PH热榜 | 2026-07-14

Hey Product Hunt! 👋

I'm Steven Cen, and today we're launching ClawTeams — an AI team platform
built specifically for e-commerce operators.

The frustration that led to this: we kept seeing smart sellers use AI tools and still
end up doing all the coordination work themselves. They had AI assistants — but they
still had to be the manager. That's exhausting.

So we built ClawTeams around a different idea:
→ You set the goal. The AI Team Lead manages the rest.

Get 800 bonus credits ($8 value) with your first top-up of any amount. No minimum required.

We'd love your feedback — especially from sellers who've tried other AI tools and hit
walls. What made you give up on them? What would make an AI team actually useful?

@stevencen Love the "zero micromanagement" vision! 👏

You mentioned that high-stakes decisions wait for human approval in Slack/Discord. How granular are those safety guardrails? For instance, can a seller set custom approval rules—like automatic sign-off for minor ad budget tweaks, but requiring human approval for price changes or launching new campaigns?

Congrats on the launch! Turning one ecommerce brief into a coordinated team for research, listing copy, creative, and review feels genuinely useful.

Huge congrats👏 to shipping. one que what happens if the team lead agent runs into a direct roadblock like an API returning an expired token error from a connected store? does it flag a human immediately in discord or try to self-heal?

The one-sentence setup is especially appealing for ecommerce. Starting with rough product information and asking for a complete listing package feels far more natural than building a workflow first.

Congrats! @s_cen

Congrats!👋👋 This could be especially useful for lean ecommerce teams juggling product research, listing copy, visuals, launch campaigns, and distribution at the same time.

Quick update from the ClawTeams team — thank you all! 🙏

We've been blown away by the warmth and depth of the feedback since launching today. Landing at #1 Day Rank with 600+ points means a lot, but honestly the comments have been the real highlight.

The questions here have been incredibly sharp — from guardrail metrics and cross-run memory, to platform rate-limiting and account-safety edge cases. It's exactly the kind of thoughtful conversation we hoped to spark, and it's already shaping how we think about the roadmap.

To everyone who upvoted, commented, or gave us a shoutout: thank you. This is just day one, and the reception has us more energized than ever. We're reading every reply and would love to keep the conversation going. 💛

Congratulations! The product feels built around how work actually gets delegated, reviewed, and delivered.

Amazing AI agent team for e-commerce!

Congrats launch！！！amazzzzing product~

Ecommerce work is full of tiny handoffs between research, copy, design, and operations. Nice to see a product trying to coordinate the whole chain from one request.

I run a small fleet of Claude agents that operates my own product (eng, QA, growth), so this is close to home. The problem that bit me wasn't planning or approvals — it was rule drift: every incident adds a constraint to some agent's instructions, and months later the rules contradict each other in ways no single edit caused. Does ClawTeams reconcile new seller-added constraints against existing ones, or does the Team Lead's rulebook just grow? Congrats on the launch.

Which integrations are available today beyond Slack, and is Microsoft Teams support already production-ready?

"high-stakes decisions wait for your approval" - is that threshold something the team lead learns from your past overrides, or a fixed list of action types you configure upfront?

Promising! Gonna try this out @s_cen

the drift-catching part makes sense for one agent going sideways, but what about two specialists stepping on each other mid-task - if the pricing agent drops a price right as the ads agent is calculating margin for a new campaign, does the team lead serialize those or do they just race?

the approval-boundary answers in this thread are solid, but they're all about controls you set. the risk I don't see addressed is the platform side - Amazon/Shopify etc flag bulk automated listing or pricing changes as suspicious and can suspend a seller account over it, independent of whether the change itself was a good idea. does ClawTeams rate-limit or pace its actions to stay under those platform-level detection thresholds, or is that entirely the seller's problem to monitor?

Congrats on the launch, the goal decomposition flow is clean.

The question I keep coming back to with agent teams is what the team lead remembers between runs. Planning is the easy half. The expensive half is not re-litigating a decision a specialist already made last week, and not acting on stock numbers that went stale two runs ago.

Does the team carry state forward across goals, or does each new goal start from a clean context?

Congrats on the launch Steven!

The per-action approval controls look well thought through. The harder problem with goal-driven agents is the goal itself: "increase Q4 revenue 20%" can be hit in ways you'd hate — deep discounts that gut margin, or heavier email that lifts this month's revenue and burns the list next quarter. Each step can look low-risk and pass its guardrail while the sum quietly optimizes the wrong thing.

Curious whether the Team Lead is measured only against the stated goal, or against guardrail metrics too — a margin floor, a send-frequency cap, brand constraints — so it's steered away from technically-correct-but-bad paths, not just blocked on individual high-stakes actions. That constraint layer feels like where the trust actually lives. Congrats on shipping, Steven.

Looks great

Does the platform support shared team memory without exposing sensitive context across unrelated projects?

How are budgets set and enforced across a long-running project with multiple agents working in parallel?

Honestly the "one goal, zero micromanagement" thing actually works. I tossed in a goal to cut ad spend waste and woke up to a slack thread with three specialists already debating keyword bids. Kind of eerie honestly.

Would love to see a quick weekly recap card in Slack showing what the team actually did and what it cost in API credits, so I can tell at a glance whether I'm getting real value or just noise.

The approval boundary is the product here. For e-commerce operators, the useful version is not “agents do everything”; it is clear ownership, budget/risk limits, and a decision log in Slack when a step touches inventory, pricing, customer comms, or ad spend.

Parallel work is presented clearly without overwhelming the user with technical agent details. Nice product thinking.

Love the concept of treating AI agents like actual team members. One thing I'd find super useful is a simple performance dashboard per specialist showing what they shipped, time saved, and cost vs hiring a human for the same task, so I can actually measure ROI over time.

Congrats on launching! "Zero micromanagement" is a bold promise for a multi-agent setup - how do you handle the failure case where one specialist agent goes off track? Does the team lead catch it before it reaches the customer, or is there a human-in-the-loop checkpoint?

Hey Product Hunt 👋

I'm Zvonimir, co-founder of Pazi. Some of you might know me from Pythagora / GPT Pilot - one of the first open-source coding agents, 33K GitHub ⭐

We built Pythagora thinking people wanted to build software. 100k users later, we realized building was just a means to an end. The moment the app was done, people didn't celebrate. They got stuck. "Now what?"

The app was never the point. They didn't want to build applications. They wanted to be entrepreneurs.

Every experienced founder knows the hard part comes after the build - getting it in front of people, seeing what lands, changing it, trying again. A business is just that loop, run a few hundred times until something works.

So for the last 6 months we've been building Pazi to run that loop with you.

You bring an idea. Pazi builds a team of agents around it and starts moving - and it doesn't wait for instructions. It tries something, sees what lands, adjusts, keeps going.

Here's one real idea that one of our users started with: a prompt pack to sell on Gumroad.

Pazi built her a team - a growth strategist, a content writer, a researcher - and got to work. The pack itself. A cover to drop into Canva, with the hex codes. A posting calendar: which platform, which day, which hook, which call to action. And competitor pricing research she never asked for, because she'd mentioned in passing that she had no idea what to charge.

Today, she pays for Pazi and for no other AI tool.

We looked at the first 529 ideas people brought us. ~75% of what they actually wanted wasn't "build me an app." It was some version of get me customers. So that's where the agents start.

Two years ago, "vibe coding" sounded absurd - many said that only real engineers write real code. The exact same shift is about to hit the way companies run.

👉 Soon, no one will run their business operations by hand either.  👈

Today is day one. Autonomous companies are rough in some places, just like early vibe coding was too.

And if you're someone with an idea and no team: this is the best moment in history to be you. That gap used to kill things before they started. It's about to stop mattering.

What would help us most: take an idea you've been sitting on, run it through Pazi for real, and tell us where it broke, what surprised you, and what was missing. My co-founder Leon and the team are here all day, reading every comment.

Come build with us 🥂

The part about getting stuck after the build is so true. Making the thing is often the easy part, getting people to actually care is where it gets messy. Curious how Pazi handles that first week after launch.

"The app was never the point. They wanted to be entrepreneurs." is a really strong insight.

As a founder, I can relate to how much harder everything becomes after the product exists. getting customers, finding positioning, creating content, doing outreach, learning what works, then repeating the loop... that is where most ideas slow down. "Vibe coding for business operations" is a very clear way to frame Pazi :)

The proactive part is what I would want to test most. how does Pazi decide what it can move forward on independently, and when it should stop and ask the founder before spending money, contacting someone, or changing direction?

