PH热榜 | 2026-08-01

Pretty good idea .. Up until now I only see this in gmail, but with this product it is possible to expand.

Draft mode is the right default, and it also makes precision the entire product. If I open the folder and 4 of 15 drafts are worth sending, I stop opening the folder by week two and the 4 good ones die with the rest. I'd trade recall away hard for that, something like only flag threads where I asked a direct question and got silence for five days, because two right nudges a week beats fifteen maybes.

Follow-up is the cheapest revenue sitting in most inboxes and almost nobody does it consistently. Nice to see someone build for that specific gap instead of another inbox rewrite. Draft mode as the default is a smart call too.

Love the idea. One question though—how does NudgeForMe decide which conversations actually need a follow-up versus threads that are intentionally left unanswered? Curious about the intelligence behind that

Congrats on the launch @chitreshsingh

Draft mode before any send is the right call. What I want to know: does it look for actual signals in the sent message, like a question or an explicit request, or is the detection purely time-based? A thread that ended with 'thanks, talk soon' should not queue a nudge. Those are easy to misread as cold.

So, basically this app will also ready my OTPs and Sensitive Codes in my mail box I guess?

the detection quality questions in here are all good ones, but the thing I haven't seen asked yet is what happens to the actual email content once it's scanned. reading sent conversations well enough to tell "thanks, talk soon" apart from a real dropped thread means an LLM is seeing the full text of my client emails and partnership deals, not just headers or timestamps. is that content processed and then discarded, or retained/logged anywhere on your end? for a tool sitting on my entire sent folder that's the question I'd want answered in writing before connecting a real inbox, not just "draft mode by default."

Ooh this is nice. Follow ups are a huge case for AI!

Anything the Snoooz team creates you can count on being top quality and probably the best support you've ever received for a product. I've been with them for over a year and @victoria_dash is an absolute super star. These guys are experts with anything email and at the end of the day, email is our biggest relationship management and sales system because it's personal. Wishing you guys so much success with this launch. From a very happy Snoooz user ❤️🚀

Great product. I have already implemented this into my daily use. It found 2 two deals I neglected to follow up on. Just paid for itself in the first week.

Bruh gets access to my mailbox? So I am not only giving away data to them but also their AI providers? Interesting.

Congratulations on the launch. This looks very interesting. I'm going to sign up for it soon.

The tone matching is the part everyone's asking about, but I'd worry more about false positives, threads that look unanswered but weren't actually waiting on anything, like a "sounds good, thanks" that just didn't need a reply back. How does it tell that kind of silence apart from someone actually going quiet on you?

This looks built for warm threads, and what I'd worry about is someone pointing it at a cold list. No reply is the normal case there, so it drafts a nudge for nearly everyone, out of the one mailbox they cannot afford to burn. I keep cold sending on a separate warmed domain for that reason, so does it tell a quiet thread apart from one that was never warm?

Draft mode protects you at draft time. The risk lives at send time.

The draft exists because nobody replied. Then you open the folder tomorrow and some of those people have answered, just not in that thread. On the phone, on LinkedIn, in a new thread with a different subject. Sending just checking in to someone who already replied is the one follow-up that actually costs you the relationship, and it is worse than never following up at all.

So the check I would want sits at send rather than at generation. Re-verify the condition that created the draft, right before it goes out: nothing new in this thread, no newer message from that address anywhere in the mailbox, no meeting with them since. Cheap to run, and it is the whole difference between a stale draft and an embarrassing send.

Does it re-check anything at the moment you press send, or is the decision made when the draft is written?

I have a question, right now draft mode is auto enabled with Gemini specially if you are using google workspace, so what is differentiator?

Starting in draft mode instead of auto-send is the detail that'd make me actually trust this with real leads.

Hi everyone!

This is another @DeepSeek moment.

DeepSeek-V4-Flash-0731 keeps the same architecture and size as the preview model, but the jump in agentic performance is hard to treat as a normal update.

Terminal-Bench 2.1 moved from 61.8 to 82.7. DeepSWE went from 7.3 to 54.4 (simply CRAZY). Flash now beats V4-Pro Preview on every benchmark shown in DeepSeek’s release table, while activating far fewer parameters.

