PH热榜 | 2026-06-11

@chloesamaha definitely feeling this pain, is there a way for my more manual tasks to get done by bond?

@chloesamaha Congrats!! :)

@chloesamaha Congrats on the launch team. How do you handle exec inaction (I know about it and just haven't done anything and so task age is increasing) and task priority?

@martijn_bonte yess!

@martijn_bonte In the era of AI agents, aren't we all becoming executives? 🤔

@martijn_bonte Definitely. Right now we're focussed on founders and executives just because they have the pain-point more than anyone else. But we're definitely rolling this out enterprise very soon :)

@jan_willem_denys early-stage is honestly where it shines at the moment. When you're wearing 10 hats nobody's catching the dropped balls but you...Bond does that part. Keeps the few things that matter today up top and handles the chasing/follow-ups in the background. Works whether your team is 2 or 200.

@jan_willem_denys Thankss Jan-Willem, lovely question.
While larger teams get a lot of value from Bond, we actually built it with the belief that early stage founders need a chief of staff even more!!

Many of our earliest users are founders with very small teams, who use Bond as an AI chief of staff before they can justify hiring one.

@german_merlo1 Appreciate the support Germán!

@german_merlo1 appreciate the comment. Every high performer deals with the daily anxiety of not wanting to be the bottleneck, scared of forgetting something. Now they have Bond that remembers everything so they don't have to.

@michael_jankie Thanks a ton Michael!

@michael_jankie really happy to have you as an early adopter!

@matthiasrossini Haha thanks! We're actively looking into WhatsApp, though Meta isn't exactly famous for friendly integrations lol. We'll get there.

Honestly we dogfood so hard at this point that we barely talk to each other internally anymore. It's just our individual Bonds talking to each other now. #culture

@michal_giernatowski Bond surfaces things based on timing, context, and relationships. E.g. A calendar event coming up triggers prep; a message from a key contact flags for a reply, and drafting kicks in when the action is clear but needs your voice (a reply, a follow-up). Autonomous actions are reserved for low-risk, high-confidence tasks where intent is clear. Anything touching the outside world (like sending an email) goes through an approval step first. The exec stays in control; Bond just removes the friction.

@michal_giernatowski Great question, because getting this dial wrong in either direction kills trust fast.

The way I think about it is a spectrum based on stakes and confidence:

• Surface when it's something you need to know but only you can judge: a risk, a decision waiting on you, a thread heating up. Bond brings it to you, doesn't touch it.

• Draft when there's a clear action but it needs your voice or judgment: a follow-up email, a nudge to a teammate, a reply. Bond prepares it and hands it over for a yes/tweak/no.

• Act for the safe, reversible, low-stakes stuff: organizing, cross-referencing, marking things done when it sees real completion signals, prepping context. The plumbing, basically.

@anshchopra thank you Ansh!

@nicklinck yess!! When Bond has context on the whole team it naturally starts connecting the dots: who's blocked, what's dependent on what, where things are slipping. Game changer for async teams.

@nicklinck Let's make it happen my guy

@millwiller Great question! It's actually both. Bond uses a combination of rule-based signals (recency, source, sender relationship) and strategic context, like your stated goals, priorities, and responsibilities, to determine urgency. So a cold email from a stranger and a message from a key customer about a blocker will score very differently, even if they arrived at the same time. The due dates reflect that combined signal.

@chloesamaha  @millwiller It's much closer to the second one, and that contrast is exactly the right way to frame it.

The rigid version ("all emails get a 48h due date") is what makes most tools useless, because it treats a note from your biggest customer the same as a newsletter. Bond doesn't do flat rules like that. It scores each item on things like importance, urgency, and effort, and weighs that against who it's from, what it blocks, and your actual goals and priorities. So a task that moves the needle on a key objective outranks something that's merely due soon.

Urgency does evolve with time, so genuinely time-sensitive things climb as deadlines approach, but time-pressure is one input, not the whole formula. The goal is to sort by impact, not just by clock. A truly time-bound thing rises, but it never lets "soon" masquerade as "important."

Does that make sense?

@sanreds Thank you, sanreds ! Great questions.

1. We don't see Bond as replacing one specific tool. It sits above the tools you already use and helps turn all that scattered context into a self-managing todo layer. So less "another app to maintain," more "an orchestration layer across the stack."

2. The ROI is mostly time and follow-through: fewer missed follow-ups, less context reconstruction, faster delegation, and less chief-of-staff/EA busywork. For execs, even saving a few hours a week can justify the cost, but the bigger value is often avoiding the one thing that would have slipped through the cracks and ofc staying focused on the right company goals is gold.

3. Today we're most focused on founders, executives, chiefs of staff, and lean leadership teams where the coordination load is already painful. That tends to be scaling startups and SMBs more than huge enterprises right now, though the product naturally becomes more valuable as the org gets more complex.

Hope this helps 😊

@ding_hao Definitely augmentation.
A lot of our users actually are chiefs of staff, they're brilliant, but also quite expensive...
Bond takes the heavy, repetitive layer off their plate (tracking what's slipping, surfacing blockers, drafting routine follow-ups) so they can spend their time on the high-judgment work that actually needs a human.

So it works alongside existing EA and chief-of-staff workflows, not instead of them, the human stays in the driver's seat and keeps the final say. 🏎️
Our moonshot is for Bond to clear more and more of the routine load over time, while the person running it always controls how much.
Personally, I'll always believe in empowering people with AI, not replacing them. 🌞

@xie_yujin Haha hear me out: there's an inverse correlation between someone's seniority and their screen size. Juniors run three ultrawides like a NASA control room, while actual execs are making million-dollar calls from their phone in the back of an Uber.

So yes, mobile matters a lot. We focused on desktop first to get the core right, but there's already a fully functional mobile PWA today. Making that buttery smooth is next on my personal priority list, because that's where most of our users actually live.

@xie_yujin Yes, so currently Bond is a progressive web app, you can add it to your home screen and it runs like a native app (full-screen, app icon, shortcuts to Chat / Todos / Replies).
On the go you can also use BondBot in Slack on your phone, plus push notifications for briefings and routines. But don't worry a truly native app is on the roadmap, and we already have some great UX ideas 🔥.

@carlos_leonardo1 This is the exact tension we designed around.

There is one shared company brain. That's what makes Bond more than a personal assistant. But access to it is scoped per person, so you only ever see through it what you're already allowed to see. Your DMs, private channels, and personal email feed your view and only yours. A colleague's Bond can't pull them, and yours can't pull theirs.

It's enforced at the data layer, not left to the AI to decide. Just like a teammate can't open a Slack DM they're not part of, their Bond literally can't return it either. Sensitive sources get flagged private the moment they're ingested, and every insight carries a source citation so nothing ever shows up "from nowhere." So: shared brain, shared context, but strict per-person walls on what each person can actually see.

Is the bigger worry for you something sensitive of yours leaking to your team, or their private context leaking to you?

@annemarie_leys Love this question.

The blunt version: we won't stop until every executive is unemployed from their own to-do list.

The thing we're building toward is you waking up to a single message from Bond that says: "you picked up 10 new todos since you left yesterday. I already handled 6 of them myself and delegated the other 4 to your team. You're all good." That's the whole dream. Not a dashboard you check, but a chief of staff that just runs the back office of your life while you sleep.

And we're honestly very, very close to that. More than people expect.

So to your actual question: it's both, but the order matters. We go deeper on the core workflows first (discovery, prioritization, delegation, follow-through) because that's where trust is won or lost, and then broader across the whole stack once the foundation is rock solid. Depth earns the right to go wide :)

@kutlwano_melamu Thank you Kutlwano 🚀
Social media integrations aren’t currently supported, but they’re definitely on our roadmap. We think they’re especially interesting for founders and creators who have a lot of important context living in DMs and comments. Stay tuned, lots more integrations are coming. 🏄

@gugan_ananth "reducing background cognitive tax" — that's exactly it. Time saved is nice, but knowing your not missing anything is such a blessing

@gugan_ananth  Exactly! We all want our time back, because wasn’t that what AI promised us in the beginning?

