PH热榜 | 2026-05-19

Hey Product Hunt! 👋

I'm Gia, maker of PollyReach.

I'm the kind of person who'd rather text than call. But some things still require a phone call — and when the restaurant is full, you have to try the next one, and the next.

In Japan, I wanted to book a small izakaya. No online reservation, just a phone number. I don't speak Japanese.

That's when it clicked: AI speaks 50+ languages. Why can't it just make the call?

So we built PollyReach — Your Agent gets its own real phone number and a real voice. It handles real conversations the way a personal assistant would — gets interrupted, responds naturally, waits on hold, navigates IVR menus, and knows when to push back and when to hang up.

It calls for you

• Polly finds the number, dials, navigates phone menus, handles the conversation, and confirms the booking. You get a summary + recording + transcript.

It answers for you

• Polly picks up 24/7 with a natural-sounding voice — screens spam, talks to real callers, takes messages, tags priority, and sends you a summary. You decide what's worth calling back.

It works for your business

• One of our users manages 80+ rental properties — his AI assistant handles tenant calls, follows up with vendors, and sends him a daily report.

Get started — just send this to your agent:

clawhub install pollyreach

Everyone here gets 200 free credits + a free phone number. Try it and let me know what your first call is.

Really excited to share this with the PH community. We'll be around all day answering questions! 🙌

What languages are actually solid right now? Curious about Japanese specifically since that's the founding story.

Do you have a demo call recording somewhere? I want to hear the natural interruptions in action.

This feels like one of those products where the demo probably converts people instantly. Would love to test it on real world calls.

This is one of those products where the demo can be simple but the real test is annoying real life. Booking a table sounds easy until the place is closed, the number is stale, the person asks a follow-up question, or the bot has to negotiate between "7pm" and "we only have 7:45."

I’d love to see examples of failed or partial calls, not just successful ones. A useful agent should say “I called, they didn’t answer,” “they only had 8pm,” or “I wasn’t confident enough to confirm,” instead of pretending the task completed.

Congratulations

Congrats! great team！！！

Aren't you afraid that some people may use it for harmful purposes? E.g. automating the frauds (against banks and its users, or "behaving like relatives")? How can we protect / avoid such an usage?

Finally, an AI that can actually wait on hold instead of hanging up after 10 seconds. Take my credits.

Does Polly learn from previous calls with the same vendor, or is each call fresh?

That's a very good tool. I tried to use it to book a restaurant for me and guess what? it succeed! Damn good.

Congrats!
Quick question: Can Polly transfer a live call to me if it realizes it's out of its depth?

How many simultaneous calls can one Polly agent handle? Asking for the property manager.

Is there an API or MCP for developers? Would love to plug this into my own agent.

Congrats on your launch! I have a small question: Does it integrate with anything yet (calendar, email, Slack), or is it standalone for now?

One scene keeps playing in my head: strolling through the streets with an AI companion, like the movie Her finally stepping into real life. She just gets it, makes the call, and the table is already waiting when we show up.

PollyReach made that dream real. So cool!

Congrats! curious can you give it specific instructions before a call, like "only book if there's outdoor seating available"?

Where do you find the number? Actually, I want to know the limit. What kind of tasks would be impossible to complete?

Interesting concept. My main question is how it handles calls where the other side gets suspicious or asks to speak to a real person — does it disclose that it's an AI? Would love to know more about how that's handled before fully committing.

Loving this
The free phone number + credits to test is the right move, i been waiting for something like this that doesn't require API setup. Just installed 🎉

@gia_xu is it possible to connect it to an existing, real phone number so it can serve as my "assistant"?

How does it handle situations where the restaurant asks for a credit card to hold the reservation — does the AI pause and hand off to you, or is that not in scope yet?

From your terms:

• Prior Express Consent for non-marketing automated calls

What does that involve? How would I get consent prior to calling a restruant like in your example?

它不仅成功打通电话（餐厅不接受在线预订），还能灵活应对 "只有吧台座位" 的情况，帮我争取到了 7 点的最佳位置，全程用自然的日语对话，餐厅完全没发现是 AI 在沟通

Hi! How does Polly handle conversations that start to become negative or unproductive?

Feels less like one of those robotic AI callers and more like an actual assistant handling stuff for you.

The “book me a table” use case is honestly super relatable 😂

Congrats on launching this useful tool! I'm curious whether the numbers PollyReach has would be considered valid ones. I know many cell phones would automatically filter seemingly suspicious calls, so I would like to make sure the calls will actually get through.

Hello - what sort of compliance do you have right now? Can it be used in healthcare?

Hey Product Hunt! 👋

I'm Yash, co-founder of Drizz. Along with my co-founders, Partha and Asad, we've spent the last year building the mobile testing tool we always wished existed.

💡 How it started

I once spent 3 hours debugging a broken test,  only to find a designer had moved a button 12px to the left. That was the moment I knew legacy systems was broken. Not "could be improved." Fundamentally broken.

XPaths that shatter when a UI changes. Test suites that need a dedicated engineer just to stay alive. We lived this at Amazon and Coinbase. Teams shipping slower than they should because testing was a tax, not a tool. So we built our way out of it.

🚀 One sentence in. A complete test out.

Drizz is a Vision AI mobile testing agent. You write what you want to test in plain English,  just the goal. Drizz reasons about your app, generates the full test, and executes it on a real device.

"Search for stays in Spain for 8th–10th April for one adult"

→ Drizz reads the app visually, builds every step, runs it on a real iPhone. No selectors. No scripts. No maintenance.

What Drizz does differently:

✍️ Plain English authoring: describe the goal, Drizz builds the test

🔮 Vision AI execution: reads your app like a human, not the DOM

🩹 Self-healing tests: UI changed? Drizz adapts automatically

📱 Real device execution:  iOS & Android, not simulators

🔗 CI/CD native: GitHub Actions, GitLab, Jenkins, Azure DevOps

📊 Results from teams using Drizz:

⚡ 10× faster test authoring vs Appium

🛠️ 67% less maintenance overhead

✅ 20000+ test cases automated without a single selector

🤝None of this happens without the full Drizz team

Every line of code, every design call, every customer conversation that shaped Drizz,  that's the whole team showing up, every single day. Huge love to everyone who's been in the trenches with us. You know who you are. 🧡

🙏 Our ask today

Not asking for upvotes, asking for honest feedback.

→ What does your mobile testing setup look like right now?

→ What would make switching to Drizz a no-brainer for your team?

Drop a comment. We read everyone, and it directly shapes the roadmap.

