PH热榜 | 2026-06-15

Hi everyone, I'm Tomer, co-founder at Novu.

Dima covered the what. Let me cover why it matters.

The agents people are building no longer just answer questions in a chat box. They send approvals, ask follow-ups, escalate, and report back. For any of that to work, the agent has to reach a real human on the channel that human actually uses, hold the thread, and pick the conversation back up when they reply. That last mile is where most agents stall, and fail to deliver value.

Novu Connect is that last mile. Connecting Agents to Humans on every channel. And it runs on infrastructure teams already trust in production, with SOC 2, HIPAA, and ISO 27001. Not a brand-new unknown stack, the proven one, now two-way.

Easiest way to feel it: open the terminal you already have Claude in and prompt Add an agent to my app https://novu.co/agents.md. A timer starts, and under 60 seconds later your agent is talking on Slack, Teams, WhatsApp, Telegram, or email. Then message it back and watch it hold the thread.

We built this for you. Tell us what is missing, we are listening all day. 🙌

Hey Product Hunt 👋 I'm Dima, CTO and co-founder of Novu.

For four years we've built the open-source notification infrastructure that 40k+ developers use to reach their users across Slack, email, push, and more everyday. Then AI agents arrived and we kept hitting the same wall: agents got incredibly good at thinking, but they still cannot properly communicate to the people they work for. Every channel has its own webhooks, identity model, and threading, so you rebuild the same plumbing for Slack, then Teams, then WhatsApp. It is weeks of work that has nothing to do with your agent.

So we built Novu Connect.

It gives your agent a voice across multiple through unified API. We are naming this layer ACI, Agent Communication Infrastructure. MCP connected agents to tools. A2A connected agents to each other. ACI connects agents to people.

The line we will not cross: we never run your agents brain. Bring Claude Managed Agent or your custom code with AI SDK/Langchain, and Novu just carries the conversation.

Pro tip, try it without leaving your terminal: open the terminal where you already have Claude running and prompt it: Add an agent to my app https://novu.co/agents.md. No account needed, try it out.

I'll be in the comments all day. Would love your honest, technical feedback. 🙏

Standard text is fine but interactive elements like slack buttons or whataap quicl replyes really drive engagmetns for bots, can our agents pass intereactive ui payloads through Novo connect ?

Damn this is clean😎

How do I manage audit logs and see what my agent actually sent?

This solves a real pain for us honestly - can I see delivery status, like did the human actually get it?

Is there any hope of integrating with Linkedin chat messaging? either corporate or personal?

Just tried the Claude prompt. Mind = blown. What’s the latency like on WhatsApp replies?

Congrats! 🙌

Love the idea of ACI, it solves a real pain point of building and doing all hassle with the integrations layer, which is a huge distraction where at the end you want just to ship the agent quickly.

I will be definitely using it for my side projects to create, automate, research, part of the work I do manualy. And I am waiting for the scheduling support!😊

I am curious what others have alredy build on top of this?

Super cool. Can I use this to allow me to chat with my bot via Whatsapp?

Emil here, I work on the marketing side at Novu.

The founders have already covered the what and the why, so I'll just add the one thing that made this click for me.

For months, we kept saying "connect agents to channels." But the longer I sat with it, the more I realized connecting a channel is the easy part.

The hard, unglamorous part is keeping one conversation coherent when a user starts on Slack, replies from WhatsApp, and finishes over email. For the agent, it should be a single thread. Same memory, same context. That continuity is the whole point, and it is genuinely hard to build.

The moment it stopped being a slide and became real for me was watching an agent go live on an actual channel in about two minutes, no API key, no plumbing.

If you're building agents, I'd love for you to try it and tell me where it surprises you or where it breaks.

Start free: novu.co/connect

So, which channel would you put your agent on first?

Congrats on the launch! The "last mile" framing really resonates — connecting agents to humans on the channels they already use is exactly where so many agent projects stall. Love that you're building on infrastructure that's already SOC 2/HIPAA/ISO ready rather than a brand-new stack. Curious: when an agent spans Slack + WhatsApp + email for the same user, how do you decide which channel to reach them on first, or is that fully configurable by the maker?

love what novu has been building for a while now — super clean product and strong vision.

aci is especially exciting, had a chance to connect with @tomer_barnea1 and it really feels like the future of notifications. big congrats to the team on the launch 🚀

Congrats on the launch. The piece I'd want to understand before adopting a notifications layer is failure semantics: when a downstream provider (email/SMS/push) is degraded, does Novu retry with backoff and fall back to the next channel automatically, or is that orchestration something I define per workflow? Dedupe across channels is the other one — stopping a user getting the same alert by email AND push.

Strong launch. The hardest part of agent UX is meeting users inside existing workflows; Slack/Teams/email threading + identity resolution is exactly the unglamorous glue. Curious how you handle conversation state when the same user jumps channels?

Excited to try it. What's the free tier look like for someone just experimenting?

Incredible launch, team Novu! 🚀 'Bring your own brain, Novu handles the voice' is the perfect line. It’s awesome to see a trusted open-source notification stack evolve into a robust two-way conversational layer for AI. Bookmarking this for my next build!

This solves a real pain for us honestly, can I see delivery status, like did the human actually get it?

Congrats on the launch! 🚀

The "-second rule for terminal-to-channel connectivity is a huge win for developer velocity. Running npx novu connect to bridge agents with Slack or WhatsApp that quickly sounds like a dream for anyone building in the AI agent space right now.

Can't wait to see how this simplifies multi-channel notifications! Just curious, how do you handle the auth/handshake for the more walled garden channels like WhatsApp within that minute?

Good luck today!

Congrats on the launch. The “agent-to-human last mile” framing is strong, especially because teams already live in Slack, Teams, WhatsApp, and email. The part I’d love to understand better is the governance layer: when the same user talks to an agent across multiple channels, how do you handle identity mapping, audit logs, and permission boundaries so the agent has enough context without exposing too much history to every operator?

Congrats! Curious how does Novu Connect handle context continuity if the human switches mid-conversation from one channel to another

This is huge for human-in-the-loop workflows. being able to ping a manager on slack for an approval button while the agent talks to a client on whatsapp is exactly what we need. awesome launch🙌

Smart. Can multiple agents share one comms layer or is it one per agent?

Hey Product Hunt 👋

I'm Serdar. I built Wobo with my co-founder Taha. Here's the rundown.

What it is: It's Tinder for jobs. You swipe right on a role, and Wobo applies for you on the company's own site.

Why we built it: Even finding the right roles meant checking LinkedIn, Indeed, then digging through company career pages one by one. Then for each one came the grind: read the full description, research the company, spend 20 minutes refilling the same forms, tailor a resume and cover letter to boost our odds, then pray it wasn't a scam or a ghost listing.