@marie_bontinck Honestly, the thing that surprised early users most was that BOND tracks commitments they made, even when they were buried in messy notes, transcripts, or events.
It helps executives avoid dropping any balls. 🎾

People expected help with email and calendar triage, but the “oh, wow” moment was BOND resurfacing a specific piece of work they had forgotten was still on their plate, with the source quote or context to back it up.

@can_erden Thanks Can! 🙌
Yes, Bond can connect to custom workflows via MCP today! and API based integrations are on our roadmap.

For your co-founder setup, absolutely. Each person can have their own tools, workflows, and context tailored to their role while operating within the same organization. So don't hesitate to try it out 😇

@nishikant_takawale the product team really is customer obsessed. Genuinely want to deliver an 11-star to do list experience!

@workout097_collab great question!! Avoiding "noisy aggregator" is the whole product. Our approach is to filter with context, not keywords. Bond pre-links every person, meeting, and responsibility before messages arrive, then nothing becomes a to-do unless it survives two checks: is this actually a commitment for YOU (not just your name near a verb), and does it matter vs. everything else on your plate? When it misses, your corrections stick. An aggregator of 10 tools' noise is worse than the 10 tools — Bond only deserves to exist if it's the opposite 🙏

@workout097_collab Honestly, this is the exact thing we position ourselves against.

There are a ton of AI assistants out there and they almost all fail at the same place: signal-to-noise. Anyone can one-shot a Pipedream MCP-everything integration, spin up a vector DB, and call it a day. But what you get back is noise with a tiny bit of signal buried in it.

Context is a double-edged sword. It's what creates the noise, but it's also the only place the real signal lives. That's why we threw all our effort into the company brain, an engine whose whole job is knowing what's actually real vs. what isn't.

So Bond is precision-biased on purpose. It filters out things like FYI chatter, bot notifications, wrong-owner asks, and figures out what's genuinely waiting on you, and backs every item with a source citation so you can verify it in one click. The bar to show you something is high, but the things that actually matter don't slip through, because the moment the list fills up with noise you stop trusting it, and a list you don't trust is worse than no list at all :)

@moh_codokiai Both, but the second one is where it gets interesting. Bond does pull from your tools (calendar, email, Slack, Linear) but what it's really doing is building a model of you: your relationships, your commitments, what you own, what you've said you'd do. Over time it knows who matters, what's stuck, and what you're avoiding.

Users open it first thing because it's a reliable list of what actually needs to get done today. It removes the anxiety of "did I forget something?", the overwhelm of "I have 100 things, where do I start?", and the 2+ hours wasted bouncing between tools trying to build a list.

Your morning looks completely different with Bond. Instead of waking up and spending two hours just getting oriented, you open Bond and start getting things done.

@adamcohenhillel thank you adam!

@abu_badr Here's a real example from my week: I ask Bond to pull up every lead I met with and or emailed. Then I say "draft follow-ups for all of them." Bond writes each email in my tone, with the context from the call or interaction, right in the web app chat, and I just edit or approve to send.

A task that would've taken me 2hours now takes me 5'. I no longer need to go digging through calendar events, piecing together call context, scanning sent mail trying to remember who I reached out to.

And honestly? Knowing it only takes five minutes means I actually do it and not procrastinate it (which might be one of the biggest pros of Bond that we don't talk about)

@aronlium Thank you Aron!!

@pavansethi_ Totally fair! Our goal is to get to 0 friction to test :)
But unlike most AI tools, Bond isn't just sitting there waiting for you to prompt it. It's running in the background 24/7, syncing across your entire stack every minute so you never have to.

Most users save 2+ hours a day, so it pays for itself in the first week. The discounted Beta price is $199/month. That's $6.60/day, cheaper than your morning coffee. ☕

@mike_martiez It's $199/month. Unlike most AI tools, Bond isn't just sitting there waiting for you to prompt it. It's syncing across your entire stack every minute in the background, which means real token cost on our end.

Most people save 2+ hours a day though, so it tends to pay for itself in the first week!

@sam_ruedinger Fair point. We're working on lowering that bar as much as possible. In the meantime, the setup is lighter than it looks, takes a few minutes, and starts surfacing value almost immediately. Happy to walk you through it personally if you want to see it work before committing.

@alochukwu Thanks Alochukwu, honestly, it's pretty damn fast. At most, it should take about a day to start seeing the first company insights, and the self-onboarding should already give you a strong feel for the magic. 🪄

@alochukwu Appreciate it!

There's two speeds to it. Bond is useful within minutes of connecting your tools: it maps who's who, which projects are live, and what's moving, so you get real signal almost immediately. Then it gets deeply dialed in over the first couple of days, as it learns the subtler stuff: how decisions actually get made, what you quietly care about, who gets looped in on what.

And it never really "finishes," which is the point. Companies change constantly, so a static model would go stale anyway. It just keeps adapting in the background.

@natalia_iankovych Great question, Natalia 🙏
Phone calls are definitely on our radar, there are a lot of executive-assistant use cases there, from reservations to chasing information, but Bond doesn't make autonomous phone calls yet.

Today we're focused on the highest-volume written workflows first: email, Slack, calendar, meetings, todos, and follow-ups. That's where executives lose the most operational time, and it's also where we can keep the approval loop and audit trail very clear.

On languages: yes, the underlying AI can work across multiple languages, and we already think of Bond as needing to support international teams. The main thing we're careful about is quality and tone especially when Bond is drafting on someone's behalf, so we prefer to expand language support deliberately rather than claim every language is equally polished on day one.

What language would you want us to support ?
Thankss

@fran3cc Congratulations on the launch!

The combination of routing, observability, cost control, and evaluation in a single platform is really compelling.

I'm curious, for teams already using multiple AI providers, what has been the biggest pain point that pushes them to adopt Respan AI Gateway first? Is it reliability, cost visibility, debugging and tracing, or reducing engineering complexity? Also, which feature tends to deliver the fastest ROI after implementation?

Wishing you and the team a very successful launch!

@fran3cc Congrats on the launch! For teams that already have a multi-provider setup, what’s the simplest, lowest-risk way to try Respan Gateway in production what safeguards do you provide to ensure no customer-facing downtime or cost surprises during the transition?

@fran3cc the evals piece is what makes this interesting to me. most teams have routing and fallbacks figured out, but almost nobody has a real answer for "how do I know this model is actually performing well in production" beyond eyeballing logs. curious how you handle eval drift over time as user inputs shift.

@workout097_collab Totally fair. The 2 lines are for getting traffic into the gateway and traces showing up, not pretending production agents are easy.

In real agents, the first things that break are rate limits, long tool-call chains, cost spikes, and not knowing which model/tool/prompt version caused the issue.

@diana_nadim2 Hi,

For non-deterministic outputs, we don’t rely only on exact-match evals. Teams can evaluate outputs with a mix of LLM judges, rubric-based scoring, semantic checks, structured/schema checks, and custom pass/fail criteria depending on the task.

For prompt changes, yes, the goal is to make regression testing easy. You can keep a dataset of representative inputs, run a new prompt or model version against the same cases, compare scores against the previous version, and catch quality, latency, or cost regressions before rolling it out!

@dipanshu_kushwaha5 Totally agree. AI reliability is still one of the hardest parts of putting these products into production.

@1mirul Thank you! That’s exactly the goal.

Teams should be spending their time building better AI experiences, not wiring together gateway logic, traces, evals, monitors, and cost controls from scratch.

@priyankamandal Thanks a lot! We put a lot of focus on making routing and multi-model management as seamless as possible. Glad to hear it’s coming through in Keywords AI. We’re also excited about how the traces and spend-limiting features help teams keep everything under control in real time.

@jimmy_lee12 Absolutely thats a crucial point. Tracing back to the customer and deployment context is often where simple gateway setups fall short. For production incidents, having that full visibility really makes a difference otherwise its tough to pinpoint the root cause. Thats something we’re keeping in mind with Keywords AI, making sure traces carry enough context to be actionable.

