PH热榜 | 2026-06-18

Hey Product Hunt! 👋 Louis here, co-founder of Upstream.

When I was at Algolia, I was drowning in 200 emails a day. I was never sure where to focus, and getting through my inbox took forever. I knew it didn’t have to be this way. Instead of creating a new messaging protocol, I bet on rethinking the tool everyone uses: email.

We built the first email client where agents do real work alongside people. We designed Upstream to be easy to pick up and fun to use. 

📥 Take the noise out of email
Agents automatically triage your inbox, separate signal from noise, and surface only what deserves your attention

✍️ Every reply starts with a draft
Upstream drafts replies for every conversation that needs one. It also changes your voice based on who you’re talking to and the context of the convo

⏰ Never miss a follow-up
Agents track open loops and remind you when needed, making sure important conversations don't fall through the cracks

🔎 Ask anything about your work
AI can instantly find information across your inbox and connected tools: receipts, meeting notes, introductions, decisions, conversations, and more

🔗 Bring all your context into your inbox
Agents use information from tools like Notion, Calendar, Drive, Granola, and other sources to produce better answers and drafts

👉 As a special thank you to the Product Hunt community, you get one free month of Upstream Pro if you sign up at upstream.do and use the code PH26! 👈

We are all so excited to hear what you think! Please do reply here and let me know 

Huge thanks to our hunter @garrytan for hunting us 🙏

Impressive how far you’ve taken the product, well done on the launch @louislecat , @jontiret and team!

The voice-matching per audience is what caught my eye — does it learn over time from how you edit the drafts, or is it mostly set from the initial analysis?

Where do you guys draw the autonomy line on triage vs sending? Those two might fail completely differently IMO. Anyways, great work!

The human + agent inbox positioning is interesting. Email has a lot of hidden context, so I like that you are not framing this as just another summarizer. How do you decide which actions should stay manual versus agent-assisted?

The UI looks absolutely beautiful and lightweight. Since it's built to clear out backend grunt work, does Upstream integrate directly with standard business tools like CRMs or project management platforms to log customer conversations automatically, or is it strictly focused on the email client layer?

This looks awesome for a busy founder email. Congrats on the launch!

the "agents do the grunt work" angle is interesting because most AI email tools still make you manage the AI. curious how the triage actually works in practice, does the agent learn your priorities over time or is it more rule-based out of the box?

HI @louislecat my question is, most tools that connect to slack and notion hit token limits or lag heavily when syncing large workspaces. how often does upstream index the connected apps for fresh context? such a great launch

I really enjoyed the UX for AI reply and integration with Granola. Congratulations on the launch

Great idea! How do you keep the agent activity from burying the messages I actually need to read myself?

@louislecat Congrats on the launch! The agent alongside you model is the right call, the inbox tools that have tried to fully automate away from the user have always hit a trust wall. The real design challenge is making the handoff between human and agent feel invisible. Curious how you're thinking about the moments where users want to override the agent without breaking the flow, that's where these products usually get complicated. Adding to my stack to test.

Great product!

I've been using it for a few weeks and I find myself in my inbox more than usual!
Great auto-complete and I've been using the label features a lot!

Late to the thread but loving what I'm reading — especially Jonathan and Louis confirming nothing goes out without approval. That's the right instinct.

The follow-up I'd love your take on: how do you keep that approval step from becoming its own inbox grind? At high volume, approving every single draft can quietly recreate the exact load you're trying to remove. The teams I've watched do human-in-the-loop well tend to tier it — once the agent has earned trust, it auto-handles the truly low-stakes stuff ("got it, thanks") on its own, and reserves explicit approval for anything that actually commits you to something.

Curious whether Upstream leans that way, or keeps a hard approve-everything line on principle? Either way — congrats on #1. 👏

Upstream changed the way I looked at mails from something annoyingly and chaotic, to structured and organized
@louislecat congrats on the launch, and great to see where you and the teamm took the product in the last month!

Use it myself, the AI assistant is simply better than Superhuman!

The human + agent inbox model is interesting - most tools treat agents as a bolt-on to an existing workflow, and it creates friction. Building the inbox around the assumption that agents will be first-class participants from day one is a different design choice. What's the biggest behavioural change you've seen in users when they start treating the AI as a teammate rather than a tool?

Congrats !! Looking forward to trying it out!!

Contratz

This is awesome! Congrats on the launch!

Congrats on the launch!

My new favourite email client. Insane how many times I’ve simply hit ‘send’ on a draft!

Pretty awesome inbox management. So much saved time!

Wonderful product !
Once you try it, you keep it forever !

Best team, best product!

My new favourite email client. Insane how many times I’ve simply hit ‘send’ on an auto reply!

Can't wait to try it!!!

Hey Product Hunt 👋 I'm Scott, part of the founding team at Honestly.

Why we built it:

Bots and AI agents now generate more internet traffic than humans.

Your customers talk about your products every day across social media. But as AI-generated content, sponsored posts, and fake reviews become more common, finding authentic customer feedback is getting harder by the minute.

Today teams either spend hours scrolling social feeds themselves or rely on expensive tools that surface more noise than insight - but this felt wrong to us.

What Honestly is:

Honestly finds & verifies all customer conversations about your products, so you get honest insights built on data you can trust.

Simply enter a product by its name on our platform, and Honestly will help you:

• Build better products by understanding exactly what customers love and dislike

• Improve marketing by uncovering trends and high-performing content

• Identify competitor weaknesses and market opportunities

• Generate custom reports tailored to your business needs

What we believe:

As AI-generated content continues to flood the internet, authentic customer conversations will only become more valuable.

The companies that win will be the ones that can find real consumer data online, understand it quickly, and convert it into informed product and marketing decisions.

That's what we're building with Honestly.

Product Hunt offer (first 500 signups only!):

As a thank you to the Product Hunt community, we're offering a special deal to the first 500 signups that use code HonestlyPH26 when signing up for early access on our website:

🎉 7-day free trial
🎉 25% off for LIFE

We'd love to hear about:

• How you currently gather customer feedback from social media

• What’s most frustrating about finding real customer opinions today

• How AI-generated content is impacting your customer research process

Thanks for checking out Honestly 🙏

Excited to hunt Honestly today.

@Honestly helps brands, retailers, and growth teams understand what people are really saying about their products across Reddit, TikTok, X, YouTube, & Instagram.

Instead of relying on broken 5-star ratings, synthetic surveys, or dashboards full of vanity metrics, Honestly turns real internet conversations into product, creator, affiliate, and competitor intelligence. Teams can see which products are getting organic traction, what customers actually mention, which creators are already driving demand, and where competitors are winning attention. No more manual scrolling through social media or filtering noise using decades-old tools that promise perfect data but can’t deliver.

What stands out here:

• Find real product opinions from Reddit, YouTube, X, TikTok, Instagram, and other social platforms

• See customer language, sentiment, attributes, and recurring themes by product

• Instantly detect creators, affiliate mentions, sponsorship gaps, and partnership opportunities

•Get detailed reports about questions you have on popular product features, important customer preferences, popular social trends, and more

If you work on product, growth, ecommerce, affiliate, or competitive intelligence, Honestly is worth checking out. It helps teams replace guesswork with evidence from the real internet.

As a founder, I care a lot about what people actually say when they are not filling out a feedback form or answering a survey. Reddit threads, TikToks, comments, and random posts often contain much more honest product feedback than the “official” channels.

The hard part is separating real signals from noise, especially now that so much of the internet is either AI-generated, sponsored, or just recycled takes. So I like the idea of focusing on verified conversations and turning them into something a team can actually use.

Curious how Honestly decides what counts as authentic or trustworthy feedback. Do you show the original source and context behind each insight, or mostly the summarized findings?

Also, how well does it work for early-stage products that don’t have a huge amount of mentions yet?

The gap between what consumers say in a formal research setting and what they actually think is one of the most underrated problems in brand strategy. Reddit and TikTok are genuinely where unfiltered consumer truth lives - this is an interesting angle on that. Curious how you handle conflicting signals when Reddit and TikTok audiences have very different takes on the same product?

Interesting idea. How do you determine whether a comment is actually authentic versus just engagement bait or bot activity?

the bot/fake review problem on social is so bad right now that "verified authentic" is doing a lot of heavy lifting here. curious how you handle edge cases where someone posts about your product without naming it directly (like describing a feature or a bug). does Honestly still pick those up?

Hey @scott_davidson_jr ,qq like a fake reviews are an absolute nightmare on marketplaces right now. does this track conversations happening inside closed loops like reddit threads and discord, or is it strictly pulling from public apis like twitter/x? great to see launch

This is the stuff founders are too scared to look at and most need to. Reddit especially will say to your face what no focus group ever would. Does it show you threads as they happen, or is it more of a periodic thing?

