PH热榜 | 2026-05-22

Hey Product Hunt 👋

I'm Yunhao, co-founder and CEO of @TestSprite.

Today we're shipping TestSprite 3.0 — autonomous end-to-end testing that actually understands what your app does, not just what your code says.

✍️ Here's how it works:

1. Point TestSprite at your live web app or API endpoints. Drop in your PRD or product spec if you have one (strongly recommended — it sharpens what we test).

2. We send a swarm of AI agents to explore it in parallel — clicking through every feature like real users.

3. They generate full backend + frontend test suites from what they saw. Then run, debug, and auto-heal them for you.

🚀 What's new in 3.0:

• Parallel exploration fleet — dozens of agents map your app before a single test is written. As far as we can tell, no one else is doing this yet.

• Backend gets serious — multi-dependency integration tests, Dynamic Variables, Auto Cleanup, and a Data Flow view that makes debugging an agent-generated test feel like debugging your own code.

• Frontend tests auto-heal when your UI drifts. Auto-auth keeps nightly regression sane.

• Accuracy up ~40% — and it really shows on complex enterprise projects, where most agents fall apart.

• Coverage jumped from ~20 to 50+ meaningful cases per run.

• CLI for Claude Code and Codex users — coming soon, right inside your terminal.

🎉 Launch day offer: Sign up for our Starter package today and get your first month free.

Come tell us what you'd want TestSprite to break on. We're reading every comment today 🚀

Every piece of community feedback has shaped what we have built, and we are deeply grateful for that!

The goal has always stayed the same: learn and build from what users actually need. We hope this latest version shows how much we truly care. ❤️

Interesting! Auto-heal for UI drift sounds great in demos and gets dangerous in CI. How do you separate the two without a human in the loop? Congrats on the launch

Congrats on the launch!
one thing i would love to understand better is how the AI differentiates between actual product bugs versus harmless visual inconsistencies. Is there some confidence scoring or prioritization involved?

The autonomous debugging part sounds wild 🤯 How many retries does the system usually need before stabilizing a test?

Super cool launch 🚀

Love seeing testing evolve alongside AI coding tools. The autonomous workflow + CLI support for Claude Code/Codex users feels especially timely.

Most testing tools still feel like they need a whole setup manual before you can even start, but TestSprite making it conversational is a smart move. Love the frontend + backend coverage in one flow too.

what’s been the hardest type of bug for TestSprite to catch so far?

I am especially interested in the collaboration aspect around generated tests and debugging flows. Can multiple developers review and refine generated suites together inside the platform?

the auto-heal for UI drift is quietly the best feature here. tests breaking because someone moved a button is why most teams stop maintaining their test suite after month two

The upcoming CLI integration sounds perfect for modern AI-assisted coding workflows. Are you planning support for local-first testing environments directly from developers’ machines?

Congrats team!
One of the hardest parts of AI generated systems is transparency and debugging. How explainable are the generated test decisions when developers want to understand why a workflow failed?

Most automation tools still require a lot of brittle manual scripting, while this feels much closer to real user behavior. Can TestSprite also simulate different user roles and permission levels during exploration?

Congrats! The data flow debugging view sounds particularly useful for developers trying to trust AI generated tests. can engineers manually edit or optimize the generated flows directly inside the platform afterward?

Congrats! how transparent is the reasoning process behind generated bug reports?

Congratulations team!
The backend testing upgrades seem particularly strong for API heavy products. Does TestSprite support testing asynchronous systems like queues, webhooks, or websocket driven workflows as well?

Congrats! the CLI integration for claude code and codex users sounds exciting for developer workflows. Will developers eventually be able to generate or run targeted test suites directly from the terminal?

Generating 50+ meaningful cases automatically is a massive jump compared to traditional approaches. How do you internally define “meaningful” coverage versus repetitive or low-value test cases?

Congrats on the launch!
The parallel exploration fleet really caught my attention because most testing tools still operate sequentially. How do you coordinate multiple agents without them duplicating the same testing paths repeatedly?

E2E testing is not just about generating scripts — it is about understanding which workflows actually matter to users and the business.Curious how TestSprite separates meaningful coverage from noisy coverage. For complex enterprise apps, do the agents prioritize test cases based on the PRD/spec only, or also by app structure, failure severity, role-based flows, and real usage patterns? and can it simulate real scenarios like live meeting and chats ?

The backend integration testing upgrades sound powerful, especially Dynamic Variables and Auto Cleanup. Can TestSprite also manage dependent API chains with temporary tokens and session-based workflows?

Congrats on the release! Looks great!

Appreciate everyone checking out the launch today ❤️

This release took a lot of iteration, especially around autonomous frontend exploration and regression reliability. We wanted TestSprite 3.0 to go beyond simply generating test scripts, and actually understand how users move through an app before creating tests.

Excited to hear your feedback and keep improving it from here!

It feels like AI coding tools are moving faster than testing infrastructure right now. Do you see TestSprite becoming part of the default AI development stack?

I’ve been with the @TestSprite team for a while, and 3.0 feels like the version where the product got much more practical.

Now, instead of guessing at what your app does, TestSprite sends out a fleet of parallel AI agents to explore it first — almost like real users poking around — then generates tests from what they observed.

This is a key differentiator from most “AI testing” products that feel like asking a coding agent to squint at your app and hope for the best.

TestSprite is pushing toward something more concrete and ambitious: hundreds of parallel agents that explore, generate, and run tests at scale, and then Ralph Wiggum healing broken features.

Also: the CLI story is spicy! 🌶️ The team described an internal “Gundam mode” where Cloud Code + TestSprite CLI + AWS CLI can keep coding, testing, checking logs, and iterating for days.

