PH热榜 | 2026-04-30

Congrats @peter_tribelhorn

I’ve seen a lot of startups struggling with demo videos so definitely much needed tool! How does this compare to screen studio for product demos?

Hey Hera team! Congratulations and good luck with the launch. I tested your app a bit, can't say I got WOW effect at the first touch, maybe I need to experiment more with some real tasks. But definitely love the clear path to the first result and the UI

Screen studio videos all look the same now ngl. ready for something with more range

Hera sounds a bit funny as read by a Ukrainian..:) Congrats on the launch!

The "taste baked into the AI, not the user" call is the right one — most prompt-driven creative tools fail because they outsource judgment to people who don't have it (yet). Adjacent use case: I host the ModeLoop Podcast on financial modeling (https://open.spotify.com/show/0m...), and announcement videos for episode drops are exactly the recurring "ship fast, look good" need that's underserved. Quick question — does Hera handle audio-first content well (waveforms, episode covers, podcast clip animations), or is the current opinion-set tuned mostly to product-launch motion? Would buy day-one if so.

I’ve tried learning After Effects three separate times and I still can't get my easing right. 😂 The fact that you’ve handled the how it looks part so I can just focus on what it shows is the exact tool I need. Does it allow for custom brand font uploads, or are we limited to the baked-in presets? @peter_tribelhorn

Big congrats! Do the videos come in different aspect ratios for social, or is it mainly 16:9 for now?

I've been a video editor for years and motion design has never been my strongest area. This looks super useful

Just tried it and honestly, this is AMAZING. First try, 0 prompt, 0 effort and the video generated is great !! Congrats @peter_tribelhorn @chia_lun_wu1 @hyung_lee @garrytan

This is the sleekest animation tool we have ever used. Way better than the replit slop

Congratulations

Congrats on the launch. Is it possible to prompt and fix certain clips?

200K animations in 2 weeks is wild. What's the #1 use case you're seeing, social content, product demos, or something you didn't expect?

The director/designer/animator agents thing is interesting, didn't expect that approach.

Good job Peter! I was able to quickly make a decent video and with a few edits it's just what I need. What new features are coming?

hey PH! 👋 Lee here with Harris and Jerome — co-founders of Jupitrr AI. Pumped to launch again in this new AI coding era.

We've been building video tools for 2 years, and if you're a founder or consultant, you already know the pain:

You know you should be posting on LinkedIn, Instagram, YouTube, TikTok. Every VC, every founder you admire is doing it. But between shipping product, talking to customers, and closing deals, the idea of making a video always slips.

So we rebuilt everything into one workflow.

VideoOS takes you from idea to published video in minutes.

Here's the loop:

→ Find trending videos in your niche (so you're not guessing what works)

→ Remix them into your own briefs (fully collaborative with your team)

→ AI writes your script in your voice (we studies all your previous transcript)

→ Record with our Teleprompter App — line-by-line, auto-trimmed (we're the first to do this)

→ Auto-edit with subtitles, B-roll, and music

→ Publish directly to LinkedIn, YouTube, TikTok, IG
→ Cross-platform analytics (so you know what to improve on next)

We want to make making videos less overwhelming — no jumping in and out of Notion, Canva, CapCut, Buffer just to get one video out.

We want you to be consistent, because your builder story deserves more attention.

Built for solo founders, consultants, and small marketing teams who want to show up on video without it eating their week. (Not the AI slop kind.)

Do try it out. Would love any feedback! ✌️

the line by line teleprompter is the actual unlock here. one bad take in the middle of a 2 min recording means redoing the whole thing, which is why most founders quit after take 4 and never post. cutting that loop changes the math on shipping consistently.

does the voice indexing work with a small corpus, or does it need a decent backlog of transcripts before it stops sounding generic? curious where the floor is for someone just starting out

congrats on the launch ✌️

Absolute banger of an update! Jupitrr never stops leveling up and VideoOS feels like a whole new era. This trio of founders is built different! Can't wait to see where they take it. Big Kidos! 🚀

Hi PH folks!👋🏻 Harris here, one of the co-founders of Jupitrr AI!

We started as an AI video editor that helps businesses create videos.
As time goes by, we realized editing was never the real problem.
I mean, there are loads of video editors out there, but business creators still struggle.

The real problem is that most people can't stay consistent with video — not because they lack talent, but because they don't have a system.

We're seeing users are on 6–8 different tools: research here, scripting there, recording somewhere else, editing in another app, publishing manually to each platform. It's exhausting so most give up too early.

So we rebuilt the whole product.

VideoOS is a video engine.

It learns your tone of voice, helps you research topics worth making, writes scripts that actually sound like you, guides you through filming, edits, and publishes to every channel — all in one place.

We've been building Jupitrr for 3 years and have ~200k users, and we're always listening, always shipping🚢

Would love your feedback — especially if you're a marketing team or small business owner who wish to gain exposure and trust through videos!

Super excited to finally put this out here 🚀

That trio-level founder energy is exactly what makes an update feel like a new era rather than just a patch. Curious — when you're shipping this fast, how do you decide which user requests hit the cutting room floor vs. get fast-tracked into the next build?

Does it only support english content?

Hey PH! 👋 Jerome here, CTO and the third co-founder of Jupitrr AI.

Lee and Harris covered the why. Let me share the how — because building VideoOS broke a lot of our assumptions about what an AI video product even is.

The hardest part wasn't the AI. It was the orchestration.

A single video in VideoOS touches: trend scraping → semantic search across your past transcripts → voice-cloned script generation → a native teleprompter app that auto-trims line-by-line as you record → multi-track auto-editing with B-roll/subtitles/music → cross-platform publishing → unified analytics.

That's ~7 systems pretending to be one product. Stitching them so it feels like one button took most of the year.

A few things I'm proud of we shipped:

→ Line-by-line teleprompter recording — no one else does this. You read one line, it auto-cuts, you read the next. Bad take? