PH热榜 | 2026-06-29

Hi Product Hunt Makers 👋 Back here again, I'm Long, CEO of Spira AI.

We spent the last month studying 40+ launches that hit top 10, and one pattern stood out: almost all of them flatlined within two weeks. The products were great, but the teams went quiet after launch and put their heads back down to build. So I built the thing I wish they'd had.

Spira AI is for Product Hunt makers who'd rather ship features than post content. We're running our own channels on it this week, and we've onboarded a few other AI startups to grow in front of the audience they were built for.

Here's how it works:

1. Drop your Product Hunt or website link. Spira AI reads it and builds your Brand DNA: your voice, design system, audience, USPs, and generate your content strategy.

2. You get a suite of agents tuned to that brand, posts already queued for your approval. One runs your brand account, one clones your founder voice, and a team of AI influencers amplifies the launch.

3. From there they run on their own. Spira AI pulls live trends, reads your performance data, and adjusts as it goes. 🚀

I built this after watching too many founder friends burn time and money figuring out what to post. (And honestly, so my GTM intern can stop doing it manually and go work on something better 😄)

One thing we're testing: when Spira AI reads your link and builds your Brand DNA, how close does it need to feel to you before you'd trust it to post on its own? Tell me where it nails your voice and where it misses! Try it for free today.

The "growth agents" framing covers a lot of ground, so I'm curious what these agents actually do when they hit the hard parts of social growth: figuring out which content to double down on versus which formats are dying on a specific account, or knowing when to engage with a thread versus when jumping in looks spammy. Most tools in this space are good at scheduling and surface-level analytics but hand the strategic judgment back to you. Does Spira make actual recommendations on those calls, or is the "agent" layer mostly automation on top of posting and engagement tasks you've already defined?

the trend detection piece is what makes or breaks this. creating content is the easy part now, knowing what to create and when is still the hard problem. most social media tools automate the posting but the content strategy is still entirely manual. if the agent can genuinely catch trends before they peak and not just react after they've already saturated, that's a real differentiator. curious how it handles brand safety though. fully autonomous posting with no human review feels risky for any brand that can't afford a bad post going live at 2am.

That line about freeing up your GTM intern hit home, that's exactly the work that quietly eats a small team's week. For me to trust the auto-posting, the Brand DNA would need to nail not just voice but judgment: knowing which trends are on-brand to jump on and which would feel try-hard. Getting the voice right is table stakes now. Getting the taste right is the hard part.

Curious how much of that "taste" the agents pick up from performance data over time vs. what you set up front (Have tried this a bit here and there and ended up with random output).

Congrats! @justin2025 @llma @djdmkim94 @zun_wang2

Good catch – anything with the Product Hunt theme has a chance to be featured, and here you have a clearly defined audience. Good move :) Wishing good luck with the launch! :)

Congratulations! building brand DNA automatically sounds like a huge time saver for early stage teams. I am curious what happens when a startup pivots or changes its messaging? can Spira AI quickly relearn the new positioning without starting over?

congratulations on the launch!
i really like the idea of turning a product hunt or website link into a complete brand DNA instead of starting from scratch. how do you make sure the AI captures subtle brand personality and does not end up sounding too generic across different startups?

Hello and congrats to your launch! I wanted to ask is there some "promo code" for PH? I really want to use it in action or test it in action right now - But! I dont pay for something i cant try first. But web and UIUX looks really great made! I will be launching my e-commerce project tomorrow so wanted to try yours rn <3

Wish you all the best!

Vojta

Another brand marketing agent tool. I want to know what is the difference between you and other tools?

Congrats @llma . I am Leo, founder of Flaq AI. Spira AI is an awesome product for product hunt makers.

Been waiting for something that keeps the momentum going after launch. Upvoted!!

This is a really interesting direction, especially for Product Hunt makers.

A lot of founders put huge effort into preparing the launch itself — the product page, assets, comments, outreach, and day-one traffic — but the momentum after launch is usually where things start to fall apart. You get one big spike, then the next day everyone goes back to building, and the social distribution layer becomes inconsistent again.

That’s why I like the positioning of Spira here. It’s not just “generate some posts for your launch.” The idea of creating a brand agent, founder/persona clones, and AI influencers that can keep planning, creating, and publishing across TikTok, Instagram, X, and LinkedIn feels much closer to the actual problem: staying visible after the launch window.

For small teams and solo founders, this could be especially valuable because social media is rarely just one task. It’s strategy, content angles, platform adaptation, timing, consistency, and iteration. Most makers know they should be posting more, but they don’t always have the time or mental energy to turn one launch into weeks of content.

What I’d be most curious to see is how well Spira captures the real voice of a founder or brand. If the agents can avoid generic AI content and actually preserve product context, personality, and platform-native tone, this could become a very useful growth layer for early-stage products.

Overall, I think this is a smart product for a real founder pain point. Product Hunt launches are not just about launch day anymore — they’re about turning that attention into long-term momentum.

I have been searching for this app for months finally someone built it!

• My problem was never launch day, it was week two when I stopped showing up. If Spira fills that silence it earns its spot. Upvoted and watching.

I like that you can actually edit the Brand DNA instead of being stuck with whatever the model guessed. That little bit of control is what would make me trust it.

My problem was never launch day, it was week two when I stopped showing up. If Spira fills that silence it earns its spot. Upvoted and watching.

I burned out being my own marketing team and just went quiet, which killed a product I actually liked. Wish this existed a year ago. Wishing you a strong launch.

The brand profile it generated was detailed enough that I could see exactly what it would and would not get right. That kind of visibility is underrated. Nice build.

Watching it extract my tone of voice and get it close was the moment I stopped being skeptical. Most tools flatten everyone into the same corporate voice. This kept mine.

Congratulations! The Social Studio idea where drafts are already waiting instead of a blank box is the right call. Empty editors are exactly where my motivation dies. Smart UX choice.

Cloning the founding team rather than a generic mascot is a real differentiator. People follow people. Curious how much training material it needs to sound like me.

I wanted to love this but it didn't get my brand voice or look at all.

Congrats on the launch team, had to leave comment on the landing page. Loved the easter egg feature with the hidden hover bio cards. Super dope!

