PH热榜 | 2026-07-16

It's 11pm the night before your final, and your tutor texts you: it noticed your professor allows a cheat sheet, and it can build one around exactly the topics you're weakest on. Want it?

Hey Product Hunt 👋 I'm Scott, founder of Paradigm. That text is Clover, and moments like it are the whole product.

Another one: you say you want to ship your first app. Paradigm spins up a real cloud machine — terminal, Claude Code, the works — and walks you through it keystroke by keystroke until something is live.

Not courseware with a chatbot stapled on — one tutor that builds whatever the moment calls for (a spoken conversation, a coding sandbox, a full desktop) and remembers everything you've ever done with it. On the canvas, in your texts, the night before the exam.

Every interaction is graded as it happens, so what you walk away with is proof of skill, not a transcript. Anyone can build a course here; ours currently run from "Speed-run ICS 31" (UCI's intro programming, crammed) to "Let's count to 10!" (pre-K).

I started building this as a college freshman in 2023 — some of you might know me from OpenClaw or ShellBench. Almost everything that actually mattered in my education happened outside a syllabus. So I'm building the university I wish existed.

(One team fact I can't not share: the engineer behind our agent pipeline is also one of China's most-streamed underground rappers — 100M+ streams. The album and the agent pipeline shipped the same year.)

I'll be here all day, reading every comment.

👉 My ask: bring the weirdest thing you've ever wanted to learn — your grandmother's recipes, a 40-page anatomy atlas, a niche cert — and tell me where the path it builds breaks.

How does the AI tutor decide what to teach next as I progress?

Juggling separate apps for notes, practice, and tutoring is exhausting. Does Paradigm bring these together?

My biggest problem with self-study is not knowing what to learn next. Does Paradigm solve that?

Love that the path evolves with your progress instead of being fixed on day one.

How does pricing work once the 1,000 free tokens run out?

Can I chat with the tutor over text/iMessage, or only inside the app?

Congrats on the launch! 🚀 That bit about everything that mattered happening outside a syllabus really got me. Honestly everything I use day to day I picked up by building stuff and breaking it, never from a course. Proof of skill over a transcript is way more useful anyway.

Congrats on the launch, #1 spot is well earned. I spent a good 15 minutes poking around the site before commenting.

The interactive widget on the landing page is a really smart touch. I typed "learn technical SEO for Next.js sites" and it started sketching out a path, which genuinely got me curious.

One honest piece of feedback from that session: every CTA on the site (Open in Paradigm, Start learning, even Name your price) lands on the exact same generic signup screen, and it never mentions the goal I just typed. All the momentum the widget builds dies right at the highest friction step. A single line like "sign up and we'll finish your technical SEO path" would carry that motivation through. Also the typed goal only lives in sessionStorage, so anyone who opens their email verification link in a new tab will feel like their goal got lost.

Tiny one: the 404 page's Return to Catalog button points to /home/catalog, which logged out visitors can't open.

The name your price tuition idea is fun, curious how tough Clover is as a negotiator. Upvoted!

An AI tutor that meets you across text, app, and chat feels way more natural than opening yet another dashboard.

The hardest part of learning to code alone is getting stuck with no one to ask. Does the tutor help there?

This genuinely looks like an exceptional product. I chose a reasonably technical (and probably not hugely mainstream) subject and it created a well structured course for me, with a blend of knowledge and practical walk throughs. As ever with Product Hunt, I am working through today's launches... but this is firmly getting an upvote and I will be coming back. I normally have a comment or query, but everything makes sense so far (I even found how to change the voice, which was my only irritation!)

This idea is simply brilliant. Having goals is not the hard part. It's creating a path and then adapting the path when life happens.

Taking your "tell me where the path breaks" ask seriously: I hand-built exactly this for myself this month — a 12-week path where every week has to produce a real artifact for my actual job, with an LLM mentor that quizzes me against those outcomes — because topic-sequenced courses never stuck. Graded interactions cover what happens inside Paradigm; my paths always broke at the boundary, where the proof lives outside the platform (a dashboard shipped at work, not a quiz). Can the tutor take messy external artifacts as evidence and adapt the path around them, or does grading stop at its own sandbox?

Interesting concept. As someone teaching myself new skills constantly while building a SaaS, the "getting lost" part is real.

This product is very cool, but during the course learning process, I really wish there was a "teacher" who could help explain confusing terms or other doubts at any time. I hope this feature can be supported in the future!

This product is very cool, but during the course learning process, I really wish there was a "teacher" who could help explain confusing terms or other doubts at any time. I hope this feature can be supported in the future! I encountered a BUG: after creating a course, the course disappeared when I returned to the homepage.

Amazing idea - exactly what is needed to help me cross the "valley of despair" when I am trying to understand that new thing.

Your product sounds really nice! Congratulation on launch! I`m learning serbian language at the moment, so testing Paradigm on creating course for learning serbian verbs. First impression is great, would check in more detailed way and share my experience.

This feels like having a patient 1:1 tutor that actually adjusts to your pace.

The coding sandbox built directly into the lessons is a really smart touch.

I like that it covers academic exams and professional skills in the same place.

Congrats on the launch! Are you guys diagnosing weakness passively from what I'm working on or users basically sit an assessment first?

most learning tools are just a fixed playlist with a progress bar. building the path around your actual goal is the part that makes this worth trying

Can I import my own study materials or textbooks into a learning path?

Would this work for brushing up on math before starting a data science course?

Does the coding sandbox support multiple languages like Python and JavaScript?

The idea of going from curiosity to mastery without dead ends really resonates. Most courses lose you halfway.

This looks great for lifelong learners who just want to explore a new subject.

👋 Hi Product Hunt!

We're MoonMath, the team behind Zro.

Over the last year, we've been obsessed with one problem: making AI inference faster and more efficient. As we worked with developers building AI products, we kept seeing the same tradeoff:

Use closed-source AI APIs and give up privacy and control.
or
Run open-weight models yourself and spend time managing infrastructure.

We built Zro so developers don't have to choose.

With Zro, you get a fast, OpenAI-compatible API for open-weight models with:

🌏 Multi-region hosted inference
🔒 Zero data retention & zero training on your prompts
🏢 Optional on-prem deployments
⚡ A serving stack optimized for coding agents and long-context workloads using MoonMath's in-house inference technology.
🛠️ Built for easy setup with the most popular AI coding agents

We're launching early because we want feedback from developers building real AI products. We'd love to know:

Which models should we add next?
What features are missing from your current inference provider?
What would make you switch?

🎁 Product Hunt launch offer

Use code PRODUCTHUNT to get 1 month of Zro Pro free.

Limited to the first 100 users.

We'll be here throughout the day to answer every question. Thanks for checking out Zro, we're excited to hear what you think! 🚀

I've been using Zro for about a month, it's great!
Very fast and reliable

Private inference is the box teams check last, usually right after their proprietary code turns up somewhere it should not. The part I would pressure-test is latency ten tool calls deep in a real agent loop, not on a single prompt. That is where most self-hosted setups fall over.