@zvonimir_sabljic1 @leon_ostrez
This is a superb idea. I am at a stage where I am managing multiple things, and I know I can spring up some agents to delegate work, but I simply don't have enough time. The next stage in my build journey is launch, marketing, and customer acquisition, and I feel Pazi can help. Love the bright orange. Kudos on the launch.

the "vibe coding but for business operations" framing is a good hook. one thing I don't see covered yet: when the team does outreach - emailing a prospect, DMing a creator, whatever - whose identity is it going out under? if it's sending from my own email or socials, a bad batch of outreach doesn't just waste time, it dings my actual deliverability/reputation with real people. is there a sandboxed identity it uses until you're comfortable, or is it your accounts from day one?

Love it! I run an influencer marketing agency as well as an influencer discovery platform, and following up with creators and clients usually takes a lot of time and efforts from my team. This include scheduling outreaches, setting up sequences and sending newsletters. Do you think Pazi can help us to sort the communication for both the companies simultaneously? Right now, we are using a mix of 3 tools to sort the communications with users, creators and clients.
PS - Really love the UX and I believe in the solution you are offering. Congrats and good luck with the launch!

A dashboard showing what each agent is actively working on right now would help a lot. When I come back I want to see a quick "live status" view instead of just results, so I can jump in and steer things before they go too far down a path I don't want.

Congrats on the launch! Coming from an ML research background, I'm curious how Pazi handles the agents' confidence in their own outputs. For example, when a growth strategy or piece of content is based on a shaky assumption, does the system surface that uncertainty to the user, or is it something you only catch after the fact by checking the dashboard? Cool direction either way.

The strongest part of this is focusing on the loop after the build. For founder work, I would want every agent step to leave a trail: assumption, action taken, result, next choice, and where it stopped for approval. That is what keeps “business operations” from turning into busy-looking motion.

Congratulations on the launch. It's refreshing to see someone optimizing for outcomes instead of optimizing for prompts.

I'd probably trust this more if it occasionally explained why it chose one direction over another.

This could become the AI equivalent of hiring your first operations person without actually making a hire. Haha

I wonder what happens when the strategy isn't working. Does the team completely pivot on its own or ask first?

One thing I'd enjoy is watching every completed task on a timeline, almost like following a teammate's daily progress. Love it!!!

The Gumroad example made the product click immediately for me. That's a much more relatable demo :D

How does Pazi decide what to work on first if someone has five different priorities competing for attention?

Congrats. This feels like where AI is heading next. Not just creating things, but actually helping people build businesses around them.

Love the branding, will certainly give it a look!

Congrats on the launch! When Pazi's agent team builds out first outreach/content for a niche idea, how does it handle domain-specific compliance stuff (e.g. regulated industries) - does it know when to flag "this needs a human review" vs just shipping it?

Congrats on launching. The line that got me is "every time you come back, something has moved." As a solo founder I already run my startup close to this way, handing pieces off and checking results, so one coordinated team instead of me stitching ten tools together is the dream. Real question: when an agent takes the next step on outreach or content, how much sign-off do I get before it actually goes out? "You stay in control" is exactly where I'd want to feel the guardrails.

the "team of agents" framing is the interesting part - when Pazi sends real outreach on your behalf, is there a review step before anything external actually goes out, or does it just fire once it decides that's the next move?

the research-grounded-vs-guessed question above is the one that matters most imo - when Brave search and browser-use turn up conflicting pricing data for the same niche, does the agent reconcile the sources or just pick one and run with it?

Well done! I like Pazi because it was super quick to get started and I could feed it data from Excel or connect it to my database. It didn't ask for a bunch of redundant information. I'm just getting started but I really liked how it identified the "core activation" moment. I was surprised how accurate it was.

Love it! My app is built, ready to launch, but I feel very unsure on where to start, social media handlers are taken (including Product Hunt), videos recorded an paid for, iterations, PlayStore published, AppStore under review, yet I feel insecure... I will give this a go, so far exhausted my first 5k credits and intend to go for a bit more testing, thanks @zvonimir_sabljic1

Congrats on the launch! "Every time you come back, something has moved" is the angle that sold me — most agent tools still just sit and wait for prompts. Curious how you balance agents acting autonomously vs. asking for approval on things like outreach?

Congratulations on the Launch🎉🎉. Visited your website, absolutely loved the interface and functions, i am just worried about AI Hallucination, and what can be done to prevent it?

Lowkey need this!!!

For me personally, coding is the easiest part of building a product, but content writing or marketing-related tasks are really too difficult. They are often the parts that take me the most time. The appearance of Pazi just helped me out 🥳, thank you so much!!

Congrats on the launch, Zvonimir. The "build vs. figure out what to actually do with the business" gap is something I run into constantly between managing a dealership and building a property management site. Honestly, half the challenge isn't the build itself, it's knowing which of the ten things I could do next actually moves the needle. Curious how Pazi decides what to prioritize when someone gives it a vague goal like "grow the business." Does it ask clarifying questions first or just start proposing a plan?

Hey Product Hunt 👋

We're Shiv and Himanshu, makers of Goose Ads Remixer. Excited to launch it today!

What is Goose Ads Remixer? 🌟 ✨

Goose helps you create on-brand Meta ad creatives using patterns from ads already performing in your niche.

Paste your website, choose the ads you like, and Goose creates a complete batch using your real logo, product images, messaging, and brand style.

The problem we faced ⚠️

Creating Meta ad creatives was painful for us. Agencies were expensive, generic AI tools produced low-quality work, and doing it ourselves took far too much time.

Solution ✅

We noticed something simple: ads already performing in your niche are public. They reveal which hooks, offers, copy structures, and layouts are working right now.

So we built Goose Ads Remixer.

Goose learns from these winning patterns and creates new ads for your brand using your real logo, product images, and messaging.

How Goose Ads Remixer works 🚀

1. Create an account: make.gooseworks.ai

2. Paste your website

3. Choose up to 15 ads from your niche

4. Generate a complete batch of on-brand creatives in minutes

Goose doesn’t copy the original ads. It learns from the principles behind what works and applies them to your brand.

Built for AI agents and humans 🛠️

You can create ads directly from the web app:

👉 make.gooseworks.ai

Or access Goose through your AI coding agent using our CLI:

npx gooseworks install --all

Special for Product Hunt 🔥🔥🔥

1. Your first 10 ads are free.

2. We’re also giving Product Hunt users 50% off our Starter and Pro plans. Use coupon code: "PH50"

We’d love your honest feedback, especially if you run Meta ads. Which niches, ad formats, or creative styles should we add next?

Hey PH 👋 I'm Soham - one of the makers of Goose Ads.

We needed creatives for our own brands, so we tried the generic tools first. We faced two major problems every time.

Problem 1 ❌ - they start from a blank prompt, so you have no idea whether the ad even works.

✅ Goose starts from ads already proven to convert in your niche and remixes those, so you're building on what's winning, not guessing.

Problem 2 ❌ - even when the idea was right, the output had a tell. You couldn't always name it, but you knew instantly a machine made it - or it just looked like something you'd never actually spend money running.

After seeing enough, the tell was always the same five things 👇

1. The logo close but subtly rebuilt instead of your real one

2. Text that looks right until you actually read it

3. Stock-y product instead of your actual product

4. Brand colors drifted a few shades off

5. A layout that's "an ad" but not yours

✅ So that's what Goose kills 🪿

It pulls your real logo, real product shots, and real messaging, then reviews its own output against your brand and regenerates in a loop until it's right. Nano Banana Pro, GPT-Image-2 and Opus checking each other's work.

We're live today and want to make it sharper ✨

If you run Meta ads - what's the one thing that instantly makes you clock a creative as slop? That's the exact thing we're going after.

Hey PH - I'm Shiv, one of the makers of Goose Ads.

I thought I could add a little more color about the company we're trying to build.

The key insight we are building Gooseworks around is that AI agents like Claude Code, Cowork, Codex, etc are becoming the operating system for knowledge work.

But these agents need the right tools to do the work. That's why Goose Ads is built natively for AI agents as first-class citizens.

We're building the skills, tools and integrations that enable your AI agents to do real work, starting with the #1 problem for companies – growth.

Goose Ads is a starting point towards that vision.

We're looking forward to hearing what you think and how we can make it better.

Please feel free to use this launch to share your thoughts and feedback to help us improve.

honestly the first 10 free is what got me to try it, and the remix-on-your-own-logo part is genuinely cool. one thing i'd love though is a quick way to tweak the tone of the generated copy before exporting, like a little slider for playful vs professional. would save me having to rewrite every batch that comes out a little too salesy.