Congratulations on the launch!
Being able to pull data from so many sources that are scraping protected is very impressive!

Later would you guys also consider getting information from Facebook, maybe Facebook groups?

It's really an interesting idea. There is so much customer feedback spread across different platforms now, but actually finding what is real and useful is becoming harder, I saw the other day about an influencer talking about the dead internet theory and how it's full of bots and AI generated stuff. I can see how this could help teams save a lot of time. Curious, how do you decide which conversations are trustworthy enough to include in the insights?

The Reddit + TikTok combo is the right pick — those two are where "what people actually think" lives, vs the polished stuff on X/LinkedIn. Curious how it handles small/new products with low mention volume — does it surface a single thread or does it need a baseline?

This is really interesting — the AI detection angle is what stands out to me. How accurate is the verification layer when it comes to short-form content like TikTok comments where tone is really hard to read even for humans?

I think this is incredibly useful for customers and product builders. AI detection is incredibly hard. How are you able to filter out AI reviews from real customer opinions?

Can't help but hear the beats of a New Order track; now my whole brain is dancing. 😂

Anyway, like the idea because it's kind of a way to find collaborators and create some affiliates for them.

Great work! Honestly seems like it is primarily geared towards larger companies since a demo is required to move forward with the platform. However, we are a small team. Do you have any self-service options for startups or small teams? The competitive analysis features seems compelling our this category of clients.

How are insights looking? Does it also come with recommended actions? Seems very interesting tho

Congrats for your launch

Surveys give you sanitized feedback. Reddit threads give you unfiltered reality. The gap between those two is usually where the real product work is. Congrats on the launch!

Great launch video - congradulations!! Was clear & engaging - not an easy task to pull off!!

In the video, you mentioned "no more noisy, outdated social listening tools" - how does Honestly specifically differentiate itself from other social listening tools out there that seem to be doing something similar?

Do you plan for predictive capabilities? For example, when launching new products.

Yeah! Love this. Reddit and TikTok not only are trendy but one of the best way to get honest feedback. Love to see you helping on this! Wish you all the best team!

Been doing this manually for months - searching Reddit threads for honest feedback on our tool is a real time sink. The verification layer to weed out AI-generated posts is what makes this interesting, because without it you'd just be surfacing more noise. One thing I'd love to know - how does coverage hold up for niche B2B products that don't have huge Reddit communities? That's usually where the signal breaks down for us.

This looks useful, especially now that it's getting harder to separate real customer feedback from AI-generated noise. How do you verify that a conversation is authentic, and what signals do you use to filter out fake engagement? Congrats on the launch!

this is another way of doing "social listening" - or entirely different?

Reddit is the hardest of these sources to pull from cleanly. It's hostile to automated access, and half of what reads like a "real opinion" there is planted marketing. How are you handling both sides: getting the data reliably without tripping Reddit's defenses, and separating the astroturfed stuff from the genuinely useful threads?

I like that Honestly looks at what people say when they are not filling out a survey. Reddit and TikTok conversations can be messy, but that mess is often where the most useful brand insights actually show up.

Interesting product and the verification angle makes a lot of sense given how much AI generated content is flooding everything right now.

Quick question for you. When you say Honestly discover conversations across Reddit, TikTok and the others, are you covering all public posts or working from a selected set of accounts and communities? Asking because for niche products the most valuable feedback often lives in smaller subreddits or accounts that are not particularly large but are very relevant. Would love to know how deep the coverage goes.

Finding real customer feedback is getting harder!! Love the focus on helping teams separate real customer conversations from all the « noise »!! Congrats on the launch 🚀

which social platforms are currently the most valuable source of insights??

@Tabstack by Mozilla keeps cooking.

Today, the team is launching not 1, not 2, not 3, but 4 new features. Introducing:

• Tabstack CLI for quick automation and scripting - view source code on GitHub

• Tabstack MCP server that gives your AI assistant direct access to Tabstack - read docs

• Tabstack agent skill for your @OpenClaw or Hermes agent

• Tabstack for Raycast - an extension for scraping data without leaving @Raycast

Go scrape something today.

The unified API abstraction on top of scraping is clever. We've hit the selector-maintenance problem building data pipelines where a single HTML change breaks weeks of work. Does it use headless browser pooling or something more lightweight for dynamic content, and how do you handle rate-limiting per domain when multiple callers share the same API key?

4 new features and its our 4th launch! How fun is that! 🚀

Try all the tools and tell us what you think. Really curious but don't have anything to build right now? Here are a few ideas of things to build:

• Spec watcher that alerts you when TC39 proposals, Node.js, or TypeScript ship breaking change—one API call per source, diff the rest yourself (repo)

• Competitor intelligence monitor that runs weekly, extracts structured data from any product homepage, and pings you when something changes (repo)

• Podcast prep agent that researches a host's last 20 episodes and returns a one-page brief before you record

• Vendor due diligence tool that pulls pricing, HN mentions, changelog velocity, and Reddit complaints into a structured brief before you sign a contract

• CFP discovery agent that scrapes Papercall, Sessionize, and conference homepages and returns open calls filtered to your topic areas

• Dependency security monitor that reads CVE databases and package changelogs for your exact stack and tells you if you're actually affected—not just "vulnerability found"

• API docs watcher that diffs the docs for any API you depend on and tells you what changed since last week

• HN front page tracker that extracts structured data daily—title, score, domain, category—and builds a dataset over time for content strategy

• Job posting intelligence tool that monitors hiring pages for companies you care about and extracts structured signals: what they're building, what stack they're moving to

Back at it! Love the idea of creating a competitor intelligence monitor that runs weekly, especially knowing it doesn't train on my data.

The underrated part is not scraping; it is giving the agent a stable contract back from the web. Agents get much more useful when the web step returns schema and citations instead of a brittle browser transcript that has to be re-interpreted every run.

Very interesting approach. Most web extraction tools eventually struggle when sites change their structure. How does Tabstack handle schema reliability over time without developers constantly updating extraction rules? Is there a point where human intervention is still required, or is the adaptation fully automated?

For launch/community ops, I would use the MCP server to turn docs, changelog pages, and competitor pages into a weekly pre-release brief. The edge case I would test first is source freshness. If a teammate asks the agent to re-check one URL after a cached extraction, can Tabstack force a fresh read for that source while still using cache for the rest?

"URL + schema in, clean JSON out" is exactly what I keep wishing for. I drive a lot of browser automation for my own agents and the thing that always bites me isn't the first run, it's the site quietly changing its DOM a week later and everything breaking silently. Does the schema-based extraction hold up when a page's layout changes, or does it need re-tuning? Mozilla-backed and "never trained on your data" is a strong trust angle too. Congrats on the launch.

The schema-first approach is interesting. Have you found that users spend more time defining the schema they want, or cleaning up the extracted data afterward? Curious where the bottleneck usually ends up.

Built at Mozilla definitely got my attention. Curious how well it handles websites that change their layout frequently.

One API call for structured JSON, markdown, and browser automation is a solid combo. Does the schema validation handle edge cases well when a site's layout changes, or does it need manual updates?

Congratulations on your launch today!

the 'no scraper to maintain' pitch hits different when you've actually spent time babysitting selectors after a site redesign. schema → JSON output that reliably matches is the right abstraction. curious what the rate limits look like at scale - the mozilla backing + no training on your data is a genuinely good differentiator

Congratulations on the launch! When you say one API call for any tool, are there really not restrictions of tools that can be used? If not, what tools have you seen be used the most?

This is pretty cool.

Congrats on the launch. The MCP + schema path is a nice fit for agents that need web context without owning brittle scrapers.

The thing I’d test is provenance across a multi-source run: exact URL, fetch time, cache/nocache state, and which fields came from which source. Do you return that beside the JSON, or mostly through citations today?

It looks very interesting! How does it compare to other AI scraper solutions currently on the market? What specific use cases did you test during development, and which ones is the agent optimized for?

The MCP server angle is the interesting bit for me. When an agent pulls structured web data this way, how do you handle pages where the schema is slightly wrong or the DOM changes mid-run?

The MCP + agent skill angle is the part that stands out to me. Scraping is usually treated as a one-off API call, but making it usable directly inside an agent workflow feels much more practical. Curious how you think about guardrails when agents browse or extract from messy pages.

Hey Product Hunt! 👋 I’m Sudipta, cofounder of Floworks.

Quick intro: I’m an IIT-Kgp grad. My first startup failed — and what stung was the reason. Not the product. We just couldn’t crack sales.

So I started Floworks and got into Y Combinator. Different company, same ghost: sales never got any easier.

We were spending hours building lead lists, only to discover half the people had changed roles, the signals were outdated, and the best accounts were hiding in plain sight.