I am using GLM 5.2 through ZRO in my daily workflow for a while now. The speed and quality have been impressive so far.

Good luck with the launch!

honestly the no-retention thing is a really thoughtful call, most teams talk about speed but skip the privacy piece entirely. nice to see it baked in from day one

"Your code shouldn't be someone else's training data" is a strong line. Which open models are you running under the hood, and is EU hosting the default or a paid tier? Congrats on the launch

The API speed of response is really fast. Been used for a while and very satisfied by the quality of the API.

Fastest API I have ever used so far.

Additionaly, you can try PRODUCTHUNTMAX for 50% off the Max Plan

private + coding agents, finally. how much latency do you give up to keep it private?

the zero-retention, no-training-on-your-code angle is the whole pitch for me. most devs just assume their prompts end up as training data somewhere and live with it. if you can actually prove retention is zero, that's a real reason to switch, not just a nice-to-have

Really like the commitment to zero request retention, that's a rare stance these days and shows the team actually thought through the privacy side instead of just slapping it on a landing page.

The tradeoff between closed APIs and self-hosting open models is very real. most small teams want more privacy and control, but not the operational burden of running inference infrastructure themselves.

The OpenAI-compatible API, zero retention, and optional on-prem path make Zro feel practical for teams that want to start simple and keep more control as they grow. Curious how you handle routing across regions for long-context coding workloads without hurting cache efficiency or latency consistency?

Latency felt noticeably snappier than what I usually get from other inference providers, and I love that nothing gets stored. Refreshing to see a team actually prioritize both speed and privacy at the same time.

"zero retention claims are cheap, anyone can audit this?"

which agents work with it?"

Private inference is one of the most underrated problems in the agent space right now. I build in a compliance-heavy vertical, so "your code never leaves your control" is a real buying trigger, not a nice-to-have.

How are you handling the latency tradeoff vs. hosted frontier models?

"on prem this early feels like a lot for a small team to support properly, are yall going setup manually right now or is there tooling for it".

"multi region's great for latency but some of us need single region only for compliance stuff, is that even an option here".

"openai keeps cutting prices every few months, not sure privacy alone convinces people to switch providers".

@Emir Soytürk makes sense, TTFT and TPS are the two numbers that actually matter for an agent loop, not just raw throughput. curious if you've published any benchmarks against the big hosted providers or if that's still coming

Congrats on the launch! Zero retention really matters once you're touching more than one company's data, I run a few different businesses through my own tool and the idea of any of that context leaking across clients or getting used for training is a non-starter for me. Curious how you think about audit logs on your end, if a customer asks "prove nothing was retained," what do you actually show them?

👋 Hey Product Hunt!

Quick correction: our Product Hunt launch offer is no longer 100% off.

We saw abuse from zero-dollar checkout flows, so we changed the Pro launch offer to first month for $1. It is still a major launch discount, but it requires a real payment and helps us keep the service healthy for legitimate users.

The Max Plan promo code is unchanged: 50% off the first month.

Thanks for understanding. We’d rather be transparent and keep things stable for everyone.

zero retention is the right pitch for coding agents specifically, a lot of codebases can't go through a provider that might log or train on the prompts. is the latency comparable to the big hosted providers, or is there a tradeoff for the privacy guarantee

Do you also support Claude? Would be interested if you supported their models.

Streaming support for token-by-token responses would be huge for chat use cases, even if it means slightly more work on the caching layer. Right now waiting for full completions before anything renders feels dated for an inference API in 2026.

Congrats on the launch! Keeping it OpenAI-compatible is a massive smart move. A lot of teams want to switch to open-weight models for privacy but dread rewriting their entire agent orchestration layer or dealing with completely different API schemas. How has the drop-in replacement experience been for early testers? Are there any specific edge cases with function calling or streaming where the compatibility layer behaves differently than native OpenAI?

Love that there's no request retention here, that's a real differentiator for anyone dealing with sensitive data. One thing that would help me evaluate it faster though: a small public latency dashboard comparing Zro to other inference providers on common open models. Even just a simple weekly update with p50 and p95 numbers across regions would make it way easier to decide if it's worth migrating workloads over.

One of the builders here 👋

GLM 5.2 is honestly a beast. I have been using it daily since we first launched ZRO internally, including to improve ZRO itself.

The speed and quality are really impressive. Give it a try!

🚀 Albato just got a major upgrade!

Hey everyone 👋 Wenddy here from the Albato team — and we’re beyond excited to introduce a new chapter in no-code automation.

We’ve rebuilt the Albato experience to make automation smarter, faster, and more intuitive than ever.

Here’s what’s new 👇

🤖 Albato Copilot — Describe what you want to automate in your own words (in any language!), and Copilot instantly builds your workflows for you. It’s like having a personal automation expert at your side, available 24/7.

🧠 AI Agents — Go beyond workflow creation. Simply describe a task, and AI Agents can understand your intent, choose the right actions, and execute multi-step automations across your connected apps with built-in guardrails.

🧩 Canvas Mode — See your workflows come to life in a flowchart-style editor that gives you complete clarity and control. Explore complex automations with ease in one clean visual space. (And yes, drag-and-drop functionality is coming soon 👀)

⚙️ Test Step — You can now test each step individually and instantly view real data samples in mapped fields. Validate and debug step by step, configure using real data, and save time while fine-tuning your workflows with precision.

🔗 Connection by Link — Connect client apps securely without sharing credentials. Generate a link, send it to your client, and the connection appears instantly in your account — perfect for freelancers and agencies managing automations safely and efficiently.

📤 Automation Sharing by Link — Share complete automations with a simple link. Collaborate faster, distribute proven workflows, and help others get started in seconds without rebuilding automations from scratch.

🌍 1K+ Apps Integrated — We’ve reached a major milestone with over a thousand connected apps and refreshed triggers and actions across our library.

And we’re not stopping here...

🔜 Coming soon:

Teams functionality — create multiple workspaces, manage subaccounts, and assign roles effortlessly.

This release is all about giving you more control, flexibility, and power to build automations that work exactly how you need them to.

We’d appreciate your support and are here to answer any questions!

Let’s redefine no-code automation — together ⚡

the natural-language branching logic part is the piece I'd want to stress test before trusting it on anything important. with a normal if/then builder you can see exactly what will happen, with an AI agent interpreting the branch conditions there's more room for it to misread intent on an edge case. is there a way to preview or dry-run a workflow before it goes live on real data

1000+ apps is a lot. curious what happens when one step in the chain breaks — does it retry or just die?

Hey team, thats huge.!I have been trying to build such an automation system (text to automation) and failed utterly :D Great technical work. What percent is deterministic and what percent of workflows is still random? AI is flexible, but not really deterministic. How did you solve this problem?

Congratulations on the launch!

This looks solid. Does the Copilot remember context across multiple automation requests in the same session?

Really great and helpful product!