That’s the bit I’m looking forward to trying!

Congrats on the launch! 🎉 The exploratory-testing-by-parallel-agents framing is the interesting move — QA traditionally treats exploration as the part you can't automate, only deterministic scripted runs. One question: when auto-heal fixes UI drift, how do you prevent a 'false pass' where the heal silently masks a real regression? Curious where the line sits between adaptation and over-tolerance.

As a solo dev building small SaaS tools, testing is always the first thing I skip when rushing to ship. The "90% cost reduction" claim is bold — curious how it handles edge cases in backend API testing specifically. Does it work well with lightweight stacks like Vercel + serverless functions?

Curious how TestSprite handles visual regression for design tokens specifically — if a colour or spacing token changes upstream in a design system, does the agent catch that drift across every affected component automatically, or is visual diffing still a separate manual step on top of the functional tests?

Really curious — when the agents explore an app, do they sometimes discover workflows or edge cases that developers themselves overlooked?

Parallel agents for testing is the right approach running tests sequentially is the bottleneck nobody talks about. Does it work for iOS/mobile apps or mostly web? That's where I'd love to use this.

This sounds awesome, particularly PRD to test case development! I gotta connect the MCP to Brief.

Hey Product Hunt 👋 Rahman here, Co-Founder of Cleo. I'm 19, building this from the US.

Rahim told you why we started. I'll tell you the line we refused to cross.

Right now there are a hundred tools claiming to be your "AI PM." Be honest - most of them are bullshit. Either it's just another chatbot with a new logo or it's a thin wrapper that "remembers" by dumping everything into a database and praying. No source, no confidence, no way to know if what it's acting on is real or hallucinated. That's not a teammate. That's a liability with a nice landing page.

I've shipped products solo since I was a teenager, and the thing that always broke wasn't the code - it was the coordination. Standups, follow-ups, "wait, what did we decide last week?" That overhead scales worse than any infra problem, and one day you realize you've stopped being the founder and become the PM of your own company.

Every assistant we tried was great at answering and terrible at remembering us. New chat, start over. Re-explain the team, the tone, what matters. By the time you've briefed it, you could've just done the work yourself.

So Cleo doesn't ask you to start over. It lives in your Telegram, Slack, and Linear, learns your team and your voice, and runs the PM work day after day - in your tone, not a robot's.

But here's the part I'd stake the whole product on: an AI that runs your team has no business hiding what it knows. So Cleo doesn't. Every fact it learns shows its source, its confidence, and a button to correct it. Five trust levels, from just watching to fully operating. You decide how much rope it gets - and you can always see exactly what it's working from.

No black-box memory. No "trust me." Just a teammate you can audit.

I'm here all day. Tell me where it still falls short — that's the comment I actually want. 🙏

Try the bot: https://t.me/try_cleo_ai_bot

The chat-first sidebar with structured memory cards below it is a pattern I haven't seen done well before. Most AI tools force you into either pure chat OR pure dashboard. Cleo manages both, and the seam between them is actually invisible

I've known Rahim since school. Watching him ship has been wild - they killed their own launch in May because telemetry wasn't right, rebuilt in in 5 days, and kept every paying customer. That kind of operating velocity is rare. Pulling for them

What’s the biggest bottleneck you faced while building this?

Having an AI PM directly in Slack and Telegram sounds like a massive time-saver for lean teams. Great concept!

Big congrats on the launch! The transparency model (source + confidence + fix button on every fact) is the right instinct, but I'd push further: there's a failure mode where an auditable AI quietly becomes a verification queue. If I'm confirming and correcting facts to keep Cleo accurate, I've just swapped standup overhead for memory-maintenance overhead. What's the design that keeps the audit asymmetric, where I only look when something's wrong, not to keep it right? Does Cleo surface only low-confidence facts for review, act silently on high-confidence ones, and shrink the queue as it learns my corrections?

Nice launch! Quick q - is Cleo built specifically for agencies, or would it work for a small SaaS team too? Most of my "PM" pain is internal standups and follow-ups, not client stuff.

Good luck!

Curious how Cleo handles context across long projects — does it maintain memory between sessions or does each conversation start fresh? The PM bottleneck is real for small teams building fast.

How do you pull in offline context easily for remote based teams? Ideally something low lift but I would imagine capturing every single meeting would be adding a lot of non-related content.

what is the difference between Cleo and handing over a few docs for guidelines on any ai agent? genuinely interested.

also, quick feedback on your comment, when you use AI to write it, remove the em dashes and make it personal! good luck!

Congrats on the launch! Curious, how does Cleo handle situations where the team has conflicting updates or unclear blockers? Does it just surface them or does it actually try to resolve the ambiguity?

What stands out about Cleo is that it doesn't just track tasks — it actually runs sprints autonomously. Curious how it handles prioritization when multiple stakeholders have conflicting requests. Does it have a built-in negotiation framework or just default to whoever shouts loudest?

Congrats on the launch! How is it different from OpenClaw with some PM skills? You mentioned OpenClaw doesn't remember what you said yesterday, but from my experience it remembers quite well.

Big Congrats on launch! Which AI models are you using? And how is company security handled?

the confidence scoring model is the right approach. most AI tools just act on everything with the same certainty and you only find out it was wrong after it already sent the message. curious how the 30 day correction window works in practice when the team structure changes fast

Very cool launch, congrats!