So today we’re launching something close to my heart: @Jesse — a modern prospecting tool for GTM teams.

Think of it as a search engine, but for finding your next 100 customers.

Apollo and Clay scrape data once, then sell it back at a premium. By the time it reaches you, it’s stale — titles changed, companies pivoted, signals came and went.

Jesse searches the entire live internet on every query and hands you fresh leads.

We’re an anti-database company.

What’s more 👇

1. Apollo gives you 8–10 filters. Clay makes you wire up data sources. Jesse just works.

   Write in plain English:

   🔎 “Looking for healthcare providers who just opened chemotherapy centres”

   🔎 “Find me 100 kids’ play facilities that just started a football academy and are seeing high footfalls”
   

2. Jesse tells you which signals to look for.

   Building a fintech for debt collection? Jesse surfaces:

   🎯 VPs of Risk at mid-sized NBFCs ($5M–$50M ARR) posting about defaults / hiring underwriters — want real-time alternative credit-reporting APIs

   🎯 VPs of Compliance at digital neobanks (50–200 emp), newly appointed, replacing manual KYC/AML with a consolidated real-time API

   🎯 CTOs at Series A/B fintechs (100–500 emp) tired of juggling vendor APIs — want one unified identity + credit + AML API

3. AE with named accounts every quarter?

   🕵️ Jesse finds the right people inside those accounts by reading decision-makers’ real online activity — posts, public appearances, forums

4. A daily barrage of fresh leads, delivered to your inbox every morning. 📥

Since we launched, the adoption has been incredible:

• 🚀 1.2k+ teams signed up

• 🌍 From a real estate agent in Florida, to a fintech founder, to 4 Fortune 500 companies and one of the largest logistics companies in the world

• 📈 50%+ WAU/MAU

• 🔥 A constant rave across sales communities

These are still early days. I get a ton of messages from Jesse users on what to improve, and I personally read and act on every one.

I’d love your feedback — especially from founders and sales leaders who’ve lived the pipeline grind. What’s the most painful part of finding good leads for you? Happy to answer any questions! 🚀

Thanks, Kevin, for hunting our product. Hi everyone, I am Sarthak, one of the Cofounders at Floworks.

We built Jesse to help sales teams run precise outbound.
Jesse is an AI agent that connects to 180+ data sources and the entire internet.
It finds you the perfect list with any filter or intent signal - all from a simple prompt.

What Jesse knows about your prospects:

• Funding rounds

• Page visits

• Products they're using

• Team's preferred tech stack

• Intent signals no one else has
  

And everything else there is to know

One of the great testimonials we received from a customer:

"It's like having an insider in each prospect's office."

Please take Jesse out for a spin, and share your valuable feedback. I am sure with all your support, we will fix what is broken in outbound.

Congrats on the launch! The biggest pain point with tools like Apollo or Clay has always been how quickly the data decays. Doing live web research at query time sounds like a game-changer for timing-sensitive outreach. Looking forward to taking this for a spin! Do you have any n8n/MCP integration to be used with Clade code?

Hey Product Hunt, Ritesh here, one of the makers.

We built Agent Jesse around one uncomfortable number: B2B data decays about 2% a month, so any static lead list is partly wrong the day you export it.

So Jesse doesn't query a snapshot. It retrieves against the live web at query time, ships every row with the source URLs it came from, and runs a self correcting loop that keeps re-sourcing until it hits your target.

As of this week it also goes where you work: an API key drops it into n8n, and an MCP server connects it to Claude, ChatGPT, or Cursor, so you can just ask for leads in plain English.

We are still actively shaping what comes next, so I would love to hear: where does your current lead data let you down most?

Happy to go deep on how the retrieval or verification works.

The pitch makes sense if the core problem is list staleness, but I'm curious what "live internet" actually means in practice for sales data specifically. Are you pulling from public web sources in real time, or is there a database that gets refreshed on some cadence? That distinction matters because a lot of "live" tools still have a 30-90 day lag on job changes and funding rounds, which is exactly when the timing-sensitive outreach breaks down. Also wondering how Jesse handles signal prioritization, whether you can filter by something like "title changed in the last two weeks" versus just searching by current role.

As someone who has used Jesse first-hand, I can say they are really, really good. I’ve tried many products that promise to use intent signals to build lead lists with just a prompt, but the list quality usually turns out to be quite poor. Most of them feel like LLM wrappers on top of traditional lead databases.

Jesse felt different. They seem to understand real intent and match it with actual buying signals. We used to run email campaigns to hundreds of thousands of leads, mostly with poor reply rates. But when I used Jesse, it clearly helped us identify which leads were actually relevant - allowing us to go much deeper on personalization and get ridiculously high reply rates.

Highly recommended, and congratulations to the team on the launch.

So proud to see Jesse out in the world today. 🚀

I've watched the team pour everything into this, late nights, endless user calls, and a stubborn refusal to ship just another stale database when the whole industry said that's just how it works.

The live-internet search is genuinely wild, but what I'm proudest of is how obsessively we listen to users. Every piece of feedback gets read and acted on, and it shows in how fast this thing keeps getting better.

If you've ever burned an afternoon building a lead list only to find half of it was wrong — this one's for you. 🙌

Searching the live internet instead of pulling from static lists is a big deal for anyone doing outbound. How fresh is the data — are we talking real-time crawls or is there some caching involved?

Cant wait to try this. Have had my own share of struggles with stale databases.

I have seen my fair share of stale leads and contacts so great job for focusing on an unsolved problem. What markets do you focus on? Are there ones that are stronger or weaker? Interested in learning more. Congrats on the launch.

Does it have MCP integrations yet?

I think the live web search can be a huge delta in tools like these. It can solve for the freshness of the data always there in the tool.

I think this is a cool concept. Looking forward to testing this out. All the best for the launch, guys.

I've been running into this exact problem with Apollo so Jesse looks like a fresh gust of air.

Is there an additional verification layer?
Because many small businesses will not have their information updated on all channels, I wonder how any conflict is handled and whether there is a way to check the validity of that information that turns up through searches.

We were using Clay at Requestly.

Shifted to Jesse 3-4 weeks back. Much simpler. You just tell what kind of company you are looking for and it finds.

Happy with it so far. Congrats on launch!

Congrats on the launch Sudipta & team! Looks very cool. How do you determine that an email is "verified"? Also is there an API integration as well?

The stale-list problem is even worse down-market, Sudipta. I prospect local businesses (insurance agents, salons, clinics) and Apollo/Clay basically don't have them, or the data is years out of date. A live, plain-English search is exactly the gap there. Genuine question: how does it do on small local SMBs vs the enterprise/BFSI examples? That long tail is where every list tool falls apart. Congrats on the launch.

How do you verify the data that is pulled is verified / correct?

I did a people search but there is no contact information or LinkedIn profiles. Am I missing something? I’m on the free plan.

Congrats on the launch @sudipta_biswas4 @sarthak_shrivastava2

This is amazing - have been using it for a few days now

Interesting positioning. How do you balance 'live internet search' with getting consistent, structured lead data at scale?

Loved the concept. Curated leads and smart outreach :) Congratulations

I tried the tool, and overall I think you are addressing a real pain point, which is great.

My first point of confusion was the naming. The app is promoted as Jesse, but the website address and logo say Floworks. That may be confusing for customers and could make the brand feel less clear.

Second, I went through the process and asked for SMBs, but the results included companies with valuations around $2 billion. That made it feel like Jesse did not fully follow my target group.

Third, I tested the product on the free plan. At the final step, Jesse told me that 75 contacts were found, but I could not see any names or details. To attract more users, it might be helpful to show a sample of the contacts, or at least show names without email addresses, so users can better understand the value before upgrading.

Overall, it is a nice product with a clear use case. These small changes could make the experience more consistent and compelling.

Love the tool. Jesse has been a second GTM brain for us.

How do you verify the leads are actually newly opened and not just old pages getting reindexed?

This is interesting. Do you cache results at all, or does every search query the live web fresh?

This is such a cool product! Clay was super hard for me to use.
Is there a free trial to get started?

Who decided that AI gets to do the fun stuff while we remain the routing layer for work?

Hello Product Hunt! We’re Ellie and Vinay, the cofounders of Elvin.

We are busy people, creating Elvin for busy people

We built Elvin because we’re all overloaded. So far, AI is making it worse, not better. Every AI-generated message or newsletter becomes someone else’s noise, and that person still has to process it. And getting AI to help meant stopping to write a pile of prompts first.

Somewhere along the way, AI grabbed all the big, interesting work, and we got left as the routing layer. Sorting, chasing, following up. That felt backward.

If AI is so smart, it should figure out how it can help you, not the other way around.