👉 drizz.dev

Thanks to @rohanrecommends for hunting us, and to the entire Product Hunt community for showing up for builders. Days like today are why we build in public. 🚀

— Yash @yash_varyani , Asad @asad_abrar1 , Zaid @zaid_ahmed_ansari , Zaid @Zaid Abdul Bari, Sreetama @sreetama_chakraborty , Partha @partha_sarathi_mohanty, & the whole Drizz team

I'm building a couples app on iOS, and my whole testing setup right now is XCUITest on simulator + manual checking, archived to TestFlight by hand. Probably the exact cohort that'd never write Appium tests but might genuinely try this. The question I'd love your take on: with Vision AI driving execution, how do you keep tests deterministic across CI runs? Selectors fail predictably...locator doesn't exist, you fix it. A vision model can fail by misreading a screen state on one run and reading it correctly the next, which feels harder to debug. And there's a scary version of "self-healing" where the AI silently re-interprets what success means. What's the rerun / confidence story, and where do you draw that line?

Congrats on the launch!

We have regulatory requirements around audit trails. Every test execution needs to be traceable back to who ran it, when, what version of the app, and what passed or failed. Do you store execution history and is it exportable?

Noob vibe-developer here, currently testing the most annoying part of the building process lol. Excited to try this out!

I also liked that it runs on real devices because that’s where most testing issues actually show up 😅.

Congrats to the whole Drizz team on the launch 🚀!

Can we test mobile apps on a phone/ mobile OS as well? Something like a TestFlight Log Interceptor or Android equivalent where we can test an app directly on the phone, or is it exclusively for the desktop form factor? Also, are VS-Code compatible IDEs supported?

Congrats on the launch!

This looks like a game changing for the mobile app testing. it's been a pain to update the test cases everytime some changes happens to the ui. Writing intent and the let the drizz take over test generation and execution is amazing use of AI.
Can I modify and save the test case steps after they are generated from intent ? for example, I want to change only one specific steps form entire test steps will it allow and remeber next time whenever I write the same intent ?

Tried Drizz last month after a friend recommended it. Worked surprisingly well on our app — the onboarding tutorial inside the product could use a polish (got a bit lost on step 3) but once you're past that it's smooth.

The team is super responsive on Slack. That counts for a lot this early.

This is awesome !

"Testing was a tax, not a tool" is a great line, and the 12px button story is painfully relatable. As a fellow builder in the automation space, I love that you went vision-first. Brittle XPaths have wasted more of my life than I'd like to admit.

Congrats on the launch, Yash, rooting for the whole team 🚀

Try Out Drizz right here, we are giving free launch day credits: https://www.drizz.dev/download-desktop-app

Complexity of the world is growing and we need more than just one test solution. As a fellow builder of AI Testing tools and Tools to Test for AI - I like the approach of Drizz. Running tests on mobile was already harder in the pre-AI world and I wish Drizz makes it easier for all of us. Best wishes team Drizz.

Curious how it handles multi-step flows where the UI state depends on previous actions — like a checkout that changes based on what's in the cart. Does the intent description need to account for that, or does Vision AI figure it out contextually?

Congrats with the launch! How can one with pipeline of Maestro + Claude Code benefit Drizz?

Hey, spent some time on Drizz's page and the no-script mobile testing angle is what pulled me in. one thing I kept thinking about, how does the agent handle non-deterministic UI like loaders, ads, or A/B variants? that's usually where script-free testing falls apart in my experience.

can we try it?

When you write a test intent for iOS, does Drizz reuse that same intent to run on Android too, or do you author separately per platform? Curious whether the vision layer handles the UI differences automatically or if there’s still some per-platform config involved.

intent-based testing is the right abstraction here. one question: if the UI shifts but the intent doesn't (new onboarding screen, reordered step), does it auto-reconcile or do you redo the test? that's where most vision-ai tools fall apart for us.

Do you expose test results as structured data? I want to pipe results into our data warehouse for trend analysis. Like test pass rates over time, most-failing flows, etc.

The no flacy slectors promise is honestly the most compelling part here mobile UI tests are notoriously painful to maintain at scale. Really curious to see how Drizz performs on fast-moving apps with frequent UI changes and complex flows.

This is basically what mobile QA has been missing for years moving from brittle selectors to intent is a big shift.
The real test will be CI reliability over time, but if it holds up, it could seriously reduce maintenance overhead for teams shipping fast.

We just closed our seed round and mobile is our first platform. Haven't set up testing yet. Is it too early to start with Drizz or should we wait until we have more features built?

 I prototype in SwiftUI and sometimes my prototypes become the actual production code. Having tests would be great but I don't have time to write Appium scripts. Could I literally test my prototype with Drizz before handing it to the dev team?

I'm on the customer support team, not engineering. When users report bugs, I can never reproduce them. Could I use Drizz to set up reproduction steps?

this is a great app. and also I want to test it out in my computer to know how it works. anyway, where's the link for me to download this app?

How does one install it?

Hi everyone!

More training scale, still aggressive pricing, and a better model for long-running coding work.

That is Composer 2.5 in a nutshell.

It continues from the same base as Composer 2, and this version is trained to be better at sustained work, complex instructions, communication style, and effort calibration inside Cursor.

Targeted textual feedback helps the model improve specific mistakes inside long rollouts, while 25x more synthetic tasks push it into harder coding problems grounded in real codebases.

Looking forward to the next larger model trained from scratch with 10x more compute!

Love cursor, been a power user for a long time now, and happy to see the native model getting better and better.

Nice to see the new models and I have been using Composer this morning - sorry for the reality check - but c'mon - Composer 2.5 does not really compare well against Opus 4.7 or Sonnet.

That was a no-brainer upvote! Testing Composer 2.5 since yesterday, and I'm extremely impressed. 2.0 was already a good one, but with 2.5, I feel like there's no need for Opus or GPT-5.5. Great job!

The focus on long-horizon agentic tasks is the right unlock for real coding workflows. We've run into this building RetainSure where the model drops context mid-task once the chain gets 20+ steps deep. How does Composer 2.5 handle state management across very long multi-file agentic sessions, and is there a hard limit on task length before it degrades?

Has the team experimented with giving the agent more awareness of runtime behavior — like logs or error traces — so it can reason about what's actually happening vs just what the code says? Curious if that's on the roadmap.

The effort calibration piece stands out to me. We've had agents lose coherence around step 6 or 7 in a long agentic chain, and it's tough to know if that's model drift or context decay. Curious how the targeted textual feedback is applied: at the step level within a rollout, or on the full trajectory?

Specializing a model for code editing workflows rather than pure generation is the right call. Generic models fall apart on multi-file edits because they lose track of interdependencies. Building RetainSure, we've hit that quality cliff when an agent touches more than three or four tightly coupled files. How does Composer 2.5 handle semantic conflicts when simultaneous edits across files introduce inconsistencies?

Hey Product Hunt community 🙌

I’m Katja, co-founder of Mantle Chat.

We built Mantle Chat because we think AI at work should feel collaborative, not isolated. When AI is part of the team workspace, it becomes more powerful, easier to adopt, and more natural for everyone to use.

The idea is simple: instead of everyone using AI alone, the whole team can learn, collaborate, and work with AI together.

How we got here?

We spent time talking to modern teams and noticed one thing: teams are already using AI, but not really together.

