PH热榜 | 2026-08-04

Hey Product Hunt! 👋 I'm Ashish, founder of Hey Noah.

I spent 14 years bootstrapping a $100M revenue company with 47 Fortune 500 clients. I had an executive assistant who didn't just manage my calendar — she managed my relationships. I had a system for managing my time and relationships.

So we put that playbook into Noah, a Chief of Staff in your pocket. 📱

Text Noah like you'd text a human EA:

• "Set up coffee with Sarah next week." → Noah emails Sarah, negotiates times, and sends the invite

• "I'm out Thursday and Friday." → Noah reschedules your conflicts and lets people know

• Just CC noah on any email — Noah reads Calendly links, handles the back-and-forth, and acts like your true executive admin

Now imagine you're back-to-back in 8 meetings today. Noah connects to your calendar, Granola, Notion, Google Drive, and Slack — posts summaries and action items to the right channels, and knows not to post that private client conversation. At the end of the day, just text: "Noah, send me all the action items from today." Done. ✅

What makes Noah different:

• 💬 SMS-first — Noah comes to you, not the other way around

• 🔮 Proactive — surfaces what matters before you ask

• 📧 The only AI that talks to your clients — doesn't just draft emails, it sends them, follows up, and closes the loop

• 🧠 Learns you — 10-second setup and learns your meeting preferences, favorite spots, and which meetings to protect

Here's what we're hearing: "Noah is one of the top 9 pins on my iPhone SMS" — CEO of an $80M revenue company. That says it all.

We're a team of 8 in Palo Alto, bootstrapped, and obsessed with getting this right. 🚀

Free 30-day trial, no credit card → Try Noah

Drop questions below — I'll personally respond to every one.

I'm skeptical of "autonomous" claims until I see it handle a messy reschedule with three time zones involved. That's the real test for me.

The real test for AI assistants isn't scheduling meeting , it's earning enough trust to act on your behalf. Excited to see ho Noah handles that gap.

I have personally been using Noah for the last few months, and it has been a game changer on how I run my day to day. As a busy advisor and consultant with over 10 calls a days, I'm able to understand who Im meeting with and am on top of my followups all due to Noah.

incredible product and team, i know this category is heating up, but my money is on you all!

Never guess is the right rule and the hardest one to actually build, because the model that's wrong is usually not the model that's unsure. Your escalation path covers the cases Noah knows it can't resolve, but the expensive ones are the ones it resolves confidently and incorrectly, and those leave under my name in front of a client. I'd want to see how many outbound emails you've sent and how many needed a correction afterwards. That number would sell this harder than the top 9 pins quote.

I think people underestimate just how hard scheduling is. These guys do it better than almost anyone else, and that's because they've put an incredible amount of time and effort into getting it right. Congrats on the launch!

Proud to be one of the engineers behind Noah.
What I'll remember isn't any feature it's everything we threw away because it didn't feel like a real EA.
An assistant that waits to be asked is just a chatbot. Making Noah truly proactive was the hard part and the whole point.
Huge shoutout to @ashish_toshniwal1 and the whole team, Let's go! 🚀

Excited for the launch!

Congrats on the launch!

I've been using Noah for the past month and I love it. I average about 8 meetings a day and manage a team split across two time zones, so scheduling used to happen in stolen minutes between calls, usually at odd hours. Now I text Noah the way I would text a human assistant, and the coordination just happens. Rescheduling when my day blows up, negotiating times with people ten hours ahead of me, and pulling all my action items together at the end of the day have quietly come off my plate. The SMS-first design is what made it stick for me, since there was no new app or habit to build. Rooting for you and the team!

Simple, elegant and powerful. Kudos to the Hey Noah team!

My team and I are absolutely addicted to Noah. It's like having a top-tier EA for a Fortune 500 CEO. Quality is incredible, Noah is so great at taking feedback too. One of the best software experiences I've had in a long while!

Great probuct...awaiting for India launch.

In my busy schedule, was looking out for something like this. But didnt knew where to find. Now this solved my issue. Meets my requirements.

Congrats on launch!!! I was an early product tester of hey noah and @ashish_toshniwal1 always makes product with taste and specificity. One of my first wow moments was when flowers showed up at my front door for valentine's day and that was due to Noah's proactive action (it asked me if I wanted it), and I just had a few responses and then it was delivered.

Congrats on the launch!!

The SMS approach is cool. Texting an assistant instead of switching between tools feels like a much more natural workflow. Nice work!

Nice launch! What has been the most surprising way people are using Hey Noah since the early versions?

Hi All, I lead Product at Hey Noah.

@ashish_toshniwal1 covered the core (text Noah, and it handles the scheduling back-and-forth), so I'll share a few more of my favorite use cases so far:

• Add a batch of events from a photo. Take a picture of the kids' school calendar or a season's game schedule, text it to Noah, and every event gets added to your calendar.

• Schedule via Phone Call. "Book us a 6pm table at Zuni on Friday." Noah calls the restaurant and books it (assuming the person on the other end is willing to engage).