So we built Wobo. Now all of that is one swipe.

How it works

1. You tell Wobo what kind of role you're after, and it builds your Wobo Persona, a digital twin shaped by your background, experience, and your voice.

2. It scans the market daily and brings you the best matches, including jobs that never hit LinkedIn or Indeed.

3. You swipe through them and keep the ones you like.

4. Wobo applies on each company's own site for you, with a resume and cover letter tailored to that job and written in your voice, exactly how you would.

What makes it different: The core is your Wobo Persona. Every answer it writes and every job it applies to, you can give feedback on, and it learns, so over time it sounds more and more like you. That's why Wobo applications read like you wrote them, not generic AI. And unlike tools that spam recruiters or just auto-fill LinkedIn Easy Apply, Wobo applies on each company's real site, only to roles you picked.

What you get

• Pre-vetted companies, scored on real signals. Every company gets a Wobo Score before its roles reach you. No ghost listings, no scams.

• A fit score on every match, so you decide in seconds.

• The whole company in one card: salary, ratings, benefits, and culture, so you can judge it before you apply.

• One swipe replaces a 26-minute application.

Who it's for: Anyone job hunting right now, from internships to senior roles, who's tired of drowning in applications.

Try it: It's completely free, on web and iOS. I'd love your honest feedback: what would make you swipe right, and what's missing?

100k+ people use it now, and my sister even found a job through it, which still feels strange to say. I'll be here all day, and a big thank you to Rohan for hunting us.

Serdar (and Taha)

Congrats on the launch! The UI is super smooth and very intuitive. I had a quick question regarding geographic coverage—is Wobo currently limited to the USA, Canada, and the UK?

I’ve been navigating the job market for about 3 months now, so I’m eager to see how much this can optimize my search efficiency.

Hey all 👋 Taha here, co-founder at Wobo. Serdar's post already lays out the full picture, so rather than repeat him I'll jump straight to the part I'm personally most excited about.

You saw the swipe, and it really is fun. But under the hood we also do Autopilot, and honestly I think that's the most important thing we've built. Once your profile is set, you can fine tune everything (the roles, salary range, locations, company type, how aggressive you want it to be) and then just hand it over. Wobo applies in the background to real roles at top tier companies, in your voice through your Persona. No swiping, no forms.

And here's the part I love most: making the job hunt fun is one thing, but not having to spend that time at all is the actually fun part. You set up your profile once, barely touch it after that, and Wobo keeps running while you get on with your life. It all lives on web and mobile too, so you can check in straight from your phone whenever you feel like it.

We're always looking at what else we can add to make this even smoother, and we genuinely read every comment here and take your feedback to heart, so please keep it coming 🙏

Big thanks to @serdaraksoy for building this with me, and to @rohanrecommends for hunting us.

the part that actually matters is applying on the company's own site instead of just spamming linkedin easy apply. that's what real recruiters pay attention to. curious how the Wobo Persona learns your voice over time... does it get noticeably better after 5-10 applications or is it pretty accurate from the start

Job searching can be exhausting, so I love the idea of simplifying the process and helping people discover opportunities they might otherwise miss.

The swipe-based experience feels fun and approachable, and having applications tailored to each role is a great touch.

Company-site applications are the interesting part here. When I swipe right, can I preview the tailored resume, cover letter, and long-form ATS answers before Wobo submits them, or is the review only at the job-card level?

The company-site application part is what makes this feel more useful than a job board wrapper. If it can keep applications tailored and natural-sounding, it removes a lot of the worst job hunt friction.

I'd go for a different swiping method tbh, the whole tinder swipe experience is giving people PTSD since 2020, especially for GenZ, otherwise the idea is awesome :)

@serdaraksoy Congrats!! How are situations handled where questions are asked beyond just typical resume and cover letter drops? "What excites you about this company?", "Tell us about an accomplishment you are proud of", etc. Over time I suspect more companies will begin adding in these bespoke questions to try and prevent people from mass auto-applying, so am curious how Wobo handles these "edge cases"!

One thing I'd love to understand: what happens when Wobo answers a custom application question incorrectly or misses important context? If the user only sees it after submission, that could be a pretty painful failure mode. A pre-submission preview of the full application would probably be the first feature I'd want.

Congrats on the launch!

How do you handle application flows and custom questions without it getting stuck or submitting something wrong?

Does it customize the resume and cover letters? Or does it just use the resume that is uploaded

Congratulations on the launch! I'm curious to know what criteria other than skill are used as the basis. Is there a value match or something more individual?

@taha_keles Hey guys, nice solution.
by the way, do you have a promo code for the purchase of Wobo Autopilot ?

Interesting take on job search automation. The part I’d want to trust most is “in your voice” for application questions. Do users get a review step per application, or can they set rules for when Wobo applies autonomously?

Congrats on the launch. Applying on the company’s actual site instead of spamming Easy Apply is the part that makes this more interesting.

The main trust question for me is review control: for high-stakes fields like work authorization, salary expectations, relocation, or “why this company,” can candidates mark certain answers as “always ask me before submitting”? That would keep the swipe flow fast while avoiding the few mistakes that can really hurt an application.

Hey, looks like you are using fake reviews front and center on your app which is definitely illegal or breaking some TOS of Product Hunt. You didn't even do a good job faking the reviews. Any LLM can immediately identify a 90% fake rate. Not to mention this review seems like a real one. Want to clear that up for us?

How does Wobo handle the wildly different form structures across thousands of career pages?

Love the "swipe right and wake up to interviews" concept — the fact that it applies on the company's actual career site instead of just LinkedIn Easy Apply is the real differentiator. The Persona feature sounds like the key to keeping applications from feeling generic. Congrats to you and Taha on the launch! One question: when Wobo applies to many roles for the same person, how do you avoid a candidate looking spammy to companies that share applicant tracking systems — is there any pacing or per-company limit built in?

interesting approach to the application side. curious how you think about the other half of this — as AI makes applying easier, the signal-to-noise ratio for employers gets worse. every applicant starts to look polished. the bottleneck shifts from "can this person write a good cover letter" to "can we actually tell who's good at the job." feels like there's a matching problem that outlasts the application problem.

Honestly love the cheekiness of this, "get paid to wait" is a great hook :)

One thing I'm curious about on the technical side: since the integration lives in the status line, it gets the JSON payload Claude Code passes on every refresh (model, current working dir, session info...). To serve and attribute ads you presumably phone home pretty often.

So my real question is around privacy: what exactly does the status line script send to your servers? Do project paths / workspace dirs ever leave the machine, or is it stripped down to an anonymous impression ping? For folks working under NDA or on sensitive code, that's the make or break detail.

Not a gotcha, genuinely rooting for the idea, just think being upfront about the data flow would build a lot of trust here. Congrats on the launch!

I've had this running for a few days and it seems that after the first day the earnings are severely nerfed? The first day I ran this for a couple of hours which made 5.40 USD for 514 events, which was pretty promising. A few days later I've ran this for a whole day, and I earned 0.43 USD over 1587 events, which seems like the app is just going to make it harder for me to earn the closer I get to the 10 USD cash out threshold. I don't know if the extension has worked for anyone since I haven't seen anyone cash out yet, and even though its "free money" the extension requires constant baby sitting, asking me to reload the window or sign in again and again. Are you guys having an incident or something or do the earnings just tank after the first day?

This seems like a troll idea, but SUPRISIGLY good.

I think the next piece of validation is "what does a dev do when they're waiting for Claude".

Because a lot of people I know (eg. me) just start doomscrolling or clicking around testing bugs on their product or whatever they're building.

Good luck Gabe, congrats!

Great idea, but it currently only supports Stripe payouts, which are not available in some countries. I’m also experiencing some issues with logging in.

I wish it supported PayPal or direct bank payouts as well, and that the login issues were fixed.

By the way, the idea is really good, and I genuinely like the product.

It made me think.. why isnt it a thing like in all software products?? For instance we can put Ads in the free space of every OS toolbar

This is a funny but surprisingly logical idea. Developers already spend a lot of time watching agents think, so turning that dead time into something opt-in and revenue-sharing makes sense.

The hook is clever: agent wait time is basically the new compile time.

The trust boundary is the product, though. For developers under NDA, the key question is what leaves the machine when the status line refreshes. Is the ad request limited to anonymous impression metadata, or can things like workspace path, repo name, model, current directory, or session context ever be sent upstream?

A public “exact payload we collect” page would probably help adoption more than any growth copy here.

"Get paid to wait" is such a fun hook — turning agent dead time into revenue is genuinely clever. Congrats on the launch! How do you keep the sponsored status line feeling useful rather than just another ad?

The dead time angle is smart — waiting for Claude to compile is basically the new coffee break. My only concern is the same as others raised here: what data exactly gets sent to your servers during those sessions? A clear answer to that in the docs would probably convert a lot of fence-sitters. Rooting for you either way, clever niche.

The wait state being monetizable is a clever reframe. You're paying for that idle time regardless, so turning it into a revenue split actually fits how agent workflows already feel. Curious how you keep the sponsored line from getting visually noisy when someone's mid-task.

This is a fantastic idea. I ran CodeFund (previously Code Sponsor) a handful of years back in an effort to fund OSS. Advertisers would pay top dollar knowing that they were getting their products in front of their exact target market.

Running an advertising company is difficult, especially when it comes to these types of audiences. As much as I hope this project succeeds, it's going to be a huge challenge if you plan on the developer retaining privacy.

For reference, see ethicalads.io and how they manage to balance privacy with advertising data.

Great idea, but it seems I have to auth the vscode extension via Google even if I have registered with some other email address. So I am not able to use it except with email addresses that are not tied to any Google account?

Oh, also my Claude Code CLI is running on an external host (dedicated Debian machine), I am connected via SSH from my vscode at home. I get a lot of timeout and other messages.

After 3 times reloading the window and signing in Kickback is now "green"!

Clever and slightly surreal market: monetizing agent wait states. I like the 50/50 revenue share. How are you thinking about keeping sponsored status lines useful to developers rather than just another ad surface?

This is a hilariously brilliant idea. Watching Claude Code spin or waiting for massive agent loops to finish is the new "waiting for code to compile". Flipping that dead time into an ad-supported revenue stream is incredibly clever.

Out of curiosity, how does the integration work under the hood? Is it injected as a custom terminal reporter or an environment wrapper? Looking forward to trying this out!

Happy to hunt AutoEdit today! 🙌

Video editing is one of those tasks that still consumes countless hours, especially for creators, podcasters, agencies, and content teams working with long-form footage. What caught my attention about AutoEdit is that it doesn't try to replace editors, it removes the repetitive work that slows them down.

Built directly inside Adobe Premiere Pro, AutoEdit uses AI to understand the content of your footage and automatically remove silences, filler words, bad takes, restarts, and repeated sections while helping structure a cleaner rough cut. The result is less time spent on timeline cleanup and more time focused on storytelling.

If you're creating videos regularly, I'd love to hear your thoughts on how AI can improve the editing workflow. The makers are here all day to answer questions and gather feedback from the community.

Congrats on the launch! Most of these editors cut silently and you only catch the bad take after export. How does Claude tell a deliberate dramatic pause from dead air?

does this work with DaVinci Resolve too? looks sick but i mostly edit in resolve and capcut

Building this as a Premiere-native plugin is the right call — keeping editors in the tool they already live in. Cutting the boring cleanup while leaving creative decisions to humans is a smart line to walk. Congrats on the launch! Curious how it handles deliberate dramatic pauses vs. actual dead air?

Is there a free version to try out? Or do you have to pay right away? just tryna test it before committing

Congrats on the launch! I like that AutoEdit focuses on removing repetitive editing work rather than trying to replace creators. That feels like a much more practical use of AI.

the no-external-upload thing is actually the main win here - i've abandoned two AI editing tools just from the friction of moving footage in and out. context-aware cuts instead of just silence detection is a smart angle too, a long pause isn't always dead air

Hey Product Hunt 👋

I'm Devin, a content creator and video editor.

I built AutoEdit because I was tired of spending 20+ hours every week manually cutting silences, bad takes, filler words, and repetitive talking-head footage from my YouTube and Instagram videos.

Most AI editing tools force you to upload your footage somewhere else, wait for processing, then export everything back into Premiere Pro. As editors, we already spend enough time moving files around.

I wanted something that worked directly inside Premiere.

So we built AutoEdit Creator Mode.

It uses Claude AI to analyze your footage, identify weak takes, filler words, awkward pauses, repeated points, and automatically creates a rough cut directly on your Premiere Pro timeline.

The goal isn't to replace editors.

It's to eliminate the repetitive editing work that nobody enjoys doing manually.

A few things that make it different:

⚡ Runs directly inside Premiere Pro
⚡ No uploading footage to external editing platforms
⚡ Uses Claude AI to understand context, not just silence detection
⚡ Builds an editable rough cut directly on your timeline

We're still early and actively improving the product based on creator feedback.

I'd love to hear from other editors and creators:

What's the most annoying part of your editing workflow that you wish AI could handle for you?

I'll be here all day answering questions and collecting feedback. Thanks for checking us out 🙏

Interesting. Does this tool have any competition?

As someone who occasionally edits long-form content, removing silences, filler words, and bad takes is always the most repetitive part. How much manual review do creators typically do after AutoEdit generates the rough cut?

Congrats on the launch! Very interesting product especially since editing content is so time consuming. Are you planning on having any integration with the CapCut if that’s possible?

Does it work well with multiple speakers and overlapping audio for podcasts or interviews?

It's great to see another spin on automating the business work out of editing content without getting in the way of creativity. Nice plugin! I wish I had more video to edit :)

Wow! It's really impressive. Time saver but most important is it always gonna edit better than us. Not subjective cutting and great timing. Wish you all the best guys!

Premiere-native is a smart wedge. Removing restarts, filler words, and bad takes while keeping the editor in control solves the right boring part. Does AutoEdit preserve enough timeline structure that editors can audit each AI cut quickly?

Congratulations

This looks really powerful for repetitive tasks and I'm excited to try it out.

As a Premiere Pro user this is right up my alley, the "remove the boring repetitive work, keep the creative decisions" framing is exactly the right line to walk :)

My one practical question before trying it: how far back does the plugin support go? I'm still on the 2023 version (not everyone jumps to the latest each year), so I'd love to know the minimum Premiere version you support, and whether features like the silence/filler removal and caption generation work the same across versions or only light up on the newest releases.

Also curious: does it run the analysis locally or does the footage get uploaded somewhere for Claude to process? That tends to matter a lot for anyone editing client work under NDA. Congrats on the launch!

Nicely made, team

What might a tool like Hyperframes or Remotion fare against Autoedit? OR is it completely different categories?

As someone who has done video editing, I can definitely relate to how much time gets spent on repetitive tasks like cutting pauses, filler words, and retakes.

What I like most is that AutoEdit works directly inside Premiere Pro. Not having to upload footage to another platform and then move everything back into your workflow is a huge advantage.

Hey y'all, just posted another fun tutorial on the plugin on our channel! lmk what you think!

https://youtu.be/5TBk4OEj9Zw?si=c102aP56JYfM15YU

Congrats on the launch! Curious how does AutoEdit distinguish between filler pauses and intentional dramatic beats?

Congratulations, looking at your website, it seems like a lot of work! I am an independent developer and have written several websites, but the traffic is average! I envy you, your website has more traffic than mine! Regarding registration, why doesn’t it support google login?

👋 Hey Product Hunt!

I'm Sina, founder of PandaProbe.

A while back we launched the open-source version here — the response was incredible. Today we're back with what many of you asked for: PandaProbe Cloud — full-stack tracing, evals, and monitoring for agents, with zero infrastructure to manage.

Here's a pattern every agent builder knows: you ship, it looks fine in testing, then quietly misbehaves in production — nobody knows why. Once agents start chaining LLMs, tools, APIs, MCPs, and sub-agents, debugging becomes archaeology. Logs tell you something happened — not why, not whether quality regressed, not how the session held together. And solving that shouldn't mean building your own agent engineering stack.

That's PandaProbe Cloud: ship better agents without the ops overhead.

What you get
🔎 Tracing — full agent executions captured as sessions, traces, and spans.
📊 Evaluation — score traces and sessions using SOTA agent-specific metrics.
⏱️ Monitoring — schedule recurring evals to track your agent's health in production.
☁️ Fully managed — we handle the infra. You just connect, ship, and improve.

Who it's for
🧑‍💻 AI engineers debugging agent behavior across LLMs, tools, and workflows.
🏗️ Platform teams monitoring quality and reliability without owning more infra.
🔬 Builders experimenting with agents who want to iterate faster.
🚀 Startups who want production-grade observability from day one.

Quickstart:

☁️ Cloud signup: https://app.pandaprobe.com/

🤖 Run: npx skills add chirpz-ai/pandaprobe-skills --skill '*' --yes

💥 Then ask your coding agent to "set up PandaProbe".

Free to start — generous usage credits. Up and running in minutes.

Quick links
📖 Docs: https://docs.pandaprobe.com
⭐ Open source: https://github.com/chirpz-ai/pandaprobe

I'll be here all day — drop your questions and feedback below.

Thanks for checking it out 🙏
— Sina

Hey @sina_tayebati ,
qq if we start on cloud but need to migrate back to self-hosted open source later because of data residency laws, is the data schema 100% compatible?

Just curios how is it different from langfuse/smith?

Bundling tracing, evals, and monitoring into a single managed layer is a sharp call. Teams building agent pipelines typically end up with fragile homegrown span logging that breaks the moment they chain subagents. The hardest part of multi-agent workflows isn't inference: it's reconstructing what happened across tool calls when something fails silently. How does trace collection handle async fan-out across spawned subagents?

Congrats on the cloud launch @sina_tayebati

The open-source version saved me during a nasty multi-agent debug a few months back, so this is exciting. Does the eval scoring work for custom agent frameworks or is it tied to specific SDKs?

I like the focus on making agent debugging and monitoring easier. As agent workflows become more complex, having better visibility into what's happening behind the scenes is incredibly valuable.

The fully managed approach is a big plus too.

"debugging becomes archaeology" lol yeah. once one action turns into 5 tool calls and a couple sub-agents the logs just tell you the order stuff happened, not why it went wrong

how does the eval side deal with non-determinism though? same input gives slightly different traces every run, so how do you score regressions without drowning in false positives

This is so cool, congrats!

For MCP-heavy agents, I’m curious how session grouping works in Cloud. Do MCP tool calls get attached to one session timeline automatically, or is that only reliable if I pass the same session_id through each layer?

Agent observability gets messy once the same user task spans tools + retries, so the session/traces/spans framing makes sense. One thing I'd check before wiring this into prod: can evals be versioned against both prompt/model changes and tool schema changes? Otherwise a regression dashboard can get a little misleading after an MCP/API update.

Congrats on the Cloud launch. The session-as-the-unit framing feels right for agents, because individual traces are not enough once tools, MCP calls, and subagents start branching.

One edge case I’d love to understand: if an agent fans out into parallel subagents and some tool calls continue after the parent task has already moved on, how does PandaProbe decide what still belongs to the same session timeline? Is it based on explicit session IDs, automatic propagation, or both?

Most agent failures happen silently in production how does PandaProbe differentiate between a hallucination a tool call failure and a reasoning breakdown when surfacing what actually went wrong in a trace?

Looks great

Does this actually trace inside MCP tool calls or just log that they were triggered?

That's always been the blind spot for me with most observability tools

Been waiting for something like this honestly.
Quick question, the agent evals, are those pre-built metrics or can you define what "good" looks like for your own use case?

"Debugging becomes archaeology" is painfully accurate once subagents and tool calls start chaining. Making the session (not the trace) the unit of analysis is a smart angle for multi-step agents. Congrats on the Cloud launch! How generous are the free tier credits to start experimenting?