They have one tool for team messaging, another set of tools for AI, and a lot of work happening in between. Team communication happens in Slack or Teams, while AI work happens in personal conversations with ChatGPT, Claude, Gemini, and other tools. This creates a few painful problems:

• Constant switching between team messenger, AI tools, docs, and tasks

• Context gets lost every time people jump between apps

• Slack + multiple AI subscriptions quickly become expensive

• People use AI quietly, but don’t always share how they’re using it

• AI conversations stay private, so teammates duplicate work

• The AI has no real context on what the team is working on

• Prompts and responses get copied around manually

• Non-technical teammates often find AI harder to adopt

• Companies want everyone to use AI, but most AI tools still feel individual or developer-focused

The result is strange: AI is everywhere, but teamwork around AI is still fragmented.
So we asked ourselves: what if AI was not a separate tool, but part of the team workspace itself?

What we built?

Mantle Chat is a collaborative AI workspace for teams.

It brings real-time team messaging, AI models, agents, tool integrations, and shared knowledge into one place.

What you can do in Mantle Chat:

• Chat with your teammates in Slack/Discord style channels, DMs, and threads.

• When you need help from AI, you can mention (@) 60+ models GPT, Claude, Gemini, Grok, DeepSeek, directly inside the conversation.

• The AI can respond in the same thread, where the whole team can see it, discuss it, reuse it, and build on top of it.

• You can also keep private AI chats when you need to work individually.

• Mantle Chat gives teams a shared knowledge base, so uploaded docs, workspace context, and project information can become part of the AI’s understanding. Instead of manually copy-pasting background into every prompt, your team can give AI the context it needs once and use it across conversations.

• Teams can also build shared AI agents with custom instructions, knowledge, schedules, and integrations. These agents can run manually, on a schedule, or be triggered by connected tools. (For example, you can create agents for research, PR reviews, meeting notes, analytics reports, customer requests, Linear updates, Stripe events, and more.)

• Mantle Chat connects with 30+ tools, including Linear, GitHub, Google Drive, Notion, Slack, and Stripe, so AI can help with real workflows, not just answer questions.

• Give the whole team access to AI without IDEs or complicated setup

• Reduce the need for separate team chat and AI subscriptions

And honestly, figuring out AI as a team is much more fun and effective than doing it alone.

Is it for you?

Mantle Chat is built for teams that already use AI, or want to start using it together.

• For startup teams: Mantle Chat gives you one place to talk, think, plan, and work with AI. You can keep product discussions, customer insights, research, and AI-generated ideas visible to the whole team.

• For product and design teams: Mantle Chat helps you brainstorm, summarize feedback, compare ideas, write specs, analyze research, and keep AI outputs connected to the conversations where decisions happen.

• For engineering teams: Mantle Chat lets you bring AI into technical discussions, create PR review agents, summarize Linear issues, connect GitHub, and reduce context-switching between tools.

• For operations and support teams: Mantle Chat can help automate recurring workflows, generate reports, summarize requests, and create agents that run on a schedule or react to events.

• For non-technical teammates: Mantle Chat makes AI feel approachable. You do not need an IDE or developer workflow to use agents, models, and automations. You just work inside chat.

• For cross-functional teams: Mantle Chat helps product, design, engineering, operations, support, and leadership stay aligned by keeping conversations, AI outputs, shared knowledge, and workflows in one place.

See it in action:
https://www.youtube.com/watch?v=GVoFZ2dxVx0

You can try the interactive demo in the hero section of our homepage:

https://mantle.chat/home

For the Product Hunt community:

You can use Mantle Chat for free, or try the Pro plan free for 7 days with higher limits and all features.

Contact with us:

Website: https://mantle.chat
X / Twitter: https://x.com/MantleChat
LinkedIn: https://www.linkedin.com/company/mantlechat
Community: https://discord.com/invite/STzq94kdDC

Huge thanks to @fmerian for hunting us today, and to the Product Hunt team and community for the opportunity to share what we’ve been building.

Please support us today and drop a comment!

We’d love to read your feedback, any thoughts on how we can make shared AI work better for teams, and how your team is using AI today: together, separately, or somewhere in between.

One underrated benefit here is institutional memory . Shared AI conversations are way more valuable than knowledge trapped in private chats.

wonderful - good luck!

Congrats on the launch @katja_danilina !

Something on the roadmap: Mantle already has a pretty strong list of integrations. Are you planning to add HubSpot or Zendesk so agents can also help with sales and customer support workflows?

Does Mantle keep a persistent shared context across sessions — so the AI understands decisions the team made in previous conversations — or does each new session start fresh? That long-term memory layer seems like the thing that would make it genuinely more useful than just adding AI to a Slack channel.

Love the "@ the model in the same channel as your teammates" framing, that's the part Slack-with-a-bot-bolted-on always got wrong. Curious how you're handling context isolation when multiple agents are working on tasks in parallel in the same channel? Does each agent get its own thread, or do they share the channel history?

I’ve been jumping between Claude and ChatGPT tabs for months depending on what I’m working on. Never really found a smooth way to keep both in one workflow.

Mantle looks interesting for that. Curious to see if having one workspace actually solves it.

Hey Katja! You're awesome. AI and team chat integrated is a game changer cause it comes to upgrade teamwork. I'm sure many founders gonna love this!

Shared AI context across a team is way better than everyone running isolated prompts. Mantle Chat looks like it could cut down a lot of duplicated work in growing teams. We've been building in the customer success for ops-heavy SaaS teams space at RetainSure, and Mantle Chat touches on something we think about a lot. How does it handle context isolation between different projects?

Shared AI threads are the right move. The problem with private AI sessions is that your team's best prompts and outputs never get socialized. At RetainSure we've got CS reps doing similar AI-assisted tasks in silos, so we lose institutional knowledge constantly. Do you have any versioning or replay for how an AI-assisted decision was reached in a shared channel?

Congrats on launching! Wishing you a smooth start, happy users, and lots of success.

the positioning against private AI chats is the interesting angle here. the actual problem is that when everyone's prompting Claude or GPT separately, institutional knowledge just evaporates. nobody knows what their teammate already figured out. if Mantle actually solves that it's solving something real

Congrats on the launch! This looks really promising! Been looking for something like this! @katja_danilina

Congrats on the launch! The idea of combining chat, agents, and tools in one shared workspace feels very practical for real team workflows. You can also say that the @ plus agent approach inside channels is important in reducing context switching.

Right now we're paying for 4 different AI tools across a 6 person team. No shared history, no shared context, and the bill keeps growing. This looks like the first real alternative. Want to give it a proper test.

Does Mantle keep a persistent shared context across sessions — so the AI understands decisions the team made in previous conversations — or does each new session start fresh? That long-term memory layer feels like the key differentiator for async teams.

Some real stories from people using CtrlOps:

A developer noticed their server was running slow. Instead of guessing, they opened the AI terminal and typed "why is this server slow?" CtrlOps flagged an unfamiliar process consuming 90% CPU. Turns out it was a crypto miner. They identified it, killed it, and secured the server in under 10 minutes. Without CtrlOps, that miner could have run for weeks burning resources and money.