So we built Elvin: a proactive personal agent that works ahead of you. It reads across your email, chat, and calendar, identifies what needs to be done, and gets started. It's as powerful as it is proactive. It does the big jobs, building documents and graphics, not just drafting emails (though it does that too).

If you’ve seen OpenClaw, you know proactive AI is real and can figure out how to do real tasks. Elvin is that, minus the hardware, the tokens, and the command line.

How Elvin works

🔍 Elvin finds tasks in your email and other tools (or you can message Elvin)

📋 Elvin creates plans for how to get the work done for you

✅ You just say yes, and off to the races! (Or tweak the plan)

⚡ Turn on Elvin’s pre-built skills for your job role or build your own

Elvin is multiplayer

🤝 Connect with your teammates

🤖 Even add your AI Agent friends like AI Software Engineer @Devin by Cognition

Elvin’s origin story

We like making powerful tools that are usable by everyone.

Back in 2018, I (Ellie) joined Slack and was immediately overwhelmed by the number of new messages and channels. I wondered when AI would be able to help me find what was most important for me to read, and could it generate a todo list based on information coming in from all of your messages and apps. In 2018, AI was not up to the task.

In 2025, Vinay and I realized that the technology was there to build a product that could not only find your most important work but also get it done for you with self-writing AI Agents.

Elvin is available on iPhone, Android, and the web. Free while it’s in beta. No waitlist. Big thanks to @Voicepanel and @CodeAnt AI for making Elvin possible!

Thanks for having a look!

✨ Ellie and Vinay

Hey Ellie! It sound amazing cause sometimes I feel I'd take most of AI if someone else is telling what to do. Becomes a bottleneck usually, so I'm sure founders gonna love this. Wish you all the best!

Congrats on the launch! Read through Ellie's answers on the approve-the-plan model — really thoughtful, and the "reading is free, approve before acting" split is the right shape.

The one edge I'd be curious about is the recurring pre-approved skills. Per-task approval is well covered here — but once a recurring skill runs on a pre-approved plan, that's where the autonomy quietly lives, and the risk isn't the plan, it's drift:

the context shifts underneath a plan that was approved weeks ago, and it keeps executing confidently against the old

assumption. Do the recurring skills re-check themselves against current context before each run, or flag when what they're

about to do diverges from the original approved intent? That's usually where "set and forget" automations end up burning

someone.

Either way, flipping AI from routing-layer to proactive is the right direction — nicely done. 👏

2018 Ellie would be proud. This is exactly what I've been waiting for—AI that finds the work for me instead of making more work. Just downloaded.

Love the idea of AI that works ahead of you rather than waiting to be asked. How does Elvin handle situations where it gets context wrong and takes the wrong next step — is there an easy way to course-correct?

Have you guys considered the possibility of two Elvins collaborating when their users are working together while keeping each user's context completely separate?

For example, if User A and User B have an important planning meeting coming up, their Elvins could coordinate data gathering and preparation on both sides, with each user authorizing what they're willing to share and what actions to take. The goal would be to automate the groundwork on both ends and surface the best possible outcome for the meeting.

One Elvin could request data or trigger an action on the other side, which would then prompt the other user for approval before anything happens.

That line about us becoming the routing layer hits way to close to home, ty guys for saving us from our own tools🫡

The proactive part is exactly what I'd want and also what I'd be most nervous about. How much does it act on its own versus check with me first?

I like the idea that Elvin looks for the coordination work first instead of waiting for a perfect prompt. Emails, meetings, docs, and task tools are exactly where the messy “someone needs to follow up on this” work usually lives.

The approval step also feels important. For anything involving teammates, clients, or important decisions, I would want the agent to prepare the work, but not silently act without me.

Curious how Elvin decides what is worth surfacing. Does it learn from what users approve, ignore, or edit over time?

The proactive angle is compelling. Most AI tools still wait for a prompt, which keeps humans as the workflow router. How does Elvin decide when to start work on its own versus when to pause and ask for confirmation? That boundary feels like the difference between an assistant and a real work agent.

Nice idea. How does Elvin decide what counts as coordination work worth surfacing versus noise it should ignore?

Hey Product Hunt, Fryd here, one of the founders.

We spent three years on a stubborn bet: build an AI that does the work, not one more assistant that drafts something and waits for you to finish it. It worked. Viktor has been doing real jobs for 30,000+ companies inside Slack, and in the last 10 weeks it crossed a $15M run rate. Today it moves to where most of the working world actually is: Microsoft Teams.

Viktor is an autonomous AI employee. You @mention it in a channel and it does the thing end to end: closes the books overnight, reconciles the payouts and flags the one that is wrong, screens applicants and books the calls, builds the board deck from six tools that do not talk to each other. It connects to 3,200+ integrations, so it works across your whole stack, not just Microsoft's.

What we care about:
- It ships finished work, not suggestions.
- It asks before anything irreversible, and pushes back when you are about to make a mistake.
- No per-seat tax. Start with $100 in credits, no card.

We will be in the comments all day. Tell us where it impresses you and where it falls short, we read every word. The question I am most curious about: what would you hand Viktor first?

Get started: viktor.com

I'm one of the founders of Viktor, working on the technical side.

Viktor runs real scheduled jobs reliably, shows its work, and stops to ask before anything irreversible. The Teams launch was a genuine lift on the permissions and admin side, and it came out IT-friendly by design. Happy to go as deep on the architecture as anyone wants.

Viktor is amazing. Was super fast to set up, and everyone in our team adopted it quickly after seeing it work for others on Slack.
Mostly love it for user analytics, and getting insights by connecting the data from all the tools we use.

me + Viktor = ❤️

been using it for weeks now, mindblowing how good it is!

@viktor.com is the best employee you can add to your team right now. We felt it directly at 500 Stories: in April–May, Viktor helped us scale revenue from one client 5x, while keeping delivery under control across 2,074h of production, 202 deliverables

The more work moved through Viktor, the more obvious it became: this is not another AI assistant, it actually helps the team ship. So happy to see Viktor coming to Microsoft Teams. Huge congrats @fwiatrowski and team.

Wasn't sure what to expect when we set this up. Within a few weeks Viktor had built us a full pipeline dashboard to replace HubSpot, integrated our call tracking across 42 numbers, and started auto-syncing leads with clean data daily. When our call center flags an issue, Viktor's already looked into it before I finish typing.

This is the kind of AI that actually does things — not just answers questions. The more my team uses it, the more "AI coworker" stops sounding like marketing copy.

Big congrats on launch day. If you're on the fence, give it a shot — you won't regret it.

Hi everyone, I do growth at Viktor.

Slightly funny thing about launching our Microsoft Teams version here: Product Hunt is the most early-adopter room on the internet, and the reason we built for Teams is to leave that room. Most of the working world is not in a Slack workspace ranking AI tools. They are in Teams doing the accounting, running ops, answering customers. Viktor is for them now too.

Here is the thing that reframed it for me. You do not prompt-engineer Viktor. You brief him.

Plain language, the way you would brief a sharp new hire. No clever syntax, no blank box daring you to be smart. You describe the outcome and he comes back with the finished thing, not a draft to clean up.

That is also why Teams matters. The people who never wanted to learn "prompting" are exactly the people who get the most out of an employee they can just talk to.

Same Viktor, same 3,000+ tools, now living where your company already works. Brief him, see what comes back.

Would love your honest reactions, especially the unflattering ones.

I've been using Viktor since late Feb, as my personal assistant.

It's scary good. You can treat it like a personal assistant. The channel-and-threads interface of Teams or Slack is a total winner. Multiple people can talk to and observe the ongoing discussion. Tons of integrations and customization.

Viktor found me a new parking spot, 40% cheaper than my previous one. Viktor is doing ongoing medical research for me. Trip planning and event planning, of course.

Friends and family have picked it up, too. It's not just for engineers.

Strongly endorse.

I don't know the Viktor team, I'm just a happy customer.

I was made unemployed one year ago. Since then, I have become a content creator helping to empower people over 50. More recently, I have created my own company using Viktor. Between us, we are in the process of creating six SaaS products, one of which is already in beta testing. Having come from no coding experience in the past, none of this would have been possible. To be absolutely blunt, fulfilling my dream of being a business owner while in the process of helping other people would never have been possible without Viktor. I can't talk highly enough of the software, of the product.

I am not one that typically responds to requests for comment or feedback on products, but when I saw the email today come through asking for me to comment here, there was no second doubt in my mind. One message to getviktor.ai is: keep up the good work.

I get pitched AI tools constantly and rarely put my name on one. I'm hunting Viktor because I've watched it do real work, not demos.

I lead growth at Wispr Flow and advise Viktor, so I've seen it up close for months. Most AI products hand you a draft and wait for you to finish it. Viktor takes the whole job and comes back done. You brief it like a sharp new hire, in plain language, and it goes and does it.