@ali_shaheen Totally get what you’re saying. Routing can feel deceptively simple until you start factoring in output quality, speed and cost. Detecting intent accurately and dynamically adjusting to pick the right model is really where it gets tricky. That balance is something we’ve been thinking a lot about with Keywords ai making it smart enough to choose the best model for the task without slowing things down or driving up cost.

Thanks for the comment@hamza_afzal_butt. For niche or low volume keywords, Keywords AI tries to go beyond just raw search data. It looks at semantic relevance, context and related intent to surface opportunities that traditional tools might miss. The idea is to give actionable insights even when volume is low so you can still target terms that have real potential.

Thanks for the support@michelle_marcelline . Appreciated!

@kevin_huang_ynng_ Thanks so much! Glad to hear the routing and spend limiting features are hitting the mark. That gap between a simple API router and full scale LLMops is exactly what we were aiming to solve. Having traces baked in definitely helps keep live traffic manageable and its great to hear itss making a difference on your end.

Thanks for the support@punn_kam . Appreciated!

@borrellbr We started with observability because that’s the first major pain teams hit in production. Once real users are making LLM calls, you need to know what happened, which model was used, why something failed, and where cost or latency is coming from.

Evals became the natural next step because once you have the traces and data, you can do more than just look back manually. You can start checking quality, regressions, and failures proactively.

That’s also why we’re moving toward more self-driving observability. Teams should not have to open the dashboard every day just to find problems. The platform should surface the important issues, run checks, and help teams catch things before they become bigger production problems.

@mikebrandswarm Great question!

Today, we support hard-error fallback, and latency-based failover is in the pipeline. For slow-but-not-erroring providers, we know this is a real production issue, so we’re designing it around configurable latency thresholds and safe handoff behavior.

On cost, it is not reporting-only. We support both soft caps and hard caps. Soft caps can trigger Slack / email alerts, while hard caps can block requests based on the settings you configure per API key, route, or provider.

For evals, we separate eval traffic from production traffic with metadata / tags / environments, so eval prompts can be traced and analyzed without polluting normal usage metrics like customer usage, token volume, latency, or production cost reporting.

@oleksii_sekundant we supports both!

Teams can run evals as a pre-deploy gate before rolling out a new prompt, model, or workflow version. That helps catch regressions before they hit production.

They can also run evals on live traffic samples, so you can monitor quality over real user behavior instead of only testing against static cases.

@sidraarifali That’s exactly the pain we hear from teams. The first version usually works fine, but once real users, providers, prompts, and costs are involved, it gets messy fast.

@andika_fadhilah this matters a lot!

For failovers, there is usually a small retry / routing latency, since the gateway needs to detect the provider issue and move the request to another live provider. But in most cases, the end user usually does not notice much beyond a slightly slower response.

@furkan_kara1 Totally!

On Respan, we log every LLM call, including the model, provider, latency, cost, prompt/version, errors, and traces, so teams can go back and actually analyze the cause instead of guessing from scattered logs.

@farrukh_butt1 Really appreciate that!

@natalia_iankovych Yes, absolutely.

This is one of the main use cases for Respan evals. You can run the same travel-related test cases across your current model and other popular models, then compare response quality, latency, and cost side by side.

Feel free to send me an email, and our team will reach out directly. We’d be happy to help you set up the first eval suite for your travel AI!

@rishi_raj Many congratulations Rishi and team! Excited to hunt Asmi AI today! :)

How I Met the Makers

I met Rishi, co-founder and CEO of Asmi AI, about a month ago. I was genuinely surprised by the product’s actual availability, as I’ve seen many products claim to solve this problem but never truly build it.

In the past few months, several makers approached me trying to build this same solution, but they never delivered anything real.

What Is Asmi AI

Asmi AI is a personal assistant that handles your real-world chores through actual phone calls. Every morning, Asmi calls you; you share your tasks, and it takes care of the rest.

It contacts services like dentists, salons, plumbers, banks, and insurance companies or reaches out to friends and colleagues to coordinate, book, or resolve things.

Once completed, it updates you via WhatsApp or iMessage. Asmi navigates IVRS menus, waits on hold, and handles complex back-and-forth conversations effectively.

Why I Endorse Asmi AI

Rishi and his team come from a credible background. They’ve cracked something broader in consumer AI that we’ve long missed from Product Hunt leaderboards, which are mostly enterprise and business-focused.

I endorse Asmi AI because it addresses real-life pain points, when AI starts taking over our greatest daily burdens, that’s when the true AI revolution begins.

I’m someone who needs to check in frequently with many things and people, and Asmi’s ability to handle IVR calls and back-and-forth conversations is particularly interesting.

Getting updates through WhatsApp and iMessage is incredibly useful, and it truly feels like an assistant I’d want to use daily.

Give it a spin! :)

@rishi_raj Been inside this product for a while now. What keeps striking me is how much the team cares about whether the task actually got done -- not whether we handled it gracefully or gave a smart response, but did the real-world thing happen. Most tools stop at the edge of the digital world. Asmi doesn't. It picks up the phone, waits on hold, navigates the IVR, talks to a real person, and comes back with a result.

My own annoying chore, Asmi helped me with - I'd been putting off calling a specialist doctor for three months. Long hold times, 3-4 month waits, just didn't want to deal with the mental load of planning it out. Mentioned it offhand on a call with Asmi. It set the whole thing up and made the appointment. Still couldn't believe it honestly :D

@artstavenka1 Hi Art - great question! Asmi has access to the world knowledge. Before initiating any task, Asmi will come up with all relevant context required to complete that task at one go. It will first get the context from the user over a quick call or chat. It moves forward to make the call after that. If there is some new info required while it’s talking to bank or others, it will come back to you to over chat, get the context, and immediately close the loop. It is trained to always complete the task, never drop it. Hope it answers your question :)

@kate_ramakaieva Hi Kate - we totally understand your worry. People generally are very responsive to Asmi knowing it’s calling on behalf of a real customer, and there is a real order, booking or issue that needs resolution. Also, the experience is almost similar to talking to a human being given Asmi’s personality, 50+ languages it can speak in - with the authentic accents.

No, we do not copy user’s voice. Asmi has its own personality, voice and a dedicated phone number.

@evakk Thanks Evak!

Absolutely! Asmi can place calls in multiple languages, either explicitly requested by you or smartly deduced from the task context (e.g., using location cues).

This multilingual flexibility ensures Asmi works seamlessly across the globe without any hassle, making it a perfect travel companion for handling local bookings and reservations abroad.

@easton_carter You got it! We are part of people’s daily lives, and many of them now eagerly wait for Asmi’s morning call.

You can call Asmi anytime of the day. Also, for Asmi’a inbound call, timing is totally customisable. In fact, Asmi confirms this explicitly over the very first call it makes to the user.

@abod_rehman Thanks Abdul!

The beauty of Asmi is that it sits right inside iMessage or WhatsApp where you already spend your time, so there’s no new app or context-switching needed.

You can text Asmi between chats, or just call hands-free while driving, on a run, or grabbing groceries. It's always right there when you need it!

@istiakahmad Thanks Istiak, you’ll be happy to hear that calendar integration actually already exists as a silent feature! You can connect it right now to add more context to your tasks.

We deliberately designed Asmi to work powerfully out of the box without needing any integrations at all, but adding your calendar will only make the experience more personalized and powerful moving forward!

@roopreddy Asmi does really well with regional accents - we have 50+ languages available today. For noise phones lines and poor call quality - we are currently dependent on external vendors, and they handle this, but for the future, we do plan to optimize this in-house and provide a better experience to all users.

@syed_shayanur_rahmanYes to that and any real world task that consumes your time while you want to heads down focus on growing your business and spending time with your family. Asmi is here to handle it all!

@outofelement Hi Marat - Asmi does not mimic your voice. Asmi is your personal AI with its own personality, voice and a dedicated phone number. It introduces itself as your personal assistant to the world, and not a copy of you. Also, you don't need to give any data to Asmi to get it to work. Just pick up Asmi's call, pass on your tasks - it asks all relevant questions to get context and gets to work.