Been using it for a few weeks now, very impressive

Congrats on the launch! Cool concept. While scraping our site, it could only return #000000 as a primary color and I wonder if it's because our colors are OKLCH. Perhaps something to look into.

Congratulations on launch!
Goose can reduce the effort of running ads by a huge amount.

Can goose also make video ads?

So I've been using ProductHunt for a couple of weeks now. Some products get a cursory glance, some I play with for a little while. This one properly sucked me in and I have really enjoyed using it. The product feels professional and fun - and it definitely delivers results. I suspect I will need to keep tweaking my settings and improving my prompts to generate output that is the right balance between the original source and my own brand. And a couple of times a "work in progress" was better than the final output (would be good to know if I could revert back to a "work in progress" stage of the process). But overall this looks great - and I will definitely revisit. My one comment... pricing was hard to find. I discovered it in the documentation, but it certainly took me a while! Which seems a shame because (starting at $29 per month) it feels very fair pricing.

I'd love a side-by-side explanation showing what changed from the inspiration ad to the generated version. Does that happen already?

I am a performance marketer. I like this product because I was doing all of this research manually. Great ads leave clues, and this feels like it's decoding those clues automatically.

We are about to start running ads for our brand. This is a very timely launch. Congrats on launching!

I am wondering how Goose knows which patterns actually drove performance versus just looking visually similar? Does it decide on data-first or visual aesthetics first?

How do you handle ad fatigue with the remixed ads, and do you have any plans to integrate with popular ad platforms for seamless deployment?

I like the positioning because every marketer already studies winning ads, this just speeds up the creative process.

@himanshu_bamoria1Congrats on bringing Goose to PH! 🎉 Starting from winning niche patterns rather than a blank prompt makes total sense.

Right now it looks heavily focused on static/carousel ad creatives. Are you planning to expand this remixing concept into short-form video/UGC formats or motion graphic templates as well? Would love to hear how you see video remixing fitting onto the roadmap!

Hey Product Hunt 👋

Adding to what Shiv and Himanshu said — here's the itch that got me to build this.

Making ads always felt like a lose-lose: either expensive (agencies, freelancers) or a blank-canvas nightmare (staring at a design tool with no clue what actually converts). And the whole time, the ads that ARE winning in your niche are sitting right there in public. Nobody's short on inspiration — they're short on a fast way to act on it.

That's the gap we went after. Not generic AI slop, and not a copy of someone else's ad. Your brand, the proven structure.

A couple things I'd love for you to try:

• Generate a batch and see how it adapts the winning pattern to your brand

• Drop in a template or an ad you admire and remix it for your product

We're all ears — if you try it, I'd genuinely love your feedback. Tell me what's working, what's missing, or what would make this a daily tool for you.

creative process improvement over content creation is very useful - I wonder how precise the brand kit is. Most of these product rely on the brand kit accuracy for output this is creating it automatically ?

I like that this focuses on improving the creative process instead of just generating random ad designs. Using my own brading and messaging would make the results much easier to use right away.

@himanshu_bamoria1 Like the idea of learning from winning ads instead of starting from scratch. Upwards and onwards 🚀

Congratulations on the launch! Do you support creating video ads, for focusing on image ads for now?

Congrats on the launch. One thing I'm curious about, the Meta Ad Library shows you which ads are running and for how long, but not spend or performance. So how do you decide which ones are actually winning vs just live?

when Goose learns the pattern from a winning ad in my niche, where's the line between remixing the structure and just recreating that competitor's ad with my logo swapped in - is that boundary a manual review step or does the model handle it?

the nano banana / gpt-image-2 / opus checking-each-other's-work loop is the interesting bit - when the three disagree on whether an output passes the brand check, is there a tiebreaker or does it just regenerate until they converge?

I think if everybody is learning from the same winning ads then the meta feeds would really look similar over time.

the brand-drift question above is the UX risk, but there's a legal one underneath it too: "remixing patterns" is fine, but if the tool is looking at a specific competitor's ad to generate from, how do you make sure it's not echoing their actual copy phrasing rather than just the structure? a hook and a layout aren't protectable, but a distinctive line of ad copy can edge into IP territory if it comes out too close to verbatim

Just signed up for Goose Ads Remixer and tried it for my website ToolJunction. I’ve used a few AI ad creative tools before, but goose feels more polished and practical at this price point.

Starting with ads that are already working in your niche, instead of generating from a blank prompt, makes a noticeable difference. And, the outputs felt much closer to something I could actually run, and the overall experience was smooth.

The only issue I noticed was that it picked the wrong color palette for my website. Great work, team!

Most AI ad tools generate noise.

This actually helps teams iterate on what's already working. Thats a much stronger approach.

Congrats on the launch

Congrats on your launch! Excited to try this out.

finally a pgpooler alternative that actually feels modern, set it up behind our staging cluster last week and the latency under load was noticeably more stable than pgbouncer. love that it speaks the postgres protocol too

Congrats on the launch! The "without changing your app" part is the real pitch here. I run a Java/Spring stack on Postgres 17 and an embarrassing share of my production incidents trace back to connection handling — pool exhaustion during traffic spikes, session state breaking when something in the middle does transaction-level pooling, that kind of thing. Two questions: does PgDog do transaction-mode pooling à la PgBouncer, and if so how do you handle prepared statements / session state? And is the sharding routing done by parsing the query itself, or do I hint it? Genuinely curious — this layer is where I've been burned the most.

The "scale Postgres without changing your app" framing is the part that matters for teams that can't stop to re-architect — connection pooling plus transparent routing means the app stays naive. My one question as I'd drop this in front of an existing service: how does PgDog handle a single transaction that touches rows on multiple shards — does it coordinate the cross-shard write, or is keeping a transaction inside one shard still the app's job? Curious where the sharding proxy's transparency stops.

2M qps in production is the number that actually matters here. schema migrations on a sharded setup - does PgDog coordinate an ALTER TABLE across all shards atomically, or is sequencing that still on the DBA?

the config-driven instances not talking to each other is a clean HA story, but during a shard rebalance does every instance get the new shard map atomically, or is there a window where two instances disagree on where a shard actually lives?

How do you plan to handle failover and redundancy in PgDog, especially in cases where the load balancer or connection pooler itself becomes a single point of failure?

Lev — 2M qps across production deployments for over a year is a real data point, not a launch-day claim (as Gal put it below). WinBidIQ's Postgres load is basically bimodal: a nightly batch job writes tens of thousands of updated federal opportunity rows scraped daily, then daytime traffic is almost all reads — scoring, dashboard, search — from SaaS users. Right now we just add read replicas and hope the app connects to the right one. Does PgDog handle read/write splitting itself, routing reads to replicas automatically, or is that still a decision the app has to make before the query ever reaches PgDog?

Congrats on the launch! Running a SaaS on Neon's serverless Postgres — connection pooling is one of those things you don't think about until it bites you in production. Couple of questions: how does PgDog handle the connection limit quirks of serverless Postgres vs traditional dedicated instances? And is there a recommended setup for a Next.js + Prisma stack?

PgDog's connection pooling and transparent query routing is technically elegant. It removes the need to re-architect your data layer, which saves weeks of migration work. We've hit connection saturation issues under bursty SaaS workloads and this seems like a real fix. How does PgDog handle long-running transactions during a shard rebalance or failover?

Congrats on the launch. One thing that would help teams adopt this faster is a built-in dashboard or web UI for live monitoring of pool stats, query latency, and shard distribution. Most teams using pgBouncer end up bolting on tools like pgwatch2 or custom Grafana setups, so having something more turnkey would be a real differentiator.

2M qps in production is a real number, not a launch-day claim. the part I'd actually worry about with transparent sharding is cross-shard joins and multi-shard transactions, since that's usually where these proxies either quietly fall back to something expensive or just reject the query outright. does pgdog handle that at the protocol level too, or is there a class of queries you tell people to just not run against a sharded setup

I appreciate the honesty in positioning this as scaling without forcing app changes. Usually, "seamless scaling" secretly means rewriting half your queries to fit a proprietary router, so tackling this transparently at the protocol level makes a ton of sense.

Wondering what the tradeoff on latency looks like under the hood, when it's intercepting traffic on the fly, is there a noticeable overhead for high frequency, simple reads compared to just hitting standard postgres?

+1 to Andras's point — the single-use card scoped to a fixed budget is the right blast-radius model for agent payments, and the control layer around it really does feel like the product.

On the reliability side: when an agent retries a checkout that actually half-succeeded, does the single-use card give you idempotency so it can't double-charge? Agent retry loops are exactly where money quietly leaks.