The Teams launch is the part I'd flag. Teams has 320M+ monthly users to Slack's ~80M. Both are full of people doing real work, but the bigger share of it runs on Microsoft, and that's who Viktor couldn't reach until today.

What earned my trust: it ships finished work, and it stops to ask before anything it can't undo. Go hand it the job you've been avoiding.

Congrats to Fryderyk and the team. This one earned the hunt.

Slack-Viktor is AWESOME!! He is helping me write my first book. I couldn't do it without him. I've tried other AI bots. Viktor beats them all. He does the research, the organizing, the editing, and SO much more! He's become invaluable to me. Thank you, Viktor!

Viktor is the one agent I've tried that actually works reliably. It doesn't wander off track or get stuck in an infinite loop. It does the stuff you'd think it does based on what you can connect to it, but I've even had Viktor project manage (without a connected PM tool - just letting it use Slack) and it figured out how to follow up well - just the right amount of annoying that makes a great project manager - and it messaged people to clear roadblocks. Well worth the price!

This looks useful. Does Viktor adapt its behavior based on Teams permission scopes set by enterprise admins?

I have been using Viktor to run my adjunct professor job search in New Jersey and it has been genuinely useful. I gave it one task, find every college hiring in Kinesiology and Exercise Science for Fall 2026, and it came back with 19 institutions including direct department chair emails at nine of them. That would have taken me hours to pull together manually. Solid tool.

If Teams version is as good as Slack's one then 🔥🔥🔥🔥🔥!
Extremely powerful tool, have been really suprised with the range of task it could do autonomously

The approval draft is the right boundary. I’d be curious about the receipt after the action runs: who approved, what data Viktor used, which tool it touched, and what changed downstream. Does Teams get that as a durable artifact, or is it mostly in chat history?

I'm a Viktor evangelist(lol) in my circle and co-working space. When it comes to ingraining it in my operations, the value is beyond what I expected and its just a fraction of in comparison to what some of my peers have been doing (one is actually featured in one of your ads now lol), but it's already bought me so much more time to be able to stay high level and work on other levers of the business in the last 60 days. Wispr Flow + Viktor once you've got it connected to workflow and economics is the sh*t!

The Teams angle makes sense for distribution - that's where the work already happens. But I'm skeptical about "autonomous AI employee" claims when the reality is usually a lot of human review happening in the background. The real test is error recovery - when Viktor closes the books and gets a reconciliation wrong, how does it flag that vs. silently proceeding? That failure mode is what keeps most finance teams from trusting autonomous agents with anything that touches the ledger. $15M ARR is a real signal though. Would love to understand more about what the 30k companies are actually letting it do end to end without human sign-off.

This is HUGE, congrats on the launch!! Love Viktor 🤩

The "brief him like a new hire" framing makes total sense for Teams users. For teams that are already deep in Microsoft 365, how does Viktor handle permissions — does it need admin access or can individual users connect it themselves?

What’s the approval flow like for actions that concerns money or sensitive systems ?

3200 integrations plus weeks of unattended runtime inside microsoft teams is enterprise level blast radius if even one of those connections gets misused. "asks before anything irreversible" is one sentence holding a lot of weight when the agent touches this much of the stack

There are millions of people using Microsoft Teams everyday who never got truly experience what an AI employee or colleague feels like. We're changing it. We're sharing Viktor, our best AI employee (and yours too, maybe), with them.

Hi, simple question - what is a difference between copilot and viktor?

This is so cool! I just spent 10 minutes asking your agent all my questions so feel like I don't have much to comment lol. Great work. I wonder how well it'll work in a use case where it's dropped into a mobile app and can answer questions about the product or do the onboarding. Will take this for a spin :)

We started from the question "what if the video itself could respond?" Not a chatbot sitting next to the video, but the actual video. The result is a new content primitive: your existing video library becomes a two-way experience, embeddable anywhere, with no re-recording needed.
And let's also not forget about the insight behind Agentic Videos - a viewer who pauses and asks questions tells you more about their intent and knowledge gaps than a thousand passive completions ever could.
Would love to hear from L&D teams, enablement folks, and anyone who's ever wondered why video analytics stop at "watched 80%" - this one's for you.

Congrats to the team at D-ID! This looks like an exciting launch @gil_perry

Looks great! the future of video!

This combination of bringing messages across with all the expertise of our subject matter Experts directly within the Video helped us to reduce costs and onboarding time to go faster into the market with our sales reps !

So proud to be part of this exciting product we're launching today. This will distrupt the way of how people watching videos. I believe it's a start of a new era in the videos field 🎬💫

Really excited to see the various uses of Agentic videos- this is a game changer and I'm curious to see where it brings the most value! Interactiveness that allows to dive deeper than the original video simply makes so much sense.

The insight about viewer questions revealing intent is really compelling. Can creators see a breakdown of which questions come up most, so they can improve future videos based on what people actually want to know?

Congrats on the launch! With 100FPS streaming ready rendering what kind of latency are you seeing in real time conversational use cases like customer support avatars?

How well does this hold up on longer videos?
Like if it's a 30 min training video does the avatar stay context-aware the whole way through or does it start losing the thread after a while?

Okay this is a smart pivot ngl, instead of trying to make "better" videos you just made the video answer back. What happens if someone asks something totally unrelated to the content though?

Curious about the API is it built more for developers wanting full customization or can no code teams plug it into existing workflows without heavy engineering support?

This solves something I've run into constantly with onboarding videos at work. Someone watches the whole thing, still has one specific question that wasn't covered, and either sends an email that takes days to get answered or just guesses and moves on. The idea of the video itself being able to handle that follow-up in real time is genuinely useful. Congrats on the launch!

This looks really useful — clean

execution. How does it handles

user interactions?

I’d probably make the onboarding/support video interactive first. The next question for me is the handoff after the answer: if a viewer says “change my plan” or “send this to support,” does it create a ticket/approval trail, or stay inside the video session?

This works excellent as an onboarding simulation!

Given the 100FPS rendering capability are you targeting use cases like live streaming or gaming avatars as well or is the focus primarily on business communication and marketing content?

What does onboarding look like for a brand that wants a consistent Digital Person as a recurring spokesperson across multiple campaigns rather than a one off video?

How does D-ID approach voice cloning and lip sync accuracy across different languages? Is quality consistent for non English markets too?

For teams already using video for sales enablement what's the typical turnaround time from uploading content to having a deployable Digital Person ready to embed?

Love the self service studio angle how much technical knowledge does a marketer or content creator actually need to go from a script to a finished Digital Person video?

Interesting concept but I'm skeptical about real-world adoption here. Most people don't naturally stop a video to interrogate it - the passive viewing habit is deeply ingrained. The creator insight angle is the most compelling part to me, but that only works if you get enough people to actually interact in the first place. Would love to see some retention and engagement data from early pilots before getting too excited. What are you seeing from actual users so far?

VoiceOS lets me stay focused on what I’m doing while it takes care of the small distracting tasks in the background.

Give it a try!

most ai copy generators sound incredibly generic and robotic. during the interview, can i paste in examples of my own past writing so the agent actually learns my specific voice for the ads and cold outreach? congrats for launch

Yo love the idea. Mouse and keyboard do feel like an outdated interface. And the issue with wispr flow is that is just parses what i say, but i can't really control the computer with my voice.

Is there also a way to define certain actions or workflows and activate those through voice?

Been using VoiceOS for a while now and it's honestly the best product I've tried in this space. As a solo founder building my own products, I know how much grind goes into shipping something real. Congrats on the launch Jonah and the whole team, well deserved.

The confirmation-before-execute step is the detail that stood out to me. I build voice AI for older adults aging at home, and that gap between intent and action is exactly where trust gets won or lost. Curious how granular the confirmation is: do low-risk actions skip it while irreversible ones always require it, or is it one global setting? And how are you handling barge-in when someone talks over the confirmation prompt?

This could really free up a lot of time, because like you said it's so much quicker to get something done by saying it momentarily versus context switching.

Sometimes, I forget that a keyboard exists in front of me because of VoiceOS

素晴らしい音声操作ソリューションです。この音声入力のショートカットを押した瞬間から、音声入力が可能になり、入力後、即座にテキスト出力される反応スピードは感動ものでした。わたしは音声入力として使ってきましたが、今後のOSへの音声操作への期待をしています。もっとも、音声入力自体のさらなる機能性のアップや、精度の向上、使い勝手の洗練度も今後も期待したいです。シームレスなOSやアプリの音声操作も革新的ですが、そもそものPC操作や、特にドキュメント作成では、音声入力（従来のキーボードでの入力）がまだまだメインです。この点で、以前のキーボード入力からのテキスト入力は、音声アプリによって変革されました。この音声入力（テキスト出力）でストレスや不足感があると、やはり使用頻度は下がってしまいます。

The confirmation step before it executes is the smart call — most "just say it and it's done" tools scare me because one misheard command and something's gone. App-hopping is the thing that breaks my focus mid-session. Curious how it handles ambiguous commands across two open apps — does it ask which one you meant, or just guess?