@outofelement Totally get the hesitation and a fair concern.

A few things worth knowing: Asmi only acts when you explicitly tell it to - it won't make a call, send a message, or take any action without a direct instruction from you. Every call made on your behalf opens with a clear disclosure that it's an AI. You have full access and deletion rights over everything it stores. On voice: we use AI-generated voice for calls Asmi makes on your behalf - we don't clone or replicate your voice.

Happy to answer anything specific. Check out our detailed privacy policy here - https://www.asmiai.com/privacy

@alex_sh78 Thank you, Alexandr!

Let me show this to you with an example. In the case you mentioned, Asmi will call the reception saying it's calling on behalf of Alexander to book an appointment for Saturday, 2pm. Given how human like and natural the experience is - most of the people/businesses are totally fine talking to an assistant That has been our experience.

@hamza_afzal_butt Hey Hamza - yes, Asmi does that today. It can call multiple vendors in parallel and get you a good price. Would love for you to try it out!

@mehulmishra6 Hi Mehul - Asmi does not take access to your contacts for privacy reasons. If you do share contact card or contact number over chat, it should be able to read contacts and also save them for future reference.

@amna9 Hi Amna - yes, we do give users option to delete the data if they wish to. We generally store it anywhere between 30 days to 90 days as of today. And it is used for getting to know users better and making Asmi more autonomous and hassle free for them over time.

@tina_chhabra Thanks Tina. Whenever Asmi comes across a missed call or voicemail situation, either it can automatically recall in next few minutes or users can explicitly decide when Asmi should call back. Asmi is trained to handle all these task loops efficiently!

@farrukh_butt1 Thanks Farrukh!

To keep you in control, Asmi smartly analyzes the task context and always asks for your confirmation and preferences—like a max budget—before initiating anything.

If an unexpected detail ever slips through the cracks, it immediately reverts to you to gather that information and wrap up the task to completion.

@ranjan_kumar45 Thank you, Ranjan!

Those are absolutely fantastic use cases, and using it to check in on family for medicine is incredibly thoughtful.

@ragsyme Hi Raghav - Yes, it works in India for all the use cases you mentioned and more. Hope that answers your question. Feel free to shoot more anytime here!

@nuseir_yassin1 Thanks Nuseir!

Absolutely! Transparency is a core value for Asmi. After each successful task completion, users get a clear update alongside the transcript/recording, ensuring you always have total visibility and peace of mind.

@zerotox Thank you so much Abhishek!

Saving people real hours and giving them their time back is the ultimate goal. So glad you discovered us on Product Hunt today! 🙌

@krutiparekh16 Thanks Kruti. Absolutely!! It can help you keep track of the day/week, remember things, and remind you at the right moment. Further, it can pick up those chores at the right moment and complete them for you :)

@divya_kothari1 haha, absolutely Divya. Let Asmi make your life easier. You just do what you love to do!

@ankur_jeswani Thanks Ankur. You're absolutely right. Asmi is totally hassle-free, and we chose calls and iMessage/WhastApp to make it really easy for everyday consumers to use Asmi effectively

@iamanantgupta That is exactly what Asmi is designed for—handling those endless life admin chores so that both of you can offload the mental load and focus on what matters most to you.

We’d love to know what she thinks once she hands off that first task to Asmi!

@himani_sah1 Thank you so much—"chief of staff for personal life" is the ultimate compliment and exactly what we aimed to build! We'd love for you to experience that firsthand by trying it out.

@himani_sah1 Incidentally, that was the first internal title we had for the product! Your comment feels like we've closed the loop!

@igorsorokinua Hi Igor - Asmi works very similar to a human personal assistant. Just that it’s more efficient, multi-lingual, and can handle hundreds of tasks in parallel.

If any business or person refuses to talk with an AI, we do respect their choice. Asmi comes back to user with the exact situation and can share the contact details with the user for them to call the business directly.

@marianna_tymchuk Thanks Marianna. That’s true. People are loving Asmi for the same reason.

@nikitaeverywhere Thank you so much! We completely agree—the future is personal agents, and we are incredibly excited to be building that bridge into the real world. :)

@borrellbr Thanks Ignacio. We initially built Asmi for busy professionals drowning in life admin, overwhelmed parents managing endless family logistics, and students juggling chaotic schedules. Basically, anyone who is completely done with phone fatigue and hold music!

@natalia_iankovych Thanks Natalia for the interest!

We’ve actually had a couple of inbounds about this already and plan to start working on a B2B version and API over the next few weeks. For languages, Asmi natively supports a wide global range including English, Spanish, Hindi, and Mandarin, among many others.

If you'd like to integrate this into your startup, drop your email ID and we’ll loop you into our B2B pipeline as soon as it's ready!

@yulesenmiao congrats on the launch David. This is really interesting, how do you prevent the glasses from becoming another notification surface and fatiguing the user? in your cc use case, there are many, many "allow once".

@yulesenmiao 

We first built Terminal Mode for our own team, and it quickly changed how we worked internally: fewer terminal check-ins, fewer idle agents, and less context switching back to the laptop.

That’s why we’re launching it here. We wanted more builders to benefit from the same workflow.

@yulesenmiao Congratulations on the launch, David!

This feels like a very practical solution to a real problem. I use AI coding agents for development, research, and automation, and they rarely fail on the hard tasks. More often, they get stuck waiting for a quick approval, clarification, or decision while I'm away from my desk.

Two questions: have you found that most delays come from missed approvals, or are there other common bottlenecks that Terminal Mode helps solve? And the second one is, how do users avoid notification fatigue while staying responsive to the agents that actually need attention?

Wishing you and the team a fantastic launch!

@song_kirby 

Exactly. If it feels like another delayed phone notification, it fails. We treat G2 as a live state surface, not an inbox. It only shows moments that need you: blocked, waiting, approval needed.

In our Claude Code tests, the HUD updates roughly 0.5-1s after the agent enters that state, so you catch the stall instead of discovering it 20 minutes later.

@artstavenka1 
That concern is exactly right — we don’t want approvals to be blind.

The HUD is glanceable by default, but approvals are not limited to one line. You can enter the session and scroll through the full context before taking action. So the one-line view is more of a “this needs you” signal, not the approval surface itself.

For anything sensitive, the flow is: notice it in peripheral vision → open the session → review the context → approve intentionally.

@xiao_zhang9 

Ditto.

We heavily try to cut out the things that aren't absolute musts. This is how we can show you what really matters.

@gizem_ozturk 

Exactly. We’re not trying to make people watch every terminal line from glasses.

The useful part is catching the “needs you now” moment before it becomes 20 minutes of idle time. That’s where having the agent state in your peripheral vision starts to make sense.

@jocky 

Path is agent runner -(network)-> companion app -(BLE)-> glasses.

BLE write is ~100ms; the dominant cost is whatever the runner takes to flip "awaiting input." With Claude Code, it surfaces in the HUD inside ~0.5-1s of the actual block. :)

@crystalmei 

That's the bet. A single approval isn't worth a product. Not breaking flow 40 times a day is - that's why we feel it makes sense to be in your line of sight when your agents need you.

Glad it lands!

@oleksii_sekundant 

Right now we can show the full raw tail, so you can enter a session and inspect the actual output when you need detail.

That said, raw logs are not always the best default for a glasses display. The direction we’re moving toward is a distilled state layer.

@farrukh_butt1 

You bet!

Now you can be on the move more & still max out your productivity.

@stella_guan 

i‘ve been there several months ago. But now we have a game changer

@isla_wei 

In short, a plethora of issues. But I am glad to hear that it works better now!

What might seem like a simple disconnect to our end users derives from many small root causes that augment each other. We are aware & working around the clock in attempts to discovering more issues.

@wangyongyiuny 

Glad to hear that!

That's the beauty of software - the potential is limitless (sort of).

@hong_han 

nice spot, thanks!

@lizzy_leeeee 

Good question. We designed approvals to be intentional, not something that can happen from a random movement.