And how do the virtual cards hold up on merchant acceptance — do you see declines from AVS / billing-address mismatch, or fraud filters flagging freshly-minted single-use numbers at checkout?

Would love your take here @keyserfaty

The idea of debit cards for AI agents immediately raises a practical question: how granular can the controls get? For example, can a team set per-agent limits, merchant/category restrictions, or time-based approvals before an AI agent is allowed to spend?

Fixed-budget, single-use cards are the right shape for agent spending. The next thing I would inspect is the receipt trail: task, merchant, amount, retry/idempotency state, and who approved the wallet funding. Agent payments need accounting-grade boringness.

love how focused the launch page is. the spec sheet showing spending limits per agent is such a practical touch.

Cool concept. Does each agent task get its own card, or do they all draw from one shared wallet?

The single use model does limit blast radius, they don't stop a compromised agent from spending that budget at the wrong place entirely. Curious whether there's any MCC level filtering on card issuance.

Love the idea of letting agents spend autonomously. One thing that would make me trust it faster is a built-in merchant whitelist so each agent can only charge approved vendors, with a clear audit log I can scrub through when the bill arrives.

How do you plan to handle fraud and liability with these debit cards, given that AI agents can potentially be compromised or act unpredictably?

love the idea of giving agents their own spending card, but the spend controls in the dashboard look really clean. nice work on the build.

The single-use, fixed-budget design is the same instinct I lean on for one-time Stripe discounts in CancelKit — scope the authorization tightly enough that it physically can't be replayed, instead of trusting app logic to enforce "only once." Curious how you handle a session that needs several small purchases: one card per task, or a budget that decrements across multiple charges?

would love to see a spending limit per agent so you can cap how much any single ai can burn through in a day, gives a lot more peace of mind when letting claude loose on random sites

would love to see transaction limits you can set per agent, so you know basically cap how much one bot can spend in a day. that would give me way more confidence letting claude loose on subscriptions honestly

So I actually tested this with my assistant to grab a domain, and it just worked without me having to handle the payment manually. Kind of wild watching an ai check out on its own.

It would be great if you could add a simple dashboard showing every transaction your agent makes in real time, so you can quickly spot anything suspicious or set spending limits per task.

The single-use card + fixed budget model feels like the right way to make agent payments usable for companies. letting an agent buy things online sounds powerful, but the real product is probably the control layer around it: exactly how much it can spend, where, and for what :)

Also, that launch offer is wild... $500/mo instead of $5,000/mo is definitely one way to make teams decide quickly :)) Curious how approvals work in practice. can companies require human confirmation for certain merchants or amounts, while letting smaller purchases happen automatically?

the single-use + destroyed-after-first-use design answers the double-charge question well, but I'm curious about the other side: disputes. with a normal card, "I didn't authorize this" has a human cardholder behind it that the bank's fraud process is built around. here the "cardholder" is a task an agent ran once and the card no longer exists by the time anyone notices something's wrong. if a merchant double-bills or a charge looks off after the fact, who actually files the dispute and does the issuer's chargeback process even have a lane for that, or does it end up as a support ticket with Agentcard instead of a real card-network dispute?

the single-use model is clean for one-off buys, but what about an agent that needs to set up a recurring subscription - is there no card left for the renewal to hit, or does that flow issue a different card type entirely?

each human gets a wallet and cards pull from it - when multiple agents under the same human are spending concurrently, is the budget split per-agent upfront or one shared pool where whichever agent hits checkout first wins?

I didn’t think about this before, but agents having their own cards actually makes a lot of sense.

Looks interesting. Maybe I am short of imagination, but based on the current state of where we are with agents, what kind of usecases you see are the low hanging fruits where you see this as a natural fit?

the single-use + human-authorized model answers most of the security questions already asked here. the one I don't see covered: chargebacks. if a merchant double-bills or ships nothing, a human cardholder can call their bank and dispute it. who's the "cardholder" here for dispute purposes when the actual buyer was an agent - does Agentcard handle disputes on the company's behalf, or does that fall back to the human who authorized the card?

finally spun up a quick test agent and pointed it at a dummy checkout. it actually paused for confirmation before charging, which i was not expecting from a card pitched at bots.

Honestly kind of wild seeing my agent check out on its own without me babysitting the checkout flow. The spend limits give me peace of mind though, would be a nightmare otherwise.

One thing that would be super useful is a per-transaction approval prompt sent back to the human owner through chat, so I can quickly allow or deny a purchase without digging through a dashboard.

This should also be available for some influencers if they have an agent. Finally, some bots would use money more responsibly compared to those creators. 😅 [know one influencer who is wasting money literally for anything]

the plain-english translation is the actual feature here, lsof telling you 'node' on port 3000 is technically correct and completely useless. one thing I didn't see mentioned - does it handle docker-proxy processes sanely? that's usually where this kind of tool falls apart for me, a container's port mapping shows up as some generic docker-proxy pid and you're back to guessing which compose project it belongs to

The ability to open local servers directly from the app is a nice touch. Small workflow improvements like that often end up being the features people use every single day without even thinking about them.

A developer I know constantly switches between projects and ends up with old servers still running in the background. I think the would save them a lot of time, especially when debugging local development issues. I'll definitely share it with them.

The plain-English process identity is the feature. “node on 3000” is technically true but not enough to decide whether killing it is safe. Project path, launch source, and last command/context would make this especially useful for Mac dev machines with several half-running projects.

The native Mac feel is spot on, especially surfacing the owning process right next to each port instead of burying it in a menu. Killing whatever's hogging a port in two clicks is exactly the kind of fix I've wanted for years.

@C3B Congrats on the launch! 🎉 Loving the clean macOS native UI and built-in firewall blocking.

Question on state tracking: is Portero strictly showing live snapshots of active listeners, or is there any background logging to help catch ephemeral/ghost background processes that grab a port and crash before you can inspect them?

How do you handle cases where a process is holding onto a port but isn't actively using it, and what's the criteria for suggesting to kill a process?

The context labels are the actual product here — "node" vs "Vite dev server, project storefront" is the difference between killing the right process and killing your teammate's demo. Curious how you infer the project context — walking up from the process cwd looking for package.json, or something smarter? Open source is appreciated, will poke around the repo.

This looks super handy for tracking down port conflicts. One thing I'd love to see is a way to save profiles or groups of ports I'm monitoring for specific projects, so I don't have to keep an eye on a long list manually every time I switch contexts.

Would love to see a quick right-click option to copy the PID or port number straight to the clipboard, since I often need to paste those into terminal commands when cleaning up stuck processes.

Hey congrats on the launch! Q - Does it keep a log of what was running on a port historically, or is it strictly a live snapshot? Thinking about those annoying edge cases where a background task grabs a port and crashes before you can even catch what it was

The "which node, from which project" problem is exactly why raw lsof never stuck for me — mapping a port back to a working directory is the useful part, not just listing PIDs. Two implementation questions: to enumerate every process's ports and to add a firewall block, does Portero need a privileged helper/root, or does it stay inside the user sandbox? And does a blocked port persist across reboots via pf rules, or only for the current session?

the visibility ceiling is worth spelling out — unprivileged lsof only sees sockets owned by your own uid, so a root daemon listening on a port stays invisible until the app escalates. for a dev tool that's fine. tagged as security software, that's exactly the port you'd want to see.

Are you planning to support deep Docker integration where Portero queries the local Docker socket? If we could resolve a generic docker-proxy process down to the exact container name and Docker Compose service, this would instantly become a must-have tool for local microservice debugging.

Turning port management into a visual, one click experience is such a smart touch, especially building on the native macOS firewall instead of reinventing it.

@ctresb It polls every few seconds by default, but what's the battery/CPU tradeoff on a laptop with that running constantly? would be nice to know if there's an on demand mode, or if the polling stays cheap regardless of how many ports are open

the auto-shot + real agent combo raises a different question than the privacy-masking one - since it's not just reading the screen but can actually run commands and edit files, what stops a prompt injection sitting on screen (hidden text in a webpage, a doc, an email) from getting treated as an instruction the moment auto-shot captures it? not asking about intentional misuse, more about the screenshot itself becoming an untrusted input to an agent that has real write access

Hey this could actually save so much time/tokens. However regarding privacy - can you mask out specific regions of the screen (like .env files or sensitive API keys) so the overlay completely ignores them while capturing the rest of your layout? Any considerations around that?