I encourage people to really give this a try because at first glance, it might be difficult to see how vastly this improves up on the various voice tools in google/apple/microsoft systems, but VoiceOS integration with AI makes it a vastly more performative product.

I've tried several voice dictation tools out there and VoiceOS was consistently the best one, especially for devs. Excited to try the new Agent features

Wow, the video demo looks absolutely slick, Jonah !

I saw you mentioned a big shoutout to the Japanese community for native localization, but I couldn't find the full list of supported languages. Which languages are currently fully optimized for voice input and workflow execution? I'd love to know how it handles non-English/multi-language environments for daily tasks.

Congrats on the launch! 🚀 :)

Very Nice. what are the limitations please, for example can work with browser tabs or just apps? does it have vision in case for example i want to copy something from a text page to a note?

System-wide voice control that actually executes actions rather than just dictating text is a big deal. What's the most surprising use case you've seen people use VoiceOS for that you didn't expect?

i've been using VoiceOS since it was in beta and really love the product - love the "agent" feature - have found tons of use cases for it and they also released a new update on it - need to spend some more time on it but seems very promising - overall great product and great team (not affiliated with them whatsoever*) -

been using voiceos on and off for about 2 months, solid product! but needs a little bit of work, mainly i dont love having to approve so much stuff, i just want it to rip the task and ill edit results if necessary. but it's early and things like this only tend to move in one direction, which is to improve!

full control over which actions get confirmed sounds great until the default config ships too permissive and most users never touch the settings. system wide execution on mac and windows means the actual risk is in what ships as the out of the box default, not the toggle existing

Hey Product Hunt - Jaski here, developer behind Juno.

Voice is becoming the new keyboard. Juno is the open voice layer for Mac.
I built it for how I actually work: long prompts, product notes, specs, messages, emails. I wanted to speak naturally, see the transcript live, and get finished writing inside the app I was already using.
Juno turns messy speech into clean writing, rewrites selected text by voice, and creates Notes, Reminders, and Alarms.

If software hears your voice and understands your screen, it has to be local, open, and unlimited. For this category, open source is not a feature. It is the only acceptable architecture.

Hence, Juno is fully local, open source, and free forever.

Use Juno and share your feedback with me. https://usejuno.co/

If you are a tech nerd, here is how the best voice to text app has been built - Inside Juno.

Hey congrats on the launch @Juno team and congrats @rohanrecommends for the great hunt!

Wwhy should someone use Juno instead of tools like Wispr Flow or MacWhisper?

Is the main difference that Juno is local-first, offline, open source, and has no subscription? Or are there also specific workflow advantages in how it handles dictation, rewriting, and typing across different Mac apps?

Local, offline, open source, and free — that's a rare combo. Does Juno work well with longer dictations, like voice memos or drafting emails, or is it more optimized for short bursts of speech?

One of the most interesting launches today! The part some people might miss is that live transcription brings its own trust problem like a word commits on screen, then self-correction rewrites it a beat later and now I'm watching my own text flicker. I wonder how long does a token sit provisional before you lock it, and does the local agreement step ever lose a race against me already talking past it?

A key discussion for me was: how high can reliability be?

• I give very long blind-prompts (I literally close my eyes because the dancing animation distracts me) - Juno can show me the transcript, which is great

• but what if my prompt is dropped? it happens sometimes in GPT still and I was very keen to understand the speech harness - which will prevent this and other oops moments from happening.

Would love it if @jas_jaski or @dudhatparesh talk a little more about the engineering behind Juno.

The "local" part is what actually matters here, and I'm curious how far it goes. Is the speech-to-text model running fully on-device with no outbound calls at all, or does "local" mean the app itself is local but transcription still hits an external API? That distinction is the whole ballgame for anyone who'd use this with sensitive work. Also wondering how it handles domain-specific vocabulary, medical terms, code identifiers, things that generic transcription models consistently mangle.

The live transcription bit is what sells me — most dictation tools make me wait for the whole block then fix it after. As an indie dev I basically live in Cursor and Notes, so something local + offline that isn't shipping my half-baked specs to a server is a real plus. Does the rewrite-selected-text work inside Cursor's editor too, or just native text fields?

Congrats on the launch, Jas! I’ve actually been looking for a solid tool to convert audio files and voice notes without uploading sensitive data to a cloud server. The fact that Juno runs completely offline, local, and open-source is a massive win.

I’m almost sad this is currently exclusive to Mac users, because it looks incredible! What’s the main reason behind this choice? Is it due to leveraging Apple's native hardware/CoreML for local AI, or do you have plans to bring Juno to Windows/Linux users in the near future? I'd love to use it across all my setups! 🚀

whisper + live transcription locally is the chunk-size dance — too short and you lose context across the boundary, too long and the perceived latency drags. cold-start hits once, but the per-chunk recompute pattern is the long tail.

how to train this model on my local system? when i make corrections in the transcript does it learn? I have tried similar tools in the past, but I was not successfully able to train them. Like the words that I wanted in its vocabulary, I was just not able to get it to learn them. How does Juno solve this problem?

I always prefer something that's private and works locally. It's works pretty great. Great work @jas_jaski @dudhatparesh

The fully-local, offline architecture is the part I keep coming back to. I work on voice AI for older adults, and the thing that breaks most transcription is atypical speech: slower pacing, disfluencies, regional accents the generic models were never trained on. Since Juno runs the model on-device, can users add custom vocabulary or adapt it to a specific speaker over time, or is the acoustic model fixed? And with no cloud fallback, what happens to a low-confidence segment, does it surface uncertainty or just commit to a best guess?

Would love to see this for Windows and Linux users

Hey! Congrats!

The "build anywhere, deploy in Retool" approach is something that grabbed my attention.

A lot of teams are experimenting with Cursor, Claude Code, and other AI coding tools, but productionizing those projects is still a major bottleneck.

This feels like a practical bridge between AI-assisted development and real-world deployment. Congrats to the Retool team on the launch!

Most AI app builders stop at generating code.

Retool seems to be solving the harder problem: getting from an idea to software that can actually be deployed, governed, and trusted inside a company.

Curious how many teams here are already using AI-generated apps in production today? 👀

GPU-accelerated and designed to stay clean even with code agents running — that's a real differentiator. How does Otty compare to Ghostty on rendering performance and resource usage in your testing?

Love it!

Session recovery, especially for Claude Code/tmux sessions, is the detail that would make me actually try switching terminals. Agent work is full of long-running context, so losing a pane is much more painful than it used to be. Curious how far recovery goes today: process state, scrollback, layout, or all of it?

A Mac-native terminal that's actually thoughtful about rendering is rare. We've spent time debugging unicode and escape code rendering bugs in iTerm configs that just don't exist in native apps. How's the AI integration scoped: is it a sidebar assistant, inline suggestions, or does it actually intercept command output to annotate it?

The terminal only became more important once agent work moved into long-running sessions. Calm panes, recovery, and low-friction context switching matter more than another AI panel bolted on top. Native Mac feel is a real product decision here.

GPU-accelerated rendering in a terminal is the kind of thing that sounds like over-engineering until you're actually running five agent processes in split panes and watching your old terminal stutter. The focus on agent-forward UX is a real differentiator. What rendering backend are you using for the GPU layer? And does the split pane model support custom layouts per workspace?

Hey PH 👋
Avishay here, co-founder of Anima.
Buddy is already the most powerful AI agent for Figma, it is based on all of our knowledge and code around Figma-to-code and agentic coding solutions, serving 2 million users.

Today, we address the token anxiety coming with vibe coding and vibe design.
You can bring your own LLM to power Buddy, we don't charge you for chats.
Plug in your existing ChatGPT subscription, and chat for free. No AI credits.

What Buddy does, right inside your Figma canvas:

• Chat to generate designs - Screens, flows, components, and variants on the canvas.

• HTML to Figma - Paste a link or attach HTML to get editable layers.

• Claude design/artifacts to Figma - Turn what you create in Claude into Figma layers.

• Image to Figma - Attach screenshots or AI-generated concepts and turn it into Figma layers.