Love how chat-driven the automation builder feels now. One thing I'd love to see is a visual diff or version history for flows, so when Copilot regenerates a scenario I can quickly see what changed and roll back if needed. Would save a lot of trial and error.

Really like where you're taking this. The AI side feels like a natural next step for Albato

Honestly surprised how fast I got a working flow going just by describing what I wanted to the Copilot. Made connecting my CRM to a spreadsheet way less painful than the usual zapier setup.

Happy launch day! Wishing the whole team a strong run today

Been a fan for a while, great to see you all on PH today. Congrats! 🔥

The chat-to-automation approach just makes sense. Way less friction than clicking through menus. Nice work

Congrats on the launch! Been following Albato for a while, always great to see the team shipping.

The copilot chat for setting up workflows is actually pretty smooth, I described a Slack to Sheets thing and it built the steps for me without much fuss. Surprised how quickly it handled the branching logic too.

With Albato being described as an AI-powered no-code automation platform, I’m curious how the AI part shows up in practice. Is it mostly helping users build workflows faster, suggesting automations, or handling steps like dictation/agent-style actions? The categories mention AI Agents and AI Workflow Automation, so I’d be interested in where you draw the line between classic no-code automation and more autonomous behavior.

The Test Step feature with real data feels like one of the most useful improvements here. building automations is usually easy until one mapped field is wrong and you have to debug the entire workflow, so being able to validate each step separately should save a lot of frustration :)

The connection-by-link and automation sharing also make a lot of sense for freelancers and agencies. Curious how much Albato Copilot can modify an existing complex workflow. can I describe a change in plain language and have it safely update only the relevant steps without rebuilding everything?

Congrats on the launch! 🚀

Canvas mode makes way more sense than staring at a list of steps. Nice work team!

Been a Albato User for few years now and a proud multiple Albato ''LTD'' holder 😎, Great tool indeed!

@Wenddy Dias good to hear, the test-before-production step is really the part that matters most to me. guardrails are only as good as someone remembering to configure them, a dry run catches the stuff nobody thought to guard against

Congrats launch! Amazing Product~

This is a solid one. Excited to see how it evolves from here. Congrats team!

Congrats team! What's been the most popular use case for the AI agents so far?

Congrats on the launch! 🎉 The Copilot is exactly the kind of thing that saves hours. Awesome work team!

Love to see it 😍 Huge respect to everyone at Albato, you keep raising the bar. Onwards 🚀

Hi everyone!

@OpenAI and Work Louder built a little hardware control deck for Codex.

Each Agent Key shows live status from Codex, so you can see which agents are thinking, running, waiting, or done without opening every chat.

The joystick launches skills. The dial adjusts reasoning. The other keys can handle voice dictation, task switching, permissions, and follow-up work.

Of course, you can do all of this with a mouse.

But turning up the model’s brainpower with a physical dial is much more fun :)

Would love a built-in cost dashboard that breaks down spending per model and project in real time, basically a way to set budget alerts before the bill sneaks up on you. Would make planning so much easier.

Turning reasoning level into a physical dial is exactly the kind of unnecessary-but-perfect hardware idea I love :))

The live status keys are genuinely useful too. once you have multiple agents running, constantly opening chats just to see who is thinking, waiting, or finished gets annoying fast. This feels like a fun little control center for people who spend way too much time with Codex... which is probably most of us here :)

The RGB status lights are the part I'd actually want, not for the fun factor but because "is this agent stuck or just thinking" is a real question when you're running things in the background across more than one company, like I do. I'd take a software-only version of that ambient status idea in a heartbeat even without the hardware.

the tactile angle caught me off guard, in a good way. is it a physical thing or software controls?

a physical dial for reasoning level is such a fun idea. and the status lights so you're not alt-tabbing every 30 seconds to check if the agent is still working, that's the kind of small thing that actually changes how it feels to use

It would be great to see a built-in usage dashboard showing token costs and request breakdowns per project in real time. Right now pulling that data requires extra tooling and it makes budgeting for client work harder than it needs to be.

Hey Product Hunt 👋 Tarek here, CEO of River, building with my brother Sam.

I used to do the same sales demo 15 times a day. Small accounts, same questions, zero time left. So Sam cloned me. My clone sells full time now: it joins a live call the second a lead enquires, shares its screen, runs the demo, handles objections, and closes.

River lets any B2B company do the same. Build your ideal account executive (voice, accent, persona, tuned to your buyers) and give every inbound lead an instant live demo instead of a form and a five-day wait.

Talk to the ramp sales team right now: rivergtm.com/demo/ramp

Will people buy from AI? We're betting the company on it. Tell us what breaks. I'll be in the comments all day, and so will Sam. And my clone.

Congrats on the launch🚀 this is such a smart application of Voice AI to a real pain point in B2B sales! Instant lead response is often the difference between a closed deal and a cold lead, so automating that first touch is genuinely brilliant.
How do you ensure the AI's tone and pitch stay aligned with a company's brand voice, especially for businesses with more nuanced or consultative sales processes?

letting AI actually close deals feels bold lol. is the demo part usable on its own if im not ready for the full thing?

@samabi love this, went through the live demo on your site and booked a session to chat more. I think this solves a great need for founders.

The $25K deal size cutoff is the most honest design choice here. Most AI sales tools claim they can close anything, which immediately raises red flags.


Thanks for your honesty 😊

An AI that qualifies and demos consistently would save most teams real hours before it ever closes a thing. The closing claim is the ambitious part, because B2B decisions happen in rooms the AE is not in. Show me it nurturing a slow deal and I am paying attention.

This is nice. How long does it take to train River to sound like a specific company's actual sales style?

This is an excellent idea, and I’m currently facing exactly this problem, especially with multilingual products. After my new product launched, I don’t have enough customer service staff to introduce it to users. It would be a great experience if AI avatars could take over client communication and product demos. I believe this will bring tremendous help to both B2C and B2B businesses.

I tried chatting with it just now, and I found it often fails to fully grasp what I’m trying to say, and I’m not sure if it’s an issue with my expression. Besides, can it actually operate my product interface to conduct real demonstrations for users? How are real cases and practical data implemented here?

Having an AI account executive instantly join a call, share its screen, run the demo, and handle objections is honestly wild :)

As a founder, the idea of never losing an inbound lead just because nobody is available feels very compelling, especially for smaller teams where repeating the same demo can consume the entire day. The hardest part is probably knowing when the AI should keep selling and when a deal needs human judgment. Curious how River detects that moment, especially with unusual objections or larger accounts?