You need to enter the specific session first, then take the approval action there. So it’s not a passive “one accidental tap and something runs” flow.

A lot of the UX work went into making sure Terminal Mode stays glanceable for status, but deliberate for approvals.

@david_marko 
We treat it as state, not logs.

Running agents stay quiet. We surface the moments that need human attention: waiting for input, blocked, approval needed, or ready for review.

The key question is “which agent needs me now?” not “what is every agent doing every second?”

@1mirul 

Cannot agree more. The principle that inspires us is "let agents work & let humans live".

@ea_z 

Glad you are asking this; We actually put a lot of engineering efforts into this.

TLDR: Skipping all the technical deep dives, we are confidently still all-day wear & pushing two days depending your usage!

@phoenixhu 
Developers are the focus for this launch because the pain is very clear there.

But it’s not limited to coding. If your agent can do it from your laptop, Terminal Mode can keep that session visible and actionable on G2.

A fun example: I connected an agent to my smart home, so I can ask it to turn on lights, play music, and read the weather from G2. It feels like an early glimpse of G2 as an ambient command surface for agents.

  @sergeynugaev nice product! Congratulations with the launch

@mituhin Thank you Paul! It's a joy to launch a product with you. Fingers crossed!

@misha_yanovich Thanks, Misha! Excited to hear how the learning of Ancient Greek went 🙌

We do have some users already trying the app for learning languages like French or Japanese.

@wwwictor Hey Victor! Love your question.

I started my educational career with the issues of teenagers' self-identification. And honestly, it's still one of the most intriguing topics for me: how we can help young people to understand what they really love to learn and do. So definitely yes, we will be looking for ways to bring Journey Now into teenagers' routine and educational contexts.

@maria_chmir Hi Maria, thanks for your question! From my experience, there are no preferred topics. For example, I used it for French learning and for blogging habits as well. My friends use it for thinking about careers or for book writing. So the spectrum is quite large.

But I think the difference depends on the educational context. My belief is that the human presence and experience are crucial for any learning, and humans are not scalable. So Journey Now works with both situations.

When you learn on your own, it gives you the support you usually can get only from a teacher, and your Pod adds real people around your progress. And when there is a teacher or a group leader, Journey Now amplifies them. That's why I especially see a great potential for team training, corporate learning, etc.

@thamibenjelloun Hey Thami! Sure, our system analyzes every piece of feedback and updates your path accordingly. So daily recommendations are based on your actual activity, not a generic schedule. And the content itself is RAG-grounded in real sources, not generated from thin air.

@taragraphy Thank you, this really means a lot! Happy to have you following along — this is just the beginning, more good stuff coming.

@kate_prasniak Thanks for the comment, Kate! You basically get a path to your goal, which covers a month, this allows you to see the weekly scope and tasks for each day. The tasks are highly specific for your goal and usually reference a source, so you can not only try it, but also see there it's coming from and what's the background behind the idea 😌

@anna_shapkina Thanks! By the way, now you can do this together with friends too!

@madalina_barbu Thank you, Mada 🙏

@klochook Thanks, Olga! We're glad you liked it. Our design lead Misha actually did a whole LinkedIn post about this a couple of weeks ago, here's the link to it 🙌

Hey@nastassja_popova, great to see you here! We use AI to source exercises, which fit your learning goal. We basically have a set of different LLMs, which search for content, verify it and adjust new practice exercises based on the feedback you provide. Think of it as a personal discovery engine, which tries to come up with the best creative task for your upcoming day while you're busy with life.

@nastassja_popova Thanks! In a few words — it's many small parts of the system (little AI services inside it), each one or a group of them handling some piece (planning your path, writing the daily content, reflections, recaps, the chat and so on) and passing context to the others through pipelines, summarizing and accumulating it in a certain way. So everything stays aware of your goal and what you've actually been doing

@dmitriipokidov Many thanks 🙂‍↕️

@artyom_zhuravlev Thanks! If you tell it a week went badly, that feeds into how your future content and tasks get generated. But beyond that — if things really go off track, your circumstances change, or your priorities and values shift, you can do a refine of your plan. That adapts it to your needs much more explicitly.

Hey Luke@montverde, great question! We wanted to make learning more intentional, so the reward system is tied deeply to your progress. You get weekly and monthly reflections along with the progress data on how well you're moving towards your goal. If you add a close friend to your Pod, you could also cheer each other up and sent small rewards with comments, emojis to show that you're supporting your friend's journey ✨

@montverde Right now the closest thing is in Pods — you and your friends share a weekly energy bar that you fill up together. Every minute of practice, every shared piece of content, post and comment from any member adds points toward a shared goal, and if the group hits it for the week, you keep your Pod streak alive. And each week ends with a Pod recap.

So there's a streak mechanic, but it's collective rather than individual. That said, it's still a tiny piece of what we want to do here — we think a lot about a fuller streak/reward system and have it planned

@solodnev Thank you! In Pods you share your activity with each other, and the goals can be completely different — so yeah, it works great either way!

@farrukh_butt1 Exactly! And that "quietly dropping it after week two" is the exact thing we're fighting. Thanks for getting it — means a lot!

@leon_thomandl Hi Leon! More than usable. Right now we're testing the Educators Console — an admin panel where any educator or company can create their own program (or even program tracks), and the app will stick to it and distribute the knowledge to the end users (learners and employees).

Hey @fatih912, one of the first ideas behind Journey Now was to build a discovery engine or a "content brain" as some of our team-mates call it 😅 It basically runs all the time, trying to come up with the best learning content, which fits your goal. The more feedback you leave and the more exercises your complete, the better it gets in sourcing the next practice exercise for you.

@alex_egorov Thanks for the support, Alex 🙏🏻

@nickanisimov thanks, Nick!

@leonid_ivanov Hey Leonid, we do have the voice input in the onboarding when Journey Now's helping you to better understand your learning goal. You can just talk about the thoughts you have and we'll help you frame the learning goal!

@sabina_vard Wow, Sabina, happy to read such words! Thank you. The logo love will be delivered to the team :)

@praneethpike the 'all agents see everything' part is shared read context — that's the tractable bit. where parallel agents actually fall over is concurrent writes. we run a few against the same repo and ended up serializing the write step to one at a time, reads shared, just so they don't clobber each other.

whether canvas nodes can write the same space at once or it's single-threaded under the hood is the thing that'd tell me how far this scales.

@praneethpike Congrats on teh launch Praneeth. Was going to ask about shared context and then answered my own question with the "one context per goal". Makes a lot of sense.

@praneethpike I haven't tried Slashspace yet BUT I just HAD to say that I love what you said about AI, I feel the EXACT same way about AI!! I read a paper called, "Cybords, Centaurs, and Self-Automators: The Three Modes of Human-GenAI Knowledge Work and Their Implications for Skilling and the Future of Expertise" that seriously blew my mind. I am a generalist as well and found the ultimate tool in AI. What was once something I had to create a whole team for is now something I can create in an hour! (long story short I love your outlook on things, great minds think alike!)

@thisiskp_ You described much sharper than my maker comment. Very grateful to have you hunting the launch!

@1mirul Yes, most of our users have it as a default AI app on their dock. Once you try using AI this way, it's hard to go back

@thisiskp_  @ruhanirabin thanks for pointing the details on linux! good have to have your support from the early day!!

@joe_maracic always happy to see early users! thanks for the continued support, Joe!

@lfgellon Thank you Luis!

@mayaa17 thank you! I'm here to see your feedback!

@thibaultll thank you for the continued support, Tibo!!

@oleksii_sekundant the new models get metadata of the canvas and existing nodes. so when you ask a new node about something, they retrieve context that's already on the canvas. unless you tell it do different things

@jparker1455 Thank you for the continued support and helping us shape the product!

@dhiraj_patel5 the models get a graph of and metadata about the nodes. so the frontier models are smart enough now to tell that another similar generation was done in a different node, and you're probably wanting something different

@dhiraj_patel5 the agents can see a context graph, metadata of the nodes, and the latest models are smart enough to know that we're repeating or something's already been asked. so the second response will be built on top of that.

