mostly tuned for clean backgrounds and slower poses?

since it's positioned as fully on-device, curious about the ceiling on that - is there a video length or resolution where it has to fall back to cloud processing, and if so does the app tell you that's happening or is the switch invisible? the "100% private" claim only really holds if that boundary is either non-existent or clearly surfaced.

ran a flaky checkout flow through it and the root cause trace pointed straight at a race condition i had been chasing for days, which was a genuinely pleasant surprise

How does it handle flaky tests or external services like third-party APIs when spinning up the whole sandbox?

how does it handle flaky stuff like network requests or third-party api calls when reproducing bugs in the sandbox?

the "boots your whole app and actually runs it" part is the compelling piece, but that cuts both ways - it means the agent can trigger real side effects too, not just observe them. if a PR touches checkout, does the run risk hitting a payment provider's live mode or sending a real email/webhook, or is that sandboxed away by design? feels like the same power that catches real bugs could also cause real damage if it's not scoped carefully.

As a solo dev shipping AI-written code daily, this hits a real pain: unit tests pass but the actual app breaks. Question — can it catch visual regressions? In my product (image processing) the worst bugs are "output looks subtly wrong but nothing throws," and I've only been able to cover those with hand-written pixel-level regression tests.

I like the shift from analyzing code to validating behavior.

As AI writes more software, the harder problem won't be generating code—it'll be proving that the system actually works under real conditions.

Running the app instead of reasoning about diffs feels like a fundamentally different approach.

The bugs that only turn up when the thing is actually running have always been the tedious ones to chase, so this feels like a real weight off before approving anything.

"Reading code and running code catch completely different bugs" is the thesis I'd tattoo on a wall — I verify my own changes by driving the actual app for exactly this reason, because a green typecheck and a working app are different claims. The question that decides how much I'd trust it: fidelity of the sandbox. Booting the stack with Docker Compose and seeded data catches "does the happy path run," but the bugs only running catches usually live in production-shaped state — an auth-token refresh race, a connection pooler that refuses you under a specific condition, a third-party API returning something your seeds never do. Seeded data almost by definition doesn't contain the state that produced the bug. Can Qlane snapshot real (anonymized) state into the ephemeral env, or is it always synthetic seeds? That gap is the difference between "the PR runs" and "the PR is safe."

how does it actually handle test flakiness when the sandbox environment differs from CI, and is that something i have to configure myself or does it handle it under the hood

Tried it on a small project and the fact that it actually boots the app in a sandbox made a difference, caught a race condition two other tools missed. Root cause trace was clear enough to act on.

finally something that doesn't just speculate about my null pointers. watched it catch a real race condition in my db queries by actually running the stack, not just staring at the diff. genuinely useful.

How does this handle apps with heavy external service dependencies, like third-party APIs that need real credentials during a test run?

Hey Product Hunt! 👋
Excited to share something big we've been building.
At Latenode, our mission has always been to make serious automation accessible to everyone, not just developers. Today we're launching a big step toward that: the AI Workflow Builder.
Just describe what you want in plain text, and AI Workflow Builder turns it into a working automation, complete with nodes, logic, and the connections between them. It even maps the data fields between nodes for you, so you're not stuck manually wiring inputs and outputs before you can test your idea.
Go from a rough idea to a workflow you can actually run and refine in seconds.
We'd love to hear what you think, drop your feedback in the comments below. Thanks for being part of the Latenode journey!

Would love to see a version history with diffs for the AI-generated workflows so I can review what changed when I tweak a prompt, instead of wondering why a step suddenly behaves differently after a small edit.

A versioning system for AI agents would be huge. Right now it feels tricky to track what changed between iterations or roll back when a tweak hurts performance. Something like git-style commits with diffs and one-click rollback would make iterating way less scary for production workflows.

Tried it for a quick lead enrichment flow and it spun up a working agent in like three minutes. The 5,500+ integrations claim actually held up, I hooked it to Notion and Slack without wrestling with auth.

How does the pricing work once you start using AI models and running lots of runs through the orchestrator, does it get expensive fast compared to just wiring up something in n8n yourself?

the auto data-field mapping between nodes is the part i'd want to stress test. when the prompt is ambiguous about which field maps to which, say two similarly named fields coming from different tools, does it pick one silently and let you find out when it breaks, or does it flag the ambiguity and ask before wiring it up? that feels like the real difference between automation in minutes and automation that quietly does the wrong thing in minutes.

Tried it with a quick "send new Stripe customers to Slack and summarize" prompt and got a working agent in under a minute, integrations wired up and all. Surprised me how clean the orchestration view is, not the mess of nodes I was expecting.

Is the Profile DNA something I set up once per platform or does it sync across all of them?

the platform-specific angle is the part that stands out to me, most tools treat "write me a prompt" as one universal problem when the actual constraints are different per builder. curious about the intake though: if someone picks Lovable but describes something that really needs backend auth flows or a real database, does archly ever push back on the platform choice itself, or does it just take what you picked and build the best prompt it can for it regardless of fit

Tried the intake flow and it actually caught a few things I usually forget to mention upfront. The Profile DNA idea is smart, makes each prompt feel less like a starting point.

The 7-step intake caught me off guard, I figured it would just be busywork but it actually pulled out platform specifics I never would have added myself. Profile DNA is a smart touch too, saved me from retyping my stack every time.

I like that you're separating "build the product well" from "build the right product."

As AI makes software creation cheaper, I think the biggest source of failure shifts upstream. Building becomes abundant, while deciding what deserves to be built becomes the scarce capability.

That feels like the more important problem to solve.

the no-go verdict is the part I'd stress test hardest. it's easy to feel good about killing a bad idea, but what about ideas that look weak on paper because the market signal is sparse or contrarian rather than because the idea is actually bad? how do you keep vibe-check from converging on "whatever already has visible reddit complaints" as its working definition of validated, since that would quietly bias it toward incremental ideas over genuinely new ones nobody's complained about yet.

since you asked for honest feedback - the place i'd expect a weaker model to drift is the memory-write step itself, not the actual work. it'll do the task correctly but skip logging the decision, or compress it down to "did X" with no reasoning, which quietly defeats the whole point once you need to know why a call was made three sessions later. is that step enforced structurally somehow, like blocking session-end until the checklist items in SHOTGUN.md are actually filled in, or does it fully depend on the model choosing to follow the loop on its own?

I like that you're treating the AI agent as infrastructure rather than the product itself.

As models become interchangeable, I think the longer-term advantage shifts toward the operating system that preserves context, decisions, and continuity across whichever models people choose to use.

That's a much harder layer to build.

Building solo, the loneliest part is re-explaining everything every single time. An assistant that actually keeps up with the whole story of what I am making sounds like real company for the long haul. Nicely done, Krishna ;)

how does it actually persist context between sessions, does it just dump everything into the repo as markdown or is there a real vector store behind it

Ran shotgun-init on a side project and the onboarding actually felt like delegating to a real cofounder, not just feeding a chatbot context. The fact that it works across Claude Code and Cursor without rewriting anything is what sold me.

The one-line install plus onboarding flow is genuinely well thought out, makes something this ambitious feel approachable instead of overwhelming.

The one-line install and `shotgun-init` approach is really thoughtful, removing the usual setup friction that kills most frameworks before anyone even sees what they do.

One line to install and my Claude Code agent actually remembered context across sessions without me re-explaining the project each time. Felt like having a real partner instead of a forgetful tool.