The screenshot-as-untrusted-input point is the real one. What's worked for us: observed content (screen, docs, pages) is always data, never instructions, and any write/run action needs a separate human confirm the on-screen text can't trigger itself. Read ≠ act. Great ship — the real-CLI-login approach is clever.

since multi-monitor grabs one tagged image per screen, does token cost scale linearly with monitor count, or do you downsample/compress before it hits the model? curious how that plays with a fast Ctrl+Alt+Space hotkey loop

Nice, neat demo! Is this constantly recording the screen and if yes, is that something token efficiient. Also it is also paired with voice input along with screen share thing?

Since you're using the full Claude Code agent instead of just an API, what's the most surprising workflow users have discovered? Has anyone been using it for something completely different than debugging?

Always-on-top framing feels really natural when I'm jumping between editor and browser. The fact that it reads my actual screen and runs as a real agent on my plan makes it way more useful than another chat sidebar.

Really like that this sits on top of the existing Claude Code login instead of adding another agent layer.

One thing I’d want for daily coding use is a small run receipt after each screen-driven action: screenshot source, files touched, commands run, and which step came from what was visible on screen. When the input is a screen, debugging a wrong action later gets harder than with a normal prompt.

That kind of trace would make the overlay feel safer for long work sessions, not just faster.

Life saver!

Congrats on the launch! Since it reads whatever's on screen before answering, how do you handle sensitive info that might be visible (API keys, customer data, etc.) - is there any redaction or is it on the user to be careful what's showing?

Love how it pulls from your actual screen context, that's the killer feature for me. One thing that would push it further: a quick toggle to switch between which monitor Claude reads, since I often have a reference doc on my second display I don't want it touching.

I really like that this is more than just a screen-aware chat. The fact that Claude can act on the file that’s already open makes it feel like part of the workflow rather than another window to manage.

Are you considering making context and permissions configurable for each overlay? For example, one overlay could be tied to a specific project in read-only mode, while another could be allowed to edit files. That could make using multiple overlays feel much safer and more predictable.

Nice work, Jason!

Hi Jason! Let's say I'm using this floating Claude Overlay on a web browser which has a very long content - meaning that we have to scroll down to read the full page. In this case, how can we make Claude Overlay read the whole context in our page?

One thing I'd love to see is cross-monitor support for the screenshot capture, specifically the ability to pick which screen Claude reads instead of grabbing everything at once. Would make it way more useful when working across a laptop and external display setup.

finally something that lets claude see my actual screen instead of me copy-pasting everything. the always-on-top window is genuinely handy when you're jumping between docs and code.

When someone enters a domain, what does Branda look at to decide the brand direction — page copy, colors, logo, existing metadata, or all of that? The “any domain into on-brand ads” promise is simple, but I’d be curious how much control a marketer or designer gets before the final ad is generated.

A freind of mine runs campaigns for several small businesses and spends a lot of time recreating brand styles for each client. This looks like something that could speed up that process quite a bit. I'll be sharing it with them.

Hello Yahia, congratulations on the launch. I like that you're using actual brand assets instead of asking users to upload everything manually. That should make the generated ads feel much more authentic and consistent with the company's identity.

Congrats on the launch. I run a dealership, and I'm always testing ad creative for inventory, so the "generic images that don't look like your brand" problem is one I know well. I usually end up either paying a designer or settling for something close enough. Curious what happens with a site that's mostly product photos and pricing rather than a strong visual brand on the homepage, like a dealership inventory page. Does Branda pull enough signal from that to generate something on-brand, or does it work best when there's an established visual identity to draw from? Excited to try it out.

Today @Goose Ads Remixer also launched. How does Branda differ?

The no-login, paste-a-link flow is such a smart move, the whole thing feels like a party trick at first glance but then you realize it's pulling real brand context behind the curtain.

@yahia_bakour3 Congrats on the launch! 🚀 Turning any URL into on-brand creatives in seconds using Context.dev's Brand API is super clever.

Quick question: does Branda work best when pointed at a root domain/homepage, or can it pull specific product/landing page context (like promotional pricing or specific feature copy) to generate campaign-specific ad variants on the fly?

Love how it pulls the brand assets automatically. One thing I'd love to see is a quick preview grid showing all the ad variations side by side before downloading, so I can pick the strongest one without generating them one by one. Would save a lot of iteration time.

This looks really handy for quick ad creation. One thing that would make it way more useful is adding support for exporting in vertical formats like 9:16 for Instagram Stories and TikTok, since right now it sounds focused mainly on LinkedIn and X. Most brands need those too for full funnel coverage.

A Meta ads version would be a huge unlock since a lot of teams live in Ads Manager. Could you add Facebook and Instagram formats with the right safe zones, or at least let me pick the aspect ratios I need before generation?

congrats @yahia_bakour3 what happens if a website has zero clear images on the homepage does the system automatically generate geometric brand layouts based purely on the color codes it found?

A nice angle would be letting users paste an offer URL or product page and Branda scrapes the actual headline, price, and CTA so the ad copy isnt guessing at what the brand is selling right now. Hooking into the live page rather than just the homepage would make the creatives feel much more timely and accurate.

It would be awesome if Bug Shot could automatically generate clear reproduction steps from the captured session. That would make it even more valuable.

The no account requirement is a great touch .Curious does it support recording user interactions leading up to the bug as well?

Hey Product Hunt 👋

I'm a product designer. My job is to notice when something is four pixels off — and then the hard part starts, because I have to describe it. I'd screenshot the bug, type "the spacing looks off here" into a ticket, and a developer would have to guess which property I meant. The rest of the report was clerical work: reproduce it, copy the URL and the browser version, dig through the console, paste it all into a tracker. By the time you're done you've forgotten what you were actually working on.

So I built BugShot, a Chrome side panel that collapses that into one pass:

• Pick an element and fix it live. Edit its CSS right on the page — through form fields or a real CSS editor with autocomplete and color swatches. Every change is tracked as a before → after table in the report, so the developer gets the diff, not my adjectives. It resolves var() chains too, so the report says --color-primary instead of rgb(79, 70, 229).

• Capture what you need. An element, a region, a screen recording, or the last 30 seconds of the tab. Annotate it before attaching.

• Logs come along for free. Console, network, and user actions are recorded while BugShot is running, including inside cross-origin iframes.

• File it where you already work. Jira, GitHub, Linear, Notion, GitLab, Asana, ClickUp — or share straight to a Slack channel or DM.

No sign-up, no account. Everything runs locally in the extension and posts directly to your tracker — your screenshots, logs and report text never touch a server of mine. The one exception is OAuth: platforms that require a client secret (Jira, GitHub, Notion, Asana, ClickUp, Slack) have their auth code exchanged through a small proxy that relays the token and stores nothing. Linear and GitLab use PKCE and skip it entirely.

I'd love to hear where it breaks for you — especially which tracker or workflow you'd want next. Happy to answer anything in the comments.

honestly this looks super useful for my workflow but it would be great if you could save a default assignee or team channel per project so reports go straight to the right place without me having to select it every single time

The before/after CSS diff plus console/network evidence is the important bit. AI can draft the ticket, but the raw evidence needs to travel with it so the developer is not debugging a summary of a bug instead of the bug itself.

One thing I'd love is the ability to mark sensitive fields like emails or tokens for auto-redaction before the report gets sent. Even with a free tool, hitting send without scrubbing that stuff feels a little risky.

the part that stands out is the privacy claim, not the css diffing. a lot of bug tools that say everything stays local quietly still route auth tokens through their own servers, so I like that you called out the OAuth proxy explicitly instead of hiding it. does that proxy log anything at all for debugging on your end, or is it truly stateless per request

finally something that grabs console logs automatically because honestly copying those into jira tickets every single time is kind of my least favorite part of the job. the live css tweak diff idea is actually clever too

Sinhyeok — the redaction thread with Hakan and the OAuth-proxy answer to Gal are exactly the kind of detail that builds trust here. I'm coming at this from the other side of your usual user, though: WinBidIQ's customers are federal-contracting SMB owners, not developers, and when one of them hits a bug they don't know what a console log is — I get "it's broken" over email and have to reproduce it myself with no repro steps. Is there a lightweight capture flow a non-technical external user could trigger — a link they click, describe the issue in plain English, and it still grabs console/network/environment for me — or is BugShot built around the reporter already having the extension and knowing what they're looking at?

This is a great idea. I do have a couple questions. You mention CSS editing in realtime. Who is the target audience for this feature? Would they know the cascading impact of such a change? A higher level CSS change could break or alter other areas of the site. Also, does it work with minified or compiled CSS? If so how does the actual line number of the CSS update get communicated back to the developer? You mention the network traffic at the time of the bug is attached, this to me is the most interesting part. Is a HAR file being attached? Is it parsed with data analysis? In my world CSS is rarely the case but the underlying dependencies, APIs or libraries are. The HAR analysis could really help debug the issue. Great work, I like where this is going.