Sending support from the Exabase team 😌

Really proud to have worked on the design here. The hardest thing wasn't making it pretty, it was making "you can see everything Cleo knows" not feel like homework. Source, confidence, a fix button on every fact… that's a lot to surface without burying people. Took a few rounds to get the chat and the memory side to live together cleanly. Happy with where it landed

We're one of the early agencies using Cleo. The thing that surprised me: it actually learned how my team writes in 7 days. Now it drafts Telegram updates to clients in my voice and I just approve. Saving me 2-in 57 minutes a day on operational stuff.

The AI PM angle is interesting because product work is not just task creation. The hard part is explaining why a priority changed. Can Cleo show the evidence trail behind a roadmap decision, such as customer feedback, current sprint load, and launch risk?

Is the free Telegram tier enough for a 5-person team to test properly?

interesting that you started from the pain of re-explaining context every session. thats probably the biggest friction with most AI tools right now, they dont remember what matters to your team. curious how Cleo handles conflicting priorities when two people flag something as urgent, does it just surface both or does it actually help triage?

That's exactly what I needed to know — persistent memory across sessions is the killer feature for async teams. We're building Vokio (voice AI for local businesses) and the memory layer for call context is our biggest challenge too. Great work!

The transparent memory model is what caught me — showing the source and confidence level for every fact it learns is genuinely different from how most AI tools handle context. Curious: when Cleo gets corrected, does it update globally or per-context? As a solo founder managing multiple workstreams, the "operator" trust level sounds useful but I'd want to know exactly what it can act on autonomously before handing over decisions.

Hey Product Hunt, I'm Jason, Co-founder & CTO of General Compute!

The Problem

Agents are the most exciting thing happening in AI right now but the infra they run on was designed for chatbots, not autonomous workflows. When an agent has to make 20, 50, sometimes hundreds of sequential LLM calls to complete a task, latency compounds into a ceiling on what's actually possible.

Most inference providers today hit you with one of two tradeoffs:

1. ❌ GPU-based stacks – Great for training, but memory-bandwidth bottlenecks mean your agent runs slowly (~120 tokens/second)

2. ❌ "Fast" inference with catches – Some providers deliver speed but lock you into small models, limited context windows, or pricing that breaks at agent-scale token volume. Speed without intelligence isn’t worth the trade off.

After years building voice agents and real-time AI products ourselves, we got tired of waiting. So we built General Compute.

How General Compute is Different 🚀

GC is an ASIC-first inference cloud built on multiple chips, including SambaNova. SN uses a 3 tier memory architecture and dataflow, which is a fancy way of saying “It’s really fast cause we don’t have the same bottlenecks”.

• 🔹 Agent first (OpenClaw) – Agents can sign up on their own and manage their own API keys. OpenClaw can move its inference just by pointing it at our website.

• 🔹 Built for agent workloads – Tuned for both coding agents and voice AI (TTFT), the things that matter when you're chaining dozens of calls. Your agent finishes in seconds, not minutes.

• 🔹 Speed without the tradeoffs – Frontier open models, full context windows, and pricing that actually works at production scale.

Who is this for?

If you're building AI agents, voice AI ,or even just using OpenClaw or OpenCode and want faster inference, then GC is built for you. Faster inference isn't just a nice-to-have; it unlocks use cases that weren't viable before.

🔗 Get started today

Sign up at https://generalcompute.com and start running your workloads on ASICs today. We are offering $200 in free credit to anyone that signs up through the Product Hunt launch (up from the normal $5 in credit)

Congratulations to the launch.

this is a very real agent infra problem. Chatbot latency is annoying, but agent latency compounds into a hard ceiling when workflows need dozens of sequential LLM calls. how General Compute balances raw speed with reasoning quality on longer agent workflows, especially when there is large context, tool use, retries, and coding tasks. Is the biggest gain in TTFT/throughput, or do you also see better end-to-end task completion?

OpenClaw can sign itself up? That's wild. Finally someone building for a world where agents run themselves. 👏

Studied full stack development but never really got deep into the infrastructure side of things. Always assumed GPUs handled everything AI related. The idea that inference needs its own optimised hardware makes sense when you think about it. Congratulations on the launch.

Congrats on the launch! Efficiently stashing heterogenous ASICs behind a homogeneous API is a challenging and exciting endeavor :) Especially curious about the technologies powering elastic scaling with request volume and bursts. Would love to see a characterization of that in maybe a future blog post as it would certainly be useful to many service designers!

The ASIC angle is interesting, how does the model selection compare to GPU clouds? Are you running your own fine-tuned models or is it more about offering the same models (Llama, etc.) just with faster inference?

Looks insane Jason! Congrats on the launch

Curious, how accurate are the AI generated test cases currently?

Big congrats on the launch!

How about the time to set up? Is it able to run on CPU?

Love that this is an OpenAI-compatible API. Being able to just swap the base URL and get ASIC-level inference speeds without rewriting workflows is huge. Great work!

The ASIC-for-inference approach is clever. GPU memory bandwidth just isn't optimized for inference memory access patterns. At RetainSure we've been routing latency-sensitive AI calls for customer success workflows, and 200ms vs 800ms response time matters a lot at scale. How do your ASICs handle KV cache eviction for long-context requests?

Do you guys plan on adding embedding models sometime in the future

PS Agent sign up is brilliant! I'm going to study that.

Congrats on the launch. Your onboarding workflow is great. I missed clear models and pricing information upfront, and when I got onboarded I saw that you offer three models at somewhat premium pricing.

This leads me to my question: what is your value prop beyond latency? Because if you're competing on price, OpenRouter is still going to get you.