The weights and inference code are already available under MIT too, so this is not just an API release.

Then there is the price. V4 Flash currently costs $0.14 per million uncached input tokens and $0.28 per million output tokens. GPT-5.6 Terra and Luna also just got much cheaper.

Top-tier intelligence is becoming extremely cheap. I mean extremely cheap.

I think we are entering a different phase of AI. When intelligence at this level is almost free and available through Codex, what will you build? How far can your imagination go when the cost of trying is no longer the main constraint??

P.S. one slightly crazy hint👀👀:

61.8 to 82.7 on Terminal-Bench and 7.3 to 54.4 on DeepSWE for a model that's "the same architecture and size" as the preview is the kind of jump that makes me want to know who ran the eval, not just what it scored. are these numbers reproduced by anyone outside DeepSeek yet, or is it still first-party only? weights being MIT and downloadable makes independent verification actually possible here, unlike a closed API release, so I'd rather wait for someone else's harness to confirm it than take the release table at face value.

The number I care about isn't $0.14 per million, it's cost per completed task. A cheaper model that needs two retries on an agent run costs more than a pricier one that lands it first, which is why 61.8 to 82.7 on Terminal-Bench is the line that actually moves my bill. Where it gets interesting is cached input pricing, since on long agent loops most of my spend is context I'm re-sending, not new tokens.

This is likely the most cost efficient model right now, works better than kimi 2.7code, and def better option than gemini flash. 5.6 Luna and Grok build 0.1 are both very good as well

I'm really enjoying using DeepSeek.

thanks to deepseek for gpt price cut.

Where does it land on long-horizon agent tasks vs. raw benchmarks? That gap is usually where the cheaper tiers fall short.

Tapping the actual option it offered instead of guessing a keystroke is the detail that sells this. Most remote-approval tools just fire an enter key and hope.

Reading the options off the live session instead of guessing a keystroke is the part that makes this usable. What I'd still get wrong on a phone is approving the right button for the wrong reason, because I can see "edit src/auth.ts" but not what's actually in the diff. If the approval card carries enough of the change to judge it, I'll clear a queue of these from a coffee shop. If it doesn't, I walk back to the Mac anyway and the buzz just moved the interruption earlier.

The approval card answers the trust question well. What I'd want to know is the network side, is the phone talking to the Mac directly over the same wifi, or through a relay so it still works once you've actually left the house and you're on cellular? That's usually where local dev tools like this quietly stop working right when you need them most.

This is exactly the kind of thing I wish I had last week when an agent burned an hour waiting on a yes/no I never saw. One thing that would make it even better: let me set a default action per agent or per project for common prompts like file edits under a certain size, so my phone only buzzes when something actually needs my judgment instead of every little change.

the way it surfaces the actual approval prompt instead of making you guess at keystrokes is a really thoughtful touch. feels like you built it around the exact moment of frustration rather than building a dashboard first.

The actual options from the live session feels like the killer detail here. Not just “agent on your phone,” but the part where you can safely unblock it without guessing what’s happening back on the Mac. Also love the Port 22 name. Super clean concept, congrats on the launch.

Hi PH,

The Go live is taking unexpectedly long from the app store review side, will surely update and relaunch as soon as it is done

Sorry for the wait and thanks for the patience, Much appreciated!

Running coding agents from your phone is a surprisingly big unlock for reviewing and approving on the go. Congrats on the launch.

Cool Project. Now that codex , claude code and other providers are building native mobile apps, where do you see the differentiation for this product.

The five state breakdown above covers the local case well. Different question: if Codex is running on a remote box over SSH rather than on the Mac you're looking at, does the menu bar have any way to see that session, or is this strictly scoped to tasks running on the same machine as the app itself?

Local-first and no upload of prompts or source is the part I'd actually check before installing a menu bar app that watches my coding sessions.

The orange-only “act now” state is a really smart detail. Most agent status tools get noisy fast, but this feels designed around attention instead of just visibility. Also love that base mode works without Accessibility permissions.