Finally something that captures console and network logs without me messing with devtools first. Honestly the side-by-side CSS tweak is kind of genius, saved me a whole back-and-forth with the designer already.

The template-first workflow is genuinely smart, takes the friction out of starting from a blank canvas. Love that exports go straight to MP4/WebM at 4K without a watermark trap.

Congrats, ran a few tests and the export quality is great.

You tagged this "vibe coding" so I have to ask: any plans for an MCP server, so an agent like Claude Code could feed it assets and pull back an animation directly? The manual flow is smooth, but that'd be the unlock for the dev crowd you're aiming at.

The pick-template → drop-assets → export flow is the right call for a portfolio piece you want done in a minute, not a full motion-design project. As a day-one user: after I export, does Animos save the project so I can reopen and tweak that same animation later, or is each export a fresh start with no saved state? And are the 30+ templates the whole canvas, or can I adjust timing/keyframes beyond the presets?

During beta, what surprised you most about how people used Animos? Was there a template or workflow that became unexpectedly popular and changed your roadmap?

Honestly the template workflow is really smooth, dropped in some assets and had a clean animation exported in like 2 minutes. The 4K output is a nice touch for portfolio stuff.

The templates actually look solid and the export was quick, kind of surprised how clean the MP4 came out for a browser tool.

Yessss!! This is exactly what I have been looking for. Thank you for having the creatives in mind. What a beautiful seamless way to showcase my works!

Hey Product Hunt! I'm Jake.

I built Sales Studio because of a demo that went sideways. I was juggling a dozen windows on a sales call: the product in one, my notes in another, pricing in a third, and Slack pinging over all of it. My prospect saw every bit of that chaos. The demo was fine. It just didn't look like I wanted it to look.

So I built the tool I wished I had: a studio view for me, a clean stage for them.

Here's what that means in practice:

• You share one window. It always looks composed. Everything else lives in your private studio view.
• A teleprompter that scrolls as you speak. Voice detection moves your script along, so you never touch it mid-pitch.
• Displays you switch like a TV director. Windows, screens, web apps, videos, or your camera, each with clean transitions. Number keys jump between them.
• Scene layouts. Size and position your shared window over a branded background, drop your camera beside it, and it all mirrors live to the audience.
• Camera controls built in. Background blur runs on your Mac (nothing goes to a server), plus mirror, shapes, borders, and shadows.
• A tool bench you arrange yourself. Notes, checklists, a stopwatch, even an embedded web panel for looking things up mid-call.
• Freeze the audience view. Need to do something off camera? They keep seeing the last frame while your camera stays live.
• Record what they saw. Get a QuickTime-ready MP4 of the exact program output, camera and all.

It's macOS only (Apple Silicon), works with Zoom, Meet, Teams, or anything that can share a window, and there's a 14-day free trial. From $9.99/mo.

I'd love your honest feedback, especially from anyone who runs demos for a living. Ask me anything!

Finally something that fixes the window chaos during live demos. The floating script view while everyone else sees a clean screen share actually feels like how demos should have always worked.

Congrats on launching! Curious how you handle latency when multiple prospects join a live demo at once - does the screen share stay smooth, or does performance dip with more viewers?

the composited "one clean window" model is the whole value prop here, but it also means the polish depends on your Mac behaving during the call. what happens when something outside Sales Studio's control fires mid-demo - a Slack or calendar notification banner, a Time Machine popup, a spinning beachball while a heavy app loads? does the studio compositing sit at a layer that's genuinely isolated from OS-level overlays, or is there still a window where a stray notification could bleed into the "clean" shared view the same way it would without the tool?

The features are interesting but I think that most sales rep are already used to workflow they have even though it's a bit messy, what were the customers feedback ?

Right on the pain point—having to juggle multiple sources during a sales call is a major challenge and adds unnecessary stress. I built a GTM strategy for a 120-person sales team, and this would have been incredibly useful back then. How does it integrate with CRM?

Finally something that fixes the worst part of screen sharing. The notes overlay is super handy and it stays out of the way of what your viewers actually see.

the script overlay next to my actual screens is honestly such a small thing but it completely changes how a demo flows, way less fumbling around</mm:think>the script overlay next to my actual screens is honestly such a small thing but it completely changes how a demo flows, way less fumbling around

the teleprompter-style script view next to your display picker is genuinely clever, makes live demos feel way less chaotic than alt-tabbing through 6 apps.

This looks genuinely useful for anyone running demos, honestly the window juggling thing has burned me so many times. One thing that would make it way better is letting you save custom layouts or demo presets so you can instantly switch between different setups depending on the audience, kind of like a snapshot system. That would save a ton of prep time.

The split-view layout for presenter vs. audience is really thoughtful, especially keeping the script and notes out of the shared screen without a clunky second monitor setup. Whoever designed this clearly ran enough bad demos to know exactly what needed fixing.

the "flat folders, no nesting" call is the right instinct for a capture tool - nesting is where these things always turn into filing cabinets nobody maintains. curious how it holds up once someone has a few hundred notes in one flat folder though - is search the only way back in at that point, or is there something like tags/pins to cut across folders without reintroducing hierarchy

Really like the direction you're taking with Flyout. Moving beyond likes and comments to enable real conversations between creators and their communities feels genuinely valuable. Excited to see how this grows - congrats on the launch!

The ⌥⌘N hotkey sliding a full editor out from the edge is such a clean trick — feels like the kind of thing macOS should have shipped with. Nice touch on keeping everything as plain Markdown too, that's the right call.

A lifetime license sold me, and the local Markdown storage is the right call. One thing I'd love: a quick hotkey or menu bar option to drop a clip from any app into Flyout, so I can grab a paragraph or screenshot without switching focus first.

The plain Markdown under the hood is a really thoughtful choice, makes it feel like it'll last beyond any app update. The ⌥⌘N shortcut is a nice touch too.

Hey Product Hunt! 👋 I'm Orçun, the solo developer behind AutoShelf.

I built AutoShelf because my Downloads folder had become a graveyard of DMGs, screenshots, and random files I swore I'd "clean up later." The existing tools were either powerful but byzantine, or simple but limited. Nothing felt like it was built for this decade of macOS.

So I made AutoShelf: a menu bar app that watches your folders and runs your rules. Set it up in 30 seconds with a template, then forget it exists.
What it does:
- 🗂 10 conditions: file type, extension, where it came from (download URL or app), age, size, name, last-modified/opened, or just "everything." Combine multiple with AND + NOT.
- ⚡ 10 actions, chainable: move, copy, trash, rename, tag, archive to ZIP, optimize images, import to Photos. Stack them in one rule: tag, then archive, then move.
- 📋 9 built-in templates: one-click "Organize Downloads," "Auto-trash DMGs," "Clean Desktop," "Sort by Source," and more.
- 🌍 19 languages, native SwiftUI, lives in your menu bar (or a full window, your call).

🤖 MCP support (this is the fun part):
AutoShelf ships with a full MCP server baked into the app binary, so it works natively with Claude Code, Cursor, opencode, and any other local MCP-compatible client. Point your client at AutoShelf.app/Contents/MacOS/AutoShelf and you get 13 tools: create_rule, edit_rule, delete_rule, enable_rule/disable_rule, run_rule, list_rules, list_groups, get_rule, get_status, pause_monitoring, resume_monitoring, and pick_folder (which pops the native macOS folder picker). So instead of clicking around a UI, you can literally say "watch my Downloads folder and trash any DMG older than 3 days" to Claude or opencode, and it builds and installs the rule for you. Write operations are gated behind a Pro-only toggle that defaults to off, so nothing happens without your say-so.

Why AutoShelf:
- Set-and-forget simplicity: templates get you running in seconds, no configuration wizard required
- Built for today's macOS, not ported from 2012
- Scriptable end to end: MCP for AI assistants, plus a Go CLI for terminal workflows
- One-time $19.99, no subscription.

What's next:
- LOT MORE conditions and actions.

- File conversions (HEIC→JPG, MOV→MP4, WAV→MP3)
- Cloud uploads: route files to iCloud, Dropbox, Google Drive
- Scheduled downloads: pull from RSS feeds, URLs, or cloud storage
- Rule creation wizard: step-by-step guided flow

Have a feature idea? I'd love to hear it → https://useautoshelf.com/support
Happy to answer any questions, feedback welcome! 🙏

A freind of mine downloads hundreds of design assets every week and spends way too much time sorting them into folders. This seems like something that could remove a lot of repetitive work from their routine. I'll definitely send it their way.