Model	Context	Input / 1M	Output / 1M
DeepSeek V3.2 Reasoning deepseek-v3.2	32k	$3.00	$4.50
DeepSeek V3.1 Reasoning deepseek-v3.1	128k	$3.00	$4.50
MiniMax M2.7 minimax-m2.7	160k	$0.40	$2.34

WordPress 7.0 is moving in the right direction. On behalf of the Tutor LMS & @Kirki team, I’d like to wish continued success to WordPress and its future. Long live open source.

Big release! I pushed it across my sites and everything held up quite well, no issues. The high contrast pass is a real win for accessibility, I love it. The AI is the most interesting part. IMO, it is less about what shipped today and more about what plugin developers, hosts, and agencies will build on top of it. Congrats to the core team and everyone involved in the release - developers, testers, leads, docs, polyglots... I will miss someone but not intentional!

Running WordPress for my app's website curious what the headline improvements are in 7.0 for site performance and SEO. The block editor has come a long way but page speed is still where I feel the most friction.

I have been using WordPress lately strictly for practice and upskilling my design abilities. I normally visualize my layouts in Figma first, but seeing the new dashboard and expanded design controls in WP 7.0 makes me want to jump right into the builder.

Having those AI tools integrated right into the core experience is going to be incredibly helpful for my learning process. I am really excited to test out the new blocks and see how far I can push the native customization. Congrats on the Armstrong release!

Great idea! as someone with ADHD I've been trying to build myself an interstitial journal, the body doubling via the notch is a great idea. Downloaded and excited to try this :)

distraction usually starts as a tiny “just one notification” moment and then quietly breaks the whole flow.

I like the body-doubling/witness angle because it feels more human than another blocker.how iPromise knows when to intervene versus stay quiet. Does it learn each user’s work patterns over time, or does the user define what “focused” means for each promise?

The trust killer for this kind of tool is the first false positive I guess? Buddy nudges me while I'm actually deep on something that just looks like drift (HN thread that's real research, Twitter that's competitor recon). How does the current heuristic handle 'looks like distraction, is the work'? By all means, good launch great work!

Body doubling is underrated for solo devs working alone means no ambient accountability. Curious how the AI version compares to actual co-working sessions. Does it react to what you're doing or mostly just stay present?

The notch as a surface for ambient companions is a genuinely good call — it's one of the few places on the Mac where a glanceable element doesn't feel like it's stealing real estate. Quick question: is the "buddy" reacting to anything (active window, time in app), or is it purely a timer/animation? I'm trying to figure out whether this works for me as a vibes nudge vs. an actual focus signal.

as a fellow mac indie — curious how you handle the notch UI on non-notch macs (external display, MBA M1)? does it gracefully fall back to menubar, or hide entirely?

I love all the animal avatars! The design is super neat. Is it possible to adjust the "tone" of your buddy? For example, I personally would find it more intriguing if it was very blunt and sassy to me when I get distracted, rather than giving a polite, long paragraph...

Hey Todd! As great as this is - it poses a serious privacy/security threat. E.g. I read "Your screen data never sends to us", but how do you independently ensure this to users?

lovely, cant wait to give this a go. how are you handing the screen context, is it local? the better question would be if it's a one time payment or a subscription?

hey todd! this is right up my street :) body-doubling shrunk to the notch, smart! the hard part is presence without surveillance. does it sit there as an awareness signal, or does it actively nudge when focus drops? and how do you draw the line. congrats on your launch, and good luck with it!

Congrats on the launch! Personally, my biggest distraction is basically a "quick check" loophole. I'd open a new tab to look up one piece of documentation and end up in a completely different rabbit hole.
Quick question about the ambient sound: Will we be able to import or bring our own custom soundtracks or background audio loops or do you have/are you planning an integration with external players like Spotify or Apple Music?

Congrats on launching! As someone who's sat in virtual study rooms this is intriguing, also an awesome name :) I guess one thing I got from people-populated study rooms was the feeling of reciprocity (the other people there also doing their own work) rather than surveillance. Curious what it will feel like to not have that, will take this for a spin.

Good job with the copywriting for the name of the product :)

Scheduling is solved by a lot of tools already, now what makes the analytics side actually useful here? Like, does it help you understand why something performed well, or is it mostly just tracking numbers you could see natively on each platform?

Managing social posting as a solo dev is always the first thing that slips. Does it generate the content too or just schedule what you give it? The content creation part is where most of my time goes.

Useful launch for teams that need recurring distribution across social and Telegram. The hard part for campaign workflows is keeping the same message adapted per channel instead of blasting identical copy everywhere. Do you support per-channel variations from one campaign brief?

Useful launch for teams that need recurring distribution across social and Telegram. The hard part for campaign workflows is keeping the same message adapted per channel instead of blasting identical copy everywhere. Do you support per-channel variations from one campaign brief, or is the workflow mainly scheduling separate posts?

This is a smart problem to solve right now. Especially with how fast Al tooling is moving.

Nice! We're covering a similar problem with Worvi — automating WhatsApp Business conversations for small businesses. The multi-channel angle is key, most solopreneurs are drowning in managing DMs across platforms manually. Does Auto Posts handle WhatsApp or just Twitter/LinkedIn/Telegram for now?

Would love to know how this compares against traditional QA workflows.

Love seeing a tool built specifically with smaller teams in mind! Having Telegram and Pinterest included alongside the usual platforms is a huge plus. Congrats on the launch!

great job guys!! I have one question - do we also get this comparative analysis on how an individual post doing on different social media platform? Maybe a post doing well on Insta but not well on X? does this kind of data available?

Looks cool, So will it post on all socials automatically by creating different set of creatives and posting them all over Insta,Fb pages,Linkedin Pages and X
also does it require any confirmation to post on my behalf regularly.
one confusion as your product says "Plan your content and schedule posts across all platforms" if so what difference it makes since mostly all platforms provide with content planning and future schedule options

I love the idea of easily turning existing content into a polished book. Perfect for creators who have loads of content but haven't had the time to publish a book yet!