Hello ProductHunt Community,

this launch has been the biggest thing we have built to date. Generating beautiful visuals from your shipped work without you having to write a single prompt.

For this to work as smooth as possible we reverse engineered Figma and built our own codec to allow you to paste the output directly into Figma or Paper and make changes to it before posting.

I hope you can give it a try, give us valuable feedback and share what you think!

Interesting and cool idea! One thing I am really missing is some examples. I see you have a standard change log on your website, but I don't see anything about the image generation in particular. I also wonder if those images include only texts or also images/screenshots?

congrats on the launch!!

i have been using notra for my projects beacuse it made writing changelogs/blog posts sooooo easy and now this new image gen feature is just what ive been missing

great launch and thanks for the beta access! :)

This is great! Congrats on building and launching. Excited to start using it at Emdash. Are you considering to build motion graphics in the future as well?

Turning merged PRs into launch content is a very real founder pain. The “in your voice” piece matters more than raw generation here. Are you pulling context from PR titles only, or also diffs, issues, and changelog history?

Congrats on the launch @dominikkoch! Honestly image generation like this quality is going to be such a time saver for us to actually commit to promoting everything our audience finds useful to know about when we ship at Inth.com and c15t.com. It's been a joy to see Dom work behind the scenes on this.

P.s thanks for using c15t as one of the guinea pig brands you were testing on!

I like that Notra starts from shipped work instead of asking the team to write another prompt or rebuild the story manually.

The Figma/Paper paste workflow sounds especially interesting. Curious how much control teams have over the tone and format of the generated outputs, especially if they want changelogs to feel more human and less like release notes.

Really useful direction. :)

Congrats on the launch; great product Dominik :D

Do you think that Code should follow brand images, or brand images follow code? Cuz it feels like code almost doesn't do "as good as a job" at capturing a brand's image as just another image does.

Any examples?

Turning merged PRs into on-brand visuals is such a smart use of work that's already happening. Love that you reverse-engineered Figma so teams can edit the output before posting — that control is key. Congrats on the launch! Any plans to extend this to motion graphics or short video down the line?

Good one! The thing I'd want to stress-test is sites that render client-side or lazy-load on scroll. Does the capture wait for the DOM to settle? Do you end up grabbing half-loaded components?

The local capture point is the part I'd trust most here. In PM-to-dev handoff, the hard bit is usually keeping the mock close to the existing design system instead of generating a pretty but impossible screen. Curious how strict the asset reuse is — if I ask for a new component, does it prefer composing from captured tokens/components before inventing new CSS?

"A paragraph and a prayer" perfectly captures the PM-to-dev handoff pain — turning a live page into an editable mockup is a great fix. Love that it's local-first and open source too. Congrats on the launch! Are you thinking about adding annotations or a diff view to make handoff even cleaner?

This sounds super interesting especially grabbing a sites design features, colours, fonts, icons etc... sounds like a much easier way than sending over a complete css file

The "no Figma needed" angle is smart – designers will use it anyway, but this opens it up to everyone else. How does it handle sites with heavy animations or scroll effects?

The live-site-to-editable-mockup flow is a great bridge for non-designers and fast product iteration. Auto-extracting typography, colors, and components is the practical bit. How clean is the exported HTML on complex responsive pages?

The PM-to-dev handoff use case is very real. A quick editable mockup from the actual live page would remove a lot of vague screenshot feedback and make small UI changes much easier to explain.

Bad and good news:

Figma is releasing their chrome extension (have you heard?) which seems to do similar as MockPilot, except just exports to Figma.

Bad news: competitor.

Good news: validation??

Anyways, congrats on the launch :)

This look good especially for demo presentation, I have a question, can't it be developed fully without the hassle of going through all these processes ?

The asset library extraction is the part that grabs me, pulling a live site's typography and colors into something reusable so edits stay on brand is really handy. Curious whether the exported HTML comes out as semantic markup or mostly positioned divs. Congrats on shipping.

This is a great product, i particularly love the UI and UX, i found it easy to use and to set things up fast

building specifically for MENA instead of trying to be "global" is the right move. every sales tool I've used is built for the US market first and everywhere else is an afterthought. the WhatsApp integration is smart too since that's actually how business gets done in most of the world outside the US. Arabic dialect awareness across Gulf, Levantine, and Egyptian is a nice touch

Congrats on the launch! It's interesting to see a sales platform built specifically for MENA instead of treating the region as an afterthought.

Hey Product Hunt! Excited to share Sulsaly with you today. The idea came from a frustration I kept hearing from founders and sales teams across the Arab world: the tools everyone recommends ( Apollo, ZoomInfo, Lusha ) were built for the US and European markets. They have almost no coverage for MENA contacts, don't support Arabic, and completely ignore WhatsApp, which is how business actually gets done in the Gulf, Levant, and Egypt. The problem is simple: a sales rep in Riyadh or Cairo has no systematic way to find decision-makers, get their WhatsApp numbers, and reach out in Arabic at scale. Everyone is either buying expensive Western tools that don't work here, or doing everything manually. Sulsaly fixes this in 5 clicks: find your ideal prospect in MENA, get their verified WhatsApp and LinkedIn, generate a personalised Arabic message with AI, and send it, all from one platform. We built it specifically for the Arab world: Arabic UI, Arabic AI-generated messages, Gulf/Levantine/Egyptian dialect awareness, and a contact database that actually covers the region. Would love to hear from anyone doing sales in MENA, what's your biggest challenge right now?

Building MENA-first instead of treating the region as an afterthought is a smart wedge — and leaning into WhatsApp + Arabic dialect awareness is exactly what Western tools miss. Congrats on the launch! How do you keep the contact database fresh and avoid stale numbers?

What I'm curious about is the AI-generated Arabic messaging. Getting the language right is one thing, but getting the tone, dialect, and cultural nuance right is much harder. Have you seen strong reply rates from actual buyers?

How are you verifying decision makers and keeping numbers up to date without burning people with stale contact data?

Most sales intelligence platforms claim global coverage, but MENA has always felt underserved. If Sulsaly can consistantly deliver high-quality decision-maker data in Saudi Arabia and Egypt, that alone could be a huge advantage.

The regional focus is smart. Sales tooling often underestimates local channels and data quality, especially WhatsApp/LinkedIn workflows in MENA. How do you keep verification fresh across the contact database?

Privacy-friendly analytics for makers is a good lane, especially if the dashboard stays fast and understandable. Funnels + journeys + AI insights sounds useful. How do you balance AI summaries with keeping the raw metrics transparent?