Hello Orçun, congratulations on the launch. The Downloads folder example immediately felt relatable because mine also ends up full of screenshots, installers, and random files. I like that the focus is on creating a rule once and then letting everything happen automatically in the background.

the MCP server baked into the binary is a smart move, that's a much easier install story than a separate npm package. one thing I'm curious about: if two rules both match the same file (say one moves it and another renames+archives it), what decides execution order? and can a rule created through the MCP tool by an AI client end up conflicting with one a human set up in the UI, or does it validate against existing rules before creating a new one?

Congrats on you launch 🎉

I'm really glad to meet my favorite app on PH again!

Thanks to AutoShelf, my download and desktop keep clean, just as I intended. The new version's UI is even more refined. I'll continue to support you.🫶

I appreciate that this isn't trying to become a full file manager. Instead, it focuses on one problem and solves it well. Those are usually the tools that quietly become part of someone's daily workflow without demanding attention.

the way rules chain together with live previews before saving is a really thoughtful touch. makes it feel less like config and more like actually designing a system.

The MCP + CLI support is what makes this more than another folder-watcher — it means an agent can drive the organizing, not just a background rule quietly firing. Two setup questions: through the MCP server, can an agent create and modify rules and query where a given file ended up, or is it limited to triggering rules that already exist? And are the rules stored as a portable file I can version and move to another Mac, or are they locked in app state?

The MCP thing is sick. i told claude 'auto trash DMGs after 3 days' and it just made the rule for me. didnt even open the app. thats how all mac apps should work honestly

Yo this is actually nice. I tried Hazel a while back and returned it in like an hour, way too much for what I needed. AutoShelf I literally set up 'dump everything in Downloads into folders by type' in like 2 minutes and forgot about it. menu bar icon sitting there doing its thing. good stuff.

Set up a rule to sort my downloads by file type in about a minute, and it just kept working in the background. Wish I had this years ago, my desktop is finally usable.

The little rule preview that shows exactly where each file will land before you hit apply is such a thoughtful touch. Clean execution on what could have been a frustrating setup experience.

As someone whose Downloads folder constantly fills up with installers and screenshots, this feels like a practical solution. The one-time pricing is also refreshing. Curious how AutoShelf handles conflicts when a destination file already exists—does it skip, overwrite, or prompt?

No way Trump Accounts made it to Product Hunt :))

I had to double-check this was a real launch. honestly, I might be way too European for this one, but seeing a government-backed child savings account presented like a startup product is definitely a PH moment :)

Trump Accounts are such a huge get for @Robinhood. Man, nice job @jgebbia and @vladtenev! The iOS app is basically Robinhood under the... hood.

If you want to learn more about this, check out this segment from the All-In Podcast.

Financial literacy starts early, and tools like this can help families better understand investing. Best of luck with the launch! 🚀

Hey Product Hunt,

I built Breva because I wanted a tiny reset ritual that lived on my Mac instead of my phone.

Most breathing apps pull me into a separate device, a subscription funnel, or a full meditation experience. I wanted the opposite: fewer features, less ceremony, and a quick way to reset from the Mac I’m already using. Breva is intentionally smaller: pick a rhythm, choose 3, 5, 10 minutes, or unlimited, and follow the orb, tones, and trackpad taps.

The unusual part is the haptic guidance. If your Mac trackpad supports it, Breva can quietly tap the breathing rhythm under your fingers, which makes it easier to follow without staring at the screen.

What is in this first version:
- Coherent, Easy Calm, Balanced, Calm, Box, and 4-7-8 Sleep patterns
- 3, 5, 10 minute, and unlimited sessions
- Visual, audio, and trackpad haptic cues
- Menu bar quick actions
- Local session history and preferences
- No account required
- Oh, and it's free :)

I would love feedback on two things:
- Which breathing pattern feels most useful during the workday?
- Does the trackpad haptic guidance make the session easier to follow?

Thanks for taking a look.

Brilliant idea. Hate how much I grab for my phone when I need a brain break. Great alternative.

the haptic tap idea is genuinely clever, using hardware that's already there instead of another screen to stare at. curious about interruptions though - if a Slack ping or calendar alert fires mid-session does the orb just keep going underneath it, or does it pause/reset? the whole point of a reset break is not getting yanked back into work mid-breath, so how it handles that matters more than the pattern selection imo.

this idea is great! as someone that's always had a passion for mindfulness, it's cool to see an app like this. would you ever consider a mobile app? i know the app uses a trackpad but it would be convenient to have a mobile equivalent as well. great work!

Congratulations on the launch! Keeping the reset on the Mac rather than the phone is a smart choice since the phone is usually the thing you're trying to step away from. What I'd be curious about is whether people actually come back to it on a stressful day, or whether it slips into the background once the novelty passes.

The trackpad haptics as a breathing pacer is such a clever use of hardware that's just sitting there — eyes-free guidance is exactly what's missing in most breathing apps, you either stare at an orb or you're lost. Did you run into limits with the Taptic Engine on the slower patterns like 4-7-8, or does it handle long holds fine? Local-only and no account is a great call btw — building in the health space myself and I'm convinced that's the right default for anything body-related.

Bro this is wild!

rooting for this to go viral.

The expanding orb is such a nice touch, way more intuitive than those rigid square timers I've tried before. Appreciate that it all runs locally too, no account or signup friction.

love that the orb actually responds to your finger on the trackpad instead of just playing a generic animation. running everything locally with zero setup is a really thoughtful touch for something meant for quick resets.

does this work with an external Magic Trackpad too, or is it built-in-trackpad only? a lot of desktop Mac setups don't have a haptic trackpad plugged in by default so curious how you're handling that gap.

The multi-voice part caught my eye. When VocalVia turns a document or article into audio, how much control does the user get over which sections use which voice? For example, can someone mark quotes, headings, or different speakers before generation, or is that handled automatically? Since the tagline mentions editable audio, I’m also curious whether edits happen at the text level, the timeline level, or both.

Editable script before audio is the right call — most TTS tools skip exactly that step. My question is about fiction: in a novel excerpt the speaker is usually implied, not tagged ("she said" disappears after the first exchange). Does the outline step attempt dialogue attribution so characters land on distinct voices, or is fiction a manual-reassignment job today? That feels like the gap between "documents" and "books."

this is one of the more thoughtful takes on document-to-audio I've seen, mostly because the editable script step exists at all instead of piping straight to TTS. one thing I didn't see covered yet: pronunciation of jargon, acronyms, and author names in research papers. that's usually where TTS breaks the listening experience even when the sentence structure is fine - a mispronounced term every few minutes pulls you right out of it. is there a way to correct or lock in a pronunciation once, so it applies consistently across the rest of that document (or future ones with the same terms)?

I believe that editing the script will take time. If I have to review the content anyway, I might just read the PDF instead. Were there users that could relate to my feedback ?

The editable-per-segment part is what stands out to me over plain TTS — being able to refine one speaker's line without re-rendering the whole thing is the tedious bit everywhere else. Turning a research paper into a two-host format is a genuinely nice use case. How much control do you get over pacing and emphasis within a segment, or is picking the voice the main lever right now?

Multi-voice from a single document is a nice unlock — script studios usually make you assign voices manually shot by shot. How are you handling voice consistency across a long document, especially if the same "character" voice needs to reappear later? That's been one of the trickier problems building voice tooling alongside video generation.

Is it primarily for generating podcasts?

@Zoey Congrats on the launch of VocalVia! 🚀 Solving the "saved articles I never actually read" problem by turning them into editable multi-voice podcasts is brilliant.

Quick question on handling complex structures: when processing dense documents (like 30+ page research papers with heavy inline citations, tables, or math code), how clean is the initial script translation? Does it automatically distill complex tables into natural conversation, or do you find users usually need to manually edit those segments first?

The multi-voice output caught me off guard, it actually feels like a real conversation instead of that robotic single narrator you get elsewhere. Refining one segment without regenerating the whole file is a lifesaver for long papers.

Dropped a 40-page research paper in and got a pretty clean two-host audio rundown in a few minutes. Editing individual segments before export is a nice touch, especially for cleaning up citations.

Hey! Editable scripts before audio generation is the part most TTS tools skip, and it's the part that matters. Does it keep speaker assignments stable when you re-edit a segment, or does the outline regenerate?

A timeline or chapter marker system would be really useful so I can jump back to a specific section in long papers without rewinding through the whole audio.

Finally tried it on a 40-page research paper and the multi-voice narration actually makes the dense sections easier to follow than my usual text-to-speech. Editing individual segments before export is a nice touch too.