Is it tailored to a specific genre like, for example, an autobiography? Or can I write sci-fi with it if I want?

Prosed is such a unique name. What was your inspiration for naming your product this?

hey elise! great product but also want to add i love your design style. this is def the underrated long game for makers! how do you tighten the voice without flattening the personality? good luck with your launch :)

The assembly-block framing is very real. The risk I’d watch is that “sounds like you” can mean two different things: surface tone, or preserving the actual judgment behind the old essays/posts/podcast moments.

For a book workflow, I’d love a source-to-chapter map: which passages came from the creator’s original material, which connective tissue Prosed added, and where the author marked “not me” or “not my point.” That would make the manuscript feel less like an AI rewrite and more like an editable spine for the body of work they already built.

What’s the most unexpected use case you discovered?

@Prosed reminds me this thread @naval posted on Twitter/X in 2018 on how to get rich:

If you can't code, write books and blogs, record videos and podcasts.

No more excuses. Thank you, @rails!

Text-only lists? Into it. I already stash movie recs + food spots in Notes, so this clicks. The item->other lists hop made sense after a minute. Feels more discovery than social. Would love more hyper-local stuff (cheap eats, late-night spots) and “comfort rewatches.”

The idea is pretty cool. Having media in social networks sometimes can be too much. I remember when other social networks were about chat. The idea kind of reminds me of forums and BBS's.

The "no media" constraint is such a bold bet. Do you find that lists actually create more meaningful connections than photos do, or is it more that they attract a different type of person altogether?

the brainrot framing is doing real positioning work here because it names the specific feeling the product is designed against. but i'd want to understand the mechanism. most social networks that promise to be different from doomscrolling still end up optimizing for engagement once they have users and engagement metrics tend to reward the same things everywhere. what's the specific design decision in moop that makes it structurally different rather than just aesthetically different

This feels so comforting and less chaotic to go through new social updates. Love the idea of simply listing things in a very minimalistic way which is very uncommon atp.

I love the idea. This could at least help with better focus :)

We would love to get your feedback on one particular feature: external links.

It's been debated in our team whether or not to allow those, so that you can link an element to Google Maps, an Amazon page, a song, etc...

In its current form, do you feel like moop is lacking that? Or do you prefer the clean, internal-only scrolling experience?

I like the design.

Does the design have a specific name?

How long did it take to get the first working version live?

Interview graveyard vibes here. Notes in Notion, gut in the PRD. If it really pulls themes and spits PRDs with quotes, that's useful. The MCP angle is neat grounding chatgpt on real calls. Keen to try on my next batch. Solo PM here, time savers matter.

Indie iOS founder here, doing a lot of customer discovery ahead of an App Store launch, so the "interviews die in transcripts no one reads" line hits home. The MCP server is the smart unlock; an agent citing real users instead of hallucinating them is the whole game.

Answering your q: the gap I'd flag is upstream of synthesis: interview quality varies wildly, and a confident-but-shallow interview produces clean-looking Nuggets that are actually just the customer agreeing with my leading questions. Does Nugget do anything to flag low-signal interviews (interviewer talked 70% of the time, all closed questions, sentiment too uniformly positive), or does it treat every transcript as equally trustworthy evidence?

I’ve personally been in situations where we did multiple customer calls, captured tons of feedback, and still ended up writing PRDs mostly from memory because nobody had time to go back through long transcripts. A lot of valuable insights just get lost in docs and scattered notes.

Really like how Nugget focuses on connecting customer conversations directly to product decisions instead of letting that context disappear. The MCP integration is especially interesting because AI tools become much more useful when they can reference actual user feedback instead of assumptions.

Congrats on the launch, Brodie!

Turning user feedback into roadmap decisions is one of the hardest parts of solo dev you get scattered signals and have to find the pattern yourself. Does it work with App Store reviews or mostly structured interview data?

In EdTech the buyer and the user are different people - e.g. the teacher decides, the student uses it. Do interviews from both groups get synthesized together, or can you keep them separate and see where they diverge? Congrats on the launch!

hi brodie, the right pain to solve! how do you weight one confident interview against five lukewarm ones? and what stops the AI from inventing patterns from a thin sample? very cool - congrats and good luck!

The MCP piece is a smart wedge because the evidence can travel into the actual spec-writing environment instead of dying in a research repo.

One thing I’d want in every generated PRD is an evidence receipt: strongest quotes, counterexamples, last-heard date, interview-quality score, and which asks were intentionally left out. That last bit matters because product teams don’t just need “customers said X”; they need to see the judgment call behind why X became roadmap material and Y didn’t.

How are early users reacting to the product so far?

Love the MCP server → Cursor setup.

Does it auto-pull context from all past interviews, or do you select specific ones for each session?

interesting... isn't this similar to ReadAI/Other notetakers?

Hey Folks! 👋

I'm Nikita, co-founder of SuprSend. We've been building notification infrastructure since 2022 - used by 500+ companies to send customer engagement notifications across email, SMS, push, WhatsApp, Slack, MS Teams, and in-app.

Today we're launching the AI-first notification platform.

The problem: notification infrastructure has too many moving parts. Workflows, templates, preferences, tenants, channels, routing logic. New customers spent weeks just learning how things connect before writing code. We tried better docs, better UI — it helped but didn't solve it.