• Take actions in Figma - Arrange files, organize layers, run batch edits across screens

• Iterate existing designs - Select a frame and ask for variations, new breakpoints, image swaps, or a new theme

• Clone any website into Figma - just paste a link in the chat

• Parallel chats - run multiple agents at once

• Design system aware - reuse your components for brand consistency

• Image generation and online research - built in

Who it's for:
Anyone who wants to design in Figma with AI.
Designers, product teams, marketing teams, engineering, and leadership who want AI in the canvas.

Anima has been leading the design-to-code space for 8 years, and we're starting our fight in token anxiety today, opening design and code for everyone.

We'll be here all day - Ask anything, share feedback, questions, and hot takes 👇

What I like most is the “use the AI you already pay for” angle. :) AI credits inside design tools can get annoying very quickly, especially when you are still exploring, iterating, and throwing away half the outputs.

For a product team, the most useful part feels like turning rough ideas, screenshots, or existing pages into editable Figma layers instead of ending up with a static image you have to rebuild manually.

I’m especially curious about the design system awareness. If a team already has components, colors, spacing rules, and brand patterns in Figma, how reliably does Buddy reuse those instead of creating something that looks close but still needs cleanup?

Also, how does plugging in a ChatGPT subscription work in practice? Is it using the user’s existing account directly, or API keys?

Bringing your own LLM is a smart move for teams already paying for ChatGPT. Can Buddy work with design systems already set up in Figma, or does it start fresh each time?

The import-anything feature is what got me - tested dropping in a component from a live site and it came through without the usual formatting mess. Bring-your-own-key is the right call too, removes the biggest friction point for anyone who already pays for ChatGPT. Agent still has some rough edges on complex multi-section layouts but the core import flow is genuinely solid. Good launch!

Congrats on Buddy’s launch 🎉 A free Figma agent sounds super useful & excited to see how it helps designers speed up their workflow!

That's amazing! Saves so much time when iterating design ideas

Congrats on the launch! 🎉 So excited someone is finally making AI actually work inside Figma. I’ve been waiting for Figma to solve this themselves, but honestly this looks like it. The bring-your-own-LLM approach is smart too. Trying it now!

Hi Product Hunt! 👋 I'm Ashley, founding team at Adapt.

AI at work is still happening in solo-player mode. We're changing that.

No one wins when AI power users operate in silos while everyone else struggles to do more with less resources and limited exposure to what "good AI" looks like.

Knowledge stays scattered across docs, stuck inside tools, and buried in dashboards that haven't been updated in months. Access is painful. No central source of truth exists for you or your agents.

Adapt is the company brain that finally fixes that. It pairs your tools, knowledge, and skills with one universal agent your whole team can use, just by tagging @Adapt wherever you work. Your entire team can use frontier AI with a shared context layer that's already set up for them. Shared integrations, shared knowledge, shared workflows - everyone can work AI native, without the slop.

As a marketer, I can tag Adapt in Slack to help me build a HubSpot workflow or vibe code a new email template. Or optimize ad campaigns across 4 platforms without 1000 clicks. In the same workspace, our engineers use Adapt to manage infra, write, and review code. New team members can onboard autonomously with Adapt.

As workflows become patterns, Adapt builds loops proactively and on a schedule. Your busywork gets automated, and your expertise becomes the differentiator. The whole team benefits.

We built Adapt as a system for working AI native together, not to replace your co-workers.

Our team is in the comments ready to respond to any of your questions today.

And you get $100 free credits just for signing up - so go try it out!

Adapt is amazing. Easy to setup, and everyone in our team adopted it quickly after seeing it work for others on Slack.
We use it to work across support tickets and production logs to see what's really happening. It's a superpower to triage across a bunch of different tools.

Hi Product Hunt. I'm Jim, co-founder of Adapt.

I believe every company will have an AI brain. It's inevitable. The only question is build or buy?

Work is scattered across too many tools, and people are the glue. A company brain ends that. It lives in Slack, always runs on the best intelligence available, and gives AI its own computer to actually do the work, not just talk about it.

Our mission is to make every company instantly AI native, and that means every worker. Getting everyone there is an imperative, and the window is short.

Here's what the AI native way to work really means, and why no one should get left behind: https://adapt.com/blog/ai-native-company

We look forward to building Adapt with you. Thank you!

We’ve been using this at wander.com for the last 6 months. A shared agent, with shared context layer and integrations has been a great unlock. Requires complete hookup of data sources for the magic.

As a founding engineer at Adapt, I’ve been dogfooding since day 1.

Adapt has changed what I’m able to accomplish so much that I would greatly miss it otherwise. For me, the best part is being able to get an answer to anything we’ve connected right in Slack. In the middle of a chat about a bug, I can tag Adapt and get a single reply with relevant PRs, logs checked, infra health, a new Linear issue created and a new PR with Adapt’s knowledge base updated for the next time. It’s awesome.

We’re proud to share this with Product Hunt. Seeing what @jimbenton has created in Adapt Apps blows my mind. He’s spun up internal dashboards pulling data from 5 different sources and created beautiful interactive decks just by telling Adapt what he wants.

The best way to get the most out of Adapt is to be upfront with your goals. Don’t hold back, just tell it what you do, what you need, and ask how it can help you accomplish that. We’ll take it from there!

The shared context layer concept is interesting — making AI useful for the whole team rather than just power users. How does Adapt handle sensitive info in Slack that shouldn't be accessible to everyone who tags @Adapt?

Hello Product Hunt! I'm Sean, one of the co-founders at Adapt, working on all the technical stuff.

When I first started building out Adapt, I knew I wanted a system that was horizontal. I wanted a single "intelligence" connected to everything that I could interact with. My day-to-day used to involve hopping from service to service, trying to piece together bits of information, and generally losing efficiency from all the context switching. Now almost everything I'm doing goes through Adapt. Feature request comes in from a customer? I have Adapt pull in the relevant context from all previous interactions with that customer, find any related Linear issues, find similar feature requests from other customers, and then finally open up a PR implementing that feature and deploy a preview environment to test it.

Behind the scenes, we're spinning up sandboxes for the LLM to use. Every interaction with integrations happens through the LLM writing and executing whatever code it needs to for accessing those integrations. We're not limited by what happens to be exposed through an MCP server, we're interacting with these services directly through its API or working with a browser inside the sandbox. This architecture lends itself to supporting the vision of being a horizontal platform capable of connecting to anything.

We're excited to be sharing this with everyone!

anyone in the workspace being able to tag this and trigger infra changes, ad spend, or autonomous onboarding is a wide blast radius for a low barrier action. shared context across that many tools is powerful, the access control behind who can actually trigger what matters just as much

Hey team! I'm Matt, I'm one of the product engineers working on Adapt.

We use our own Adapt agent at Adapt, my personal favorite use case is having Adapt automatically do quick reviews / suggestions of my code on GitHub. I've set up these proactive triggers so that anytime someone asks for a PR review on the #eng channel, Adapt immediately hops on it for a review. (Adapt doesn't actually approve the PRs, just makes suggestions).

Right now, I'm working on Adapt's agent memory, working on helping Adapt remember things about your company and its operations, that way it gets more intelligent the more you use it.

Happy to answer any questions about the Adapt product or any of the work that I do here!

Hi folks! I'm Bruno, a Product Engineer at Adapt and a big fan of the product as someone who dogfoods the platform in my day to day workflow.

I've been using Adapt to monitor logs and alerts, provide understanding and context for production errors, brainstorm solutions, write code, and review code. It has been a wonderful coworker so far.

If you have any questions about how Adapt can help you and your business work more smoothly and productively, just let me know!

I've been a big fan of Webflow since it's release so I'm excited to see what Ploy is able to do!

the cmp issue called out above is the same root cause as most vibe coded security gaps, ship fast, gate later, except later usually means after eu regulators or a user notices first. autonomous site changes plus untracked compliance is a rough combo to scale on

Hey PH 👋

We’re really grateful for the support Locofy has had since the early days — a lot of what we are today is because of this community.

We’re back with a very different Locofy: more developer-first, more modern, and built for how people actually ship frontend today.

Earlier, Locofy was heavily designer-centric and enterprise-focused. Over the last couple of years, we heard consistent feedback from startup devs and freelancers that the workflow wasn’t flexible or accessible enough — so we’ve made a deliberate shift.

Now Locofy supports:

* Figma → pixel-perfect frontend code using proprietary design-aware models + LLMs
* Agentic workflows built for developers (use Locofy directly from CLI, Cursor, Claude Code, other Copilots)
* Self-serve plans with flexible, affordable pricing for individuals and small teams

Why does Locofy sit between Figma and Cursor/Claude instead of just using them directly with Figma?

Because AI coding agents are incredible at logic and iteration, but they still struggle with UI fidelity and understanding Figma’s structure. That leads to broken layouts, inconsistent spacing, and a lot of manual cleanup.