Nice launch, Khalid 👏

I run a small digital agency and we ship most of our client sites in Next.js, so a lightweight, privacy-first tracker is exactly the kind of thing I keep looking for — GA always feels like overkill and the cookie-consent overhead is a pain to justify to clients. A clean script that just gives me visitors, sources and funnels without dragging down performance would be an easy sell.

Two questions before I try it:

• How heavy is the tracking script, and does it play well with a Next.js App Router setup / SPA-style route changes?

• Is the data EU-hosted, and is it GDPR-compliant out of the box (no cookie banner needed)? That's a real selling point for the clients I work with.

Will give it a spin on one of our sites. Congrats on shipping 🚀

Congrats, Khalid! Building analytics from a founder's perspective often leads to a much better user experience than enterprise-first solutions.

The privacy-friendly approach is a huge plus. More founders are looking for actionable insights without compromising visitor privacy.

I want to ask for those who are just building their app and still in validation stage, does the performance metric include these categories that I mentioned earlier ?

Congrats on the launch! The product looks really clean and I like the focus on making analytics easier to understand.

I'm curious, what made you decide to build Reignat when there are already so many analytics tools out there? Was there a specific problem you kept running into?

Congrats on the launch. Privacy-friendly analytics is a very real need for indie sites, especially when GA feels too heavy. One thing I’d be curious about: how do your AI insights stay tied to the raw metrics? For makers, the most useful flow might be “traffic source changed → affected pages → likely reason → suggested experiment”, while still letting us verify every step from the underlying data.

Privacy-friendly analytics built specifically for makers is a great lane — GA really does feel like overkill for most indie projects, and skipping the cookie banner is a real selling point. Congrats on the launch! How do you balance the AI-powered insights with keeping the raw metrics transparent and trustworthy?

Structured guidance beats a blank chat box for founder workflows. The decision memory + “kill weak ideas early” angle is useful if it stays honest. How does Fonda push back when a founder keeps steering toward an invalidated idea?

That "11pm when the big calls can't wait" line got me, that's literally my build window once the kids are down. Building solo, the thing I miss isn't another idea, it's someone who'll tell me an idea is weak before I sink three weekends into it. My honest worry: most AI leans agreeable, so how do you keep Fonda from just nodding along with whatever the founder already wants to build?

I really like the idea of having a structured guide for founders instead of just another chatbot. Starting a business can feel overwhelming, especially when you're doing it alone.

The focus on giving one clear next step at a time sounds particularly helpful. Sometimes that's exactly what founders need to keep moving forward.

I already have a rough idea, so the main question for me is where Fonda lets you enter the journey. Can you bring your own idea and start around Validate, or does every project need to begin with the profile and matched-opportunity steps?

Congrats on the launch! I like the idea of focusing on the next actionable step instead of overwhelming founders with endless advice.

A structured 14-step journey with decision memory is a much better fit for founders than yet another blank chat box. The "honest when your idea is weak" angle is the part that really matters. Congrats on the launch! How do you keep Fonda from just agreeing with whatever the founder already wants to build?

I built Synopsule because every meeting tool I tried wanted my audio in their cloud,
an account, and a bot in the call. I wanted the transcript without the trade.

So the whole pipeline runs on the device that recorded it:
- Whisper transcribes locally. The model ships in the app, so it works offline on first launch.
- On the Mac it captures both sides of a Zoom/Meet/Teams call by tapping system audio
directly. No bot, no screen recording, no virtual driver.
- Speaker recognition runs on-device too. It labels who said what, remembers returning
voices across recordings, and learns from your corrections. Raw audio is deleted once
the transcript exists.

Includes MCP functionality so you can integrate your meeting transcripts with Claude, Codex, and ChatGPT.

Summaries are the one thing that can touch the cloud, and only if you ask. Bring your
own key, or run a local model on the Mac so even that stays on your machine. Only
transcript text ever leaves, only when you choose.

It's $1.99 once for both Mac and iPhone. Transcription shouldn't be a subscription.

The hardest part was diarization on long, messy recordings, and making speaker
recognition improve locally without a cloud model.

I'd love feedback, and I'm here all day. What would it take for you to trust an
on-device transcriber over the cloud one you use now?

Nice, so it works fully offline with Whisper for transcription and some sort of a local Apple LLM?

Fully on-device with no bot in the call and audio that never leaves the machine is exactly the privacy stance meeting tools have been missing. A one-time purchase instead of a subscription is a nice bonus. Congrats on the launch! How does the on-device speaker labeling hold up on long calls when people talk over each other?

Hey Sunny, this is indeed very welcome. I am currently with Granola and has tonnes of meetings captured there. Is there a way to migrate them to Synopsule?

Love the product positioning. Many transcription apps focus on AI summaries first, while you're focusing on accurate transcripts and letting users decide what happens next.

No bot in the call is the detail that stands out to me. Getting transcripts from Zoom/Meet/Teams without uploading audio or adding another participant feels much cleaner for real meetings.

Does speaker labeling work offline too, or does it need extra processing somewhere?

The on-device, no-upload stance is the part I keep coming back to. I work on voice AI for aging-in-place, and the privacy bar gets brutal once you are handling conversations that involve someone's parents or their health. Running Whisper locally so the audio never leaves the machine is a real answer to that. How is the on-device speaker recognition holding up with overlapping talkers or weaker mics? That is usually where local models start to struggle for us.

on-device is the version people want once they think about who's in the room. the part demos skip is diarization holding across a 60-min call when people talk over each other — no cloud gpu locally.

On-device transcription with summaries only on request is a refreshing privacy stance for meeting tools. The one-time Mac+iPhone purchase also stands out. How well does speaker labeling hold up in noisy multi-person conversations?

Hey Product Hunt 👋

We built AEVS at Fetch AI because as AI agents start doing real work, it’s becoming harder to answer a simple but important question:

What did the agent actually execute? 🤔

Chat history can tell you what an agent said. Logs can help, but they’re often scattered, incomplete, or easy to lose. For agents that call tools, trigger workflows, move data, or interact with external systems, teams need something stronger: a verifiable record of execution ✅

That’s what AEVS provides.

AEVS is a drop-in SDK that records every AI agent tool call as a tamper-evident execution receipt

Each receipt captures the tool used, inputs, outputs, status, and timing, so teams can verify what an agent actually did without relying only on chat history or fragile logs.

The goal is to make agent execution more trustworthy, auditable, and production-ready 🔐

It’s designed for builders working with agent frameworks like LangChain/LangGraph, MCP tools, and custom agent stacks 🛠️

You can add AEVS to your existing workflow and start generating verifiable receipts for tool calls without rebuilding your agent from scratch 🚀

We’re especially interested in use cases where agents perform actions that matter: customer support workflows, financial operations, internal tools, compliance-sensitive tasks, data pipelines, and API automation

I’d love feedback from agent builders, infra teams, and anyone thinking about trust and verification for AI systems 🙌

One thing we're particularly excited about is how easy AEVS is to adopt.