Hey Product Hunt! 👋

I'm Denis. I built AlterSend because sending a big file to someone is still way harder than it should be. You upload it to WeTransfer or Google Drive, wait, hit a size limit, send a link, and your file ends up on a server that isn't yours. It's slow, it's capped, and it's not private.

AlterSend sends files straight from your device to the other person's. No upload step, no account, no size limit, end-to-end encrypted. Works anywhere - same room or another country.

We released a month ago and it's going better than I hoped: 7,000+ downloads, almost 1,000 GitHub stars, and a lot of great feedback from users that's helping me decide what to build next.

How it works:
• Each transfer creates a random 32-byte key - that's the join code you share with the receiver
• Devices find each other through a public DHT using a hash of that key - the key itself never leaves your devices
• The connection is encrypted end to end. If a direct connection isn't possible, a relay passes the encrypted stream along - it can't read your files and doesn't store anything
• When you disconnect, everything is wiped — nothing is stored anywhere

It's open source (Apache-2.0), free, and there are no accounts.

I'd love honest feedback - especially: what would it take for you to stop using WeTransfer, Drive links, or USB sticks? I'll be in the comments all day. 🙌

no account or cloud step is honestly so refreshing, sent a big video between my laptop and phone in like seconds.

How do you handle firewall and NAT traversal in the peer-to-peer file transfer process?

This looks really clean, love that there's no account or size limit involved. One thing that would make it way more usable for me is the ability to resume an interrupted transfer if someone drops off wifi for a second, since flaky connections mid-download are the worst.

The no-account, code-or-QR pairing is such a refreshing call back to simplicity. Love that it's truly open source too, not just lip service.

Almost everybody is logged into google drive and I see that the difference between AlterSend and Drive is that nothing gets stored. It's great but there's also a downside if someone wants to delete the file later, how often does that happen ?

Denis — the Hyperswarm + blind-relay fallback you explained to Aymen above is a clean answer to NAT. The harder case I run into with WinBidIQ's audience (SMBs teaming up on federal contract bids) is government and defense-contractor networks that don't just sit behind NAT — IT policy actively blocks outbound P2P/DHT traffic at the firewall, full stop. Does AlterSend's traffic pattern look enough like ordinary HTTPS/WebSocket to get through a typical corporate proxy allowlist, or is a network locked down like that just a hard no regardless of the relay fallback?

the "your files never touch the cloud" pitch is solid on paper but real P2P is never that clean once both sides are behind restrictive NAT/CGNAT - like on separate corporate wifis. what happens then, does it fall back to a relay server (still E2EE, just not direct), or does the transfer just fail and you're back to a USB drive?

Sent a video between my phone and laptop in seconds, no signup or upload wait was genuinely refreshing. The QR code pairing feels way cleaner than emailing myself a link.

Love the no-account, P2P approach - feels way cleaner than the usual cloud shuffle. One thing I'd want though: a quick way to pause or throttle transfers, since big video dumps tend to saturate my whole connection and wreck whatever else I'm doing online. Even a simple slider would do the job.

Direct transfer usually means both devices need to be online at once. How do you handle sending something to someone who isn't around until later, or is that out of scope by design?

my most repeated two-step is crop/reformat then caption - I record a quick screen demo at whatever aspect ratio the app gives me, then need it as a vertical clip with captions burned in before I can actually share it anywhere. the "no timeline" pitch really clicks for that specific case, since a full editor feels absurd for something that's really just two operations. is aspect-ratio conversion (16:9 to 9:16, with smart crop rather than just adding bars) one of the 26 tools already, or still on the list?

The no-timeline framing is spot on — as an indie iOS dev I mostly just need to trim + caption a quick clip, and opening a full editor for that is overkill. Love that most of it runs on device. Are the captions auto-generated, and does the on-device export keep the original resolution?

Love the "no timeline" framing — most video tools assume you want a full editor when really you just need one clip trimmed or one shot dropped in. Curious how you're handling model/rendering under the hood for quick jobs like this — are you generating anything AI-side, or purely editing existing footage? Ask because balancing speed vs. quality is the constant tension on the generation side of video tools too.

Is this mobile only? Or there's a web version too?

One thing I'd love to see is a way to save a custom sequence of edits as a preset. I often do the same trim, mute, and caption steps for short clips, and being able to chain them and reuse the flow with one tap would be a huge time saver.

Tried it on a phone for trimming and adding a quick filter, and it handled the clip without uploading anything. The on-device processing makes it feel way faster than the editors I usually use for this kind of small job.

A keyboard shortcut for trimming would be amazing, like hitting a hotkey to set in and out points without scrubbing with the mouse. Also would love to see a simple batch queue so I can line up a few clips and walk away while it processes them one after another.

The pipeline that lets you keep editing the output of your last action without re-importing is a really thoughtful touch, the kind of small detail that makes quick clip work feel fluid instead of fiddly.

I’ve been using Mojave Paint since it first appeared on Reddit and I love it - I bought it straightaway. John regularly updates the app and adds new functionality, and it now does ~95% of what I need from graphics software. Being spectacularly old I appreciate the 90s aesthetic of the interface, but mostly I appreciate that it’s lightweight (9.2MB installed), fast and does almost everything I need it to do. Based on the last few months’ progress, I doubt I’ll need to use any other software by the end of the year. Unless you’re a graphics professional - in which case you probably need the bells and whistles other, vastly more expensive, software provides - I thoroughly recommend Mojave Paint.

love that you kept this purely about the craft—no AI gimmicks, just a real bitmap editor. the handmade-from-seattle approach reminds me of the old shareware era, which feels like exactly what mac image editing needs right now.

I get a little depressed when I see Reddit threads of people asking for AI tools to make screenshots for their iOS app. Just make them yourself! You're not a great designer, so what? Ugly is better than AI-loooking. And Mojave Paint is just one of many tools you can use to make graphical assets. I'm not saying it's better than other options, but it's an option, and a pretty darn good one if you're on an Apple Silicon Mac.

'If you're looking for AI features, look elsewhere' is a great line. I spend my days compositing images with ImageMagick math ops, and the older I get the more I value tools that do exactly what you tell them. Is there any scripting/automation hook, or is direct manipulation the whole philosophy?

I really appreciate the intentional "if you're looking for AI, look elsewhere" stance. Do you see that as a temporary product decision, or a long-term design philosophy? In other words, are there AI features you believe could genuinely improve image editing without taking away the sense of craftsmanship?

finally a paint app that just gets out of the way. loved how the brushes felt on my m2 air, no lag, no weird cloud sync nonsense, just pure editing.

There is a real place for small, fast, native tools that preserve direct manipulation. Especially for App Store assets and product screenshots, “I made the tradeoff myself and can adjust it in 30 seconds” beats a generated image that looks polished but does not respect the product.

Downloaded it yesterday and the layer blending feels really snappy on my M2, no lag like in some of the bigger apps. The brush engine has a nice texture to it that reminds me of older Mac tools.

Love the no-nonsense approach, especially the stance on AI. One thing that would make this way more useful for me: native support for layered TIFFs and PSDs so I can round-trip work with older projects without flattening everything on import.

Love the no-AI stance, that alone sets it apart. One thing that would make me reach for it over Pixelmator though, native support for opening and saving indexed-color PNG files with a real palette editor. Most editors downsample or strip the palette these days, and it's a pain when working with retro game art or pixel assets. Would be a perfect fit given the Mac heritage angle.

Played around with it for ten minutes and the live grid where everyone is carving out cells at once is oddly addictive. Fun that you can just ask Claude to whip up a loop and copy other peoples patterns with a click.

I did not know I needed a room full of strangers quietly building a beat alongside me, and yet here we are. Theo, this looks like the kind of place I could happily lose a whole evening.

the way the grid pops in as other people lock in their cells around yours is genuinely satisfying, feels like a tiny shared studio.

Have you considered adding a way to lock down a row or column of cells so you can lay down a signature pattern without someone else overwriting your kick or snare mid jam. Would help a lot for keeping a groove steady while the rest stays collaborative.

To all those hunters visiting the page now (I see you on Umami 👀😅) would appreciate any questions / feedback. To those who have upvoted, a big thank you, it means a lot. Keep them coming!

the "the chain isn't decoration" framing actually lands, most onchain gimmicks are a UI backed by a chain that didn't need to be there, this one would genuinely stop working without it. question on the Claude flow specifically: when Claude hands you back a pre-loaded link, is that loop just sitting client-side until you commit it, or does generating the link already claim/reserve those cells on-chain before you've even listened to it?