In base mode with no Accessibility permission, orange for input needed has to be inferred from session metadata, and a long tool call looks identical to a prompt sitting unanswered. If that's a timeout threshold I'd want the number documented, because a 90 second web fetch showing up as needing me is the false positive that gets the whole menu bar ignored by Thursday. Five states only pay off if orange is never wrong.

You asked which signals help without adding noise, so here's the one I'd add: the state that isn't in your list.

You have thinking, unread result, input needed, error, idle. There's nothing for I don't know. White for idle and white for the metadata went stale or Codex isn't where I expected are the same pixel to me, and they mean opposite things. Idle is fine. Lost is not. If the menu bar can't separate those, then a quiet menu bar tells me nothing, and quiet is the state it's in almost all the time.

Second one, on blue. Elapsed time is a weak proxy for stuck. A long legitimate task and a hung one look identical behind a spinner, and what separates them isn't duration, it's whether anything has changed recently. Blue that decays into thinking, but nothing has moved for six minutes would tell me more than a timer, and it costs you no extra permission since you're already watching that metadata change.

The five states are well chosen otherwise, and orange being the only one that means act now is exactly right. Noise usually comes from having two colours that both mean maybe.

Every state in your list is instantaneous, and I think that is the gap. Thinking at 30 seconds is healthy. Thinking at 30 minutes is hung. Same colour today. @asadmalik901 gets at it from the permission side, that a long tool call and a prompt sitting unanswered look identical. Duration in state separates both, and it needs no Accessibility permission, because you already know when the state last changed. So I would not add a sixth state. I would age the five you have, and let orange escalate itself once thinking crosses your own p90. That also answers your noise question, because the threshold learns itself instead of you picking one up front. The missing-state point above is right. I just think the missing thing is an axis, not a state.

Wow, it looks so real. I think I need to convince my friend to go to Japan with me :D

Very creative! Love the idea

This is such a nice idea, and the no-seam loop detail is clever. My one worry with running a live 3D scene as background for hours is the opposite of ambience - fan noise and battery drain on a laptop. Does it throttle rendering when the tab loses focus, or pause entirely? Also curious if the time of day in-scene syncs to your real clock, since a night ride home would feel pretty different from morning rush.

Very liminal vibes from this!
Takes me back to my trip to Tokyo

this is such a specific vibe. curious whether it loops or eventually stops at the end of the line.

this is such a specific vibe. tried study music streams before but this feels more immersive. curious whether the ride loops when you hit the end of the line or if it just stops.

the ambient train sound is actually what gets me. i have been using coffee shop noise generators as work background for years and this feels more specific. does it loop continuously, or do you eventually reach the end of the line?

100 points on Product Hunt! 🎉

A huge thank you to everyone who supported Yamanote 3D, shared feedback, or simply took the time to check it out. I’m excited to keep improving and expanding the experience. If you enjoyed the project and would like to support its continued development, you can optionally buy me a coffee here ☕

https://buymeacoffee.com/vergastadigital

Thank you again! 💚

Sparklines from raw script output is a genuinely useful primitive. Most widget tools stop at a static number.

Dale's monitor point above is the one I would build for, and I think it needs one thing the terminal gives you for free. A terminal shows you the error. A widget shows you the last good value, forever. If the script exits non-zero at 3am, the widget does not go blank, it just keeps displaying Tuesday and looking healthy. Glanceability is the whole product, and a glance cannot tell fresh from frozen. So the feature is not a refresh interval, it is a staleness contract. Last-updated on the face, or the widget visibly decays when the exit code is not zero. Does it do anything today when a script fails, or does the previous output just persist?

Since these scripts end up needing real credentials, an API key to pull sales numbers or check service health, that's the part I'd ask about next. Does TerminalWidget just shell out and leave secrets entirely to however the script handles them, or is there any credential storage involved once that output starts mirroring over to the iPhone widget?

API health in your own list is the case I would design around, because there the widget is not a display, it is a monitor.

Widgets get read as current. That is the entire point of putting something where you glance rather than look. But the refresh cadence is not yours, particularly on iOS, where the system decides when your timeline reloads and will throttle it when it feels like it. So a green API health number on a home screen might be four hours old, and it will still be read as "fine right now", because that is what a glance means.

Which makes the age of a value part of the value. Two things I would want as a script author. A way to declare a max age, so that past it the widget renders stale rather than the last known number. And a rule for what happens when a script exits non zero: keeping the last good value is both the best looking option and the most dangerous one, because a broken monitor and a healthy service then produce an identical widget.

On Murat's point above I would argue the other way. A widget that acts on things is a widget that can be wrong expensively. Showing me something is the right scope for this. It just has to be honest about when it last knew.

We don't have a dashboard problem anymore. We have an automation problem. The winners of the next decade won't be the apps that show information—they'll be the ones that act on it. Do you agree or disagree?

this caught my eye as a macOS user. before buying i want to know: when the widget runs a script, does it run as my normal user, or is there some kind of sandbox that could limit what it can access?

love the idea of running scripts as widgets. one question before buying: does the widget process run in a fresh shell, or does it pick up my .zshrc environment? i have a few PATH additions my scripts depend on.

The macOS widget side is what I would use first. Quick question on script execution: when the widget runs its refresh, does it get a clean shell environment or does it inherit the PATH from wherever it launched? I have scripts with custom PATH entries from .zshrc and need to know if those carry over.

the sparkline widgets are genuinely a nice touch, like having a tiny dashboard without needing to alt-tab or pull up a full terminal app. one-time purchase for a universal app is honestly refreshing these days.

Hey Product Hunt 👋

I'm Peter, the maker of Terminal Candy. I spend basically all day in the terminal, and I got tired of it being a grey box. So I built a terminal you live inside — pick a skin, draw the terminal square right onto the art, and get to work.

What makes it fun:

🎨 Skin it into anything — Game Boy, cassette deck, Pip-Boy, or your own image

🛠 Skin Builder — magic-wand the background off any picture, draw the terminal window on it, pick a palette. Loads live, no restart, no compiler.

🌈 84 built-in palettes — Dracula, Nord, Tokyo Night, Solarized, CRT greens, and more

🖥 Real CRT effects — GPU-composited scanlines, phosphor glow, curvature, grain

⌥ ⌥Space global hotkey — summon/hide from anywhere, no Accessibility prompt

🛍 Community skins — browse, install with one click, and upload your own

🔔 Agent alerts — sound + dock bounce when Claude Code or Codex actually needs input. Stop babysitting the window.

It's a real terminal underneath (full VT emulation via SwiftTerm), 100% native macOS — no Electron. macOS 14+, Apple Silicon.

Pricing is simple: 14-day free trial, then $10 once. No subscription. (Fun twist: publish 10 approved community skins and the $10 is on the house.) 🍬 PH treat: first 100 hunters — code STOPBABYSITTING gets it for $7.

I'd genuinely love your feedback — what device would you skin your terminal into? What's missing? I'm here all day reading every comment. 🙏

Everything in this thread already matches how I work. Some nights I have more than one agent running at once, not a single terminal. A dock bounce tells me something needs me. It does not tell me which window.

If three terminals are each running a different agent, does every window get its own badge, or does it collapse into one bounce no matter how many are actually waiting on you?

The Skin Builder turning any image into a working terminal, not just picking from presets, is the actual product. Everything else is a coat of paint on top of that.

Agent alerts is your last bullet and it's the only one that makes this a daily driver instead of a weekend toy. I'd move it to the top. Anyone can picture phosphor glow, but "the dock bounces when Codex is actually waiting on me" is the sentence that gets me to pay the $10, and right now it's sitting under the Game Boy skins.

Audit trails in BI tools are one of those things nobody misses until compliance asks. Timely add, congrats.

The AI query attribution is the part that would actually unblock us. When someone on our team asks why a dashboard number changed, being able to pull the exact query the AI ran at a specific time is very different from guessing from logs. One thing before I connect a production workspace: does the audit log store the full query text, or just metadata like a hash or query ID? For a SIEM export use case, having the raw SQL attached to the log entry makes the difference.

The retention-follows-your-policy part is what I'd want to dig into before rolling this out on anything financial. Compliance usually forces a retention minimum, but a log full of query text can end up holding customer PII well past when a deletion request should have wiped it. Is retention global, or can specific fields get redacted before the entry is even written?

Logging the SQL the AI ran is the necessary half. The half I'd push on is whether you store the answer it gave in words right next to the query, because a correct query with a wrong summary on top is the failure that actually reaches a decision, and nobody goes back and re-reads the SQL to catch it. Six months later the artefact a review needs is the sentence someone acted on, not just the statement that produced it.

Building the citation manager yourself instead of fighting a Word plugin is exactly the kind of problem you understand well enough to actually fix. Good call on skipping accounts too, one less thing for a student to worry about losing access to over a summer. Teaching yourself Swift and shipping this while still in school is a genuinely good first release.

Finally a writing app that doesn't force me to sign up for yet another subscription just to format a bibliography. The built-in citation manager alone sold me.

This feels very practical for students. Academic writing gets messy fast when the writing app, citation manager, PDFs, and export tool all live in different places.

I like the buy-once angle too. For students, no account and no subscription is a real feature, not just a pricing detail.

the part I'd want to poke at before trusting the citation manager: which styles ship built in, and what happens when a style guide updates? APA and MLA both revise their rules every few years, sometimes mid-degree, and that's the exact kind of maintenance burden that made the Word plugin route painful in the first place. if the formatting rules are baked into the app, a style update means waiting on you to ship a new version rather than a library auto-updating. how are you planning to keep up with that over the buy-once lifetime of the app?

Native long-form writing app for Mac and iPad is exactly the kind of thing I keep looking for. Quick question on the sync: if I start a draft on my iPad and pick it up on Mac, does it sync automatically in the background or do I need to manually open iCloud Drive each time? Nothing breaks the writing flow faster than having to chase a file before I can continue.

No accounts and no subscription is the real feature for a student tool, not the pricing footnote it usually gets treated as.

Keeping Stripe data live in Sheets kills a whole category of manual exports - genuinely useful. Congrats on shipping.

I have connected Stripe data to Sheets manually using the API before, so curious how this handles it. Does it pull via webhook events in real time or run on a polling schedule? And on the auth side: what is the minimum Stripe token scope it needs? I tend to create read-only keys for anything that touches billing data, so knowing the exact permissions up front matters.

The thing that breaks every Stripe to Sheets sync is not freshness, it is history changing underneath you. A charge from March gets disputed in August. Does the March row mutate in place, or do you append a correction? Mutating is what people expect, and it quietly rewrites a number someone already reported to a board. Appending is honest, and it breaks every SUMIF they built on the range. There is no clean answer, which is exactly why I would want to know which one you picked and where that is documented, before I point a deck at it. Raffay's IAP point is the same shape. A third of the revenue picture is worse than none unless the sheet is labelled Stripe only right on the tab.

We run subscriptions through Stripe plus Apple and Google IAP, so a Stripe-only view is maybe a third of the real revenue picture for us. Is multi-rail on the roadmap at some point, or is this staying Stripe-specific by design?

The eval gate that @galdayan pulled out of you is the right idea aimed at the wrong moment. It fires on edit, and the failure mode here is that there are no edits. An expert uploads once, the agent earns, and the pitch is explicitly that they step out of the loop. Meanwhile the expertise decays. Eighteen months on, the agent is confidently selling a 2026 answer and the expert has no signal it went stale, because the money is still arriving. Revenue is the one number that will never tell you your knowledge expired. So what re-triggers the eval when nothing has changed on the expert's side? A decay on the score, a re-cert prompt, something tied to the world moving rather than the file moving?

Congrats on the launch!

Love this direction. One thing I'm curious about—if an agent starts drifting from the original goal midway through a long workflow, how quickly can the orientation bring it back on track? Congrats on the launch

the per-message pricing makes more sense to me than subscriptions for this kind of tool. one thing i want to know: what happens to conversation state between messages? if someone asks a follow-up 3 messages later, does the agent still have that context?

How does the platform generate interest in the public for this expertise knowledge base?

Hi everyone,

I am Meghna, one of the co-founders.

Would just like to introduce some of the really cool features we have on Kopai.

1. We have a custom Gen-UI library that we made. Our chat feature or any agents use it asa part of their answer generations. Do give it a try. It is particularly helpful for data visualisation as well as game visualisation.

2. You can actively earn from your agent. So please do share with your friends and connections.

the part I'd want to understand before uploading my own knowledge base: when the agent gives someone a wrong or outdated answer under my name, who's actually on the hook for that, me or Kopai? per-hour consulting has a built-in correction loop, the client pushes back live and I clarify. per-message, the buyer just gets an answer and leaves, so a bad take can sit there generating "instant answers" indefinitely before anyone notices it's stale. is there a review or flagging loop on the expert's side, or does it rely entirely on buyers reporting bad responses after the fact?

Tl;dr: Our agents have seen every office building in the city — they know exactly what’s available, what it’s going to cost you, and how to negotiate the best possible deal. You get human support backed by the best real estate AI anywhere. And it’s all done at no cost to you. We are live in San Francisco, New York City, and Boston.

Start your search at tandemspace.com.

Why office leasing is broken

There are thousands of office spaces in a city and no single view of them. What exists instead is memory. A few hundred people, each carrying their own map of the city, built by walking it.

Those maps are good. Brokers who have been doing this for twenty years know their buildings cold, know the landlords, know what a deal should actually cost. But it's one person's map of a market that changes every week, and there is no version of this job where one person keeps up with all of it.

So the process runs on relays. You spend an hour explaining what you need. Your broker spends the week calling the brokers they know, who call the landlords they represent. Eight options come back. You like one. The calls start over. Nobody is doing anything wrong. It just takes days to move a question through a chain of people, and you are one of a dozen searches your broker is running at the same time.

The economics of this market were built for bigger deals. Commercial leasing grew up around full floors and ten year terms, and everything in it is sized for that. A 3,000 square foot suite on a two year term takes nearly the same work and returns a fraction of it, so smaller spaces get less attention across the board. It's why so many of them go unphotographed, unlisted, and unmentioned. That's most of the market by count.

The people in this industry are not the problem. This is just what a market looks like when it runs on memory:

You're seeing only what your broker chooses to show you, prioritized only as much as your deal size warrants, with no way to verify you're getting a fair shake.

How Tandem works

On paper, Tandem is an office leasing brokerage like any other. Same license, same tenant rep fee, same person meeting you in the lobby. The difference is what that person knows, and how quickly they can act on it.

Every Tandem agent works with an agentic brain behind them. It's watching the market constantly: what came available this morning, what a similar unit actually leased for last month, which landlords are flexible on term, what's about to come back online. Knowing the market stops being one person's job, so it doesn't stop when they go home, and it isn't limited to the buildings they've personally walked.

What that means for you:

• You see the whole market on day one, not on the third callback. Your agent already knows what fits before you finish describing it.

• Answers in minutes, over email, text, or phone, on anything from zoning to what the freight elevator situation is.

• You know what it should cost. Your agent can see what comparable units actually went for, so you go into a negotiation knowing where the number should land.

• You don't have to talk to anyone if you don't want to. The same market is browsable at tandemspace.com. Compare asking rents, book a tour for tomorrow, bring your agent in whenever you want them.

• A 3,000 square foot requirement gets the same attention as a 30,000 one, because your agent isn’t paid more for a bigger deal.

The result: clients who search with Tandem see 30%+ more spaces than in a traditional search, and finish about 6x faster. At no cost to you.

Start your search with Tandem

We’re proud to have placed more than 500 companies into offices, including Cursor, Upwork, Apollo GraphQL. But, we’re even prouder that our average time from first tour to signed lease is 27 days, versus an industry average of ~180.

If you’re in the market for an office now or anytime in the future, get started for free at tandemspace.com, or email team@tandem.space.

Artem's incentive point is the whole thing and I would push harder on it. AI does not fix that conflict. It hides it. A human broker's conflict is at least legible. You know who pays them and you discount what they say accordingly. An agent that has seen every building in the city arrives sounding objective, and the recommendation lands with no visible interest attached to it. Same incentive, tell removed. Which makes this a structure question rather than a model question. Are you paid by the landlord as a percentage of rent, like everyone else? If so, what stops the agent from being extremely good at finding me a building I can slightly overpay for? The version of this I would trust puts the fee on every single recommendation.

Hello Product Hunt 👋

Forms haven't changed much since 2005. The way people talk online has. People hit a wall of fields, bail halfway, and you never learn why. We didn't want to be stuck between a static form and an AI agent that might not get the exact answers we need. So we built unquestion.

Unquestion replaces your form with an AI that actually has a conversation. Instead of a static page of inputs, your visitor gets a friendly back-and-forth that adapts to every answer, and you get clean, structured data on the other side. Not a messy chat log. Actual columns you can use.

Here's what's happening under the hood:

• You build it like a flow, no code. Drag-and-drop nodes, questions, statements, and booking links. Each question takes free text, choice buttons, or both.

• The AI runs it, not a rigid decision tree. It reads each answer, decides if it's good enough, asks a smart follow-up when it isn't, and routes based on what people say ("interested → jump to demo booking"). You can set rules in plain English, like "must give full name", and it enforces them mid-conversation.

• Messy input → structured output. It extracts the exact answer per question and auto-tags responses, so you get a clean table and one-click CSV, not transcripts to comb through.

• It speaks any language and matches your brand with multiple themes and fonts to choose from.

• Share it however you want. Embed it on a page or send it as a single link.

Try it free today 🎉

Use voucher code PH250 to get 250 free credits.

We'd love your feedback! Drop a comment below or reach us anytime at hello@unquestion.ai

Love the idea behind it! Do you see unquestion as a competitor to the google forms??
Also how do you balance asking enough follow-up questions without making the conversation feel too long???

The zero-intervention part is what I'd want to know more about. I made myself approve drafts for weeks before letting my own posting run unattended, because that's where quality quietly slides. Four products publishing daily with nobody reading first, has one ever gone out that you wished you'd caught?

@rahul_verma60 - Quilly looks really promising, however I host and maintain my own website, which has some key bespoke integrations. Are you planning to offer a version that works with self-hosted solutions?

Was building something like this but you did it better and now I don’t have to! Good work!

Hey Hunters! My name is Mike 👋 and today I'm launching Pokey, a tiny menu bar app that turns your cursor into a little hand that actually taps what you click.

How it started
The idea came from a super random moment. My friend and I were recording a quick screen video for a neighbor, explaining how to use their AC. Someone joked, "This needs one of those MrTinyHand pointers!" (if you know the viral IG videos, you know). Seeing how much joy apps like @SlapMac brought people recently proved we definitely need more weird, fun apps for Mac.

So I sat down, built a rough prototype of Pokey, and posted a quick, unedited demo online. To my complete surprise, the video blew up overnight with 35k+ views and brought in 30+ organic sales 🙃 Turns out, people just want to bring a little fun and personality back into their calls.

What Pokey actually does

It gives your cursor different "moods", from a polite pointing hand to a cat paw, or even the middle finger for those meetings that definitely should have been an email.

Best of all, it adapts to your energy! Gentle clicks get a soft, subtle nudge, but intense clicking scales up the impact for when you really need to make a point.

Discount

Use code "NOMOREBORING" to grab Pokey for $2.99 (first 99 hunters only).

Life is way too short for lifeless pointers. Poke it like you mean it 😀

One question before you go

Fun way to spice up routine team syncs, or a bit too informal for your work calls? Where’s the line for you?

Haha, I downloaded it. This might actually make me sound way more convincing in meetings. Thanks! 😂

Finally, not some AI-focused product, or how to make myself 10% efficient, but a fun one!
That will definitely bring more fun to group calls, especially in the corporate environment.

@toshipepe - having all your AI Ganet usage available in one place is genius. Thanks for building this and hope today's launch goes well!

I use Claude Code, Codex, Cursor, Hermes, Grok Build and a few
more on the regular. Each one only knows about itself. I wanted
to see token usage per session, where I stood against limits, and
what each project was actually costing me.

Basically, I wanted analytics for my token usage.

Tokimeter reads the files these tools already keep on your
machine and puts it in one place.

Desktop is covered, not just the CLI. Claude Code and Codex both
write their desktop sessions locally, and Cursor's desktop agent
too. That was the piece I most wanted for myself.

You can slice it by project, day, tool or model, and
tokimeter trace walks through a single session, what it
cost and which models it used. If you bill clients, report --md
gives you a per-project file to attach to an invoice.

Token counts come from your own files so they're exact, anything
estimated is marked with ~. The README has a table showing how
well each integration is verified, including the two I haven't
confirmed against a live request yet.

The CLI is MIT and does all of the above. No account, no
telemetry, completely free, and open source. Pro is $4/mo for the things a local tool can't do,
history that survives log pruning and sync across machines.
Ignore it and nothing changes.

Try it without installing:
npx tokimeter report

Install:
npm install -g tokimeter && "$(npm prefix -g)/bin/tokimeter" setup --auto

Github:
https://github.com/toshipepe/tokimeter

Would love to hear what you think, and which tools I should add
next.

Huge congrats on the launch, Anıl! 🚀 That awkward silence during presentations and town halls is such a real problem. The fact that attendees can join instantly via QR code without downloading an app or creating an account is a massive UX win. I'm definitely going to try this at our next team all-hands. Best of luck today!

This looks great. The fact that people can join with a QR code and no app install is a huge win for live sessions. 🙌

Hi Product Hunt 👋

I’m Zoe, a PM at ViiTor—and also a K-pop and VTuber fan.

I started caring about this problem after missing too many great moments in livestreams. Fan-made subtitles often didn’t arrive until the next day, while general translation tools struggled with fast speech, slang, and the way people actually talk on live video.

That’s why we built ViiTor Translate: to help you understand those moments as they happen.

Turn on Floating Subtitles, choose your languages, and ViiTor displays translated captions directly over the video or livestream you’re watching. It works with YouTube, TikTok, Bilibili, and other supported platforms.

The part I’m most excited about is context. ViiTor uses our own speech technology to process live audio, while Fan Dictionary lets users add fandom-specific names, phrases, and community terms that ordinary translators often miss.

Our goal is to make live translation feel less literal and more aware of what people actually mean.

K-pop and VTuber streams are some of our favorite use cases, but ViiTor can also help with international videos, meetings, classes, and face-to-face conversations.

ViiTor is available on iOS and Android, supports 20+ languages, and is free to try.

For our Product Hunt launch, we’ve also introduced lower-latency subtitles and a new dual-language mode ✨ You can follow translations faster while keeping the original captions visible.

I’d really love to hear what you’d try it with first:

Which livestream, creator, or recurring translation mistake should we test next?

"Keep up with speech and context" is the interesting pairing, because those two pull against each other — context needs you to wait for the sentence to finish, and keeping up means committing before it does.

Japanese is the case where I'd expect that to break most visibly. The verb lands at the end, and negation lands after the verb, so a subtitle that commits early can say the opposite of what the speaker meant and only find out three words later. Does it revise a line it already displayed, or hold back until the clause closes? Revising is more accurate and much harder to read.

Hey @willblacklock, I'd love of this app can make me combine my letterbox and goodreads all into one! Do you support porting of any information from letterbox/goodreads? Or will all users have to start building their profiles from scratch?

Great job @willblacklock !

I’ve watched this one grow from the very first version, and it’s genuinely replaced the mess of notes on my phone.

Favourite bit is the explore feature. Being able to open a hand-picked collection like the top films of 2025 or the critics' top 25 TV shows and add straight to my watchlist means I'm actually working through the good stuff instead of scrolling for twenty minutes and giving up. Recommendations from friends and colleagues go in the second I hear them, and now books are in there too it's the only list I keep.

Really looking forward to the friend-matching side, being paired with people whose taste actually overlaps with mine is the thing I'd like most.

What's next on the roadmap?

finally something that gets into the actual math behind creator income instead of just throwing around vanity metrics. the churn calculator alone made me rethink my pricing tiers

This type of complex calculation is what we need calculator for now days.

The UI needs a bit optimization though. Best of luck!