✅ ~2 lines of code to integrate

✅ Supports LangChain, LangGraph, and MCP today

✅ CrewAI support coming soon

You can even tell your coding agent:

Read https://aevs.fetch.ai/llms.txt and add AEVS to my agent

and it can help integrate AEVS directly into your workflow.

Our goal was simple: make verifiable agent execution accessible without forcing developers to redesign their existing stack.

The receipt model makes sense for agent workflows where a tool call can change customer state, like support ops or billing automation, and chat logs are not enough. Capturing inputs, outputs, status, and timing as a receipt feels useful, but the edge case I’d worry about is sensitive payloads. Can teams redact specific fields before the receipt is written while still keeping the proof tamper-evident?

This resonates. I ship an MCP server for ad-account monitoring, and the audit log of every tool call (tool, sanitized params, latency, result) turned out to be the feature people trusted most, especially once agents could trigger actions on live accounts. The piece I'd add: receipts answer "what happened," but for actions that move money or change customer state, the bigger win is gating the write before it executes, not just proving it after. Receipt plus an approval step is the real combo. Are you thinking about a pre-execution hook, or is AEVS purely the record after the call fires?

Congrats on the launch! As agents become more autonomous, knowing what they actually did becomes just as important as what they said. The execution receipt concept is really interesting.

Congratulations on today's launch.

@mehul_g001 Interesting idea, but I wonder if builders are solving for a problem that only appears after agents reach a certain scale.

Most teams I know are still struggling to get agents to work reliably.

At what point do customers start saying “I need verifiable execution receipts” instead of “I just need my agent to stop breaking”?

Would love to know what your early adopters look like.

@mehul_g001 The concept makes sense for high-trust environments, but I’m curious where the boundary is between observability and proof.

If an agent calls the correct tool but submits the wrong parameters, or reaches the wrong conclusion based on valid execution, does AEVS help teams understand why the decision was made, or only prove that it happened?

It feels like the next challenge for AI systems isn’t just verifying execution, but verifying intent and reasoning as well.

This hits a real trust gap for agents: chat history is not an execution log. Tamper-evident receipts for tool calls could become table stakes. Are receipts designed to be portable across agent frameworks, or mostly SDK-local at launch?

Can AEVS verify why an agent made a decision, or only what action it executed?

So ive opened it in the background, will it ping me or send me a notification when a plane travels above me? BTW nice one.

The postcode-first approach is what makes this click. Instead of opening a full flight tracker, you get a simple answer to what’s above you right now, which feels much more casual and fun.

@chris_ward9 Excellent work on this. Every time I spotted a plane flying above, I would always say to myself - "Wish I could know what flight that is." I can do that now.

Sending this to my friend who always wanted to be a pilot and nerds over this. Thanks for building!

The retro radar aesthetic is a perfect fit for this. Way more fun than a boring list of flights.

Love this. As a pilot, history buff, and aviation geek, I totally nerd out on this kind of stuff. Very cool UX.

Hi everyone!

MiMo Code is based on @opencode, but puts more effort into handling long-running sessions.

Instead of letting the main agent manage its own memory, it uses a separate writer subagent that periodically writes structured checkpoints on a schedule.

A key design choice is that these checkpoints are triggered relatively early, well before the context window fills up. This avoids having to do heavy compression under high context pressure.

When the window eventually approaches its limit, the system rebuilds the working context from the accumulated checkpoints and project memory, rather than relying on increasingly unreliable summarization.

It also runs background processes to extract reusable patterns from past work over time.

The early checkpoint idea makes sense: summarize while the context is still healthy, not when the window is already under pressure.

The hard part I’d worry about is negative memory. If a checkpoint captures a wrong assumption during a long refactor, how does MiMo Code prevent that from being promoted into durable project memory? Do you keep task-local checkpoints separate from reusable MEMORY.md-style knowledge until a later validation pass?

Interesting angle. For agent memory, the thing I'd want to see is how it stays reviewable: what gets remembered, what can be reset, and how a dev can tell when old context is hurting more than helping.

The checkpoint subagent idea is the most interesting part to me. Long-horizon coding agents fall apart when context management is implicit. How are you deciding what gets written into long-term memory versus kept as task-local state?

Long coding sessions usually fall apart when context gets compressed too late. The separate checkpoint subagent feels like a practical fix: less magic, more reliable memory when the task stretches for hours.

Long-term memory in a coding agent is the hard part. When a function gets refactored acroos 10 files, does MiMo-Code update its stored understanding automatically, or do you need to manually invalidate stale context?

The separate writer subagent for periodic checkpointing is clever. Triggering early before context pressure builds means you're compressing while you still have bandwidth, not scrambling when the window's nearly full. The naive approach of letting the main agent summarize itself tends to fail exactly when you need it most. When the background process extracts reusable patterns, how do you filter signal from noise across sessions?

the memory architecture is the part that matters. every AI tool I've used loses everything the moment you close the session and you start from scratch every time. the checkpoint approach where it saves what it learned before the context fills up instead of scrambling to compress after is a smarter design. curious how this scales when projects get really large

Wrong memory seems more dangerous than no memory on long refactors. If a checkpoint records a bad assumption, can I mark it as wrong before it gets promoted into MEMORY.md, or is cleanup left to the later Dream pass?

Hey Product Hunt 👋

We built Dropmatico because we were tired of the hourly ritual. Open a marketing image, resize for Instagram, save as. Resize for LinkedIn, save as. Resize for X. Resize for our Amazon listing. Resize for Etsy. Multiple exports, multiple chances to mess up a dimension or format, every single time we posted.

So we built the tool we wished existed.

Drop one image, and Dropmatico exports it ready for every platform you post on... 90+ presets across 34 platforms, including the marketplaces other tools ignore (Amazon, Etsy, Noon, Flipkart, Ajio). Or run a batch of 16 images through one platform's spec. Resize, compress, and format conversion all happen in one pass, on a real-time canvas with unmatched security. Files never touch a server.

A few things we obsessed over:
Real-time canvas: you see the crop as you adjust, no "submit and wait" loop.
Client-side processing: up to 100MB per file, no uploads, no queue.
Bulk Processor: 16 images at a time through any preset.
Format conversion: HEIC, PNG, WebP, AVIF, JPEG, all in the same pass.

A small story we're proud of: last month we shipped "Keep Original Size." A user in Canada wrote in and said, "there was no way to say, just keep what I uploaded." Their photos were already at the right dimensions, they only needed the format flip and the compression. So we added a workflow-level toggle that skips resizing entirely. Shipped 10 days after the email landed.