Another team spent 2 days manually debugging a production issue. Logs, SSH sessions, trial and error. Nothing. They connected the server in CtrlOps, asked the AI terminal what was wrong, and it pinpointed the issue in minutes. A misconfigured environment variable that was silently breaking things. Their own team could not find it in 48 hours. CtrlOps found it in one question.

This is what gets me excited. Not the features. The moments where someone solves a problem they thought required a DevOps expert. And they do it themselves.

Got a server horror story? I would love to hear it.

ok so the file manager sounds boring, I know. But I was doing everything through a separate SFTP client before this. separate login, separate window, separate headache every time.

now i just open it inside CtrlOps and edit configs directly. for someone managing multiple client servers, this is honestly the feature i use the most. more than the AI stuff even.

Managing Linux infrastructure through natural language sounds powerful but a bit terrifying from a security standpoint. Having a terminal assistant help debug server configurations could save hours of parsing logs. What kind of guardrails or confirmation steps are in place before destructive commands execute?

The “spreadsheet with random server IPs and commands” line is too real 😂

Really like what you’re building here.
Making server management simpler without needing full DevOps knowledge is a big win for a lot of developers.

Congrats on the launch @parth_makwana07, @hiren_kalariya & @daxesh_italiya 🚀

finally something that replaces my mess of ssh tabs and random bash scripts. the playbook feature is underrated, set up my common fixes once and now its just one click. great launch guys..

I have 100 servers in production, can I manage them using this application?

This hits way too close to home. The "bash" and "bash (2)" terminal tabs alone gave me flashbacks 😅

The pain point you're solving is so real — server management has always felt like it was gatekept behind one person who "just knows" how everything works. The moment that person is unreachable, the whole team is paralyzed.

What really stands out to me is the plain-English terminal idea. Lowering that barrier means developers can actually own their environment instead of depending on a single DevOps hero. That's a huge shift in team dynamics, not just tooling.

The "named servers instead of IPs" detail is small but brilliant — it's the kind of UX decision that shows you built this from real pain, not from a whiteboard.

Congrats on the launch! Can't wait to see how teams adopt this. 🚀

genuinely curious how the web search feature works in the

AI terminal. does it pull from the actual docs or just

general search results? asking because we work with some

niche tools and outdated answers are a real problem.

What’s the biggest time saver in practice, is it the scripts library or the AI generating the commands?

HR person commenting on a server tool, I know.

But whenever someone leaves the team, we need their server
access gone immediately. Before this it was a whole back and
forth with tech. Now I check SSH management myself and flag
it in 2 minutes. Offboarding got so much easier, honestly.

The AI-assisted debug loop for Linux servers is something we've wanted at RetainSure for a while. Chasing down intermittent issues across multiple EC2 instances usually means a lot of context switching between logs, metrics, and SSH sessions. Does CtrlOps maintain state across a debugging session, so the AI can reason about what it already tried before suggesting the next fix?

Making infra management conversational is clever but the hard problem is safe command scoping. An AI that can debug is also an AI that can accidentally drop a table. We've been careful at RetainSure about giving AI systems any write access to production infra. How do you scope what CtrlOps can actually execute? Restricted user, sandboxed session, or something else?

The approve before execute thing is what sold me.

Every other AI tool just runs stuff, and you find out

What happened after.

Yeah I've done this exact thing. Wrong tab, wrong server, restarted nginx thinking I was on staging. Took down prod for an hour. The part where it shows you exactly which server you're on before anything runs is what got me.

The preview step is the whole game when AI touches live infra. CtrlOps gets it right: ask in plain English, see the exact command before it runs, approve. Been running it alongside ClawMetry and the fit is natural. Congrats Hiren and team 🚀

This gonna be the best experience for someone like me who don't like tinkering around CLIs 🫠

Hey, went through CtrlOps's site and the AI-on-the-Linux-shell angle is what stuck with me. one thing on my mind, how do you handle destructive commands, is there a confirm gate before things like service restarts or migrations or does the agent just send it? blast radius is the part I'd want to understand here.

this is exactly what i wanted last week debugging my railway worker at 11pm honestly 😭

q> does it work with hosted platforms that hide raw ssh access (railway, render, fly)? feels like that's where most indie devs are landing now instead of raw VPS. TY

Congrats on the launch. The "AI suggests, you approve" framing is strong, especially for something as sensitive as servers.

Curious what users ask for first once they trust the AI terminal: safer debugging, deploy checklists, reusable scripts, or monitoring?

the fact that i dont need to install any agent on my servers sold me immediately. got it running on our staging env and already caught 2 issues before they became outages. will be moving prod over soon

Whenever I read about AI systems managing servers, it always scares me :) Is there no risk it could accidentally delete something?

Yeah I've done this exact thing. Wrong tab, wrong server, restarted nginx thinking I was on staging. Took down prod for an hour. The part where it shows you exactly which server you're on before anything runs is what got me.

as a solo founder wearing the devops hat, this fills a gap i didnt know i needed filled. one dashboard to rule them all. solid launch

the file manager feature is the one nobody talks about

but everyone needs. separate SFTP client is such a

pain when you just want to edit one config file.

Hey everyone, my name is Adish and I'm one of the founders of Mosaic, the company behind Motion.

Motion is our new frontier video agent made for tasteful motion design. Here's how it works:

1. Give Motion a prompt — include any context like product links, X threads, YouTube videos, personal assets, and more. Motion will reference styles, incorporate research, and storyboard scenes.

2. Motion creates the video — it can orchestrate entire explainers, launch videos, logo animations, or even add motion design to your own talking head videos. Motion has taste in its visual animation.

3. Edit everything — resize, drag and drop, modify elements, or iterate with chat. Tell Motion to iterate on selections or add new elements until you’re satisfied. No need to re-generate entire scenes or deal with hallucinated artifacts.

To see what Motion can do, check out our launch video, which was made entirely with Motion.

Motion is in early access but we are giving everyone that finds us on ProductHunt free invite codes. Sign up for Motion today with access code phmotion.

For support, please jump into our growing Discord community and if you're interested in API access or end-to-end produced launch videos, please schedule a quick call.

Congratulations, Adish! Another awesome product by you!

every other video tool makes you start over if one frame is off. how granular is the editing.. like can you swap out a single element mid-scene or is it section level

🔥🔥

I have used this to edit my podcasts, trailers , Short videos perfect!! .. Saves a lot of time.. and very good at motion design!! Highly recommended.

Thrilled that Motion is finally out! We've seen some stunning videos being created from our beta testers, such as launch videos, product demos, and more. Join our Discord community for the latest updates.

Motion is here for anyone who wants their videos to actually look good, not just “done.” 
I’m part of the team behind Motion. We built it around the real taste and workflow of motion designers and video editors, instead of just making another generic AI agent that edits videos.

Our goal is to capture that crafted, human motion-design feel in an AI-powered tool, so the results look like something a professional would actually be proud to publish.