Do we have to provide our own warmed up TikTok accounts or does Spira provide handle account creation, warming up, etc? Looks great!

This is exactly what I've been looking for! I never have enough time to keep up with my social media

the "in your voice" part is the whole challenge — autonomous social agents usually flatten into generic engagement spam. how are you keeping the persona clone actually sounding like the founder, not like every other AI?

Congratulations on the launch, @llma !

As a small startup founder, I am always searching for scalable solutions to automate our branding machine without agency level cost. And Spira AI is exactly what we are searching for. We have been using the service for building our founder's personal brand on Youtube and the result is very promising.

Curious moving forward, what level of content edition do you plan to support for power users? Do you plan to support some level of AI-empowered video editing, similar to CapCut, for power users to fully control the content, in addition to the end to end automated video generation pipeline?

Congrats on the launch! How do you plan to handle feedback in these early days?

Hey PH community, Romeo here from Receiptor AI 👋

Last time we launched, you made us Product of the Day. That still gives us chills. Thank you.

Here's the problem we've been obsessed with since: your receipts and invoices don't live in one place anymore. They're in your inbox, your other inbox, WhatsApp, the glovebox. Every one of them is money — a deduction, a record you'll need if you're ever audited. And catching them all is still a manual, dreaded, end-of-quarter scramble.

This year, we asked one question: what would it take for you to actually trust an AI agent to run that workflow the way you would? Not just collect documents and dump them somewhere, but manage them. Catch its own mistakes. Learn your habits. Ask when it's unsure. Work without needing you there.

Today, we're back with the answer: Agent Mode

⚙️ What's new in Agent Mode

• 🧠 Memory — remembers your preferences, vendors, and past decisions

• 🔁 Pattern recognition — learns how you work and writes its own rules

• 🙋 Asks when unsure — when something's ambiguous, it asks once instead of guessing, and never asks twice

• 🩹 Self-healing extraction — every extraction is math-validated, catching and correcting its own errors

• 💬 Ask from anywhere — query your expenses in the app, on WhatsApp, or right inside Claude via MCP

💚 Why people stick with Receiptor

• ⏳ Save hours — no inbox digging, no manual entry

• 💰 Capture more deductions — nothing slips through

• 🧾 Always audit-ready — documents clean, sourced, and in the right place

• 👻 Works invisibly — set it up once and forget it's there

We built this for SMBs who got burned by "good enough" AI — so we want your honest feedback: ask us anything, tell us what's missing, and if it earns it, show us some love.

🎁 Try it → 14-day free trial, all features. Use PH2026 for 30% off any plan for a year.
👉 https://receiptor.ai

Huge thanks to our hunter @rohanrecommends, and to everyone in this community who's been with us since day one.

The part that stood out to me is being able to ask about your expenses right inside Claude. Most tools make you log into a dashboard to see your data. Letting you just ask from wherever you already work feels like a real shift. Curious if people use it for quick mid-month checks ("how much did I spend on X?") or mostly at tax time. Congrats on the launch, @romeobellon @luigi_receiptorai @gvantsa_garmelia @rohanrecommends !

Every founder has some version of the Sunday night “sort out receipts before the accountant” ritual - if this actually kills that, it’s already a win.
The interesting part is what happens after: once everything’s clean, can you actually see where money goes across vendors, categories, time?
Also curious how this handles real-life messiness - like one person running two companies with overlapping cards and receipts.
Congrats on the launch!

Auto-posting to Xero/QBO is the bold part. The edge cases that bit us when we built similar classifiers were refunds, partial payments, and split transactions, where the model is confident and wrong and someone only catches it at reconciliation weeks later. Do you bias toward precision and route the ambiguous ones to a review queue rather than chase full automation from day one? The reversal cost on a bad post tends to dwarf the time it saved.

Do you expect the interaction between the agent and human to feel humanlike or purely transactional? Congrats on the launch!

Bookkeeping that runs itself is the dream for solo

founders. The amount of time spent manually

categorising receipts and reconciling accounts is

embarrassing when you think about it. If Receiptor

genuinely handles this autonomously that's a

significant time unlock. Following closely 👍

The "only asks when it needs more context" line is what decides whether this is genuinely hands-off or just a smarter inbox — what is the confidence threshold where it auto-categorizes and posts to Xero/QBO vs flagging for me to confirm? And since I can query it from inside Claude and ChatGPT, is that an MCP server you expose or a hosted bridge — does the receipt data live in your cloud as the source of truth, or write straight into my accounting software?

Congrats on the launch! I've been struggling with keeping track of receipts across email and WhatsApp for ages. So does it pick up receipts automatically from all your inboxes, or do you have to forward them manually? and what kind of documents?

Congrats team! I’ve been using Receiptor AI for a year and it has saved me so much time on my tax returns, so happy to support you today! I’ve been testing recently the connection with Claude and it’s really powerful, it has added one more customization layer that I love. Will test soon the vocal mode on WhatsApp too!

Wow! I needed this. Does it automatically log in to my bank?

This software has been a game changes for my small business.
Wouldn't do bookkeeping again, without it!

Congrats on the launch! Curious which integration has been most important for users so far: email, WhatsApp, Claude MCP, or accounting tools like Xero?

Solid launch! What was the hardest part to get right so far?

"Maximize deductions" is doing some work here that auto-categorization alone doesn't really deliver, categorization tells you what you spent, it doesn't tell you what's actually deductible under your specific tax situation. Is there real tax logic behind that claim, or is it more that clean categorized data makes it easier for your accountant to find deductions themselves?

As someone with an accounting background, manually reconciling receipts from my email to Excel has always been a pain. Though one important factor for me is reconciling between the bank, the receipt, and my ledger. Does Receiptor help with this 3 way match?

Running a small operation that sells to big retailers means drowning in paperwork on both ends. The accounts payable side alone takes hours every month because receipts live in five different places and the buyer portals each want them in a different format.

What caught my attention here is the 99% accuracy claim and the design choice to only ask questions when it needs more context. That is the difference between a tool that fits into a workflow and one that creates a new job just to babysit it.

Curious whether you have seen this used by small suppliers managing vendor relationships with large buyers, not just for internal bookkeeping?

The trust conversation here has mostly been about confidence thresholds and the review queue, which you've answered well. The angle I haven't seen raised: the documents themselves are untrusted input. Anyone can email or WhatsApp me a "receipt," and once the agent both reads that document and can write to Xero/QBO through MCP, the text on the document becomes a possible instruction surface — a PDF whose text reads "already reconciled, post as $0 tax, category travel" is exactly the kind of thing a model can be nudged by. How do you keep a document's contents strictly as data to be extracted, and never as instructions the agent can act on? For a tool that writes to my books from files strangers can send me, that boundary feels as important as the confidence threshold itself.

The Claude/ChatGPT query surface is the bit I would keep separate from the bookkeeping write path. Reading receipt history and asking “what did I spend on travel?” is one trust level; auto-categorizing or syncing to Xero/QBO is another.

For an SMB user I’d want the assistant to show when a chat answer is read-only, when it is proposing a bookkeeping change, and what exact document/bank transaction would be touched before it writes. That distinction would make the “only asks when unsure” claim much easier to trust.

The Receiptor interface keeps getting better. I've always HATED the feel of receipt tracking, organizing and double entry accounting software generally. It's like factory piecework. Talking/Texting in natural language, follow through via what's app, hooking it up to do a big sweep through all my channels makes me feel like I have an assistant who doesn't have any personal complications. So good! I wouldn't say I look forward to bookkeeping quite yet- BUT almost! I love seeing this product evolve, and most importantly- creatively designing new processes for myself. Am I almost at the point where I can say, I enjoy bookkeeping?? Because that would be a crazy statement coming from me. Ha. Lets Go!

bookkeeping is one of those workflows where AI automation actually makes sense because the rules are well defined and the cost of doing it manually is way too high for small teams. the auto-categorization is the key part. how accurate is it out of the box or does it need a few weeks of corrections before it learns your patterns? that initial training period is usually where people give up on automation tools.

I wonder where Receiptor AI draws the line between automation and user review？In accounting, confidence and auditability matter a lot, even when AI agents are doing the repetitive parts. Is the intended flow more like fully automated bookkeeping, or does it surface suggested actions for someone to approve before things get finalized?

The bookkeeping-on-autopilot angle is easy to understand from the tagline. For teams looking at Receiptor AI from the Accounting or Productivity side, where does the human review usually happen? Is the product meant to fully automate receipt handling, or more to prepare the bookkeeping work so someone can approve it faster?

Cool Romeo! It's sounds super interesting. Wish you all the best on this impressive launch!

The "self-healing, math-validated extraction" detail is the part that stands out to me — most receipt tools just OCR and hope, so having the agent catch its own arithmetic errors is a smart trust signal for something running unattended. When it does correct itself or reclassify, does that correction become part of the audit trail you can show an auditor, or does the document just quietly end up in its final state?

the "asks once, never asks twice" design is the right call - most agentic tools interrupt constantly and the interruptions kill user trust fast. curious about the pattern recognition piece: how many transactions does it take before it's confident enough to categorize correctly on its own? and what happens when a categorization error from 6 months ago surfaces at tax time - does the agent know it was wrong, or does the user eat it?

Really useful too for entrepreneurs, where bookkeeping requires alot of discipline.

agent mode for bookkeeping is the right unlock and also where "agentic" actually has to mean something. for chat copilots the worst case is a bad sentence. for autonomous bookkeeping the worst case is a misclassified deduction that an auditor catches three years later.

real question is what does agent mode do when it's not sure. does it pause for a human, queue the ambiguous one for review, or guess and flag? that decision rule is the whole product. good luck on the launch.

Congrats on the launch guys!

Does it support multiple languages and currencies?

👋 Hey Product Hunt,

I’m Saoud, founder of Cline.

What we’re launching today🚀

Cline is built to be the best agent harness for open-weights coding models. Today, we’re launching ClinePass: a $9.99/month subscription that gives developers low-cost access to top open coding models across Cline’s CLI and IDE Extension.

Try ClinePass for $1.99/month at https://cline.bot/product-hunt (discount available only for the next 15 days)

We recommend you to use ClinePass together with Cline CLI: install via npm i -g cline on your terminal!

Why ClinePass❓

Open models are getting seriously good for coding. They’re becoming more capable, more flexible, and often much more cost-effective for real development workflows. But using them well is still harder than it should be.

The challenge is twofold: open models are spread across providers, accounts, and API keys, and even after setup, developers still have to figure out which models are actually good for agentic coding, which ones hold up across long-running tasks, and which harness brings out their strengths in real development workflows. Heard about GLM 5.2 but not sure where to try it? Trying to keep up with Kimi, DeepSeek, MiniMax, Mimo, and every new coding model release? Want to know which ones actually perform well inside an agent, not just on a benchmark?

That’s why we built ClinePass: ClinePass gives you simple access to leading open-weights coding models, directly inside Cline’s agentic development workflow.

What you get with ClinePass 

🔷 Access to best-in-class open-weights coding models
🔷 Models curated, tested, and benchmarked for agentic coding
🔷 $9.99/month for reliable access with 2-5x API rate limits (to be clear: we are trying our best to explore what we can afford to offer here, and some of those pricing/limits are subject to change and we will keep it transparent with the community)
🔷 Use it across Cline’s IDE Extension and CLI
🔷 Models included: GLM 5.2, Kimi K2.7-Code, Kimi K2.6, Deepseek-v4-pro, Deepseek-v4-flash, Minimax-M3, Mimo-v2.5, MiMo-V2.5-Pro, (more to come!)
🔷 No lock-in: keep using any providers and models with Cline

Where we’re headed ✨

We see this as another step toward a more open, flexible, and developer-controlled future for AI coding. Developers should be able to use the models they want, inside the workflow they already trust, without being locked into one provider or one closed system.

We’re excited to share ClinePass with the Product Hunt community today and would love your feedback!

the $9.99 flat unlock for cline plus open weights is the right counter to per-token paranoia. once you stop watching the meter the agent can actually think for ten more seconds instead of being told to be brief. that changes what people let an agent attempt.

curious how rate-limiting works once a model has a slow day. soft-degrade to a smaller model in the pool, or just queue? for autonomous workflows the worst behavior is silent failure to complete.