@yogesh_prajapati3 Thanks Yogesh. This is how human thinks and most AI products don't take this into consideration

@ianhxu the node title generally solve this problem. And all the new nodes will get a metadata context of the graph so the new nodes will know what they're asking for

@dmitry_isaevski You're spot on. Every other tool is trying to store on the cloud. We took a page from Obsidian and went with storing everything locally. This gives compounding benefits as a knowledge / context system

@abhijit_patil4 thank you for your continued support Abhijit! So happy to see you here

@younas_awais the straico fam has been very helpful in shaping the product! thank you for your continued support

@snarayan Means a lot coming from you, Satya! We'll keep improving it to impress you more :)

@farrukh_butt1 Yes exactly! the problem of losing our context as humans is severely overlooked. Everyone's too excited about all the possibilities with AI but very little innovation on the interface layer to match the human's intelligence

@hello thank you Farooq! great to have your continued support

@eugzolotarenko Thanks Eugene, it's great to have your continued support!

@odeth_negapatan1 Yes Odeth, infact it doesn't need to 'relearn'. If users make the selection and preference distinctions for different roles clearly, it will be able to distinguish and take the right calls without having to 'relearn'

@diana_nadim2 Hey Diana, yes it can - if you make the distinction clear for each role, Claude will be able to remember it clearly.

@ragsyme Thanks Raghav!

@roopreddy Glad you liked it Roop!

@syed_shayanur_rahman Thanks Rahman, yes exactly. We wanted to codify judgement that's the reason for developing all our skills.

@busra_seker1 exactly. Reliability is where Tabstack shines in that regard.

@mark_toubman2 exactly what I said when I was interviewing to join the team! Such an awesome win.

@oleksii_sekundant Tabstack uses the rendered DOM plus a model, not selectors. For JS-heavy pages it renders the page in a real headless browser, then a model maps the rendered content to the JSON schema you define. The part that solves your selector pain: extraction is schema-driven, not selector-driven. You describe the meaning of the data you want, not a DOM path, so when a site reships its markup there is no selector to break. The model re-infers structure from the new render against the same schema.

On lazy-load behind scroll: the rendered path handles it. After navigation it waits for the network to go idle and for JS to render, then scrolls the page in passes (with short pauses) to trigger lazy-loaded content before it reads the DOM. So data that only appears as you scroll down gets pulled in.

The honest boundary: that scroll pass is bounded, so it covers typical lazy-load-on-scroll, not endless infinite feeds. For unbounded scrolling, or "scroll, then click or interact, then read" flows, the automate endpoint drives a real browser and can scroll as a deliberate step, then return structured output.

@anand_thakkar1 This is exactly the tradeoff we built around: no, it's not a real browser on every request. Extract and generate give you three effort levels, and you pick the tier:

• min: plain HTTP fetch, no JS. Lowest cost and latency, for static or server-rendered pages.

• standard (default): balanced handling that covers most pages without full browser rendering.

• max: full browser render that executes JS and handles lazy-loaded content. For known heavy SPAs.

So a real browser is the heaviest tier, used for the pages that need it rather than the default for every call. Responses are cached too, so repeat requests for the same page don't re-fetch (unless you use nocache).

Honest tradeoff: the lighter tiers are faster and cheaper but can miss content on the most dynamic pages, which is exactly when you reach for max.

@erkan this is the main reason Tabstack exists. The maintenance pain you are describing comes from selector-based extraction: you pin to CSS paths or DOM structure, and the moment a site reships its markup, those break.

Tabstack is schema-driven, not selector-driven. You define the JSON shape you want, the meaning of the data, not where it lives in the DOM. A model reads the rendered page and maps content to your schema. So when a site redesigns its layout, there is no selector pinned to the old structure to break. The same schema keeps returning the same shape against the new markup.

Honest boundary: this insulates you from layout and markup churn, not from the content itself changing. If a field genuinely disappears from the page, changes in meaning, or moves behind a new interaction, you may still need to adjust. But the day-to-day "they shipped a redesign and my extractor broke" maintenance mostly goes away.

@furkan_topcuoglu exactly. Tabstack takes a different approach, helping avoiding the constant scraper management fixes. I had an audit tool reviewing developer documentation and the pricing page was a constant pain point, so many varying needs and ways to get pricing across different brands. I implemented Tabstack (when I was interviewing) and realized how much more effective it is. I haven't had to adjust my app in this area since.

@malithmcrdev thanks for the continuous support, Malith! let's spread the word on LinkedIn, repost this

Thanks for the support, Alina! Help us spread the word on LinkedIn, repost this

@corey_haines aww thanks for the kind words!

@borrellbr for sure the marketing! hahah. Nah, I'm just biased. We have had a lot of challenges from getting agents to navigate the web to ensuring we're great stewards of Mozilla's manifesto for data privacy and transparency. It's been a really fun adventure, though.

@srbiv might have something else to share here that could be fun to hear.

@borrellbr Great question, two things come to mind:

- Building a product that puts privacy first means we're often making decisions with incomplete data. We have to get creative with the few signals we do have to understand our users' needs.

- In the same vein, we're tuning the system to use the most efficient fetching strategy for any given URL. Even with continuous learning and tuning, we don't always get it right, so we give the caller the ability to control the effort of the fetch. If they aren't happy with the results, they can increase the effort and we'll spend more time and resources to get back as much content as we can.

@fabian_maume Good catch, and the single agent string is intentional. Every request identifies as Mozilla-Tabstack/1.0 with a contact URL on purpose, so site operators can see exactly who is accessing them and reach us directly. Identifiable, predictable access is the posture we want right now. If there's a specific case where the single UA is blocking a legitimate extraction for you, tell us more about it, that's genuinely useful for how we prioritize.

@natwar86 that's amazing! Feel free to provide us a review if you ship anything on Product Hunt! I would love to support your efforts across the board so reach out if I can help in any way.

@chris_davis23 Heavy JS rendering is handled through the effort parameter on the extract and generate endpoints:

• max: full headless browser rendering. It executes JavaScript and waits for dynamic content to load before pulling data. This is the setting for SPAs (React, Vue, Angular, Next.js client-side), lazy-loaded content, and pricing or product grids that only appear after JS runs.

• standard (default): lighter JS handling that covers most pages.

• min: static HTML only, no JS, for lowest latency.

So for a JS-heavy site you set effort: 'max' and extract against the fully rendered DOM:

const data = await client.extract.json({
  url: 'https://example.com',
  effort: 'max',
  json_schema: { /* the shape you want back */ }
})

The same effort control applies to markdown extraction and the generate endpoint. And if a page only reveals content after interaction (click, scroll, log in), the automate endpoint drives a real browser to do that first, then hands back structured output.

@natalia_iankovych Hi Natalia! Yes that absolutely works.

@varun1jan Thanks very much Varun!

@thamibenjelloun Thanks very much Thami!

@maximilian_arnold Cheers Max!

@borrellbr Thanks Ignacio! We both grew up with chronic conditions so experienced the struggle of waiting months for an appointment and years for answers

@roman_burdyga Definitely an easier interface for many than manual logging

@gabriella_anjani Thanks very much Gabriella! Really appreciate it

@marianna_tymchuk Thanks so much Marianna!

@david_marko Yes, I think part of the reason why I developed this app is because of the back pain I was getting sitting so long in front of my computer. To answer your question... in the tests I've done, yes, it will point to the failure node but the response is usually provided in the format of long form text, which I now realise might not be the most convenient when people really want a quick look at their phones.

I've made a note of that for the next update... should be an easy JSON schema fix.

Thank you so much!

@thefamoushesham congrats on the launch Hesham. Quick diagnose and resolve sounds like a winner.

@kripanshu_shekhar Thank you — really appreciate it and let me know if there are any features you'd like added.

@thamibenjelloun It actually does. It will absolutely suggest a fix and will provide JSON that you can export with the tap of a button to your n8n instance. It will also pinpoint the failure node but the response is usually provided in the format of long form text, which I now realise might not be the most convenient when people really want a quick look at their phones. I've made a note of that for the next update... should be an easy JSON schema fix.