So we shipped 6 AI-native interfaces:

→ MCP Server - 24 tools inside Cursor, Claude, Windsurf. Build workflows, create users, manage preferences, debug delivery - all from your IDE. One prompt does what used to take 20 lines of API code.
→ CLI - notification config as code. Pull, version in Git, promote via CI/CD. Your notification stack ships like the rest of your code.
→ Agent Skills - domain knowledge loaded into AI context so it doesn't hallucinate your API.
→ SuprSend Agent - AI copilot in the dashboard for PMs and ops.
→ Slack Agent - trigger, debug, manage from Slack.
→ Claude Extension - use SuprSend directly inside Claude.

What makes this different from just "adding AI" to a product:

→ Every AI action follows notification best practices automatically: channel selection, frequency capping, preference cascades, spam avoidance. You don't need to be a notification expert. The AI already is.
→ Works across your entire team: engineers in Cursor, PMs in the dashboard, ops in Slack. One platform, every interface.
→ Production-safe: destructive operations are restricted, updates do read-before-write, production environments can be locked to read-only.

Results we're seeing:
→ Integration time: weeks → hours
→ Template creation: ~8 hours → 5 minutes
→ Test setup: 15 min manual work → one prompt

Free tier available - try it: suprsend.com/ai

Would love your feedback — what's missing? What notification problems are you still solving manually?

Happy to answer anything here! 🙌

I use the SuprSend Agent daily now. The biggest time saver is delivery tracing - when someone says "I didn't get the notification," I used to dig through logs, check preference settings, channel availability, workflow execution. Now I just ask the copilot "why didn't user X receive the confirmation" and it traces the entire path in 10 seconds. Used to take 15 minutes.

As a marketer, the Slack Agent is surprisingly useful. Quick lookups that used to mean opening the dashboard. "Show me delivery stats for my last campaign" or "what channel is working for campaigns sent to the free tier users" — now happen right in the Slack channel, visible to the whole team. For the 20 quick checks you do daily, it's way faster than switching apps.

I have tried the new agent and Slack App Integration. Honestly, big time saver for me.

Tried building a workflow through the agent and it was surprisingly quick. What I liked is it actually thinks about edge cases I'd probably miss on my own.

The analytics part is what sold me though. Earlier, if something failed, I'd be digging through logs trying to figure out what went wrong. Now I just ask and it tells me — right inside Slack, which is even better.

So now at EOD I literally just ping it — how many notifications went out today, how's engagement looking, did anything fail, and if yes why and how do I fix it. Done in like 30 seconds.

Template creation was our biggest bottleneck. Designing email HTML, creating channel-specific variants for push and in-app, testing with dynamic variables, iterating on copy, testing, observing the stats - a single notification template used to take 6-8 hours end to end. With all these AI tools, I describe what I want and it generates the template with the right structure. Down to under <30 minutes now!

The Agent Skills approach is smart. Pre-loading your notification schema into AI context means the model isn't guessing field names at inference time. At RetainSure we've been building customer-facing alert pipelines, and channel fallback logic when primary delivery fails is the hardest part. Does the MCP server surface delivery failures synchronously to the calling agent, or does it require polling?

Six AI-native interfaces for the same underlying infrastructure is the right call. The Agent Skills piece stands out: loading domain knowledge into context so AI doesn't hallucinate your API schema is what most platform teams miss. How are you handling schema drift when notification configs evolve between agent sessions?

The hardest part of multi-channel notifications isn't the send logic, it's the orchestration layer that decides which channel wins when user preferences conflict with urgency. The AI routing angle is interesting here. How do you handle deduplication across channels when a user might receive the same notification via email and push within seconds of each other?

The Slack Agent is surprisingly useful. Quick lookups that used to mean opening the dashboard - "show me delivery stats for this tenant" or "what preferences does user X have" - now happen right in the Slack channel, visible to the whole team. For the 20 quick checks you do daily, it's way faster than switching apps.

Love the versatility Suprsend gives me. Must have for any early or growing startup

How does SuprSend AI handle notification templates that use custom web fonts? Wondering whether the AI copilot is aware of font fallback stacks when composing email templates — and whether there's any guidance on when to embed vs. reference a custom typeface given email client rendering inconsistencies.

Multi-channel notifications get hard when teams need different tone per channel. Does SuprSend AI adapt the same message differently for email, Slack, and in-app alerts?

Love this. I've been searching for a competent ebook reader. Is it possible to ask it about a word you don't know? And are you planning a Google Play launch eventually?

Hi Phil! the missing ebook feature! when you ask a question, does it answer from this book only or pull from wider AI knowledge? those go very different places fast. how do you handle translation that flattens the author's voice? this seems great, though as an Android user I'll keep an eye out for when it lands in the Play store! congratulations on shipping and good luck :)

what types of books does it help with? for example can i use books i have downloaded from amazon? do they need to be in specific format?

What kind of teams are seeing the most value from this?

Hey, congrats on the launch! This might be a dumb question, but does a device need powerful specs to run your app locally?

Multi-step AI workflows are powerful when they include clear checkpoints. Can a buildpipe workflow pause for approval before a step changes code, posts externally, or calls a paid API?

hey buddy!
The idea looks very interesting for productivity.

Just my concern for recurring workflows like Hacker News summaries or daily briefings, do you maintain a searchable history of previous runs?
For example, if I saw a story in the morning and later wanted to revisit the ai summary or outputs, having run history would be super useful. It is what we usually do.

But at the same time, storing every pipeline output/log long-term could also become a storage challenge locally specially for heavy AI workflows.