Locofy solves that by translating design intent into structured, production-ready frontend code first and acting as the missing frontend layer between Figma and Cursor / Claude Code etc— so Cursor and Claude can focus on what they do best: building features, logic, and product flows on top of a solid UI foundation.

This launch is really us correcting course, and building for the way frontend is actually being built now.

Appreciate everyone who’s been part of the journey 🙌

proud to see how far this product has come. Huge congratulations to the team on the launch, and wishing you all the best for the journey ahead! 🚀

Congrats on the launch. The "agents struggle with UI precision" framing is spot on. One architectural question: does this need the Figma source as the anchor, or can it help when you're editing an existing/running app where there's no design file to read from? Curious where the boundary is, since a lot of agent UI work happens on screens that never had a Figma.

Bridging the Figma AST to idiomatic React components is a hard problem. Most tools spit out nested divs that are painful to maintain. We've dealt with this exact friction building UI from design mocks. Do the agents understand shared state and component boundaries, or does that still require manual cleanup after generation?

Congrats on the launch! Excited to see how design-aware models hold up against messy Figma files in the wild vs. clean design systems.

The design-to-code agent direction makes sense, especially if it keeps design intent visible instead of only generating files. I’d love to understand how teams review the generated code when the Figma source keeps changing.

I've been trying the new workflows, and they feel much more developer-friendly than before.

The CLI workflow is probably my favorite. Being able to generate code directly from a Figma URL and merge it into an existing codebase without leaving the terminal feels super natural.MCP is also a great addition. Having Locofy available inside Cursor and Claude Code fits much better into how I build products today

The combination of design-aware code generation + AI-powered refinement feels like a natural fit for modern frontend development.

Does it stay fully offline, and can you selectively share a milestone without revealing your home area?

The “spot the gaps you haven’t explored yet” idea is a nice motivator. It makes movement feel more like discovering your own map, without turning route history into another social feed.

I like the idea, especially because I love travelling and active lifestyle. What will be the price in the future?

How would you compare Tiles to Bump (by amo)? If someone is already using Bump, what would be the main reason to switch to Tiles? Is it the local-first approach, Apple Health/workout imports, privacy, GPX/FIT/CSV support, or something else?

Also curious about monetization. Since the app is currently free, do you plan to keep it that way long term, or are you considering some kind of paid tier, subscription, donations, or other sustainability model in the future?

Hi Product Hunt! 👋 I'm excited to share Tiles with you today.

I originally built Tiles to solve a personal problem: I had years of hiking trails and trips recorded on my Apple Watch, and wanted a way to see all of it in one place. Once I built my demo, I realized I kept visiting the same places over and over. Tiles became an easy way to see nearby areas that I'd never explored before. It pushed me to break out of my routine and explore new places.

As I shared early versions, almost everyone asked for the same thing: "How do I add friends?", "where can I see other people Score?" I wrestled with this for a while. I always imagined Tiles as offline and local first, but everyone I showed it to wanted some kind of social feature.

I did not want to build another fitness social network, so I tried to keep the fun parts of sharing without turning the app into a feed. Tiles has shareable cards for scores, milestones, and year-in-review style stats, without needing to upload your full route history anywhere. Does that feel like the right tradeoff, have you had to make similar decisions between privacy and social features?

For the Product Hunt launch, I set up an early explorer offer: 3 months free of Tiles Pro.

I’d love to hear your feedback or questions. How would you use Tiles: as a memory map for old routes, a way to find new places to explore, or something else entirely?

hah, that's cute, like playing a game or uncovering a HOMM3 map

Hey Product Hunt 👋

I built Cliptop because I wanted clipboard history to feel as fast as copy and paste itself.

I use the keyboard for almost everything on my Mac, and most clipboard managers still felt like they made me pause: open a window, scan a list, reach for the mouse, pick something, then get back to what I was doing.

Cliptop is my attempt at making clipboard history feel invisible. Hit a shortcut, search what you copied, press Enter, and keep going. No context switch, no digging around, no breaking flow.

It opens from the notch/top edge or by pressing Shift+Cmd+V, but the main idea is speed: recent copies, previews, Pinboards, and paste actions that are quick enough to use constantly throughout the day.

It’s local-first too, so your clipboard history stays on your Mac. Private iCloud sync with the Cliptop iOS app coming soon as well, but that will be an optional feature.

I’d love your feedback on the flow: does Cliptop feel fast enough to become muscle memory? And what keyboard-first paste actions would you want next?

Thanks for checking it out!

So refreshing to see a unique approach to solve clipboards. Congrats Raz!

Super useful app! Congrats on the launch :)

Such a useful app Raz! Thanks for making it. Love the UI!

Maybe I finally found a substitute for my neverending pile of text files. Congrats on the launch!

I've been using Cliptop for a week and what can i say, it's super conveniemt.
I didn't even know i needed some of the features before i discovered them ;)
COngrats on the laucnch, Raz!

Hey Product Hunt 👋

I’m Noa, co-founder of Genie.

My path here has been a bit unusual. I dropped out of high school at 16 and moved to the Bay Area after my work gained traction with companies like Google, Facebook, IDEO, and others. At 19, I exited a startup and later became the founding designer at Turing. Since then, I’ve had the opportunity to work as a technical artist with brands such as Puma, Microsoft, and OpenStore.

You can find courses I’ve created online where I teach design and coding, along with tutorials on YouTube that have reached hundreds of thousands of people. My obsession has always been the same: learning and understanding the most exciting technologies shaping the future.

That obsession is what led us to build Genie, and, by extension, products like Mentions (launching on PH today!) and GenUI (stay tuned!)

As we explored the AI landscape, we realized how influential products like OpenClaw have become. At the same time, we noticed that most people struggle to set them up, and that agents often break down without persistent memory, context, and continuity.

We thought there was an opportunity to solve those problems directly: persistent memory, social behavior, social cron jobs, and on-device models that improve speed while reducing token usage. The goal isn’t just to make AI more capable, it’s to help create the kind of product culture Brian Chesky often talks about when discussing Airbnb.

Apps are becoming agents, and the way we interact with software is fundamentally changing. it only makes sense that being able to reference your friends inside a prompt, access shared context, and weave together agentic workflows could unlock an entirely new social layer of computing.

So what does all of that mean for the actual product? Well, Genie acts a bit like that mutual friend everyone has: the person who understands both sides, makes connections, and brings people closer together.

That becomes incredibly powerful when you consider what it enables: more personalized interfaces, awareness of how people actually think, and a deeper understanding of how AI fits into everyday social life, not just professional workflows.

A few things I’d love to hear from everyone:

•⁠  ⁠What’s the most useful thing ChatGPT can’t do because it doesn’t know your friends?

•⁠  ⁠If AI knew both you and your friends really well, what’s the first thing you’d ask it?

Hi Product Hunt 👋

I’m John, also Genie co-founder. 

I haven’t had the fortune of being on a PH Launch Team since @Nokia Z Launcher (with other fun gigs after that like being the first PM hire at @Bumble , then a PM at @facebook , then @Citizen ), but it’s good to be back!

AI has gotten immensely powerful as it became more personal. And yet to us it felt like it was missing a key component of who we are.

We don’t live in an isolated bubble. Our friends, our family, our spouses, our housemates…they help shape what we do and even who we are. But all my other AIs know about them are stale comments I made about them from weeks or even months ago.

The neat thing about Mentions is that it’s an AI that is designed to be multiplayer from the ground up - to have separate private and public-facing memories and models designed to make use of them. This hasn’t been easy (there’s a reason everyone else defaults to just trying to make a “single-player” experience work well), but as you and people in your circle use it, you can start to see why it was worth the effort. When Mentions knows those around you who help define your life, it feels immensely more useful, more surprising, and more fun than AIs that were designed to exist in a social silo.

There are a lot of ways AI can be made better where we think we can make some meaningful contributions despite our tiny size. Making your AI socially aware is one of them, and we’d love to hear what you think of it. We’d also love for you to give us a follow so you can be among the first to see the next things we have coming.

Mobile-first TTS with ElevenLabs voices is genuinely handy — I make music and keep needing quick voiceovers for promo clips without booting up a whole studio setup. The on-phone voice cloning is the surprising part. How many seconds of reference audio does it need to clone something that sounds decent?

voice cloning on a consumer app is the one feature in this space that needs guardrails baked in from day one, since the risk isnt a data leak, its someone elses voice saying something they never said. worth locking down who can clone what before this scales past your own testing

Congratulations on launch! What about some texts which have related graphs/images as well. Sometimes, we need to read, observe and verify at the same time? Are there any plans to handle that as well?

Thats very useful because sometimes our CFO gets 70-page contracts and reading through that its pure madness, especially when its from chinese vendors☠️ Does ur product has something like a bulit-in translation feature for this typa stuff?