Honestly this is the kind of thing sales teams would try immediately if it saves even a few hours of back-and-forth.

the demo and objection-handling part I can see working well, but "closes" is doing a lot of work in that sentence. does it actually have authority to agree on pricing or contract terms on the call, or does closing here mean it gets the prospect to a signed-ready state and a human still has to finalize the deal

Soon we'll have agents selling to agent buyers. Seriously though, very cool

Congrats @tarekabi & @samabi on your launch! This is a great product that addresses one of the most difficult aspect of running a business. I am just wondering if your AI agents have options for African voices. Say a local product built in Africa for Africans wanted to use River, would that fit in naturally?

the way you framed this around "no lead ever waits" is honestly such a sharp angle, most voice ai stuff feels gimmicky but the demo call flow sounds like it was actually thought through

Hey Product Hunt, George here 👋🏻

Nitrosend came out of running Obooko (producthunt.com/products/obooko, #2 Product of the Day here in February). Running a free ebook platform with 840,000 readers on its list, entirely through an AI agent, convinced me the dashboard era of email is done. It's also my third email company: SmartrMail (acquired 2022) and Sendicate before it, with six billion emails sent between them.

Six weeks ago we launched Nitrosend as headless email you run from Claude or Codex instead of a UI. Today it goes further: the agent gets an inbox, and get send for itself. Point any agent at nitrosend.com/SKILL.md and it signs up, connects your domain, sorts billing and sends its first email. no human required.

The bit I'm proudest of is Agent Inboxes (beta): real addresses for your agents on the domain you own, created BY your agent, like ernest@yourcompany.com, sylvia@yourcompany.com. They send, receive and hold 1-1 conversations with your customers, and escalate to a human when it matters. Gmail actively fights this. we built the place where it's native. Because very soon this will be table stakes for your agent, so we're shipping it now.

Coming real soon: Agentic Outreach, where your agent researches prospects, writes personally and works the replies. Get a look alike audience from your current customer list, zoom in on your ICP, give them personalised messaging. No start up founder I know likes doing spray and pray outbound, but some still do it as the tools aren't good enough to really make it relevant. We're solving that.

In development: set a goal for your brand's email and Nitrosend does the rest.

Agentic signup and onboarding are shipped, headless full stack newsletters, flows, transactional email all live sending and delighting our first users (including me on Obooko!). Agent Inboxes are beta by request (ask in app and we can turn it on for you), Outreach is coming soon, goal-based is in dev.

Super excited to ship this, ask any question in the comments!

The "agent points at SKILL.md, signs itself up and connects the domain" onboarding is slick, but the thing I'd test first is deliverability once an agent is the one hitting send. When an agent ramps a fresh domain, does Nitrosend enforce its own warmup and rate limits (plus SPF/DKIM/DMARC verification) before it'll send at volume, or can an agent torch a new domain's reputation on day one? That guardrail is what would make me comfortable letting an agent operate unattended.

Love the idea - i) Is it visible on the receiving end that its agentic? ii) i it EU compliant i.e. opt out options per default ?

Giving agents their own inbox sounds narrow until you realize half an automated workflow is waiting on a reply nobody is watching. Deliverability and sender reputation for bots is the hard, unglamorous problem underneath this. That is also the moat if you get it right.

Finally a sane way to let agents handle email without me babysitting every send. Pointed one at the SKILL.md file and it had my domain sending in minutes, which honestly surprised me.

Tried it out and sent an email from Claude in about 3 minutes. Pretty useful, I'm not a fan of logging into a clunky dashboard just to send an email when everything else is via API/MCP.

the self-signup part is the piece that gives me pause - an agent that can sign itself up and start sending on my domain without me in the loop for each step. what's stopping that from tanking domain reputation if the agent gets a little too enthusiastic with the marketing send? is there a rate limit or review gate before it actually hits real inboxes

Congrats on the launch! I'm wondering if companies would actually use AI agents to send mails because for me it's not about the technology but more around trust that there should be a human verifying. Have you seen customers with similar perspectives ?

This is really smart. Will definitely try it out for our startup.

kinda wild concept. whats stopping this from turning inboxes into agent spam over time?

Seriouslyyyy impressive product

The beta inboxes on your own domain worry me. A wrong 1-1 customer reply now comes from the company's real domain, so the blast radius lands on the sender, not Nitrosend

agent inboxes that escalate 'when it matters' is the real product. deciding when to hand off is the hard part. what signal makes it escalate vs reply itself?

Congrats @gthartley for the launch 🙌 love the target audience lookalike concept for outbound, qq how do you handle privacy and security for AI-managed conversations?

would love to see a freshness score on each idea so we know how recently the underlying complaints popped up, since some pain points fizzle out fast and others keep showing up year after year

I have shipped over 30 projects and my problem was never building, it was picking ideas nobody asked for, so this speaks to me. One thing I keep wondering though. If hundreds of members see the same top idea with the same build spec, do we all end up shipping the same product? Any way to see how many people already pulled a spec?

The idea of pulling real build specs directly through MCP is genuinely useful, saves so much guesswork. One thing that would make this a no-brainer for me is a freshness or trend signal on each idea, something like how often the underlying complaint is being mentioned recently or whether the existing profitable players are gaining traction. Right now an idea could be mined from a five year old thread and feel current when it's actually fading.

"I could waste weekends faster" hit a nerve, I'm on the other side of that right now, built the product first and I'm doing the demand-validation work in reverse. One thing I'd push on: a repeated complaint proves the pain is real, but people complain about plenty of things they'd never pay to fix. Does the database distinguish "loud problem" from "monetizable problem" beyond listing competitors who charge? That gap is where most idea-mining tools miss.

Half of the good indie products started as someone's Reddit complaint anyway, you just made the pipeline official. How do you filter complaints that are loud from complaints people would actually pay to fix? Congrats on the launch

finally a tool that skips the brainstorming step and gets straight to building. i pulled a saved idea over mcp and it dropped the schema into my agent without me touching a prompt

the receipts angle is what sells me here, actual evidence of who wants it beats another AI-generated idea list any day. how often do you refresh the "who's already profitable" data, that seems like the part that would go stale fastest since competitors and pricing shift constantly

I'd definitely talk to users before building anything but the database sounds useful. I'm curious if multiple founders use Eureka do they all end up chasing the same opportunities ?

The "receipts" angle is the part most idea lists miss — knowing someone's already profitable off a complaint is way more useful than another AI-generated wishlist. How do you keep it from surfacing the same over-mined ideas everyone's already chasing? Curious how fresh you keep the Reddit data.

This solves a real pain — I built my SaaS because a friend complained about her workflow for 6 months, not because AI suggested it.

Funny timing, I spent this morning digging through Reddit threads about client content chaos for exactly this reason. Though my product came from pain I lived myself, not a database. How do you tell a loud complaint from one people would pay to kill? Volume alone seems like it'd surface a lot of venting.

Live MRR in the bio is a bold level of building in public. Does it update down too on a bad month, or is there a vanity floor? Respect either way, congrats on the launch

Congrats Ammar! All the best here

the banner auto-update from google analytics actually works without me touching anything, kind of mind blowing how it just pulls live numbers into the profile. wish more tools did this instead of making me copy paste screenshots every week.

Love the idea of live social proof in bios. One thing that would make it even better is letting users pick which specific metric goes in the bio versus the banner, since real estate is tight and founders might want MRR in the bio but traffic growth shown on the banner.