We’re also very excited about what’s coming next. Soon, you’ll be able to choose exactly how much control you want: dial in every detail like a motion designer, or hand more over to Motion so you can ship high‑quality videos in bulk with minimal effort.

The edit individual elements instead of regenerating the whole scene part is the unlock everyone's been waiting for.

Every other AI video tool turns into a slot machine...pull the lever, hope, repeat.

Curious about the taste claim though. That's the hardest thing to get right and the easiest to fake in a launch reel. Is there a style library it pulls from, or is it composing from scratch per prompt?

Grabbing an invite. If it can turn a product link plus a few bullets into a real explainer, it saves me a week.

Looks super helpful! Love the builder and the ability to post on social in the same flow.

So exciting to see Motion going live 🚀

Being part of the team behind it, the fast edits, huge amount of undiscovered use cases, and professional grade outputs are all things we're incredibly proud to share. Definitely give it a shot and be sure to hop into our Discord!

Team GitHits was looking for a product exactly like Motion. We were lucky meet the team Mosaic at SaaStr. We love the product and will do our launch video with it.

I used this to make my product videos, they turned out so well.
I saw increase in engagements.

love the little video!

Been testing the tool for a while… really nice output!

Why redirection from product hunt invalid access. Invite code failing

Amazing as a content creator lacking the creative bone if you will this is perfect for me. I can create the artistic high quality motion graphics that other tools lack and I don't have to pay $1000's to have a graphic designer make them.

The element-level editing is the right place to be opinionated. For product/launch videos, “tasteful” usually depends on constraints that are easy to lose: brand pacing, how much product proof appears before flourish, which animations feel off-brand, and which references are actually approved.

One thing I’d love to see is a small style-memory / “why this direction” panel: pulled from website, these references, these uploaded brand notes, then user corrections like “too kinetic” or “not enough product detail.” That would make the agent feel more like a reusable creative director than a one-off generator.

Hi everyone, I'm Gary, co-founder of Chert!

We've spent the last six months building projects on iMessage for leasing companies, DTC startups, and home service agencies. Across these use cases, we kept seeing the same problem. Founders wanted to deploy conversational iMessage agents for customer service, lead capture, and outbound follow-up, but the underlying infrastructure was difficult to set up and hard to scale reliably.

This is what inspired us to build Chert, a platform for teams to create and deploy iMessage agents, powered by infrastructure that can send, receive, and automate conversations over iMessage at scale. There's a few main features that we believe will make Chert stable, reliable, and scalable:

- Chert provides comprehensive line health checks, making it safer and more robust for outbound use cases.
- Additionally, Chert offers a scalable pricing structure that lets teams scale to hundreds of lines and thousands of messages.
- Finally, Chert integrates with CRMs like Salesforce, HubSpot, and Close and tools like Vapi and Slack, so teams can easily add iMessage into their existing stacks at scale.

Feel free to try building and deploying your own iMessage agent through the Agents page in our website. No credit card required.

Would love any feedback or thoughts!

Congrats! This is cool. Are your customers typically coming from other iMessage tools and switching, or are they mostly introducing iMessage as a brand-new channel?

Do people usually use Chert to replace sms or have you seen circumstances where people use

both sms/rcs and blue-bubble messaging?

Getting iMessage delivery working for AI agents is a non-trivial integration challenge since Apple doesn't expose a public API. We've wrestled with customer communication channel tradeoffs at RetainSure, and iMessage reach in B2C contexts is real. How are you handling message delivery guarantees and rate limits at scale when the underlying transport doesn't give you standard webhook callbacks?

@garygao Congrats on the launch! Really interesting product. Is there a way to try building an agent before talking to sales? I’d love to send a few test messages and see the developer flow end to end.

Really cool. What are ur thoughts on trust and user experience? iMessage feels valuable rn bcs

its not yet flooded with business messaging slop.

Do you support outbound or only inbound messaging?

Big congrats on the launch. What does the integration actually look like? Is it mostly configuring the agent prompt and workflow in the dashboard, or do teams usually connect it into their existing stack too?

Could see this being really useful for businesses that get a lot of inbound interest. Are people using Chert to capture new demand right now or to mainly convert old demand they already missed?

Hey everyone, thank you all for trying it out!

Congrats on this launch!

What is the difference between Chert and SMS/RCS other than blue-bubble messaging?

Love this! AI agents in a channel people already trust feels much more compelling than yet another app download. Can the agent proactively message users too, or is the product more focused on responding once a conversation begins?

iMessage open rates are hard to beat. How does the underlying send mechanism work exactly? Apple doesn't have a public business API for iMessage - is this throught Apple Business Chat, a Mac relay, or something else?

Why only iMessage? We recently had a client asking us to build a similar service, but connected to all popular messengers and social networks for handling customer requests, which would then also be sent to the CRM. We looked for a ready-made solution, but couldn’t find anything that supported all the platforms they needed.

Can I bring my own numbers?

Do you support running different prompts for different stages, like qualification vs scheduling?

Are the numbers automatically assigned or do I get to choose the number that I’m using?

Is the main product like a rest api, webhook, sdk, or integration?

Interesting launch.

I saw that Chert targets businesses like DTC brands and SMBs on the webpage.

Is Chert more of a no-code agent builder or something built for developers?

Hey, was reading through Chert's site and the iMessage-as-support-channel framing is honestly wild. one thing I wanted to understand, what's the path for non-iPhone customers, is there an SMS fallback or is the product fully iOS-only by design? that fork basically defines whether this is niche or omnichannel.

Congrats on the launch! How would love to understand more how the inbound/outbound mechanism work? Is it basically an agent for each use case? Is the skill included out of the box?

Building on iMessage means dealing with Apple's undocumented protocol quirks, and wrapping that behind a clean API is the real product. We've been building customer-facing AI agents at RetainSure and delivery reliability across channels is a constant headache. How do you handle delivery confirmations in iMessage? Does the API surface read receipts or are you inferring from response patterns?

The iMessage delivery angle is smart. B2B tools rarely nail async customer touchpoints, but buyers actually respond to texts. We're running customer success workflows at RetainSure and the biggest gap is getting responses to renewal nudges. CRM write-back via HubSpot is a nice touch. Does the agent handle multi-turn conversations well, or does it reset context between sessions?

Are you routing through registered businesses with iMessage for Business or solving it some other way? The leasing companies angle is sharp positioning. SMS open rates collapsed in the last couple years for those verticals and iMessage actually moves the needle here

Congrats！Amazing Product！！

pretty cool! Congrats for the launch!
Is there any ways for the customer to talk to a human when ai is not efficient enough?

Are your customers typically coming from other iMessage tools and switching, or are they mostly introducing iMessage as a brand-new channel?

Hey!

I'm Dominik, creator of Monocle, and I'm excited to share a new 3.5 update with you!

For those who don't know Monocle:

Think noise-cancelling, but for your screen.
Designed to actually feel like it came with your  Mac.