The flat price is the headline, but the thing I'd actually stress-test is long-horizon behavior. Open weights like GLM or DeepSeek can match frontier on a single completion and still drift over a 40-tool-call autonomous run, re-reading the same file or losing the original task. We hit exactly that building agent loops, the failure was never one bad call, it was accumulation across many. Does ClinePass let you mix models inside one session, say a stronger one for planning and a cheaper one to execute, or are you on one model per run?

I love cline! used it for a year now! Is this new subscription you can eat tokens? or is it pay as you go? Its not fully clear from the description.

For me personally I think that the consolidation angle is the real win. There is so much friction in juggling separate DeepSeek, GLM, and Kimi keys and billing pagesthat hits at the worst possible moment. One sub that just works is a clean pitch. One question: with all these under one harness, does ClinePass route to the best model per task, or do I pick per request? Curious especially whether you can split roles in a single run since that's usually where long-horizon agent runs hold up or fall apart.

the consolidation play here is underrated - managing separate DeepSeek, GLM, and Kimi API keys, rate limits, and billing pages is friction that burns time at exactly the wrong moment (middle of a coding session). $9.99 to have one thing that just works is a clean value prop. curious whether there's any intelligent routing under the hood - like if DeepSeek hits capacity does it silently fall back to GLM, or does the user explicitly choose which model runs?

The "no juggling provider accounts / API keys / billing pages" angle is the real pain point here. Two questions: with GLM, Kimi and DeepSeek under one sub, does ClinePass auto-route to the best model per task, or do I pick per request? And are the 2–5x rate limits relative to hitting the providers directly, or to Cline's default tier?

An agent that can work across the editor terminal and browser feels much closer to how developers actually work day to day.

Congrats for your launch! The SDK, IDE extension, and CLI options caught my eye. Are these meant to be equivalent surfaces for the same autonomous coding agent, or do you expect different use cases for each one?

The data path is what I'd want pinned down before routing my agent stack through this — when ClinePass proxies my prompts and codebase context to GLM/Kimi/DeepSeek, is it pass-through with zero retention, or do you log requests for benchmarking and abuse handling? And are you hosting these open weights on your own infra or reselling third-party inference, since that decides where my code actually lands and what the latency floor looks like.

Open-weights in Cline is interesting for the same reason local tools keep coming back: cost control and privacy both matter once agents become daily infrastructure. The practical test is whether teams can swap models without breaking their workflow.

the SDK option is what separates this from most coding agents. being able to embed an autonomous coding agent into your own tools and workflows instead of only using it through an IDE is a much bigger unlock. most teams don't just need an agent that writes code, they need one that fits into their existing CI pipeline and review process. how does cline handle the review step? does it wait for approval before committing or can you set it to auto-commit on low risk changes?

The open-weights angle here is underrated - when you're working on anything where you can't send your code to an external API, being able to run a capable model locally through Cline changes everything. Curious how the performance compares to the hosted models for actual coding tasks - do the open-weights models hold up on complex refactors or is there still a noticeable gap?

The curated-and-benchmarked-for-agentic-coding angle is what stands out to me here, more than the flat price. New open models drop constantly, so how fast does a model typically go from release to being added to the pool once it clears your benchmarks? Wondering how a GLM 5.2 or a fresh Kimi release makes it in.

running open-weights in a coding agent is the part most people still sleep on. which ones actually hold up on real agentic tasks, not just benchmarks?

the bring-your-own open-weights angle is the right bet — not being locked to one provider's pricing or privacy terms is underrated. how's GLM/Kimi holding up vs frontier models on the harder agentic tasks?

Great to see this live! Which use case are you seeing the most demand for?

The actual pain point this solves is real, juggling API keys and billing across five different open-weight providers just to try models in Cline is annoying enough that I'd pay $9.99 just to skip that step, separate from whether the rate limits matter.

Hi everyone 👋

I'm Francesco, founder of VisibAI. I spent years selling SaaS across Europe and watched search behaviour move from Google to AI assistants.

VisibAI tells you whether ChatGPT, Perplexity, Claude and other AI assistants recommend your business when someone asks for one. Then it shows you exactly how to improve.

The problem: for years everyone optimised for Google. Now people open ChatGPT or Perplexity and just ask for a recommendation, and most businesses have no idea whether they show up in those answers.

The solution: a score on its own does not help, so we give you the full benchmark. VisibAI runs automated queries across six AI platforms (ChatGPT, Perplexity, Claude, Gemini, Mistral, You.com) and shows you:

• a visibility score from 0 to 100

• how often you are mentioned or cited

• which competitors appear instead of you

Then it tells you how to fix it: a prioritised fix list, ready-to-use fix files (robots.txt, schema, FAQ), and an AI action plan.

The benefit: you stop guessing. You see where you stand against competitors in AI answers, and you get concrete steps to climb. Start with a free audit, run a one-off report with no subscription, or go monthly for ongoing tracking and competitor monitoring.

Who it is for:

• Brands that want to be the name AI recommends in their category

• Agencies, who can white-label the whole platform under their own subdomain and brand, and run it for every client

You can try it here: https://getvisibai.com

Built in the EU and GDPR-native, which matters to a lot of the teams we work with.

Happy to answer any questions 🙌

checking visibility across six AI engines is smart, the answers diverge way more than people expect. is the fix list stuff you ship to your site, or mostly content nudges?

The hard part with a visibility score like this is LLM nondeterminism — ask ChatGPT the same recommendation query twice and you can get different brands back. Do you sample each query multiple times and average into the 0-100, or is it a single-shot snapshot? And are the six platforms hit through official APIs or logged-in scraping, since that changes whether the result matches what a real signed-in user actually sees.

The interesting part isn't the score, it's knowing what to do next and whether it actually worked. How do you decide which queries to test so they reflect real buyer behavior? And once a team implements the fixes, what's the feedback loop? AI visibility doesn't have a Search Console equivalent, so I'm curious how you help teams know they're actually improving. Congrats on the launch!

Nice work shipping this! What made you decide to build this now?

AEO is still very opaque for many people, so this tool seems like it could be quite helpful.

It honestly hadn't hit me how much people now ask an assistant before they ever open a search bar. Thanks a lot for this product Francesco!