The trust mechanic is really good! Is there a public proof layer, a badge or a link back to the source? Or the credibility leans on people already knowing you built it straight?

Love the idea of auto-updating metrics, would save so much hassle. One thing that would make it even better is letting users schedule metric freezes for specific date ranges, like pausing auto-updates during a slow weekend so the bio doesn't reflect a temporary dip. That small layer of control would really help with how the stats represent the business.

honestly this is pretty cool, been waiting for something like this. one thing though, would be awesome if you could add a scheduling feature so the stats update at specific times instead of constantly, like maybe only refresh daily or weekly. some founders probably don't want their numbers flickering all the time

@therayess - great idea, and once my products start earning definitely something I may try out. Best of luck with today's launch. Post Hog integration would be useful and also a discounted annual plan, and possibly a life plan. I tired of more and more subscriptions.

@Ammar Rayess thanks, good to know it's flexible on the window. 30 days is probably what I'd default to for a bio stat, all-time can look impressive but it stops meaning much once the number is huge

The Stripe + GA4 combo is smart, but a ton of indie makers ship mobile apps where revenue runs through the App Store / RevenueCat instead of Stripe. Any plans to pull those in? I'd happily put live IAP revenue in my bio lol.

honestly the "authentic numbers" pitch lands, half the MRR screenshots floating around are stale or cherry-picked the one good month. connecting straight to Stripe and GA4 means it can't really be gamed the same way. does it let you pick a rolling window (last 30 days vs all-time) or is it locked to one metric format

QA never gets love so nice to see this. does it flag flaky tests or just fail them?

Hi Product Hunt 👋

I'm AbdelRahman, founder of Manta AI. I'm always here — ask me anything.

The honest version of why I built this: I spent years managing software development teams across telecom, fintech, e-commerce, and real estate. Testing was always the part that broke under pressure. Scripts failed every time the UI changed. QA teams burned hours maintaining automation instead of actually testing. And at some point, the team quietly gave up on automated UI testing altogether — not officially, just practically.

That decision always came back to hurt us.

When AI coding tools arrived and made development even faster, I knew the testing gap was going to become a real crisis for a lot of teams. So I built Manta AI.

Manta explores your web app autonomously — without being told where to look. It finds bugs in real user flows, not just unit-level failures. You can also describe a specific flow in plain English and Manta will test it for you, no script required. When your UI changes, the tests self-heal. No selector updates. No maintenance.

The runner can be deployed locally on any machine or server — so you can test apps behind a firewall, on a private network, or on localhost without exposing anything to the public internet. Or run it from the cloud if you prefer.

Free trial is live now. No card required.

If you try it, I'd love to hear your experience — what worked, what surprised you, and what you'd do differently. That's more useful to me right now than anything else.

Cool idea. Right now I'm using codex with browser or claude code with browser use to do such testing. How do you see Manta giving more value?

I ship MVPs fast and testing is always the thing I skip when I am tired, then I pay for it later. Curious how it handles login flows, that is where most of my bugs hide. Congrats on the launch.

@aelsergani the local and firewall-network angle is interesting, most testing agents assume a public URL. However I'm curious how it handles authenticated flows, like SSO or MFA prompts?

Coverage is easy to demo, trust is what makes a testing tool stick. If it cannot tell a real regression from a timing flake, teams stop reading the reports within a week. I would lead with how you hold the false-positive rate down.

Local runs against localhost:3000 are the right bet here - most agent testers die before they can reach a staging app

self-healing test cases is the part i actually care about. writing the tests was never the hard bit, keeping them from breaking every time the UI moves is. if manta handles that reliably it solves the real problem

The self-healing tests actually adapted when I tweaked a button label, which caught me off guard in a good way. Local execution is a nice touch for our staging environments behind a VPN.

Pointed it at a staging app and it picked up a broken checkout flow on its own without any setup. The self-healing part feels pretty solid too when I shuffled the UI around.

Really like the framing here, AbdelRahman — the honest "the team quietly gave up on automated UI testing" is such a real failure mode, and going after it with autonomous exploration instead of brittle selectors feels right.

One thing I'd genuinely love to understand: when the UI changes and a test self-heals, how does Manta tell "the UI legitimately moved" from "the UI broke"? A redesign should heal, but a regression that shifts the same element is exactly the bug you'd want it to catch — curious where that line gets drawn.

And on the autonomous exploration: since the agent picks its own path, how repeatable is a run? If it surfaces a bug today, will the same flow reliably reproduce it tomorrow, or does the path drift between runs?

Excited to try it on a flow behind a firewall 🙏 @aelsergani

the self-healing part makes me a little nervous rather than reassured, if a UI change is actually a bug (a button moved because of a broken layout, not an intentional redesign) does the agent know the difference, or would it just quietly adapt the test to match the broken state and mask the regression

Okay, as promised, I'll go first 😅

Mine was a weekly "alignment sync" that survived three reorgs. The project it was created for shipped in 2023. Nobody remembered why it existed, but nobody wanted to be the one to kill it — so every Monday, eight people spent 30 minutes confirming there was nothing to align on.

It finally died when someone noticed the original organizer had left the company a year earlier.

That meeting is basically why In Parallel exists. If the decisions and status lived somewhere that updated itself, the meeting would have had nothing left to do.

Your turn — what's yours? 👇

As a CPO this hits a nerve 😅 The gap between "what we decided in the room" and "what's actually on the roadmap" is where half my week disappears. Every tool captures something — notes, tickets, docs — but the decision itself always seems to fall through the cracks. Love that In Parallel sits underneath instead of adding yet another surface to babysit. Pricing per workspace instead of per seat is the right call too. Congrats on the launch, upvoted 🚀

SOC 2 Type II for an MCP context layer is exactly the boring enterprise signal I'd look for

Every comment here is about big teams so let me ask from the other side. I am a one person company and I still lose context between Claude Code sessions every single day. Is this built only for organizations or does it make sense for a solo builder too?

Congrats on the launch, Kristian! The concept of standardizing the shared context layer over MCP is brilliant. A major pain point with centralized team hubs is 'context drift'—a decision is made in Slack, the execution changes in GitHub, but the overarching context file stays static. How does In Parallel natively capture those micro-updates without forcing team members to manually edit the shared state every single day? Is it passively listening to integrations?

The portability problem is the real one: context trapped in each tool means every agent is a stranger. Building on MCP is the right bet because it standardizes that layer instead of locking it to one app. The test will be keeping context fresh, stale shared context is worse than none.

Congratulations on the launch! Solving the context problem for organizations with information moving faster than ever is a big deal! Heck, just figuring out how to manage my own personal context as a solo builder to make the most out of AI has it's challenges.

This bit me two weeks ago, a rule I set at planning stage silently disappeared before the generation step and the output invented a fact. Took a day to find where the context died. So, real question: two agents pull the same context, one updates it mid-task, what happens? Last write wins or something smarter?

Really interesting approach to the LLM memory bottleneck standard custom instructions are too static, and vector databases get too noisy. Does your self-hosted VPC setup allow us to restrict sensitive context folders so only specific teams (like HR or Finance) can query them?

One thing that would make this a no-brainer for me is a mobile companion app. Most of my context-switching happens away from my desk, and being able to quickly check on goals, risks, or ownership changes from my phone would keep me aligned even when I'm not at my laptop. A simple read-only view with the ability to flag updates would be enough.

every "single source of truth for the org" tool I've seen eventually turns into the thing everyone stops updating, then it's worse than no source at all because people trust a stale answer. how does In Parallel keep goals/ownership actually current without someone manually maintaining it, is it pulling straight from where the work happens (tickets, docs, calendars) or does it still rely on people logging updates

The "explain your company once and every agent already knows it" pitch nails the actual daily tax of re-pasting context into each new chat. Where does that shared context actually live: a hosted store on your side, or something I control and can host in my own environment? And how does it stay current: does it auto-derive decisions and ownership from the connected tools, or does someone have to curate what goes in? With permission-scoped MCP access, is scoping per-user so an agent only sees what that person could see, or is it one org-wide context every connected tool can read?

docs go stale in a week so i like the living map idea. how does it stay updated without me babysitting it?

One thing that would help us adopt this faster is a visual timeline view inside the operational map, so we can scrub through past workflow executions and see exactly where drift or failures happened. Right now troubleshooting at scale feels like guesswork.

the framing that the model was never the hard part is honestly accurate for most companies, the real question is whether the "living map" stays correct as workflows quietly drift over months, not just when the UI visibly changes.

spend months trying to get an agent to reliably click though our old internal ticketing system before giving up, this is the exact problem.

nothing in the pitch about workflows that need a human judgement call partway through, is that out of scope for now?

feels adjacent to Browser Use in spirit, just aimed at ERPs and desktop apps instead of the open web.

building stable tools instead of chasing every API makes sense, but that shifts a lot of ongoing maintenance onto graft's side to keep those maps accurate as companies change.

if the underlying business logic changes and not just the screen, does the agent, facing tool silently start doing the wrong thing instead of breaking?

approvals + audit trails baked in from day one is the part that'll actually get this past an IT review, not the AI framing.

Graft turns real company operations into a shared, continuously updated layer for agents. We've seen companies transform to the AI native era but fail in operations because of institutional knowledge, and graft solves that.

the self-healing part for legacy UI drift is the piece nobody else seems to be tackling well, most RPA-style tools just break silently when a button moves. what's the actual gap between the UI changing and Graft detecting + repairing it, are we talking minutes or does it need a human to confirm the fix before it goes back into production

hello everyone, just a quick update. I've updated graft's social media, the old one was the wrong one. Thank you.

Graft will be helping f500's transform their legacy applications into operational knowledge basis for agents to use. Agents can interact with their legacy software and help them transition into agent native software, this is a shift we're seeing but companies fail because the data is in people's head as domain knowledge so we're building the infra to make it operational by agents.

Building a receipt-scanning feature right now and even at my tiny scale, the "verification" part is the hard bit — I ended up schema-validating every LLM response and failing closed on anything unparseable, because silently-wrong output is worse than an error. Curious how you handle the case where the workflow succeeds mechanically but the result is semantically wrong — is verification rule-based per workflow, or learned?

Operations as a map agents can walk is close to what I do per-client, structured facts an AI can't step outside of. Mine is one brand, yours is a whole company, and I suspect the hard part scales badly. How do you keep the map current when the operations change weekly?

The sheer speed of this is impressive! I am not directly involved in academic research currently, but (like most people I guess) I have topics that I am curious to explore further. I spent a good while exploring the topics via this platform and was hugely impressed by the results, the UI and the speed with which they were returned. It was easy to access original content (where applicable) and I would certainly recommend.

Really glad someone finally built a search engine that doesn't choke an agent after three requests. If you could expose a streaming endpoint or webhook for long queries, agents could kick off a deep search and poll instead of holding a connection open the whole time. That would make it way easier to chain searches across Claude Code sessions without timing out.

finally a search tool that doesn't choke when my agents hammer it. the MCP integration just worked out of the box with claude code, which is more than i can say for most academic apis.

The MCP endpoint is what sold me. I keep hitting the Semantic Scholar 1 req/sec wall when Claude runs a deep lit-research pass and it just stalls. Does the free 100/min hold up for a full run, or does it throttle once an agent goes wide?

Hi @emmanueladesola , I always think more clearly out loud than on a blank page, so watching my rambling turn into something I can actually look at later sounds really nice. Replaying how an idea took shape is a lovely bonus.

Congrats on Weave. The live map feels more useful than a plain transcript because it keeps the structure of the thinking, not just the words. The pause-triggered questions are a nice touch too. Curious how you handle messy spoken thoughts: do you preserve the original phrasing anywhere, or only the cleaned-up cards and connections?

The pause-triggers-questions part is the bit I'd actually use. Half my thinking stalls because I never notice the question I'm avoiding. How does it decide which questions to float without derailing the train of thought? And does the Markdown export keep the card links or flatten it to a plain outline?

what happens if it sends the wrong email or books the wrong thing tho , who's on the hook for that.

Gave it a quick spin and was honestly surprised how natural it felt to just text an "employee" to handle a task. The idea of them having their own inbox and memory makes it feel less like a chatbot and more like an actual hire.

same energy as Thomas's comment lol, though I wanna see one finish a full task start to finish before I call it an "employee".

gonna try hiring one just for lead follow ups first, feels like the safest way to test it before trusting it with money.

no pricing past "free options" and nothing on limits for what the wallet can actually spend.

is this for solo founders who want an assistant or teams trying to cut headcount, kinda reads like both.

agents delegating to other agents sounds hard to get right without stuff going sideways. how's that actually working so far.

• cool until it does something dumb with no human checking first. how much can you actually rein it in.

an AI with its own wallet and phone number? bold move 😅

wait, they get crypto wallets?

There have been many products which claimed this before, but how do you ensure that while transferring from one agent to another, the context is maintained effectively?

This is where I felt other products in the market have not done as well as they had communicated.

"How much can you actually rein it in" is the question I'd want answered before hiring one of these. I don't run autonomous employees, but I run a monitoring tool across three businesses, and the thing that earned trust wasn't making the AI more capable, it was making its confidence visible: it grades every call it makes (Verified, Very Likely, Needs Review, Monitor Only) instead of acting the same way on everything. If an AI employee could show which of its own actions it's least sure about before it does them, not just after something goes wrong, that would make the "who's on the hook" question a lot less scary. Congrats on the launch, curious whether that kind of self-reported confidence is on your roadmap.

Hey Product Hunt 👋

I'm Muni, founder of dot.

The feedback loop for most builders in 2026: build something, screenshot it, paste in Slack, type "thoughts?", get "looks good 👍", ship it, find out three things were broken. Every week.

Now everyone's building faster than ever with Claude, Lovable, Cursor, v0, but the feedback loop hasn't caught up. You still bundle files, deploy somewhere, paste screenshots, and lose all the context.

dot. fixes that. Create a project for any URL or Claude artifact, anyone can click anywhere on the live page and pin feedback. No signup, no extension, nothing to install. Built something in Claude? Say "add dot. feedback to my artifact" and dot. turns it into a shareable page, nothing to deploy. Iterate your artifact and the review link stays the same; feedback stays attached to the version it was left on.

The part I'm most excited about: dot. is in the official Claude connectors directory, trending #8 today on it, and since going live two weeks ago, 1,200+ people have connected it. Inside Claude, you say five words and get a shareable project with feedback. Build → get feedback → iterate, all in one place.

Reviewers are always free and never need an account, you only pay for the people creating and managing projects (solo or a shared team workspace).

For Product Hunt: use code PRODUCTHUNT for 1 month free on the Individual plan.

I'll be here all day, would genuinely love your feedback 🙂. What's the most broken part of your feedback process right now?

— Muni

the click-to-pin mechanic on the live page is such a clean alternative to screenshot chaos. love how it removes the friction without making the reviewer install anything.

Honestly, the click-to-pin thing on a live page is way smoother than I expected, basically feels like leaving a sticky note on the actual site. No extension or signup hurdle is genuinely the move.

the click-to-pin approach is really clever, it removes that awkward "what tab do i screenshot" back-and-forth that always kills momentum on async reviews. loving how frictionless this is.

The no-signup reviewer flow feels right for AI-built prototypes. One thing I’d want as a builder is a tiny feedback receipt after a review: what page/state the comment was pinned on, what changed since that link was created, and whether the reviewer saw the latest version or an older artifact.

That would make comments easier to trust when Claude/v0 iterations are moving fast.

The "no signup, reviewers just click anywhere to pin a comment" flow is exactly what kills the screenshot-in-Slack loop for me. Day-one question: when I paste a Claude artifact or URL and then ship a change to the underlying page, do the existing pinned comments stay anchored to the right element, or do they drift/detach once the DOM moves? And can I mark comments resolved or export them into a clean punch-list, or does all the feedback just live in the thread?

Just tried this with a landing page I'm building in Claude, said "add dot. feedback" and got a shareable link in seconds. Sent it to a friend, they pinned 3 comments directly on the page without creating an account. This is how feedback should work. The fact that reviewers don't need to sign up is a huge deal.

Giving feedback on screenshots for a website feels like we're still living in caves. This tool is 🔥.

Congrats on the launch! Love the "no signup, pin anywhere" model — feels like the same instinct behind CancelKit (capture the reason right at the moment/place it happens, not a follow-up form later). Curious how you handle pin anchoring on pages that re-render client-side (React/SPA) — do you snapshot a DOM selector at pin time, or something more robust to layout shifts?

the click-anywhere-on-the-page commenting is genuinely clever, turns passive feedback into something way more natural than screenshot chains

Shoutout to everyone checking out amami.dev on Product Hunt!

We’ve prepared an exclusive gift just for PH community members: 3 full months of Pro membership, 100% free — no hidden fees, no strings attached.

Grab your free trial right away!

Hit me up in the comments or shoot an email to watson_zang@foxmail.com if you’ve got any questions.

Exclusive PH promo code:

AMAMI-97F1E-70C61-C30E5-B87E8-1C048-171E1

Hey hunters! 👋

We built Amami because we kept asking ourselves: "Why do I need to open a browser to check analytics when I spend 90% of my day in Cursor?"

The MCP integration means your AI assistant becomes your data analyst. Not just "here's your pageview count" — but "your pricing page got 3,240 visits last week, HN visitors stay 42% longer, and I recommend doubling down on your HN launch strategy."

Happy to answer any questions about MCP, privacy-first analytics, or our tech stack!

P.S. — Try the one-prompt setup. It genuinely feels like magic the first time your AI says "I've added analytics to your project."

out Software: dashboard.amami.dev/login

software Docs: docs.amami.dev

Shoutout to everyone checking out amami.dev on Product Hunt!

We’ve prepared an exclusive gift just for PH community members: 3 full months of Pro membership, 100% free — no hidden fees, no strings attached.

Grab your free trial right away!

Hit me up in the comments or shoot an email to watson_zang@foxmail.com if you’ve got any questions.

Exclusive PH promo code:

AMAMI-97F1E-70C61-C30E5-B87E8-1C048-171E1

From my experience: a dashboard sometimes shows something you weren't looking for - a spike you didn't think would happen. With a query-first setup, how do you surface the thing you didn't know to ask? 

I do SEO for my own product and the first 20 minutes of every morning go to checking analytics and Search Console in separate tabs. Asking Claude for the numbers inside my editor would give me that time back. Does it only track visits on my own site or can I also see which pages are winning in search so I plan content from the same place?

Staying inside your work while checking how things are actually doing, with no jump into yet another dashboard, is the kind of small relief that quietly adds up.

The pain is real, dashboards get built and abandoned constantly. This only works if answers are auditable, otherwise it is a confident guess I cannot defend in a meeting. Traceability is what separates this from a chatbot bolted onto a database.

the mcp setup was way easier than i expected, like two minutes and i was asking claude about my traffic from the editor. really nice not jumping back and forth to a browser tab anymore.

Love the MCP integration idea, this is exactly the kind of thing that makes analytics actually useful inside my workflow. One thing that would seal it for me is real-time visitor counts streaming live, so I can literally watch the impact of a post I just published without refreshing anything.

The MCP integration is genuinely clever, that part hooked me. One thing though, I'd love a way to compare traffic windows right from my editor prompt, like asking how this week stacks up against last week without opening a dashboard. That would make the AI-driven angle even more useful day to day.

finally something that fixes the annoying back-and-forth between my editor and the browser tab just to check stats. the MCP integration works smoothly, asked claude about yesterday's traffic and got a clear breakdown with actual suggestions on what to fix.

Finally, analytics I can actually query from my editor instead of alt-tabbing like it's 2009. The MCP angle feels genuinely useful for quick "is this page getting traction" checks.

How are you determining the risk that an airline will void an error fare? Is the AI looking at historical behavior from that specific airline or more at the characteristics of the fare itself? Really interesting feature either way.

@shahvyy Wait, the AI actually looks at the likelihood of a mistake fare being honored?? That’s such an obvious thing to add once you hear it, but I’ve never seen a flight deal service do it. Very cool.

Really like that each deal comes with an AI breakdown of the route, pros/cons and the “things to know before booking” — with mistake fares the price is only half the decision. One thing I’d love to know: does that breakdown also flag how risky the fare is, i.e. the chance the airline voids the booking as an error, so I know when to hold off on booking hotels around it? Congrats on the launch!

mistake fares are basically a race against the airline noticing, so speed is everything here. how fast is the alert actually landing after the fare goes live, and once someone books on it have you seen airlines honor it consistently or is there a real risk of a "sorry, we're cancelling this ticket" email a week later

The is this fare actually going to ticket? angle is the killer feature for me. Finding a $180 flight to Europe is great. Finding out the airline canceled it AFTER you booked a nonrefundable hotel is not

Going user here 👋 I don’t necessarily think I’d cancel it immediately, but the AI analysis on error fares is genuinely interesting. I’ve always been nervous about booking hotels right after grabbing a suspiciously cheap flight. Going to test FlightGlitch alongside it and see how they compare.

Cool idea team.

I want to sign up and see it, but honestly not quite willing to drop my credit card yet. My recommendation - a short free trial so I can see a few alerts would probably have convinced me.

Good luck with the launch!

Scored a $138 roundtrip to Lisbon last week that would've been $900 on the airline site, alerts hit my inbox within minutes of the mistake going live. Game changer for spontaneous trips.

Hey man! I've actually tried to build something similar before (though I gave up on it), so it's cool to see someone push it live. I kept hitting walls with the GDS APIs like Amadeus and Sabre, and even Duffel, and at any real scale the flight data pulling plus cron scheduling got iffy fast, both on cost and rate limits.

How did you end up solving that side? Are you polling routes on a fixed schedule or reacting to some cheaper signal, and did you stick with a GDS or find another source for the fare data? Really curious how you keep it fresh without the bill exploding. Congrats on shipping it.

This looks useful. Are there any limits on how many airports can be tracked at once, and how quickly are alerts typically sent after a deal is spotted? Congratulations!

hey, are you plannign to launch thsi fro EMEA fligths as well ?

If you give it access to content from a camera roll can it come up with its own ideas for content?

The thing is that camera rolls are usually messy. I have some duplicate takes, blurry clips, and random videos. How well does the AI filter through all of that?

The "no timeline editing, just match each script line to a shot from your roll" flow is the part that would actually get me to finish a video instead of abandoning it half-edited. Day-one question: when the auto clip-matching picks the wrong shot for a line, can I swap that single clip without redoing the whole edit, or does it regenerate the sequence? And does export give me a clean file I own (no watermark) plus the individual matched clips, or just the stitched render?

Abigail, the part of posting videos I dread most is the fiddly editing, and my camera roll is a graveyard of clips I never do anything with. This feels like it respects my time and my laziness in equal measure.

Congrats on launching ClipMatch. I like the focus on turning existing camera roll footage into usable short-form videos instead of starting from a blank timeline. The script-to-clip matching feels especially useful for creators who already have lots of raw footage. Curious: when the script changes after clips are matched, do you preserve locked shots or re-run matching from scratch?

Everyone's sitting on months of unused footage, so the source makes sense. What I couldn't tell from the page: does it pick up how the account actually talks over time, or is the caption style the same for every user? Asking because voice is the thing I fight daily building for ghostwriters.

Took a stack of random clips from a weekend trip and ClipMatch actually paired them with my rough script better than I expected. The auto-export for different platforms saved me a solid hour.

would love a way to lock in specific clips so the AI doesn't swap them out when regenerating. sometimes the match is already perfect and I just want to tweak the script without losing that shot

Hey Product Hunt 👋 Oleksii here, one of three makers.

For 8 years we've run Fulcrum, a dev studio — 100+ products shipped. Every client conversation started with the same two questions: how much will it cost, and when will it be done?

And for years, my honest answer was: it depends.

It depends on a pile of risks. A pile of nuances. Work everything out properly upfront — that's waterfall, and it takes forever. Go with an open check — the risk shifts to the client: they might pay more than they planned. Either way, one thing was constant: clients pay for hours.

Lately there's a third option. Founders try to build with AI themselves. They burn weeks and tokens — and rarely reach real production. Limitations show up. Slop shows up. The money is already spent.

Here's the uncomfortable accounting truth behind all three paths: unfinished work is money that never capitalizes. Half-built features, slipping roadmaps, abandoned prototypes — pure spend, zero asset. Only code in production converts money into value.

And a second problem I kept seeing with non-technical founders: they can't evaluate whether an engineer is good, can't predict budgets, can't see risks. The outsourcing model quietly transfers all of that responsibility onto the person least equipped to carry it.

That's why we built SonOf — for ourselves first. It started as an internal engine to push Fulcrum client work through faster. After ~10 engagements ran on it, we realized we'd built the product. (Yes, it's named after Son of Anton — Gilfoyle's AI in Silicon Valley. We're that kind of team.)

SonOf reads your entire product — repo, tickets, docs — as one context, and does the discovery work that used to take weeks. What you get is something the old model could never give:

→ Full visibility over scope, before you spend anything. Every piece of work becomes a ticket with a fixed price on it — $500 per story point.

→ The timeline in your hands. You approve tickets one by one — what to build, in what order, at what pace.

→ Payment lands where value lands. You're billed only when code reaches production — the moment spend becomes an asset. A senior engineer signs every plan and reviews every PR before merge. Rejected work costs $0.

See what that does to the two problems above? You don't need to judge engineers, budgets, or risks anymore.

So after all these years, my answer finally changed — not because we work differently now, but because the client finally sees what we see.

How much? It's on the ticket, before you say yes.

When? You set the pace — approve, and the engine works around the clock.

Not "it depends."

The audit of your product is free: connect a repo, see your whole map. I'll be in the comments all day.

Love the name! 😁
What do you see as the main use cases for the product? I'd love to understand where you think it provides the most value.

Sounds interesting! But how can a non-technical client verify that the estimates are adequate and the code quality is high?

Genuinely impressed that the billing only triggers on code that ships to production. The senior engineer review on every PR gave me confidence to let it touch real tickets instead of just test repos.

Charging on outcome instead of effort is rare, and customers notice, most tools bill you for trying and failing. The operational risk is disputes: what counts as shipped, who arbitrates, what happens on a half-done feature. Get that airtight and this is genuinely differentiated.

Congrats on launch! What are the numbers of how it impacted the projects you'd already delivered?

Oleksii, the honest bit here is that work only counting once it truly ships really lands with me. So many things claim progress long before anything reaches real people, and that framing feels refreshingly grounded.

pay-only-when-it-ships puts the risk on you, not me. that's the rare part. does the senior engineer sign the plan up front, or only review the final pr?

'Act on anything' is super interesting! A browser already logged into my email and bank taking actions on its own is a much bigger trust jump than remembering what I read. QQ - where do you guys draw the boundary on what it can touch before a user hands it that much rope?

Curious to see how you draw the line between remembering useful context and keeping the user in control.

Christophe, I live across an absurd number of tabs and forget half of what I read, so the idea of something that quietly holds onto all that for me is genuinely appealing.

Most AI tools feel like talking to someone with amnesia — you explain your project every single time. The "remembers everything you've read" angle could actually fix that. Does it work across different AI tools or is it tied to Trace's own interface?

one thing i'd love is a way to manually tag or highlight snippets within pages so the AI can pull from those specifically instead of just the whole browsing history