You know that feeling when you sit down to work and your screen is a wall of tabs, open apps, and windows you'll come back to later? Monocle softly dims all of it except the window you're using with a simple cursor shake.*

Nothing closes, nothing changes. The noise goes quiet enough to work, write, browse, or think again.

…just wiggle your mouse.

(*or a shortcut, menu bar click, whatever's easiest for you)

Some context:

3.0 was Monocle rebuilt from the ground up. It was the new foundation. 3.5 is what makes it actually click. Most of the features here have been on the community's wishlist since 1.0 launched, and getting them all in finally feels like the version I always imagined.

What's new in 3.5:

• App Groups: Group two or more apps so they share focus. The whole group stays clear when any of them is active.

• Stage Manager Support: Monocle now detects Stage Manager automatically and keeps your window strip visible alongside the focused app. No toggle, no setup.

• Multi-monitor Blur, Finally Fixed: The long-standing bug where only one display would blur is gone. Multi-display setups are now fully supported, even with "Displays have Separate Spaces" disabled.

• Corner Peek: Flick your cursor to any screen corner for a quick peek under the overlay.

• Cursor Reveal Effect: A new way the overlay appears and disappears, fanning out from your cursor's position. Especially cool across multiple displays.

• Blur-Free Mode: Turn blur off completely and keep just tint, grain, and monochrome. For those who prefer the classic dimmer look.

• Plus a loooong list of polish: smoother auto-hide for the Dock and menu bar, license activation across user profiles on the same Mac, new scripting commands, settings UI/UX improvements, and lots of bug fixes throughout.


50% OFF discount:

As always, to celebrate the new release, I've hidden a generous discount as a new easter egg on the Monocle website. The sale is running until May 31, 2026. Good luck!


Latest mentions:

I tried the Monocle app for Mac and it completely changed how I handle digital distractions by blurring out everything except the window I am currently using by Tom's Guide (5/4/26)

Your Mac Is Missing All of These by Snazzy Labs (4/11/26)

10 Mac Apps That Will Change How You Use macOS in 2026 by MacRumors (12/24/25)


Oh, and make sure you read Monocle's Wall of Love, it's beautiful!
https://heyiam.dk/monocle/testimonials


Let me know your though :)

Dominik

Using this for ages now. Love the concept and the implementation. Looks so beautiful that i am tempted to not keep app windows in full size 😅

the app groups feature is the one that makes this actually usable for multi-window workflows - single window focus tools always broke down the moment you needed a browser and a doc open side by side. solving that is what takes it from a neat demo to a daily driver

Why didn’t macOS have something like this from the start? :)

I hate when too many windows are visible :) I keep every window maximized fullscreen so the others aren’t visible at all - they distract me. Though I have a MacBook Air, so the screen is small. With large monitors, that’s probably not as convenient.

the 'designed to feel like it came with your Mac' line is either accurate or it isn't and you find out in the first five minutes. that's the bar they've set for themselves which is either confident or reckless depending on the execution

I've seen your project at 1st launch I think. I find it very cool, congrats for the idea! I'm going to give it a try for sure

Question: I want something lightweight, do you have any stats on how much resources it takes when it is running? Or in the background?

cheers

I’m Tyler - CoFounder of Voker, and I’m so tired of being disappointed by AI hype claims. I bet you are too.

I studied physics in college, and worked in data science, ML, and analytics until founding Voker. I’m a skeptical person by nature (I think it's the scientist in me) and my gut reaction to any technology hype is to be cautiously optimistic until I see things proven out in data.

I felt this way about LLMs when they first hit mainstream. I knew they had real potential applications, but was also worried about the lofty marketing buzz they were getting.

AI as an industry has written checks that individual builders are left to cash. Promising full automation, PhD-level intelligence, and perfect results. As someone who's skeptical of that narrative, I still believe agents can genuinely deliver, but only if teams are rigorous about measuring performance in production. Every website or product has Amplitude or PostHog for click and pageview analytics; a standard way to understand who's using it and how. Agents have no equivalent, so we built Voker.

We are the Agent Analytics Platform where you can:

- Monitor your agents
- Measure their performance
- See what users are asking
- Know for certain agents are delivering for your users
- Optimize based on real data

You install our SDK, and Voker collects your agent conversation data, automatically detecting:

- User intents (Book me a hotel in Vegas for next Saturday with a poolside view)
- Corrections (No, that room doesn’t have a poolside view!! TRY AGAIN)
- Agent resolutions (Tool Result: Room Booked... Success!)

These automated annotations are the foundation for building a holistic view of agent performance and user behavior in one analytics platform.

We asked 100+ AI founders, product managers, and agent engineers how they monitor their agents in production and the answer was resounding: by combing through individual traces (with the occasional evals sprinkled in). They all reported that they depend on customer complaints to tell them when agents are messing up. We feel strongly that there is a third leg of the agent monitoring stool missing - Agent Analytics.

You shouldn’t have to wait for users to complain to learn that a recent prompt change is breaking your hotel booking agent, or that the AI finance advisor you built is calling the wrong tool to look up realtime stock prices.

Turns out the antidote to AI hype is simple: measure your agents diligently, then iterate until you get it right.

Your users deserve better AI experiences (we all do)!

Install the Voker SDK on our free tier (up to 2,000 events/mo), and start building better agents today:
https://voker.ai/

amazing team building something that’s really needed! Congrats on the launch 💗

Really cool that we can get an idea what people are using our agent for. The downside of having a powerful agent is that you don't always understand what people use it for and where it is not meeting expectations.

Oh this looks really interesting. How much of the setup is out of the box vs customizable?

Most observability tools treat agent calls as black boxes, logging tokens but missing the decision loop entirely. Building RetainSure's AI workflows, we struggled to attribute downstream outcomes back to specific agent choices. Our logging was ad hoc and we ended up rebuilding it multiple times. Does Voker capture branching decisions when an agent picks between tool calls, or is it focused on input/output tracing?

Automatic intent and resolution detection is the right abstraction. Most agent monitoring tools just log tokens or latency, but you actually need to know if the user got what they came for. We're building AI-driven customer success at RetainSure and agent quality drift between deployments is a real headache. How does Voker handle cases where the user's intent shifts mid-conversation?

Love the brutal honesty here AI has definitely written checks that devs are stuck cashing in production. Quick question on the SDK: how does it handle semantic variations for corrections? Will it catch things like actually scratch that versus no that's wrong out of the box, or do we need to train it on our own domain vocabulary?

Congrats on the launch!
How does Voker handle intent attribution when the agent proactively redirects the user, say, a billing agent that detects the user is actually in the wrong product area and routes them elsewhere? The intent the user arrived with and the intent the agent resolved can diverge legitimately, and in those cases it's not clear whether that should register as a correction event or a successful resolution. Curious how the analytics model handles that distinction, since getting it wrong would skew correction rates significantly for agents designed to reroute.

Prompting Vibes definitely don't scale when agents start failing silently in production. Being able to catch tool errors before a client screams at us is a huge lifesaver. great job @tyler_postle

How do you determine the quality of answers? I have an AI service with its own vector database. For almost any user question, we know the answer, provide tourist attractions, and we have more of them than ChatGPT. Will you be able to understand whether these are top-tier attractions or not?