That's the loop we want to keep running on PH today: tell us what's broken in your image workflow and we'll build it.

A question for the community: what's the one image-prep task that still eats your time every week? We'll be in the comments all day.

Team Dropmatico

The privacy angle is underrated. Keeping files on the user's device instead of uploading them to servers will matter for many businesses. Congrats on the launch!

Good job on this idea, it is definitely a painpoint, at least for me.

I guess I'll be waiting for when the agent plugin comes out (if it does), right?

The marketplace presets are a smart detail. Social sizes are easy to find, but Amazon, Etsy, Noon, Flipkart, and Shopify images are where the prep work usually gets annoying fast.

One image, every platform is a strong promise. I like that you’re treating resizing, compressing, and converting as one workflow rather than separate utilities. Are the presets maintained manually, or updated when platforms change specs?

The client-side processing approach is a huge win. Faster workflows and better privacy are a great combination.

Hey Product Hunt 👋

I'm Mark, creator of Verol.

I built this tool because I got tired of manually copy pasting AI answers into Google one sentence at a time to check if they were real. It was fine for a couple of days, but it quickly became exhausting and I still couldn't trust half of what came back.

Verol fixes this by sitting right on top of the AI chats you already use. Instead of asking the model "are you sure?" (which usually just triggers another confident, hallucinated paragraph), Verol extracts individual claims and passes them through an independent backend pipeline with live web lookups.

It currently works on ChatGPT, Claude, and Gemini. And soon this list will increase. You get a clear trust score, per claim verdicts, and actual, clickable source links. Plus, your verification history never leaves your browser it stays completely local in your browser.

Out of curiosity, I also ran a small benchmark (100 identical prompts across 3 models) to see how they stack up. It’s not a formal scientific study, but the charts are in the gallery if you want to see how often confident-sounding answers fall apart under a live source check.

There's a free tier with 5 verifications so you can test it on your own workflows. I’d love your honest feedback, especially if you rely on LLMs for research, coding, or anything where a subtle hallucination completely ruins your day.

I'll be here to answer questions all day. Thanks for checking it out! 🙏

One thing I appreciate in verification tools is when they can honestly say "we couldn't verify this" instead of forcing a verdict. Admitting uncertainty is often more useful than filling the gap with a confident-looking citation that doesn't actually support the claim.

Congrats on the launch!

The plausible-sounding hallucinations are the real danger, and they're the hardest to catch because nothing about the answer looks wrong. I work on the data side of this, feeding models pre-verified facts so they don't have to guess, but that only helps when you control the source. For everything else, a verification layer like this is the right shape. The honest "couldn't find a source, here's my confidence" behavior is the part that earns trust. Asking the model to check itself just burns credits and produces another confident paragraph, like Luke said. One question on the real-time mode: verifying every response live sounds great but also expensive in latency and lookups. Do you let people scope it to certain chats or claim types so it isn't running on every throwaway message?

Independent verification layer is the right framing. Asking an LLM to grade itself is a dead end, so source-backed verdicts and confidence metrics are valuable. How do you handle claims that require paid/logged-in sources or rapidly changing data?

Honestly the part that worries me as a non-coder isn't the obvious hallucinations, it's the plausible-sounding ones I'd never think to question. When I lean on AI to research a post or sanity-check an idea, I'm just... taking its word, because I can't always check the source myself. What does Verol do when a claim isn't really web-checkable, or when the sources it pulls disagree with each other?

This is great! @mark_prod I would love to try this out soon. AI hallucinations has been the bane of my existence lately, just 2 days ago, I was working on a project that required a lot of research, I caught it hallucinating, pointed it out and it immediately changed it's direction. This got me worried if all of my previous research had been wrong and I had to start validating them all. So great job! This is really a problem I'm sure a lot of people face.

I'm curious to know though, what triggers Verol? Does a kind of kinds of question trigger it? Or does it work on the background, validating very response the AI chat gives?

Also, what does it validate against?

The "learns from every call, with owner approval" detail is what stood out to me. I build voice AI in the aging-in-place space, and that human-in-the-loop approval step is exactly what earns trust when an agent is acting on someone's behalf. Curious how Relay handles a caller asking something the business facts don't cover: does it defer to a human, or attempt a best guess?

I'm sure dentists and med spas are going to love this. Missed calls are their biggest revenue leak and they know it. Does it handle appointment rescheduling too or just new bookings?

AI receptionist is a crowded lane right now. The "answers only from the business's own facts" constraint is the part I'd want to see hold up, since that's usually where these break. What's the differentiation you're leading with against the others launching this quarter?

Turning a website into a receptionist is a clean onboarding idea. The owner-approval loop for learning from calls is smart too. How do you prevent the agent from answering beyond the business’s known f

Turning a website into a receptionist is a clean onboarding idea. The owner-approval loop for learning from calls is smart too. How do you prevent the agent from answering beyond the business’s known facts?

I'm in Berlin tried selling this Voice Agents for local business but I don't speak German....so when I visit them I can't sell them.

you think I could easily build one of this to different businesses?

And do you think I could build one Agent to explain each business what is this about, so I don't need to talk german?

This feels made for people who treat photos like little memories, not just uploads. The frame styles give everyday shots more personality without turning the app into a heavy editor.

Hey Product Hunt!

Email marketing has always forced a bad trade: either pay a designer and copywriter for every campaign, or settle for a clunky drag-and-drop editor and emails that look off-brand. And the "free" plans? Usually a 14-day countdown in disguise.

We re-built EmailFlow.AI to kill both problems. Describe the email or automation you want in plain English, and it generates a responsive, on-brand design in seconds.

Connect your website and our AI agent learns your logo, colors, and voice so the first draft already looks like you, not a template.

Under the hood you get the full stack: a 12,000+ template gallery, visual multi-step automations, and managed sending so deliverability is handled (just verify your domain).

We'd love your feedback on the AI builder and brand scan especially. What would make this a no-brainer for your sending?

The site scan for brand capture is a nice touch – that's usually the part that kills the "just describe it" workflow. How accurate is it on sites with inconsistent branding?

This hits a real gap for me. I've been piecing together email marketing with AI design tools and the handoff is never seamless. The scan-your-site-to-capture-brand step is the part I'm most curious about. Going to try it.

The text-to-email workflow is a strong hook because email design is where a lot of campaigns slow down. Curious how much control users get after generation: can they lock brand rules, reusable sections, and tone before sending?

For the site scan, how well does it pick up brand voice, like does it pull from blog copy or just visual style?

Open-source terminal-first agents are where a lot of real developer workflow will shake out. BYOK + provider flexibility is a good default. What does VEXI do to keep long-running refactors inspectable before changes hit the filesystem?