Thank you so much!

@borrellbr I think the Live Monitoring and the ability to switch workflows on and off + find out what node caused an error are the most popular features, but I'm anticipating that people will eventually fall in love with the AI Workflow Builder for the simple reason that it produces great JSON that is immediately exportable to n8n with the touch of a button.

@josh_bennett1 Honestly, the first early user was me — I run n8n myself after all. It's a useful tool if you don't want to be stuck at your computer all day monitoring workflows and it combines enough elements (I think) to actually allow you to manage your workflows without having to reach out to your computer all the time. The other key feature is the location trigger, which I've been personally used to trigger workflows when I arrive or leave a place.

I also don't think there is an app that currently lets you build workflows with AI and export them directly to your n8n instance as well as store your workflows so you can restore them to another n8n instance and produces reliable homescreen widgets that lets you see if any execution errors popped up recently.

@natalia_iankovych I think mainly it's because building and operating tend to come with different issues. You'll still have to build workflows at a desk, but workflows run 24/7 and it's unrealistic to expect to developers to remain at their desk on standby. Nodey allows you to handle the triage by sending a failure notification, telling you what node failed, allowing you to retry the workflow, deactivate so you don't keep burning through tokens and credits... or rebuild and optimise it using our AI tools. There is also a use case in that sometimes you want your workflows to be linked to the real physical world... for example, I've been using the location trigger feature to trigger workflows when I leave my home for the office. It's useful when you have workflows that are tied to physical locations.

And, in the case of manual trigger workflows that don't run on a schedule, you can still fire them from your phone without getting yourself on a desk.

There is also the Workflow Vault (which allows you to store your workflows and restore them to any n8n instance (useful if managing clients' instances)) and the AI Workflow Builder that builds the JSON for you and exports it directly to your n8n instance.

@oleksii_sekundant  The catch-and-re-run loop is exactly what's built: background polling flags the failed run, you get a notification, open the execution to see which node errored with its output, and retry (or stop/delete) it right there — plus toggle workflows and fire webhook triggers manually, via NFC tag, or by geofence. Node-level editing isn't in yet, and that's somewhat deliberate: the public n8n API has no granular node PATCH (you'd replace the whole workflow JSON), and an in-flight execution runs the version it loaded at start anyway — so "mid-flow edits" can't rescue a running execution even in theory. Where Nodey goes past pure monitoring is creation: the AI builder drafts complete workflows from a prompt and pushes them to your instance, and a debug companion reasons through why a node failed. Editing the canvas is the one thing n8n's web UI proves doesn't shrink to a phone — if I add it, it'll be parameter tweaks on a single node, not graph surgery.

@anand_thakkar1 Neither, actually — Nodey talks straight to your instance with no middleman, but over REST polling rather than WebSockets. As far as everything that we've tried indicates, n8n's push channel (/rest/push) is internal-only — cookie-authenticated for the editor UI and unstable across versions — so the public API (/api/v1) effectively mandates polling for third-party clients. While an execution is live, Nodey polls it on a tight interval (5s default, remotely tunable so I can throttle fleet-wide if needed), then drops to iOS background refresh plus refetch-on-foreground otherwise — the instance stays the single source of truth, so the app never trusts a stale local view. And deliberately no proxy: your API key and execution payloads never touch any server of mine.

@dhiraj_patel5 ABSOLUTELY! This was one of the features I wouldn't compromise on. I tested a few other apps that did the same thing on iOS and Android and I was always so confused when they didn't have this basic feature. With Nodey, you can trigger any workflow you want within the app, we have widgets that allow you to trigger workflows from your homescreen, you can trigger workflows when you arrive or leave a location, and through NFC tags by waving an NFC tag across your phone!

@moh_codokiai  Nodey's raw material is the n8n execution record — but that's richer than it sounds, because when an external service fails, its failure signature is captured in the failing node's error output: the 401 from an expired key, the 429 from a rate limit, the timeout from a slow database. So the execution detail view answers the first triage question at a glance — which node died and what the service said — by showing the full node chain with per-node status, timing, and error message. From there, failed executions get a one-tap AI diagnosis, and the Debug Companion chat can take the workflow's JSON and reason about whether the cause is your expression, your credentials, or their outage — and what to check next.

What it won't do... is probe your Postgres or your Stripe account directly: the goal is getting from "something broke" to "it's the Stripe node, 401, rotate the key" in a minute from your phone.

@bickov This is a clever idea. I like that SlimSnap doesn't just throw a screenshot at AI and hope for the best — turning annotations into structured JSON with element IDs feels much more reliable. The token savings are a nice bonus, and the fact that everything runs locally with an open MIT schema makes it even more appealing. Feels like a really useful tool for anyone building UI automation or working with Claude Code. Nice launch! 👏

@farrukh_butt1 Yes, exactly the case the schema was built for. Each element gets a unique ID regardless of how visually similar it is to others. OCR text + bbox coordinates + (if present) parent context disambiguate the duplicates. So if there are five "Submit" buttons on the screen, they show up as e_button_5, e_button_8, e_button_11 (or whatever IDs they get), and your arrow annotation points at exactly one of them.

The edge case where it still struggles: identical floating elements with no surrounding container or distinguishing text (rare but possible in canvas-based apps). For 95% of UI work, the ID + bbox + annotation combo holds up.

What kind of UI are you hitting this with most? Cursor with React forms? Claude Code with admin dashboards? Useful for prioritizing where to harden the schema.

@keirodev Yeah fair. The form demo is way too clean. Anyone'd just open DevTools for that. Real wedge is exactly your example: 40 overlapping components where "second input in the third card" is the only useful way to point at it. Picked the form because it fits in one screenshot. Wrong asset for selling the real case.

Redoing the demo on something messier is on the list. If you've got a real dashboard you'd want me to throw it at, send a screenshot and I'll post what the JSON comes out as.

@umberto_abbatantuono Hearing this a lot today. Windows port isn't in the short-term roadmap (OCR layer is Mac-native, needs a different pipeline), but if there's enough signal it moves up the list. If anyone else here is on Windows and would actually use this, reply to this comment or email hi@slimsnap.ai. That's how I'll prioritize.

@bernardjhuang Love this idea. Turning a single photo into a usable RPG sprite pack with transparent PNGs and walk animations is super neat. The fact that it's a one-time purchase with no account or subscription makes it even more appealing. Feels like a fun tool for indie devs and hobbyists who just want assets they can drop straight into Godot or Unity. Nice work! 🎮✨

@samirrashed thanks Samir!~ HUNT80 gets you your first sprite for $1 buckarooo

@ryanmt Really interesting approach. I like that Lium is aimed at scientists and analysts instead of assuming everyone wants to write code. Being able to work with complex datasets through plain English and bring together geospatial, energy, and infrastructure data in one place sounds incredibly powerful. Turning weeks of data wrangling into a conversation is a pretty compelling promise. Congrats on the launch! 🚀

@alina_tyslenok_ Thank you Alina, appreciate the kind words.

@borrellbr This is an important one. We have a few ways:

-With early access customers, we had 1:1 calls with all of them to collect feedback first hand. This is where we got conviction that we were really on to something with ~50 teams giving great reviews (and some constructive feedback of course!)
-For a more measured analysis, we have analytics built in to see usage, where people are getting stuck, how many tools they build, how many users they share it with, etc. On average, users are sharing with 2-3 collaborators which is a great sign.
-Soon we will release an evaluation harness that enables the users to track and evaluate performance themselves.

A fun one: an astrophysics researcher using our platform messaged me this morning asking to add me to their paper because Lium has had such a big impact on their work!

@borrellbr appreciate the kind words, Ignacio! 🤟🏽

@scotthburk it's usually the unglam stuff that really differentiates, right Scott! - great work, man!!

Hi @dhiraj_patel5 , great question!

We work with a wide range of complex datasets from advanced industries, including 3D seismic volumes, NDT/inspection data for semiconductor and manufacturing workflows, hyperspectral and remote-sensing imagery, and other large multimodal scientific and engineering datasets.