Hey Tyler, went through Voker's site and the "Amplitude for agents" framing is honestly the cleanest take I've read on this gap. one thing I wanted to ask, how do you detect a "correction" automatically, is it sentiment delta on the next user message or something pattern-based? that label seems to do a lot of work in the product.

Do you also handle multi-agent, multi-turn orchestrations ?

Love this Guys!! Btw, does it track visibility down by country/market too, or is it focused on global signals for now?

Great work guys!

This is cool!! How can I share these actions to other teams?

Let's go! Congrats on the launch 🚀

How long do I need to monitor a prompt before I can get an insight?

Love this feature!

What a great work from the whole team building it!

What works for ChatGPT visibility doesn't necessarily work for Perplexity. Are the action plans tailored per engine? Congrats on the launch!

Hey, went through Insights by Omnia and the brand-in-LLM-answers tracking is a smart wedge. one thing I wanted to ask, given LLM outputs vary call to call, how do you measure visibility, are you running sampled prompts on a schedule or do you have something more deterministic under the hood?

That’s helpful — the tagging is exactly the piece I was hoping existed.

The next thing I’d watch is whether the insight carries enough “why this owner” context when it gets shared. A content person, SEO lead, and partnerships/PR person may all see the same citation gap, but each needs a different next action and confidence level. If the share/export preserves that owner-specific reasoning, it becomes much easier to hand off instead of just discuss.

Strong direction — monitoring without a next action is where teams tend to stall.

One thing I’d look for in the action plan is a clean split between “create or update owned content” and “earn/repair external citations.” They usually have different owners, timelines, and success signals. Are you exposing that split clearly in the plan/export so content, SEO, and partnerships/PR can each pick up their part?

I’ve tested more than 10 similar services. Everyone focuses on internal website optimization, and very few show which data ChatGPT was trained on, even though external sources are the most important once the site itself is already optimized. Can you show external websites and specific pages on them that were included in the model’s training and where information about us should be added?

Hey Product Hunt! 👋

Sergei here, the author and CTO of imgproxy.

Almost every product I have been involved in has dealt with images in some manner. So I know the pain of making the right choice for image processing. Building your own image processing solution is a huge time sink, and it can easily be done wrong. On the other hand, using a third-party service quickly becomes a money pit as your traffic grows.

That's why we created imgproxy – a self-hosted image processing server that just works. It's packed in a single Docker image, deployable anywhere in minutes, with no third-party dependencies. Combine it with a caching layer of your choice, and you have an enterprise-grade image processing solution that can handle any load.

Since its first release back in 2017, imgproxy has been adopted by thousands of developers and companies worldwide. You probably see images processed by imgproxy every day without even realizing it: Medium, Dribbble, Substack, and many more use imgproxy to serve their images.

imgproxy comes in two flavors: a forever-free open-source version and a paid Pro version with additional features and support. The Pro version is available as a subscription and includes features such as video preview generation, object detection, SVG minification, and more. Both versions are self-hosted, so you have full control over your data and infrastructure.

Today, we are happy to announce the new major release of imgproxy, version 4.0. This release includes a lot of new features and improvements, such as:

• Better performance thanks to parallel source image downloading and processing

• Internal cache for processed images

• Support for digital cameras' RAW formats

• Image classification

• Improved SVG minification

• ... and much more!

Check out the full release notes for all the details: https://github.com/imgproxy/imgp....

Drop us a line in the comments if you have any questions or feedback. We are always happy to hear from the community!

Hi PH!

Marina here, co-founder & CEO of imgproxy.

One thing we’ve noticed over the years is that more and more teams want to own their infrastructure again.
Not because self-hosting is trendy, but because image and video processing becomes surprisingly expensive and limiting at scale when it’s fully outsourced to SaaS platforms.

A lot of companies start with third-party image APIs because they’re convenient early on. But later they run into:

• unpredictable costs

• vendor lock-in

• limited control over caching and performance

imgproxy exists because we wanted a different tradeoff: a fast self-hosted system developers can run anywhere and fully control themselves. And honestly, seeing companies process billions of images through imgproxy every day is still surreal to us.

Huge thanks to everyone who tested v4 early and helped shape this release.

Hi PH! Viktor here, developer of imgproxy!

One thing we cared about from the beginning was building imgproxy as a sustainable long-term open-source project.

Why? Because the tools developers rely on most should be the ones they can actually own.

Not "own" as in pay for a license. Own as in run it yourself, on your own infrastructure, and trust it to do exactly what it claims. No hidden calls home, no lock-in, no surprises in your bill at the end of the month.

v4 is probably our most significant release in that spirit. We didn't just add features — we practically reworked the whole codebase. Cleaner architecture, better code quality, better performance. The kind of changes that don't make great screenshots but make a real difference when you're processing millions of images a day.

If imgproxy solves a problem for you, come say hi on GitHub. And if it doesn't yet — tell us why. That's how the best features get built

With the new internal cache for processed images in v4, how does cache invalidation work when the source image changes? Does imgproxy check source ETags or Last-Modified headers, or is expiry purely time-based?

Hey PH! We’re building Haystack to help teams deal with the explosion in the number of pull requests that need to be reviewed due to the rise of coding agents.

Haystack replaces the GitHub PR review system with a queue that triages each PR before a human has to read any diffs. It looks at the diffs, the codebase, and the coding-agent conversation that produced the PR. Haystack then routes it into one of three buckets:

1. Safe to merge. This means the PR has enough evidence behind it that the team can merge it without another human’s review.

Some examples:

• A small UI copy change that includes a screenshot showing the final state

• A backend change where the author clearly tested the important paths and ran the changes in a real environment

2. Needs fixes. This means that the PR has bugs or violates a rule in your codebase and therefore the PR needs to be fixed by the author.

Some examples:

• The agent was asked to make loading a large table faster by adding pagination, but the PR still loads every result at once and “implements” pagination in the UI

• The PR silently catches an error instead of logging, surfacing, or handling it. This violates the team’s “no silent error swallowing” rule

3. Needs human review. This means that the PR could not be sufficiently verified by the author or is touching a sensitive part of the codebase (determined by user-input guidelines) and thus requires human review.

Some examples:

• The PR changes a significant amount of logic in billing

• The PR changes an important user flow like onboarding, but the author only ran unit tests and never opened the app to check the flow end-to-end. That violates the team’s rule that high-impact user-facing changes need manual verification.

Instead of starting with line-by-line diffs, Haystack immediately tells the reviewer the goal behind the PR, what design decisions the author made (informed by their coding-agent conversation), and how much the author did to verify that the pull request works (e.g. run scripts, checked the frontend, etc.).

In this way, review shifts from “what changed?” to “is this the right behavior and is there evidence that it works?”.

Here’s a quick demo: https://www.tella.tv/video/strea...