This looks super clean. Love the direction you’re taking with AI workflows 🚀

the AI calls step in the pipeline is the one i'd want to understand better. chaining shell commands and HTTP requests is well-understood territory. adding AI calls introduces non-determinism into the pipeline which changes how you think about error handling and retry logic. if an AI step returns something unexpected does the pipeline fail, retry, or branch? and how do you define what unexpected means for a step that's inherently probabilisti

Nice work keeping this local-first with encrypted keys, that's the trade-off most pipeline tools skip. Good luck with the launch!

The file-watch trigger is interesting — curious whether buildpipe can handle asset pipelines where the output is binary files rather than code. For instance, running a font build step that produces WOFF2 files and then pushing them to a CDN as a chained step. Is the pipeline model flexible enough for arbitrary file artifact workflows, or is it currently more focused on shell/AI/HTTP steps?

This is honestly exactly what I have been looking for. I spend hours in After Effects and Premiere Pro tweaking motion graphics and trying to get the pacing right. Hearing that Miltos 5.0 understands timing and knows where to cut is a massive deal for my workflow.

I already use AI tools for generating assets and conceptualizing 3D stuff so having an editor that handles the heavy lifting for animation right in the browser is huge. I do a lot of projects with upbeat funk and pop aesthetics so I am super excited to see how the motion model handles those vibes. Definitely giving this a try today.

How is the hand off to NLEs like Premiere - any ability to tweak and find tune after export? Is overlaying on other video possible?

AI video editors are getting crowded, so the workflow details matter. For marketing teams, can motionvid keep one campaign style consistent across multiple aspect ratios and short variants, or is each output edited as a separate video?

How much control do you have over the style and tone? "Describe in plain words" can mean very different things depending on whether you need something polished for a school presentation vs something punchy for social. Congrats on the launch!

the silent agent deployment is the part i'd want to understand from a legal and consent standpoint. depending on jurisdiction, installing monitoring software on a candidate's machine during an interview has compliance implications that vary pretty significantly. does Zero Assist require explicit candidate consent before deployment and how are you thinking about the liability if a candidate later claims unauthorized monitoring? that's not a hypothetical question in 2026, it's already happening in employment law

We are currently witnessing a massive arms race in remote tech hiring. A year ago, cheating meant a candidate looking down at their phone or having a hidden browser tab open.

Today, it is a completely different landscape. There is an entire industry of "Stealth AI" tools (like Parakeet AI, Final Round AI, and Cluely) specifically engineered to defeat traditional video interviews and screen sharing. These tools intercept system audio, transcribe the interviewer's questions in real-time, and project the exact code or answers onto the candidate's screen using invisible GPU overlays that standard screen-sharing software physically cannot see.

The cost of this is staggering. Companies are making $150k+ hiring decisions based on AI-generated performance, leading to massive sunk costs in onboarding and immediate productivity bottlenecks. Even worse, brilliant, honest candidates are losing out on jobs to people using software to fake their skills.

We can't rely on webcam "eye-tracking" or "gut feelings" anymore. I built Zero Assist because the only way to detect stealth AI is with forensic, OS-level monitoring. We have to restore integrity to the hiring process so that the best engineers win on merit, not on who has the most expensive hidden copilot

Hey Product Hunt! 👋 I'm Vaibhav, the maker of Zero Assist.

Remote technical hiring is fundamentally broken right now. Candidates are increasingly using stealth AI tools like Parakeet AI, Final Round AI, and ChatGPT to pass interviews. Traditional screen sharing can't detect audio-based whispers, hidden browser tabs, or GPU overlays.

I built Zero Assist to fix this and restore integrity to remote hiring. It’s a lightweight, forensic-grade agent that runs on the candidate's machine during the interview and monitors OS-level processes.

✨ Key Features:

Real-Time Detection: Catches 20+ known AI cheating tools the moment they launch.
Silent & Seamless: Runs in the background with no intrusive pop-ups for the candidate.
Live Interviewer Dashboard: Instant alerts delivered via WebSockets (<500ms latency).
Deep Forensic Evidence: Post-session audit reports with exact timestamps and process metadata.
Privacy-First: Legal compliant. Exits cleanly after the session, leaving no persistent footprint.
Zero Assist empowers engineering and HR teams to make hiring decisions based on genuine ability, not AI assistance.

I'd love to hear your feedback, answer any questions about how the detection works, or take feature requests below! 👇

just curious about privacy invasion with your product.

Love seeing products focused on improving interview integrity.

Ohh yeah, the must needed one! Facing this problem in almost 60% of interviews where we can't judge if the candidate is cheating or not. And the result, we waste our days, hours on interview rounds of irrelevant candidates who're actually caught cheating in interviews.
Hopefully it helps now. Congrats on the launch team. Good one.

Great mission and concept! As mentioned the false positives is the big concern. Have you tried testing it with "generic" sounding interviews? ones that might be flagged as AI but aren't?

Interesting problem to solve, but I'm curious about the flip side, as AI tools get better at mimicking "natural" behavior, how do you stay ahead without the detection itself becoming a false-positive machine that flags good candidates unfairly?

Hey PH! I’m Iftakhar, building DCP.

AI agents are moving from answering questions to executing real work: signing transactions, using API keys, making payments, and calling tools across apps.

But there’s a problem: agents should not hold private keys, raw credentials, or sensitive information.

DCP is my attempt to solve that. It is a local permission vault for AI agents.

The flow is simple:

agent asks  

you approve on Telegram or in the app  

DCP signs locally  

secret never enters the model context

What works today:

- desktop app

- local encrypted vault

- Telegram approvals

- Solana wallet signing

- API credential storage

- budgets and approval limits

- MCP-compatible agent flow

- open source repo

I built this because I think the next bottleneck for agents is not intelligence. It is permission.

If agents are going to act for us, they need a safe way to use wallets, credentials, and sensitive tools without taking custody.