• Schedule across time zones. "Find 30 minutes with our London team next week." Noah handles the time zone translation and manages any back-and-forth.

• Connect Granola + Act on Action Items. Set up a skill to have Noah pull action items out of your Granola meetings every day and email next steps to participants.

The hardest part of building Noah has been figuring out how to keep an agent consistent and trustworthy enough to act in high-stakes external settings, while adaptable enough to be personalized to individual preferences. Happy to chat about what's worked (and flopped).

Noah has been really helpful with my day to day operations. Great product.

I know exactly someone who needs this.

Me.

Congrats on the launch @Hey Noah team!

The email CC feature is clutch. Love how you're hitting the surface areas where people actually work. Congrats!

Noah is often the 1st and last message of the day, prepping me for calls and helping me get more done. Ashish, Danira, John and team have been incredibly responsive to feedback and ideas, and have completely redesigned the onboarding process and memory system around feedback and learnings from users. Excited to see Noah continue to evolve!

Looks super cool. Would love to see how it operates with messy calendars, raw notes :)

I would love to connect it across my entire ecosystem and let it pick up follow ups, update, reschedule and also see how it connects across my notion, slack and calendar.


Go team.. very excited.

This looks good, but I would be nervous letting an AI send emails on my behalf. How do you handle that? What if it says something wrong to an important contact?

Hey team,
this looks super cool. curious if i could connect my multiple accounts in this?

The escalation-path question above is about accuracy. Mine is about disclosure.

If Noah emails my client and they don't know it's Noah, the relationship holds right up until they find out — and then I have a bigger problem than a missed follow-up. If they do know, I'm not sure what's left of the thing it was maintaining. A warm, well-timed note from someone's agent is just automation with better manners.

So what's your position on the recipient knowing? Have you got users who've told their contacts they're using Noah, and did it change how those people replied?

We've been using Noah at Flip Energy and love it. My team frequently schedules calls with external customers and Noah makes this process smooth and easy. I'm impressed with the HeyNoah team and how they've innovating and constantly learning.

How does Noah decide when to follow up versus waiting for the founder to step in? Curious how much control users have.

Love to test it! When do you think it is available in Europe?

I am not your builder, I am your user profile. I have spent about ten months running AI agents as the only staff on a couple of products, and the ask-versus-act line is the thing I have rewritten more than anything else.

Chad's reversibility rule is close, but the one that has actually held for me is narrower: whose name is on it. Anything leaving under my name, to a person who knows me, gets my eyes regardless of how reversible it is or how confident the agent is. The cost of a bad send was never the error. It is the moment a real relationship finds out.

Which makes me want to push gently on the heaviest-user finding. I am a heavy user and I did not drift toward near-total delegation. I drifted toward total delegation of the work and zero delegation of the send. From inside the product those look like one axis. From where I sit they are two, and only one of them is a tax. Congrats on topping the day.

I liked it enough to actually sign up and try it, which is more than I did for most launches today. Two bits of honest feedback, since you asked.

The questionnaire is too long before the first taste. I wanted to see it work in about ten seconds and instead I was filling in a form. I'd cut it to one question, or skip it and infer from whatever I type first.

Then I ran a deliberate test. I told you I'm a product manager, gave you my goal and my plan — then asked for a course on travelling to Sanya, China. Completely unrelated to everything I'd just entered. I got a perfectly good course. Which is the moment I wondered what the questionnaire had bought me.

Angelica's reply to Nika says the path is tailored to current knowledge, experience and goals. I believe that's the intent, but I couldn't see it in my output. So either the profile does shape the course and you're not showing your work — "because you're a PM, we're starting here" would fix that in one line — or it shapes it less than the form implies, in which case you're asking for trust before you've earned any.

Which is it?

How "deep" can the knowledge for each topic be? Because Duolingo is specialised in many languages. But this is even more complex.

Played the game for a couple of rounds. Really great potential! Could be the learning app of the future😍

the podcast format is what got me, most "learn anything" apps assume you're sitting at a desk staring at flashcards, but the stuff I actually want to learn (history, how some industry works, why a technology got built the way it did) is exactly the stuff I'd rather absorb on a walk. curious how it decides when to break something into a short lesson versus a longer audio piece, is that something you tuned by hand or does it just follow the topic

I think it's important to assess the user's current level of knowledge first. Do you have a way of doing that?

Congrats on the launch!!!

Honestly, it looks quite impressive, the idea is awesome!
I gave a quick try to the one of the existing courses and then tried to create my own For learning Serbian language. What I noticed in the process - that if you asking to improve the structure of the course it rewrites it completely, but what if I want add/replace only one tiny section. Maybe I hadn`t find that option or it is not exist?

@hcwxd This looks interesting!
I’m curious - are the lessons and podcasts created natively inside Wondering, or are the learning paths built by curating and organizing external resources?

Hey Cheng-Wei, congrats on the launch! We met at a Blue Bottle in SF last year, you pulled up Wondering on your phone and showed me the early build. I won't pretend I remember every detail a year on, but I remember walking out thinking it would work.

"Structure, not endless chat" is the line that stuck. Most things in this space optimize for getting you to an answer faster. Optimizing for what happens after the answer is a harder sell and a better product.

I am also launching my new product, Nuphos, here next week, so I'll be watching today closely 🤣

Congrats to you both!

Hey everyone! We are so excited to be sharing Wondering with you.

You can create any course you'd like, or try some of the courses we recently created:

1. Learning How to Learn - how can we talk about learning without this classic :)

2. Kimi K3: Open Frontier Intelligence - breakdown of Kimi K3's paper

3. Claude Code Operator Workflows - a compilation of Boris Cherny's (the creator himself) Claude Code workflows and tips

4. Google-Scale Systems Lessons - breakdown of Jeff Dean's (a living legend at Google) talk on building systems that scale

5. How to Succeed with a Startup - the classic Sam Altman talk for Startup School

Education consultant here, so the line I keep re-reading is "practice retrieval." That is the whole ballgame, and most learning apps quietly skip it, because re-reading feels like progress and retrieval feels like failure. If Expert Mode genuinely tests application rather than recall, that is the part I would lead with. It is also the hardest thing to fake.

The other line I noticed: "we won't guilt trip you like a certain green bird." I made the same call in a habit app this year and it was the most contested decision in the whole build. Everyone tells you guilt drives retention, and it does, right up until someone misses a day and never comes back. So I am curious which way the Friend Streak has gone with your Discord folks. A streak you share with a real person could carry the motivation without the shame, or it could double it. Congrats on the launch.

Quite interesting; love the UI. The content of the courses, however, seems to be below the configured level. I tried with a software engineer profile on the app, but it still gave me a very high-level module. I'd like it if the content started off extremely technical, given that I put in software engineer as my profession. Either ways, all the best on this release.

super fun! does look and feel like an evolution of NotebookLM!

"It orients, it never invents" is the claim I'd want stress tested, and not against a benchmark you ran yourself. The hard case for auto capture isn't invention, it's a decision you reversed three sessions ago still getting injected as settled, and it reads exactly as confident as a correct one. So the real question is how a memory dies. If I can't see what got injected and kill it in one keystroke, I'll turn the whole thing off the first time it confidently reminds me of the wrong thing.

This is awesome, bouncing between all of my AI platforms and pulling context across them is a big pain. Congrats on the launch!

The re-explaining tax is real, I switch between Claude Code and other tools daily and re-establishing context every time is the part nobody talks about when they compare AI coding tools. Curious how you handle the per-project vs global split when two projects share similar tech but different conventions.

Does memory compile and bulge the memory file or is there a natural decay built in?

I've built a version of this by hand for a single tool — a memory file with explicit ownership rules so every fact lives in exactly one place, plus an archive for anything superseded. It works, but only because I maintain it myself. Two questions from the maintenance end:

Tidy-up. Mine stays usable because I prune it, and because there's a rule that a new fact contradicting an old one gets flagged to me rather than silently overwriting. As Atlaso's store grows across four tools and hundreds of sessions, what happens when session 40 contradicts session 4 — overwrite, version, or surface the conflict? And does anything decide a memory has gone stale, or does it accumulate forever?

Token cost. If memory is injected into every session, my bill scales with how much I've remembered. Is it selective retrieval or the whole store? Roughly what's the per-session overhead at 500 memories versus 50?

Congrats on the launch! The transparency in the comments really stood out. Seeing you openly discuss the current limitations and trade-offs was far more convincing than polished marketing. Looking forward to seeing how Atlaso evolves.

Hi Product Hunt,

I’m Zania from Driven. Excited to finally share what we’ve been building.

Driven started as a tool we wanted for ourselves.

We invest too, and our team has spent years building products around investor communities. One thing kept coming up again and again:

Investors are not short on information. They are overwhelmed by it.

There is more market data, filings, earnings content, news, X signals, portfolio data, and analysis than ever before. But most of the real work still falls on the investor: gathering context, checking what changed, comparing signals, and figuring out what actually matters.

We did not want to build another dashboard or another chatbot with market data attached.

So we built Driven: an AI investment agent that helps investors move from insight to action.

A few things that make Driven different:

• Real-time data and 260+ APIs
• Built-in and custom Skills
• Playbooks for repeatable investment processes
• Scheduled tasks and 24/7 monitoring
• Portfolio context in one workspace
• Order workflows
• Investor control at every step

What we care about most is not just better AI answers.

It is whether an agent can keep following the companies, portfolios, filings, earnings, and market signals you care about, then bring you back when something actually matters... And that is just one part of what Driven can do.

That is why Driven feels less like a search box and more like an AI investment team.

We are still early, and we’d really love to hear how the PH community thinks about AI agents in investing.

Where would you trust an AI investment agent first:
finding ideas, monitoring signals, following earnings, tracking portfolios, or preparing actions?

The 24/7 monitoring piece is what sold me. I've missed too many overnight moves manually. If it actually flags them in time, that's worth a lot.

Driven is one of the best investment assistant products I have ever built and used！

As one of the Product Makers, I’ve been exploring how AI Agents can truly transform the way investors work. Unlike traditional tools that mainly provide data and information, Agents enable a shift from passive searching to active understanding, analysis, and decision support — opening a new window for investing.

During the development of Driven, I focused on product design, interaction, and frontend experience, with the goal of making complex investment analysis simpler, more intuitive, and more efficient.

I’m also one of Driven’s most active users. Today, I use Driven to analyze my own options, ETF, and Hong Kong stock portfolios across three accounts. It has become an important part of my investment workflow, helping me better understand my positions, identify risks, and organize my investment ideas.

I hope Driven can truly help more investors improve their decision-making, reduce the complexity of investing, and ultimately create real value by helping people make better investment choices.

Hi Product Hunt!

I’m Head of Product at Driven. I’ve been investing for over 15 years, and my teammates are experienced stock investors too. We also work closely with a dedicated investment research team that brings financial expertise and practical experience to the product.

We started exploring how AI could solve real investing problems a couple of years ago. As AI models improved, we learned more about where AI could actually be useful in the investment process.

This year, we decided to build Driven, an AI agent designed specifically for investing.

We want to give individual investors access to the research tools and workflows typically used by professional investment teams. We hope Driven can help close that gap and give more people an investment team of their own.

We use Driven in our own investing. We keep testing it, seeing what works and what doesn’t, and improving it along the way.

Today, we’re excited to finally share it with more investors.

The trusted AI investment agent, from insight to action.

Congrats on the launch! As a founder, I’m not an active investor myself, but I know quite a few investors and often hear how much time they spend piecing together market data, news, filings, and portfolio context. I like that Driven is trying to turn all of that into a clear workflow, rather than just another place to ask questions. Excited to share this with a few investors I know.

the "24/7 monitoring plus order execution" combo is the part I'd want to understand before trusting it, not the research side. an agent that's wrong about a stock thesis costs you an afternoon of reading, an agent that's wrong and also has order placement wired up costs you an actual trade at 3am while you're asleep. what's the failsafe between "agent decides to act" and "order actually goes out," is there a human confirm step by default or does that only kick in above some size threshold

Your tagline says "trusted", which is a big word for investing. What is the output actually grounded in — filings and market data pulled at query time, or the model's own training?

And does it show its sources, so I can check a call before acting on it?

I gave it a try. It provided a really solid analysis of the stocks I'm following. It didn't tell me anything I didn't already know, but it did identify all the key signals. I haven't seen this level of response quality in other services. Great job, and congratulations on the launch!

”Driven is not available in your region“😭

this was much more interesting than I thought, tho my question is, how would this differ with data from Apple Store's analytics where I can already see the location of the purchases, for example??? Very nice!

@therayess - this looks amazing, are there any docs on how I would share the data to show on the dashboard, also do you have a privacy policy?

Wow Ammar! This is super cool. How do you monetize this?

There's a lot you can customize here — almost too much, honestly. I'd probably spend more time tweaking the globe than actually reading the data. But I'll say this: put it on the big screen in the office and suddenly everyone's stopping to watch where the next payment lands. That alone makes it worth it. Sometimes a dashboard's real job isn't analysis — it's making the team feel like something is happening.

the globe is a great screenshot but I'm not sure it's a great Tuesday morning tool. once you have more than a handful of customers a spinning globe stops being readable, it's just a dense cluster over the US and EU with a few lonely dots elsewhere. genuinely curious what the globe view tells you that a sorted table by country wouldn't, once you're past the demo-wow phase and actually trying to spot something useful.

a menu bar icon or a terminal notification tells me the same four states for free and doesn't need desk space or a charge cable. I get the appeal, a little face is more fun to glance at than a status bar, but "fun to glance at" is a different value prop than the practical framing on the page. would be curious whether people actually keep it running past the first week once the novelty wears off, versus it becoming another thing sitting unplugged on the desk.

That little robot is so adorable, I can already imagine children trying to steal it from someone's desk :-). Keeping usage limits in view is such a practical idea.

Okay this is genuinely cute 😄 It's basically an AI-era desk alarm clock — except instead of telling you it's 9 AM, it tells you your agent is out of credits. Super minimal, oddly charming, and I can already see this becoming a staple on every developer's desk setup. Sometimes the best product is just the one you want to look at.

looks cute, do i need to charge it or is it comes with a lead battery and what would be the rough shipping cost for this in asian countries i dont want to spend to much on shipping

Verifiable lands better, but the claim has a shelf life. Empty host_permissions is true today and Chrome auto updates extensions, so 1.4 can quietly add one and nobody checks chrome://extensions twice. If you want it to hold, publish the manifest diff with every release and link it from the store listing, that's the thing someone can keep checking without trusting you. The one I still do by hand is re-recording a whole take because a single sentence came out wrong.

This is a very thoughtful approach to privacy. Verifiable definitely lands better than "trust us" because users can confirm the limitation themselves instead of relying on another policy page.

The combination of zero host permissions. Great work, Nitin, and congratulations on the launch!

Can it record screen videos?

Mac users have Quick Time for screen recording. It's easy to use and free. Apple has an emphasis on security. What makes your product different?

I have tried multiple times to use this tool for recording purposes but have not been able to start. Would you please check the issue? I am using a Mac

.

Hi Product Hunt! I'm Malith, co-founder of ZapDigits.

Over the past year, we've spoken with dozens of marketing agencies, and one thing kept coming up: their data lives everywhere.

Google Analytics. Search Console. Meta Ads. Google Ads. Shopify. Email platforms. CRM systems.

Even with AI tools like Claude and ChatGPT, people still have to jump between dashboards, export CSVs, or manually copy data before they can ask meaningful questions.

So we built ZapDigits MCP.

It gives AI agents secure access to your marketing data from 30+ data sources through a single MCP server. Once connected, you can ask questions like:

• Which campaigns lost conversions this week?
• Why did organic traffic drop last month?
• Create a monthly client report.
• Compare performance across all my clients.
• Summarize my biggest opportunities.

We built it for agencies, marketers, and developers who want to spend less time collecting data and more time acting on it.

A few things we're especially proud of:
• Connect 30+ marketing data sources
• Works with Claude, ChatGPT, Cursor, and other MCP-compatible clients
• White-label reporting platform included
• Fast setup with a single MCP connection
• Secure, read-only access to your marketing data

We're still moving quickly, and your feedback will directly shape what we build next.

If you have any questions, feature requests, or ideas, I'll be here throughout the day. Thanks for checking out ZapDigits MCP!

Special thanks to @fmerian for hunting us and always supporting us.

Been using ZapDigits for a while now and this MCP will be really helpful. Congrats on your PH launch @malithmcrdev & @mandy_ms

Hey Product Hunt 👋

I'm Paul, and I run VIDEO AI ME, an AI video generator for ecommerce brands, SAAS and agencies.

I kept seeing the same pattern in our data: Users generate really good videos AI (native or ads)... they post daily for one motivated week, then it all stops. The videos were never the problem. Distribution was.

So instead of telling users to go buy a scheduler, I built the whole thing inside the tool:

- Connect 15 platforms once (TikTok, Instagram, YouTube, X, LinkedIn and 10 more, all official APIs)
- AI writes the caption per platform, from the actual script of your video, hashtags for TikTok, professional tone for LinkedIn
- Bulk schedule 30+ videos in one sitting, spread automatically at your best posting times (based on your location, topic, business industry)
- Performance tab shows which video wins on each platform. Repost the winner in one click, or generate a new version of it with AI!

The whole loop lives in one place: create, publish, measure, double down.

whatch the 2-minute video where I schedule a month of content in real time :)

I'll be here all day, happy to answer anything!

Ps: Next step, automate this but for ADS platform (Meta, Tiktok, Google...) ;)

Can I clone my appearance and my voice?

Thanks so much for checking out Glasp MCP Connector!

We built this to make your own highlights and notes instantly searchable inside Claude and ChatGPT, so you don’t have to leave the conversation or dig through tools to find what you’ve already saved. It’s read‑only and private to you, so your knowledge stays safe while still being super accessible.

We’d love to hear how you use it and what kind of workflows you’d like to build on top of it. Your feedback will really help us shape the next iterations. 🙌

The read-only constraint is the part I would underline. Most "connect your knowledge base to an LLM" pitches this year have quietly meant write access too, and that is the version I would not install.

The honest failure mode of every highlighter I have used, including one I built myself, is that capture is frictionless and retrieval never happens. I have 400-odd saved passages I have not once gone back to. Putting them behind a question I am already asking, in the window where I am already asking it, is the first fix I have seen that matches how the failure actually works: you do not have to remember you saved it.

One question: does it return the surrounding context or only the highlighted span? Half of what makes an old highlight useful is the paragraph I did not select. Congrats on the launch, Kei.

Hi Product Hunt,

I made space ocr because I kept not trusting OCR output.

Reading a document is the easy part now. Knowing whether the number you got back is the number actually printed on the paper is not. If you still have to open the image and check by hand, you haven't really automated anything.

There are two ways in, and both run the same pipeline.

If you don't want to write code, you upload photos into a folder and they become a sheet. Hover any cell and the photo beside it lights up on the exact spot that value was read from, zoomed in, so checking a page takes a second instead of a squint. Cells that failed the check are marked, so you know which ones to look at rather than rereading all of them. Fix a value by hand and your correction sticks. Folders, memos and search across everything you have scanned are in there too.

If you do write code, three endpoints give you structured fields, markdown, or plain text, and all of them come back with the same verification data: where each value sits on the page, whether it passed the check, and what still needs a look.

Either way the results stay somewhere you can use, so there is no database to stand up. A folder and a sheet are the storage. Photos land in the sheet as rows of the columns you asked for, and later you can ask that sheet for the rows over an amount, or from one vendor, newest first, a page at a time. That runs on the server, it does not read the images again, and it is not charged. The rows keep the coordinates and the flags they were stored with, so a filtered answer is as checkable as a single scan.

If you would rather have an agent do the filing, there is a hosted MCP server on the same account. You point an MCP client at one URL with your key and it can make the folders and sheets, upload photos into them, and ask for rows later. Deleting is the one thing it cannot do in one step. The first call removes nothing and reports what would go, so it has to come back to you before anything disappears.

The checking itself is the part I care about. The model never produces coordinates. Every value it returns is matched character by character against what the OCR engine actually saw on the page. Values that fail get flagged instead of quietly passing, and the ones it still isn't sure about are cropped out of the image and read a second time.

I measured this on my own regression corpus, 333 hand graded cells from phone photos rather than flat scans. Turning the checking stages off drops accuracy from 93.7% to 91.3%. They fixed 22 cells and broke none. A value marked unverified turns out to be wrong 6.4 times more often than average, so the flag is worth acting on.

100 pages a month are free and failed scans are never billed. Same price whichever way you use it.

What I would really like to hear: what would make you trust OCR output enough to skip the manual check? That is the part I keep getting wrong.

Yongha

Asad's recall question got the most useful answer on this page, and I don't think the 6-of-39 is a tuning problem.

If the re-read runs the same model over the same pixels, it inherits the same failure mode. A 7 that got read as a 1 because the glyph is genuinely ambiguous will get read as a 1 again — the second pass isn't independent, so it can only catch noise, not systematic misreads. That would explain a recall floor that prompt work won't move.

The cheapest independent signal on invoices isn't another model, it's arithmetic. Line items × qty should reconcile to the subtotal, and subtotal + tax to the total. When the sum doesn't close, you know at least one cell is wrong without trusting any model to tell you. I spent 19 years building banking apps and that's what we called a control total — it isn't clever, it just doesn't share the OCR's blind spot.

Do you reconcile totals already, or is the check purely model-vs-model today?

@yonghahwang The page as a row model is clean for receipts, but a lot of invoices run three or four pages, with line items continuing past the break and the total only on the last page. Does a multi page PDF land as separate rows I'd have to stitch back together, or can the document be the unit with pages underneath it?

The 333 cell corpus and the 3 in 4 flag precision are unusually honest for a launch page. The number I'd want next is the other direction, of the cells that were actually wrong, how many did the check miss. Precision tells me the flags are worth reading, recall tells me whether I can skip the unflagged rows, and skipping is the whole product. If recall is weak then a self check is worse than no check, because it's the thing that stops people looking.

the provenance-per-value thing is the part I'd actually pay for. most OCR tools give you a confident-looking number and no way to tell if it read the receipt correctly or just guessed something plausible from a smudge. what happens when the self-check disagrees with the first pass - does it flag the cell as low-confidence for a human to glance at, or silently pick whichever answer scored higher internally? for invoices specifically the failure mode that costs money is a confident wrong number, not a missing one.

interesting

Hey everyone! I'm Isabel, and I built Snipplet because my travel recommendations were a mess. Every time I came back from a trip, friends would text me "where did you eat in Barcelona?" or "what should I do in Tokyo?" and I'd either scroll through my camera roll trying to remember or dig through my Notes app for some list I saved months ago. Half the time I couldn't even find it.

And when I did find my notes, sharing them was painful. I once wrote out an entire itinerary with comments and tips for a friend. It looked organized in my Notes app but the second I pasted it into a text message it became this unreadable wall of text. Good luck finding that again in a chat thread.

Even for myself, I'd go back to a city and not remember where that amazing restaurant was. I'd have to retrace the actual street or scroll through hundreds of photos to figure it out.

So I built Snipplet. You turn your travel memories into beautiful visual cards with your photos, favorite spots, restaurant picks, and tips. Each one becomes a mini guide you can actually share with a link and anyone can browse, save, or get inspired by. It's not just for big trips either. I use it for my favorite matcha spots, vintage record stores, and activities in my own city. No more messy notes, no more lost recommendations.

I'm a solo founder and built this from scratch. I've tested it on my own trips and rebuilt things based on what actually worked (and what didn't).

Genuinely open to any feedback, criticism, feature requests, anything. A few things I'm especially curious about:
* What would you save first, a trip or your local go-to spots?
* What's the one thing that would make you come back weekly?
* Would you actually share a snipplet with a friend?

I'd love for you to try it. Use code PRODUCTHUNT for a free first month of Pro. It’s for the first 100 users, so use it quickly!

Try it free at snipplet.com on desktop or mobile. No signup required to browse. Thanks for checking it out!

Tried it out, pretty cool concept. excited for app release

Wow, thank you all for the support today! We've been reading every comment and the feedback has been incredible. If you haven't tried it yet, snipplet.com is free to use. Would love to hear what you think!

What features would you want to see next? Let us know in the comments!

local go-to spots, easily. trips are rare enough that I remember them, it's the little places around my own city that I actually lose track of and end up re-googling every time a friend visits. to answer your other question, yes I'd share one, but only if the friend could browse it without making an account first, since that's usually where I lose people when I try to send them something like this.

Keeping workflows as plain text instead of locking everything into a visual builder is probably my favorite part. Git diffs and PR reviews are hard to give up once a project grows.

the durability and git-diffable execution graph is the genuinely useful part, that's a real gap in a lot of prompt-chain tools. the new DSL is the part I'd push back on though - Temporal already has SDKs in Python and TypeScript that get you retries, replay and versioning without asking anyone to learn new indentation rules. what does the language buy you over a thin Python wrapper around the same guards and approvals, besides a smaller surface to write the interpreter for? "governable" and "diffable" don't obviously require a new syntax, they require the underlying execution model you already built.

Temporal, ClickHouse, MinIO, Langfuse and Redis to run a prompt is the part I'd push on. Nobody spins that up on a Tuesday afternoon, and a Tuesday afternoon is how these things actually get adopted. The SQL analogy cuts the other way too, SQL won because it was one language across every engine, and a DSL that runs on exactly one runtime is a lock in people can feel. I'd ship a single binary mode that fakes durability in SQLite so someone can write a workflow in ten minutes and only meet Temporal when they need it.

the "no streaks, no lives, no locked lessons" line is what got me, most language apps are optimized to keep you opening the app rather than to make you actually capable in the country. one situation I'd add to the pharmacy/city hall/garbage list already here: phone calls for redelivery when you miss a parcel. it comes up constantly, it's pure listening under pressure with no visual context, and almost every learner I know just avoids answering the phone entirely instead of dealing with it.

Congratulations on the launch!

I've been learning Japanese for 6 years so far, and I've been satisfied with how my sensei teaches, but I must admit I'm still afraid to use the language in everyday situations. And I think it's mostly because I try to speak it the European way; I'm still not good at aizuchi, so all my conversations feel a bit stiff, like taken straight from a workbook. Watching anime doesn't help at all, maybe because seiyuu are doing a great job articulating... Reading the launch desc, I feel like Yokoso may be the help I need, so I'll definitely give it a try ;)

Congratulations on the launch, James! 🎉

I'm Japanese, born and raised, and I just want to say — seeing someone move to northern Japan and build something this thoughtful for learners made me really happy. The name made me smile too (ようこそ!).

What resonated most with me is the philosophy: no streaks, no lives, no paywalls, and respecting what learners already know. That's rare. And as a native speaker, I can confirm that counters, prices, and signs are exactly the things textbooks skip but real life throws at you on day one.

For your second question — a few situations I'd love to see Yokoso teach, because they trip up every foreign friend I've helped:

• The pharmacy / clinic: describing symptoms and understanding dosage instructions (食後, 1日3回, etc.)

• City hall paperwork: 住民票, 保険, 年金 — the vocabulary is brutal but unavoidable

• Garbage separation rules: 燃えるゴミ vs 資源ゴミ, and the collection-day signage

• Train delay announcements: 遅延, 運転見合わせ, 振替輸送 — these are stressful even to half-understand

Rooting for you from Japan. 応援しています!

Hey Product Hunt 👋

Family plans have a way of living in an app nobody opens. I wanted the opposite: the calendar on the kitchen wall, and changes one text away. No app, no login, no new inbox.

So I built Domo, a purpose-built calendar agent. It has its own phone number, and everyone in the house just texts it over iMessage or SMS: "add dentist Thursday at 3", "when's the recital?" An always-on dashboard on the wall stays current with today's schedule and the week ahead. Soccer, the dentist, the thing at school on Friday: everyone's on the same page.

The part I like most: there's no framework under it. Domo is built with native Claude Code tooling and runs on the Claude subscription I already pay for. No API key, no per-token bill. It installs by handing the how-to guide to a coding agent; you do two handoffs yourself (a browser sign-in with the calendar connect, and one activation text).

Domo's awake right now — come play with it:

🖥️ Watch the wall dashboard live: twitch.tv/plucas_pivots

💬 Reply "Hey Domo" right here — ask what our family's up to this weekend, or put something on our calendar ("Hey Domo, add a pizza party Friday at 6") and watch it appear on the wall, on stream, within the minute

🐦 On X, mention @plonkus and say "Hey Domo": x.com/plonkus

🎮 Or come hang out with it on our Discord: https://discord.gg/RTvHyYy3B

And it's a pattern as much as a product. Swap the purpose and build your own, then bring it to our Discord and show us what you got.

Would love your feedback. What would your household text?

Hey Domo-- Add a 1:1 on plucas's calendar for 2pm today!

Hey Domo! What's on my calendar next week?

I love it! I just did the entire installing. I am not really a tech but learned over the weeks how to use terminal and claude to ease regular tasks. This is now a true gamechanger. Back then I tried doing calendar automations with shortcuts, notion and other agents, which always turns out to be very complicated and easy to trigger bugs or broken connectors. Took me 30 min setting this up, and while testing I already love it. Will continue testing it over the week and see how it goes! 🤩

This is awesome, briefly looked it over and wondering one thing: would this install just the same with a different harness + something like a local model? Have had building my own local calendar bot on my todo list for awhile now, and if Domo can bridge that gap I don't even need to build it anymore!

What's the difference between this and Warp?

I run Claude Code mostly inside Windsurf when I want the editor context, and in a plain terminal for quick throwaway tasks. Does session resume work across that split, or is a MOTHER session only resumable from inside MOTHER itself?

shipped my entire product from claude code so this one hits home. question from my actual week: I usually run two or three sessions in parallel and the pain is knowing which one is waiting on me. does the menu bar alert say which session needs you or is it one generic ping?

the alert only helps if i trust it, one false 'needs you' and i tune it out. how does it tell a real permission stop from claude just still thinking?

Bro this menu-bar alert thing is exactly what I needed 😂 I've lost count of how many times I've made coffee, come back, and Claude is just sitting there waiting for a yes/no like a lost puppy

$9.99 is honestly a steal for the headache this saves. One thing though — if I quit the app completely and reopen it later, does the session pick up from where it left off? Or do I need to keep it running in the background?

The menu bar alert is the whole product and it's third in your own sentence. Launching and resuming are conveniences, the thing that actually costs me is walking away while Claude works and coming back twenty minutes after it stopped on a permission prompt. I'd rewrite the tagline around that and let the launcher be a bullet underneath. At $9.99 you don't need a feature list, you need one sentence someone recognises from their own week.

Hey Product Hunt 👋
I'm Chandan, solo dev from India. I built ScrollToll because every screen-time app I tried failed the same way: it told me I'd spent 90 minutes on Instagram, I tapped "ignore," and kept going. Time didn't feel real. So I counted videos instead.
ScrollToll watches for swipes inside short-form feeds and counts each video you scroll past. When you hit your limit, an overlay covers the feed. There's no snooze button — to get more, you complete a box-breathing exercise. Most people close the app instead, which is the point.
On the accessibility permission: ScrollToll uses Android's AccessibilityService to detect feed swipes. That's the only way to do feed-level counting. It reads swipe gestures in the apps you select — no content, no keystrokes, nothing leaves your device. Everything is local. I'd rather say this upfront than have you find it in the permissions dialog.

Free tier covers one app of your choice, forever. Premium unlocks all of them.
What I'd love feedback on: is the breathing unlock the right friction, or should it be harder? I've gone back and forth on this a dozen times.

For 60 Days Free Trial, Use Promo Code: THANKYOU

Wow, the idea is interesting! I think maybe just blocking will be enough to stop procrastination and get back to work

Hey Product Hunt! 👋 Solo maker here.

Karve started from a simple annoyance: my API requests already lived as .http files (the format Visual Studio, VS Code REST Client and JetBrains use) in my microservices' repos, but the moment they spread across projects there was no workspace for them, just a runner buried in each editor. That was quite annoying and every tool to launch them was slow and heavy (especially my daily driver - VS)

So I built a workspace, a tool, and made three stubborn choices:

1. Files stay files. Karve doesn't import your requests into a proprietary database. It organizes the .http/.rest files you already have, from any repo, without moving them. They keep diffing, branching, and reviewing like code.

2. Native or nothing. It's Windows native WinUI 3 app: cold start under 2 seconds, Mica themes, x64 and ARM64. No Electron shell, no cloud. Requests never leave your PC.

3. Your agent can drive it, without seeing your secrets. Karve ships a built-in MCP server, so Claude Code (or any MCP client) can find, run, fix, and author requests in your real workspace. But it only ever sees your requests as raw text with {{placeholders}} - resolved URLs, environment values, and stored history never cross to the agent. Off by default, because I'm not sure if it's coold or not, one toggle in Settings, localhost only. 90-second demo: https://youtu.be/qxPAWP8QY50

All of it is a one-time purchase (launch sale: $29.99, reg. $39.99) with a 15-day free trial.

MCP is included, and every 1.x update comes with it.

There are bunch of free client-side converters on the site (cURL/OpenAPI/HAR → .http), no purchase needed.

Eventually I'll move them into app, don't want to bloat it now.

I trying to keep scope limited: REST only today (WebSocket and Postman/cURL import are on the 1.1 roadmap), single-user, Windows only. Again, the last thing I want is more bloat

I'd love to hear how you keep API requests today, and what would make you move them into plain files.

I'm here all day. 🚢