We previously launched Haystack as a tool for understanding large PRs (https://news.ycombinator.com/ite...). As many of you can probably relate to, the release of Opus 4.5 completely shattered our conception of how fast an engineer could craft a PR.

And as coding agents got even better from 4.5, we realized that pull requests did not scale along with our coding velocity. With each member of our team being able to pump out more than 20 pull requests a day, code review quickly became cognitively exhausting and less helpful.

After talking with other folks, we learned many feel similarly, and currently face the binary option of either not doing review at all or trying to keep up with a fire hose of pull requests.

Haystack is our attempt at a third path. We still believe in code review, but as coding agents produce more code, human reviewer attention becomes more valuable and more expensive.

Haystack helps teams spend that attention on the PRs where a human can meaningfully change the outcome of that PR. And for such PRs, Haystack shows the reviewer what the PR intended to do, whether the author showed that it works, and what design decisions need a second pair of eyes.

We’re still quite early and are figuring out whether Haystack truly makes code review better. We would love any and all feedback!

Does Haystack handle monorepo setups where different services need different review rules — e.g. stricter verification requirements for a payments service vs. internal tooling? Or is the ruleset currently global across all repos in a workspace?

Hi. This is a real problem now - AI makes it easy to create PRs, but review time is still human.
I like the idea of routing PRs instead of treating every diff the same. What signals matter most for deciding “safe to move forward” vs “needs human review”?

The signal-to-noise problem with PRs is real. Does it work across multiple repos or is it scoped to one at a time?

Any reason as to your inspiration for naming it Haystack?

Hey Sami, went through ShioriCode's page and the "long sessions die in the terminal" line basically described my last week. one thing I wanted to ask, when you unify Claude Code, Codex, and Gemini in one workspace, are you normalizing the tool-call format under the hood or just rendering each one's native output? that abstraction call feels like make-or-break.

The project-aware threads plus streaming diff review solves a real problem. Agent CLIs have terrible UX for work that spans hours across multiple files. Building RetainSure, we kept hitting context window limits in long-running sessions and had to externalize state manually. How does ShioriCode handle continuity when an underlying agent reaches its token limit mid-session?

Really exciting to see Starchild-1 pushing world models beyond just visuals into real-time synchronized audio + video generation.

What stands out:

• Real-time multimodal interaction instead of fixed offline clips

• Continuous response to streaming user inputs

• Audio-video causal rollout for more immersive simulations

• Potential across gaming, robotics, education, healthcare & more

A big step toward more natural, interactive AI systems grounded in how the real world evolves.

Read more here: https://odyssey.ml/introducing-starchild-1

Technical report: https://starchild.odyssey.ml/starchild-1.pdf

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends

This feels especially useful for product-led teams where education is part of activation, not just “support docs.”

One feature I’d love in an LMS like this: tying lessons to product moments/version changes. Example: when a feature changes, show which lessons mention it, which users completed the old lesson, and what needs a small update vs. a full rewrite. That would make courses feel like living enablement instead of another stale content surface.

Congrats on the v1.0 launch! The "self-host or use cloud, your content stays yours" stance is exactly what makes this kind of platform actually usable for serious course creators. I teach Excel for financial modeling on Udemy (https://www.udemy.com/course/exc...) and the biggest pain point with most LMS tools is exactly what you're solving — the ability to embed code execution and AI Q&A grounded in actual course material. That's the gap between "video library" and "students actually finish and apply this." Curious how your AI handles formula-heavy or domain-specific content vs. general explanations.

Hey Badr, was reading through LearnHouse and shipping v1.0 after a long heads-down build is a respect-earning ship. one thing I wanted to ask, on the in-browser code execution across 30+ languages, is the runtime per-language sandboxed or you've got a unified container approach? that piece is usually the most expensive thing to scale on a learning platform.

Hey Orçun, was on AutoShelf's page just now and the menu-bar-watcher approach to file org is what pulled me in honestly. one thing on my mind, when two rules conflict on the same file, what wins, is it order of rule creation or a priority system? in my Downloads folder almost every file matches three rules at once.

Always think if I'm bad with Mac files or if it's a disaster haha. This is a great idea to manage everything as you want.

does it work with cloud drives like Dropbox/iCloud too? BTW, Congratz for the launch

Does AutoShelf support different rules per watched folder, or is it one global ruleset? My use case is Figma exports landing in Downloads alongside PDF invoices — ideally I’d route each file type to a completely different destination without them conflicting.

Folder chaos on my Mac has been a constant battle. Does it handle nested folder structures or just top-level organization?

Congrats man, been needing something like this lol my downloads folder is cooked

Hey Jun, spent a bit of time on Hanami's page and the "no AI-generated art" stance is what made me look closer. one thing I wanted to ask, how do you source attribution for thousand-year-old works, is it museum partnerships or public-domain curation with manual research? attribution at that depth feels like the hidden labor in the whole product.

Really nice! It would be great to also have text instead of the descriptions being audio-only :)

Hello everyone. I’m Amir, founder of Papr.

We built Papr Graph after seeing AI agents fail in production. The model wasn't the problem — retrieval was. Multi-hop questions, versioned policies, relational data — flat vector search breaks on all of it.

Vector search ranks by semantic closeness. But closeness ≠ correctness. A doc saying "aspirin reduces heart attack risk" and one saying "aspirin causes stomach bleeding" rank nearly identical — they're both about aspirin. For an agent making a recommendation, that's the difference between helpful and harmful.

Papr Graph is a graph-native embedding that sits between your existing embeddings and your agent. It encodes structured signals — topic, time, intent, entities, anything you define — directly into your embedding, so ranking reflects meaning in context, not just surface similarity. It's model-agnostic, works with whatever embeddings you're already using.

We saw Papr Graph improve existing embeddings on MTEB (coding, scifact, finance tasks) by 5-20%. On Stanford STaRK (MAG synthesized 10% dataset), Papr Graph leads retrieval models with 92% hit@5 accuracy.

Getting started is free. Keep your existing stack. Add our plugin. Drop graph-native ranking into your current retrieval flow with one API call.

Will be watching this. We use a voice agent and will need to have it in other languages. Congrats on your launch!

Hey Ashutosh, went through Thinnest's site and the broken-Hindi-voice-agent story honestly hit home. one thing I wanted to ask, with BYO STT/TTS across 100+ languages, are there benchmarks on which Hindi-English code-mix setups hold up in production? Indian users switch mid-sentence and that's usually where these break.

Congrats on the launch! 🚀 Quick question: I saw a pop-up to download the app on my PC earlier, but it disappeared and I can't find the install option anymore. How can I download it now?

@zairabbas Really cool seeing you build something from a personal problem first. Does the app also suggest healthier ways to prepare desi meals like using less oil or lighter ingredient alternatives or is the focus mainly on tracking calories and macros right now?

Congrats on the launch! Love this idea. I’m not originally from the UK and tracking food from my country has always been surprisingly difficult. Most apps just don’t get it. Really nice to see someone building for this.

If you all don't want to sign up and just want to try it first, you can try it from this link:
https://calog.cc/try