Would love feedback from people building agents, wallets, MCP tools, x402 apps, or anything around agent commerce.

Website: https://dcpagent.com  

Docs: https://dcpagent.com/docs  

Download: https://dcpagent.com/#download  

GitHub: https://github.com/1lystore/dcp

Love seeing builders focus on secure AI systems early.

Congrats on the launch! One question, if we hit the daily cap does it pause and wait for next day or does it notify you?

Scoped permissions for agents feel important. Can teams issue temporary task-level keys that expire or revoke automatically after an agent run, instead of giving the agent broad credentials?

good luck, guys. let's support each other!

FYI, I REALLY want to try DCP, but downloading the installer (made sure it's the right version for my Apple silicon Macbook Air 15") it does install, but then gives me a "DCP is damaged; you should move it to trash" which I cannot get around - tried re-downloading and reinstalling, tried a few different things (running as admin, etc.) - nothing works.

This is exactly and precisely the solution I need, so, I hope I can figure it out... congrats on the launch, regardless!

The Telegram approval flow is a clever UX choice for mobile-first teams. Curious how DCP handles permission scoping for design tool APIs specifically — if an agent is accessing a design system's token API, can you define read-only vs. write permissions per tool, or is it a single permission gate per service? Granular scoping seems key for preventing over-permissioned agents.

That's definitely sounds like one of the most useful products I've seen launched in some time. I've wanted to build something with OpenClaw for some time now, but always stopped myself because of the fear of possible leaks. Would you say it is a good use case for your vault?

the harder question underneath this product is whether detecting AI use in a technical interview is actually the right goal. a senior engineer who knows how to use AI tools effectively might be more valuable than one who can whiteboard without them. curious if there's a way to configure what counts as unauthorized versus what's just how people actually work now

As someone into music tech, this feels super interesting.

Hey PH 👋

Replying to @math_biomes , but this might be useful for everyone.

Shuffle's Design CLI is the command-line version of AI Website Redesign (and design), so you can generate results from multiple leading models at once - including Claude, Gemini, and GPT.

The CLI returns:

• URL to online project preview,

• a screenshot of the result,

• and a fully editable project inside Shuffle (you can download it using CLI too)

It's especially useful for automating workflows where multiple design directions, a live preview, and screenshot generation are important.

We also included two ready-to-use demo examples:

• Design Slack Bot

• Design Outreach Automation

CLI-first approach to design is interesting — curious how Shuffle handles custom icon sets in the generated output. Does the CLI support feeding it your own component library's icon system (icon fonts or SVG sprites), or does it default to a generic icon set? Wondering how well it plays with existing design tokens when redesigning rather than starting fresh.

Multi-AI redesign from the CLI is interesting for shipping UI changes quickly. How do you prevent different models from drifting away from an existing design system when they generate sections or components? Is there a way to lock tokens, spacing, and component conventions?

when you say "ships websites better faster", in what sense do you mean? what is the difference between you and let's say gemini, claude and cursor?

I'm getting a 'TTS failed [401]' error when trying to add language content to the stories ('quota_exceeded').

Regardless, I just wanted to say that this website is such a fantastic idea! Really love the creativity here.

this is a great idea especially if you are tired but still wants to entertain the kids:)

Helping bilingual kids through storytelling is such a meaningful idea.

This is so cute ♥️ it will be so easy to be part of my niece's every days, especially when time difference makes it impossible to make video calls. I have recorded some voiceovers of fairy tales and songs, but it's not engaging for her to only listen

My favourite feature is "keep your hero" and being able to narrate every night new adventure of the Pirate Barbavento to my young Nicholas

I love this. I have four kids, 11 through 19 now, and bedtime stories are a thing we'll always remember! ❤️

The mushroom characters are super adorable! Were they AI-generated, or designed by an illustrator?

Hey Product Hunt! 👋

Every news product we tried was solving what to show us, not how it felt to read.

AI has given us smarter curation and better summaries, but the reading experience remains fundamentally... boring to us.

So we asked: what if reading the news actually felt fun?

Shroomie makes news bite-sized and interactive. We designed an immersive flow where you can effortlessly skim through your feed, react, and deep dive on topics of your interest.

Every day you read, you grow a mushroom garden. We believe the best habits aren't built through discipline — they're built through play. After every session, you walk away with something real: a garden that grew, and a mind that did too.

AI powers everything behind the scenes — sourcing stories, generating visuals and summaries, and deep-dive research. A personalization engine combines ML and LLMs to curate a feed tailored to each reader.

Check us out on App Store! This is our v1, and we're shaping what comes next with early adopters. Try it and let us know what you'd change!

this is good for micro learning, I would like to know how do you profile user or are you sending same news to everyone. and what is the frequency and how do you choose the frequency ?

Congrats!
Android when? ;)

sounds like fun actually. any way to add your own topic and sources?

Congrats on the launch! I'm in love with your mascot and couldn't resist leaving a comment 😅🍄
Where did the metaphor of growing a mushroom garden come from?

Congrats! How deep does the personalization go? Can I control what topics I see, or does it learn on its own over time?

Just downloaded it. How granular does the personalization get? Can I tell it I want more tech and zero politics, or is it more of a black box?

Why only iphone?

Congrats on the launch, I love whimsy software! 🍄 Too many news sites spam you with ads rather than making the content experience actually delightful.

Love the asking follow ups, what other AI features do you have planned?

A virtual coworker with its own computer living in Slack is fascinating — most AI agents just reply in chat, not actually execute. Curious how you handle security boundaries. Can Faby install packages or access internal tools, or is it locked to a sandboxed environment?