This looks like a really fantastic product. I just filled out the detailed form to run an audi for my site but it immediately failed. See attached screenshot. I checked the console logs to see if there was any technical feedback but I don't see any. I still want to try your product because it sounds fantastic. Let me know if there is something I should do differently or if there is a way for me to share technical feedback like checking console logs.

Really like the framing here. The thing I keep running into with AI visibility is that "not showing up" almost always traces back to plain old ranking and authority. From what I've seen the answer engines mostly pull from pages already sitting in the top 20 for a query, so a page down at position 40 can be perfectly structured and still never get cited.

Does VisibAI separate those two cases? As in "you're invisible because your content isn't quotable" vs "you're invisible because you're not ranking high enough to be in the pool yet." The fix is completely different depending on which one it is, and that's the part I'd personally find most useful.

Sounds interesting but my free test audit failed.

GEO is going to be as important as SEO was 10 years ago and most businesses haven't even started thinking about it. the multi-platform scoring across chatgpt, claude, perplexity etc is smart because your visibility can vary wildly between them. one model might recommend you and another might not even know you exist. curious how fast the fix recommendations actually move the needle. with traditional SEO you're waiting weeks for changes to reflect. how quickly do AI models pick up on changes you make to your site or content?

Smart idea. Which AI platform tends to show the widest visibility gaps for most businesses?

The 'are you in AI answers' question is one I've been thinking about a lot lately - SEO taught us to optimize for search engines, and now there's this whole new discovery layer in ChatGPT, Perplexity, Claude that most tools don't even measure. What sources does this check - just the big three, or does it also cover the AI integrations in search like Bing and Google AI overviews?

The competitor-citation angle is what makes this more than a vanity score for me — seeing who AI names instead of you points straight at the content gap. Since the six engines pull from different sources with different recency windows, do you surface which specific source got cited (a Reddit thread, a listicle, a competitor's page) so the fix list can target that, or is it focused on the on-site schema/llms.txt side?

The visibility gap in AI-generated answers is real and most brands have no idea they're invisible. Excited to see tooling for this, does VisibAI track citation sources across different LLMs or just ChatGPT/Perplexity?

the six-platform sweep is the right call - visibility on ChatGPT vs Perplexity vs Claude can look completely different because each pulls from different sources with different recency windows. a score of 60 on one and 20 on another tells you something specific and actionable. the part I'm most curious about: when you show "how to fix it" - is that primarily schema/llms.txt/content changes, or are you also surfacing the competitor citations that are showing up instead of you? knowing who's displacing you in AI answers is probably the most valuable signal for figuring out what content you're actually missing.

I like that this starts with a one-time audit instead of asking teams to commit to another monthly SEO tool. My main trust question is reproducibility: does the report show the exact prompts, platform, timestamp, and raw answers behind the score so a team can verify what changed after applying fixes?

Hey,
Congrats for the launch.

Quick feedback on my first test so far:

• 30% of my traffic is coming from GEO/AEO

• We've done quite extensive work on that and continue
  

But from what your app tell us: score 49/100

And everything is 0% , not passed, etc.
I don't believe that nothing can be found about us and we get 49/100 score. How is this related?

Like I literally didn't learn anything from it and will not be willing to go further or even paid for that yet.

Hope that's helping you guys improve!

The core problem is real. Every new context window means re-explaining architecture decisions, naming conventions, the reason you made that weird choice in the auth layer three months ago. Curious whether PMB is basically a structured prompt file that lives in the repo, or whether there's something more dynamic happening, like the context getting selectively injected based on what part of the codebase the agent is touching. Also wondering how you handle drift, because the project memory that was accurate at week two is often wrong or incomplete by month six, and a stale context file might be worse than no context file.

Every new Claude Code session I spend the first few minutes re-explaining the same architecture decisions. The local-first + MCP approach is the right call, once project memory lives in a third-party cloud it becomes a security conversation for any serious team. The part I'd want most is the impact tracking that shows which memories actually influenced agent suggestions. Context without attribution is just noise. Look forward to hearing from you.

Persistent memory for coding agents is one of those features where “what not to remember” matters as much as what to store.

The part I’d be most curious to see is a small memory diff after each session: new lesson added, old assumption updated, and which memory actually influenced a suggestion.

That would make it easier to trust local memory instead of treating it like a hidden second prompt. Also helps catch stale project decisions before an agent keeps repeating them.

How do you decide what gets written into memory, like is it automatic from chats or only explicit saves?

But remembering also has a cost. Your front loading context. aka using a lot of the available context before solving a problem. That means your runway to solve it is smaller. How do you get around that? some times you need all the context runway you have 😅

This is exactly the problem that makes AI coding feel like a conversation reset every 5 minutes. You paste context, it forgets, you paste again. The local-first memory angle is smart - keeping it in the project rather than some cloud sync feels like the right call for sensitive codebases. Does it handle monorepos where different agents might need different context scopes?

We run agents across multiple client projects simultaneously, so stale memory leaking into the wrong context is a real operational risk. The keyed fact system handling latest-wins with old value archived covers simple attribute updates, but I'm curious how it handles decisions that don't have a clean key (for exmaple, an architectural direction that got reversed mid-project without an explicit "we switched from X to Y"). Does the conflict surface in the dashboard, or does the old decision just keep scoring well on BM25 until someone manually archives it?

Local-first append-only is the right base. The thing that actually bit us running a file-based memory like this for our own agents was staleness: the agent confidently acted on a decision that had been reversed two sessions back, because the old event was still sitting on the read path. Append-only sharpens that, since both the decision and its reversal live as events. Does the read path collapse to current state, or can the agent pull a superseded decision and treat it as live?

Curious how PMB handles context boundaries in practice — is it meant to store high-level project docs, repo-specific conventions, current tasks, or some mix of those?

The re-explaining loop between sessions is exactly what drives me nuts with Cursor and Claude Code, so local SQLite memory over MCP feels like the right call. How are you deciding what gets stored automatically vs what I have to tell it to remember? congrats on shipping.

memory is the half of agent workflows nobody talks about. every demo shows the magical "it just knew" moment but never how it knew. once you ship multiple agents working in the same codebase, memory is the only thing keeping their fights from becoming bugs.

the part still missing across the space is provenance. memory says "these are the decisions." but who made each one, when, based on what context? without that you eventually get the agent equivalent of "why is this code here? git blame says steve from 2019."

local plus sqlite is the right call. ship it.