We also see teams use Lium on large, messy business data, especially in areas like marketing and finance, when the data is spread across files, systems, and formats that are hard to analyze together.

A concrete example: Lium connects to multiple NOAA sources, so you can query terabytes of climate and weather data, often stored in obscure formats, just by asking questions in natural language. To try it, you can sign up for free and turn on the Weather & Climate domain pack, which includes the data connections and a set of pre-built tools.

Hope that helps, and I’d love to hear your feedback if you test it out!

@onpilotai  The biggest hurdle I see for my audience isn't finding data, but trusting the AI to take the right action across multiple tools. How does Onpilot handle 'hallucinated actions' or safety checks before it schedules a task or sends a message?

@onpilotai congrats on the launch Gunit. How do you deal with business process exceptions, wron or just incomplete data?

@divjot_singh_sarna I am so glad you asked this question because this is EXACTLY the kind of smart response Onpilot provides. Onpilot doesn’t just identify something and trigger an action. It explains the “why” behind it first and then next action to take. Here's what a response looks like:

Thanks, Harman! Yes, that’s definitely one of the use cases of Onpilot.

Onpilot can sit on a customer facing website, guide visitors through the right questions, help qualify the lead, and then push the details into your CRM with the right context attached. You can name the agent and choose your branding as well.

So instead of just collecting a form submission, it can capture what the visitor needs, why they’re interested, how urgent it is and what the next best follow-up should be @harman_saini6 

@navjot_singh21 Thank you, Navjot! The fastest value usually comes from one painful daily workflow...let's say most often proactive risk alerts or follow-up automation.

For example, Onpilot can watch across tools like Slack/Teams, CRM, support, or ops systems, spot things like missed follow-ups, delayed tasks or risks before they become bigger issues, then explain what happened and suggest the next step.

Dashboards and opportunity discovery comes next once there’s more workflow context. We’ve found it works best when teams start small, see value quickly, and then expand.

Great question@montverde 🙌.
We built Onpilot to be customizable, but the first use case we had in mind was operations-heavy businesses, teams with workflows spread across tools like CRM, ERP, spreadsheets, emails, and internal docs.

So it’s not limited to one industry, but it works best where there are repetitive decisions, approvals, follow-ups, reports, or cross-system actions that teams want to automate safely.

@alex_iliescu Actions can be routed through approval workflows before they're executed.
But honestly, today the "two sources disagree" catch happens because the action is paused for review, not because there's a dedicated contradiction-detector auto-resolving it.
Explicit conflict-flagging is something we're leaning into next.

@tina_chhabra 

It's a hybrid by design. You define what matters upfront, but Onpilot also builds organizational memory over time. It learns your business context, key people, processes, past decisions, and how work actually gets done, so the insights become more relevant and less noisy.

As for the 3,000+ integrations, they let Onpilot connect with the tools your team already uses, pull information when needed, and take action across systems from a single place.

Thanks @borrellbr  🙌 We built Onpilot first for businesses where teams are constantly switching between tools to get work done, operations, manufacturing, service teams, sales ops etc.

The idea is simple: instead of giving every business the same generic AI agent, Onpilot lets you create an AI workforce around your actual workflows, tools, and approval process.

@natalia_est haha thanks! :) As a photo editing product, we figured we should start by being a good photo gallery first.

@tom_palmer_ux I'm thinking about the workflow, if team members can add anything to what is approved, how are versions controlled? Or do you only assign certain team members to be contributors?

@lennoxbeflying We will offer in our next builds - to embed the transcription in the video file or as an additional file to the "shared" package. I assume we will share this update in the next few days.

@qifengzheng hi, That won't happen in our case. We smartly offload the stream from RAM directly to a cache file and know how to play with the memory . Keep in mind that as a pure native app, we have direct hardware and OS access. This gives us way more flexibility than typical apps, so we don't suffer from those usual upper-layer issues.

@michaelcyger Thanks , In OwnClip, we solve this with a feature called Smart Share.

How it works: While videos record locally on your Mac we also provide a way to back them up through your private Google Drive (additional private services will be added in the near future - but the key here is you own your data ), Smart Share instantly and securely uploads the file to the cloud in the background.

You immediately get a shareable link, and the recipient can watch the video directly in their browser without downloading anything.

Rebranding: You can also fully customize the viewing page. Add your own logo and brand colors so the shared link looks completely professional and tailored to your business.

@tomer_weiss2 Hi Tomer, checked out OwnClip today. The local-first architecture and on-device AI is a really strong differentiator, especially against tools like Loom and Screen Studio that call home to the cloud.

One thing though: there is no video showing what OwnClip actually looks like in action. For a screen recorder where the entire pitch is speed, privacy, and native performance, that experience needs to be seen. Right now visitors have to take your word for it.

I make cinematic 60 second demo videos for SaaS founders. Script to delivery, done for you.

Worth a chat?

Arsh
famouslyhq.com

@alexander_arshavski Thank you very much for your feedback

@fberrez1 Hi there, OwnClip features a local-first AI architecture that runs entirely on-device using your Mac's Apple Neural Engine and other small models we use. Features like instant on-device transcription, smart audio enhancement, OCR , WebCamera effects, and others.

@piodubro looks and works great!! Super useful

@boopathi_raja007 Thank you! Currently it's Mac-only, but Windows support is not off the table! 🙂

@novamaker01 The provider allowlist only needs the external callback URL that the browser sees. Changing the target app port behind Proxee, like localhost:3000 to localhost:5173, should not require another OAuth allowlist entry as long as the callback URL stays on the same Proxee URL, for example:

http://proxee.localhost:7331/api/auth/callback

You’d only need another provider entry if the actual redirect_uri sent to the provider changes, such as a different host, port, scheme or callback path.

@leo_kayali Congrats on the launch Leo. This is very cool. Very blue ocean.

@thamibenjelloun Thank you! I appreciate it!

@divya_kothari1 the photoreal means the map matches real life 3D San Francisco in term of building, power gird, transited cables, helipads, charging stations, fog, etc.. to make an exact replica to air navigation environment of San Francisco.

@baraa_bilal Congrats on the launch Bilal. This is a really cool use case. For heavy weights, what is the safety margin? I don't want to be benching "just.one.more" and then have to wiggleslide under the bar (yes, it's happened. anyone who lifts heavy has one of these). Are notifications audio?

@st1100 Nope, you go to the gym, do whatever you want, and it shall detect it. You can still manually input what you want to do, or build a program if you want. But usually I go to the gym with a program only in my head; I don't need to write it down in an app - and sometimes my plan changes based on what machines are available if the gym is busy.

@busmark_w_nika Hi Nika! For now, it supports Apple Watch only, and the workout data is synced with your Apple Health app.

@tyler_bush Definitely on the roadmap! Thanks Tyler

@keren_dona Thanks! Honestly, the biggest challenge was noise — separating real reps from everything that isn't a rep. There are two kinds:

1.   Sensor noise. The watch IMU doesn't give you clean signals. Raw accelerometer and gyroscope data is jittery, and any attempt to compute velocity from it drifts fast. A lot of the work was filtering the signal down to something you can actually trust rep-to-rep.
   

2.  Movement noise. This was the harder one. Unracking the bar, adjusting your grip, setting the weight down — on the wrist, these can look almost identical to a rep. And it gets worse near failure, because grinding reps get slow and irregular, which is exactly when you can't afford to miscount. Telling "ugly real rep" apart from "setup motion" reliably was where most of the iteration went.

@ankur_jeswani Thanks, Ankur! For sure, something we have in our minds. Other wearables (other than the Apple Watch) should be able to produce close IMU signals that we can use for Riven.

@ferdi_sigona Great question! Currently, the app supports exercises that involve upper-body or wrist movement, as well as some lower-body movements like squats (since your body moves up and down). However, exercises where your wrists remain stationary—such as the leg curl machine—cannot be tracked automatically. You can still log these exercises manually to ensure your entire workout is recorded!