Why I built this: every new Claude Code / Cursor session I'd burn the first 10 minutes re-explaining the same things - which decisions we'd already made, which directions we'd tried and ruled out, why the architecture looks the way it does. The agent forgets all of it between sessions, after every model upgrade, every time I switch tools. Teaching it the same lesson for the fifth time ("we use pnpm, never npm") was the most demoralizing part of working with an AI agent.

So I made the memory live locally - one SQLite file over MCP - and fed it back automatically before the model thinks, instead of hoping it remembers to look. Now it shows up already knowing. No cloud, no API keys, and I can open the dashboard and see exactly what it remembers. That's the whole pitch.

The re-explaining problem is genuinely painful. Every new Claude Code session starts with several minutes catching it up on decisions already made, directions already tried, and why the architecture looks the way it does. Having MCP-backed persistent context that retains all of that locally - no cloud, no API overhead on reads - is the right shape for this problem. The SQLite approach is smart for anything touching sensitive project details. Curious: how does it handle conflicts if two different agents write to memory simultaneously on the same project?

this solves a real problem. the context re-explanation tax is probably the single biggest friction point in AI-assisted coding right now — you lose 10-15 minutes at the start of every session just getting the agent back up to speed on decisions you already made.

curious about the memory graph structure. how does it handle conflicting decisions? like if you stored "never use ORM" as a lesson but then later decided to add one for a specific service?

So if I have a session in Claude, it will have the memory to store the chat from claude and when I switch to chatgpt or others llms it wil pick up the left over work from claude?

the "no cloud, on your disk" call is the whole thing for me — local-first genuinely changes what people will put in their memory. building healthos on the same constraint. how's retrieval holding up as the graph grows into thousands of entities?

Congrats on shipping! What is next on the roadmap after launch day?

Love the local-first SQLite approach here; keeping project memory on-disk instead of a hosted service is a smart trust boundary, because sensitive architecture decisions and lessons learned never leave your machine.

Everything staying right here on my own machine is the part that lands for me, Oleksii. Repeating myself over and over has quietly been one of my least favorite parts of the day, so this feels like a real relief.

Keeping agent memory in one local SQLite file is a clean approach. Re-explaining project decisions across Claude Code, Cursor, and Codex gets old fast, so shared MCP memory could make coding sessions feel much less repetitive.

the "no LLM call on the read" part is a nice detail. most memory solutions make an API call every time the agent needs context, which adds latency and cost to every single interaction. storing it in local SQLite and letting the agent pull what it needs without a round trip makes way more sense for coding workflows where speed matters. does it handle memory conflicts though? like when two sessions produce contradicting decisions about the same part of the codebase.

PMB’s local SQLite + no read-path LLM claim is the part I’d test first. The hard bit with project memory is not storing more facts; it’s deciding when an old fact should lose authority.

Do you track expiry or conflict per memory item? For example, if a repo switches from REST to GraphQL, I’d want the old REST decision preserved as history but not injected into a fresh coding-agent context unless the current file still touches that path. The dashboard would be more useful if it shows not just “this memory helped”, but “this memory was skipped because it conflicted with newer evidence.”

@oleksiijko - Very nice. Definitely beats managing multiple .md files locally with skill integration. Will review in more detail, but looks very promising.

Hi PH 👋

I read on a Mac, a Windows laptop, and my phone, and no app could follow me across all three. Apple Books is stuck on Apple, Kindle on Amazon, and Calibre is a heavy desktop install. I just wanted to open my own PDFs and EPUBs anywhere and keep my highlights with me.

So I made ReadHere. It runs entirely in your browser: read PDF and EPUB, highlight and take notes in both, keep a per-book journal, all of it offline. No account, no upload, your books stay on your device.

It's free. To be honest about where it's at: everything is local to your browser right now. Syncing to your own Google Drive (yours, not mine) is what I'm building next, and there's a waitlist if that's your thing.

Would love your honest feedback, especially on how it actually feels to read in. Thanks for taking a look 🙏

This is such a clean and thoughtful product. ❤️ I love that everything stays on my device, works offline, and doesn't require creating yet another account. Definitely something I'd actually use.

One small feature I'd absolutely love: automatically remember where I left off in every PDF/EPUB. It would be amazing if I could close a book and, the next time I open it, it jumps right back to the exact page/position I was reading. That tiny detail would make the experience feel perfect. Great work !!

no account, no ecosystem is a stance not just a feature. the entire reading software space drifted into "we own your library now" over the last decade. kindle apple books kobo all want lock-in. choosing to not own anything of the user's by default is rare.

the per-book journal is sneakily the strongest feature. that's where re-readers live. quick question for v2: when google drive sync ships, do journal entries sync separately from highlights or together? annotations leaking across devices is one thing. half-formed thoughts on chapter 3 is another vibe entirely.

Great idea & just tested - super convenient to use! is there any way to ever make this work without wifi? ex. if i want to read the book on a flight w/o wifi

"No install, no account, no ecosystem" is doing a lot of heavy lifting in the best possible way. Most PDF tools bury you in upsells before you even open a file. The per-book journal is the feature I didn't know I needed - does it store in localStorage or IndexedDB? Curious how it survives a browser data clear.

This is awesome, congrats on the launch! Any plans to add a Calibre style sync to move books to other devices like Kindle or Xteink?

Hey Product Hunt! 👋 I'm Silvio, co-founder and CEO of Sami.

I run an ad agency managing north of $40M a year in client ad spend, and for years I watched our team spend half their time on work that should never have needed a human in the first place.

The manual grind looked like this:

• Catching budget pacing issues before clients noticed

• Pausing underperforming ads

• Protecting against overspend

• Flagging campaigns with zero impressions

• Monitoring CPLs across dozens of accounts

All of it important. All of it manual. And all of it pulling our team away from what they're actually great at: building client relationships, studying the market, and running better experiments.

We tried to stitch it together:

• Native platform rules that only work on one channel

• Zapier workflows that break

• Shared spreadsheets that are outdated the moment you close them

• AI tools that explode your token costs the moment you try to feed them campaign data at scale

Nothing gave us a single place to see everything, act on it, and automate it across every platform.

So we built that.

With Sami, you connect your ad accounts, build campaign portfolios, and get real-time pacing across every client and channel in one dashboard. Then you deploy automations (we call them SAMs) that monitor your KPIs and either notify you the moment something's off, or take action automatically based on rules you define. Your thresholds. Your metrics. Your level of control.

We validated Sami with our own money inside our own agency before asking anyone else to pay for it. One early user, Trustpilot, achieved a 12x ROI and saved 300+ hours a year in manual optimizations. Across our early users, teams are saving an average of 3 to 5 hours a week from the automations alone.

This is just the beginning. We're building toward a future where paid media runs on autopilot: fully automated bidding and budget management, intelligent campaign optimizations, and eventually the ability to build and launch campaigns end to end inside Sami. The goal is an automated future where performance marketers spend their time on strategy and creativity, not operations.

Try it, break it, and tell us what's missing. Every piece of feedback goes straight into what we build next.

🎁 Anyone signing up today gets a 14-day free trial. No credit card required. Just connect your accounts and visualize your ad budgets across channels.

Get started free at https://sami.bot/ or reach me directly at silvio@sami.bot

Huge thanks to my co-founder and CTO @yassin_gofti for building this with me. None of this exists without him. And to everyone here on Product Hunt, thank you for taking the time to check out what we're building. It means more than you know.

Much love,

Silvio 🫶

This daily budget pacing snapshot from Sami on Slack really makes the life of marketers running paid ads a lot easier 🙌🏼

One of my favorite views in Sami is the Budget & KPIs section.

You can visualize pacing across Google, LinkedIn, Meta, and YouTube in one place 🔥

This helps save a ton of time when managing dozens of client accounts or large accounts internally.

Chiming in as someone who gets daily Slack reports from Sami (we're a client of AdConversion), it's super helpful. I always know where we are on budget and what ads we need to turn off as well. These are things I used to manually do every couple of weeks, but now I'm on top of it, hands free.

Congrats on the launch, Silvio and team! 🎉

I really like the idea of having one place to keep an eye on performance and set rules. Excited to see how Sami evolves. Big congrats again!

This is literally what I've wanted for the last decade+.

This feels like one of those tools where the upside is huge, but the edge cases really matter.
A good rule saves hours, but a poorly set one could do the opposite pretty quickly - especially on client accounts.
How are you thinking about guardrails here? Is there a way to preview or stage actions before they go live?

Congrats on the launch!

The cross-platform pacing is the hard part most tools get wrong - budgets shift fine but when you have overlapping audiences across Google and LinkedIn hitting the same person, the coordination layer gets messy. How does Sami handle attribution conflicts when the same conversion gets claimed by two platforms simultaneously?

this seems fascinating! i'm quite the newbie so apologies if this is a stupid question, but does this run on mcp servers for each platform? i've seen some videos about similar applications but curious if you've built your platform on a similar foundation.

Congrants on the launch! I like the idea a lot! how is the data privacy managed when it comes to credit card details and so on? thank you and big congrats

Congrats on the launch! What would you say is the minimum spend/complexity for which this makes sense?

Congrats on the launch! 🎉
The notch has bugged me since I got my M1 too. Smart move turning the most useless part of the screen into the useful one. One thing I'm curious about; What happens when I'm docked to an external monitor? I'm working remote and haven't got access to an external monitor to test it out yet.

Honestly the notch has annoyed me for years, so turning it into something useful is a fun idea. The clipboard history and file drop are the two I'd actually use daily. The thing that stands out though is you fixing the battery drain and post-sleep freezing, that's the reason I gave up on a similar app before. Nice work, and the one-time price over a subscription is refreshing. @zack40x @Crest

niceee

Putting something actually useful in the notch is one of those ideas that sounds obvious once you see it - surprised it took this long. The translation feature is an unexpected pairing with system stats. How does it handle technical text or domain-specific terms? That's usually where inline translation breaks down.

Curious how Ad Reframe handles the jump from typical UGC formats into TV-ready creative. Since the tagline is “Convert UGC ads for TV,” are you mainly focused on aspect ratio, pacing, and framing, or does it also help with things like voiceover, captions, safe zones, and length requirements for TV spots? That distinction would be helpful for marketers comparing it with a normal video editor.

This is awesome, kinda feels like a bridge between a TikTok and a YouTube video 💯

CTV ad spend is growing fast and pillarboxing is genuinely embarrassing to watch on a 65" screen - so this solves a real pain. The key question for me is how the AI handles the expanded areas when there's motion near the edge of the original frame. Static backgrounds are easy; someone walking across frame is where things get tricky. Any samples with fast-moving footage?

The "reshoot it for CTV" conversation is one nobody wants to have mid-campaign. Really glad someone built a clean fix for this. Simple as it should be 👍

it's nice to see this product reach todays rank#9. We run UGC campaigns and the 9:16 to 16:9 problem comes up constantly whenever we want to extend to CTV or YouTube pre-rolls. The pillarbox blur workaround genuinely looks bad. The real test for us would be fast motion and shaky handheld footage — that's where outpainting usually falls apart. Curious if there's a confidence score on the output so you know which clips need a manual check before the campaign goes live.

reframing 9:16 to 16:9 without pillarboxing is the part everyone fakes with blur. do the generated edges stay temporally stable across frames, or can they shimmer?

Hey Product Hunt 👋
I'm Samson, co-founder and CPO at Intelli. We built this for a reality most customer engagement tools don't build for because they don't understand nor relate to: in emerging markets, WhatsApp is how people buy, ask, and pay. So we built an A customer engagement platform that's WhatsApp-native first, then multichannel across Instagram, Messenger, web chat, and email.
What makes it different:

– AI assistants trained on your own data, not a generic bot

– Local payment rails built in: M-Pesa, MoMo, Paystack, Flutterwave

– Broadcast campaigns, lead qualification, and support in one inbox

– Live in about 15 minutes
We're an official Meta Tech Provider serving 200+ businesses. We launched on Peerlist on monday this week and it inspired us to come and launch on product hunt too.
Would love your honest thoughts and questions. What does your current customer engagement tool get wrong? I'll be here all day.

Congrats on shipping. With AI conversations the hard part is rarely the happy path, it's the handoff: knowing when the AI should stop trying and pull in a human before the customer gets frustrated. How does Intelli decide that escalation moment? That line is what makes or breaks trust for the business owner.

Customer support AI is crowded but the multi-channel part is the hard problem - most tools work fine on one channel and fall apart when you mix WhatsApp, email, and live chat in the same conversation thread. What does the handoff look like when the AI can't resolve something? That moment is where most platforms lose customer trust fast.

Good approach. Is WhatsApp showing higher AI resolution rates than web chat in your early deployments?

Curious how Intelli handles intent signals beyond the words, do you capture any behavioural or emotional cues from conversations? That layer tends to be where leads actually convert or drop off.

Very interesting! will check this out for sure

Running community launches, the WhatsApp-native plus local-rails combo (M-Pesa, Paystack, Flutterwave) is the part most support tools ignore — buyers in emerging markets transact in-thread, not on a separate checkout page. One concrete edge case: when a customer pays via M-Pesa inside a WhatsApp conversation, does the AI assistant receive the payment confirmation as context so it can move the conversation forward or trigger fulfilment, or is payment a siloed step the agent never sees?

The native Mac client space for FTP is genuinely underserved. Most people end up on Cyberduck or Transmit and just stay there out of habit rather than because those tools are great. What I'm curious about is where "fast" is coming from specifically. Is this a smarter connection-handling layer, local caching of directory listings, or something at the rendering level? And does it handle edge cases like servers with slow PASV negotiation or broken directory listing formats, because that's where the legacy clients tend to silently fail and leave you guessing.

building on SwiftNIO instead of wrapping libcurl is the right long-term call - you get real async I/O without fighting legacy C semantics. the dual-pane layout is the correct UX choice too; every sysadmin who's used Midnight Commander or WinSCP has that muscle memory. two things I'd check before switching from Transmit: macOS keychain integration for SSH key passphrases (the workflow killer is re-entering keys on every session), and whether it handles S3/Backblaze remotes or stays pure FTP/SFTP. the FTP-only constraint is fine, but people will ask.

Native-and-lightweight is exactly why I would switch off the Electron FTP clients — the wrapper bloat is real. On the security side, where do credentials live: does it use the macOS Keychain and read SSH keys/agent from ~/.ssh, or keep its own store? And can I import existing connection profiles from something like Transmit or FileZilla, or is it all manual entry on day one?

Congrats on the launch Danny! It'll be interesting to test with our own agent. Everyone who has built an agent want to make its use as smooth as possible. Great idea!

Danny, this is a genuinely interesting direction. The thought of an assistant that can quietly handle the boring everyday errands on its own, without someone hovering over it, is the part I keep coming back to.

Great product!! At what point will the agent need human intervention, or will that not be needed at all despite the magnitude of the problem it faces?

Genuine question for the makers — when your agent needs to call a paid API mid-workflow (Serper, a data vendor, anything per-call), how are you billing it today? Shared team card? Hardcoded key? I've never found a clean answer that doesn't involve giving the agent unlimited spend.

Hey @dannyheng,

Can an AI agent onboard itself — get its own login, credentials, and access — with zero human in the loop? How does it work?

Congrats on the launch! What’s been the most surprising use case people have brought up so far?

The "no inbox, no credential store" framing is exactly what I had this conversation about yesterday on the Weavz launch. Funny that two products solving overlapping pieces of the same problem launched a day apart, feels like a category that's about to consolidate hard.

Honest answer to your question, the wall I keep hitting isn't an agent capability, it's an agent state problem. After a long run, the agent has implicit context (what it learned, what it ruled out, why it picked X over Y) that doesn't survive a restart. Mail and vault don't help there because the loss is upstream of any I/O. Curious if you've thought about that or if it's outside Mailgent's scope.

Congratulations on the launch, @nilay1101 ! 🎉

Excited to see this finally go live. Wishing you and the team a fantastic launch day and all the best for what's ahead. Looking forward to seeing it grow. 🚀

How would you monetize it?

Trust is so important with outreach now. Great extension.

I have had a Lovebox from the very beginning and I love it. I was looking for something portable to use while my husband an I travel. It's perfect for that. We also used it around the house as a special hidden surprise for love notes to each other.

The detail that got me is the "no battery, nothing to check" part. You removed every reason to anxiously refresh, so the note just exists when you happen to glance at it. I work on voice AI that keeps adult kids in touch with their aging parents, and the same thing keeps proving true: presence beats notifications, and people want to feel close without another feed to manage. On the broadcast-to-a-group idea, my one worry is that one-to-one is exactly what makes a KEEP feel like a person and not a channel. Have you tested whether a shared KEEP still feels personal, or does the magic dilute once it goes many-to-one?

Do you think we'll get a windows app soon?

Congrats on the launch! 🚀

This is a very practical idea. I can see it being useful for sales calls, client presentations, interviews, and even product demos where you need talking points but don’t want to keep switching windows.

Curious how Oculta handles different screen-sharing apps technically. Does it stay hidden across all sharing modes, like full screen, window share, and screen recording?

Hi everybody! I'm Florencia, part of the team of Oculta.

Over the past few months we've spent an unhealthy amount of time asking questions like "Does this button make sense?" and "Can we make this one pixel better?" 😅

It's surreal to finally share Oculta with all of you.

I'd love to know: what's one situation where you wish you could have private notes without anyone else seeing them? Meetings? Interviews? Sales calls? Something completely different?

I'd genuinely love to hear your ideas, so please leave a comment. I'll be here all day reading every one and taking notes for future updates!

I don't use TikTok, but my brain is definitely rotting because of IG reels, seriously need to try this. Also love the green accent colors, looks very relaxing! Good luck with the launch!

I’ve been using NerdSip for a while and really like it

It’s a nice way to learn something useful in small moments without committing to a full course or long video.

Congrats on the launch!

How do you keep the lessons from getting shallow or hallucinated when someone picks a niche topic?

This is dope!! My gf thinks I'm a genius when I'm dropping facts out and about lol. Should probably tell her... anyway! Everyone should give this app a go!

This looks amazing! I just spent the past few weeks traveling and did wonder so many times what a certain building was. Or what famous person used to roam specific streets. This will make getting the answer much easier!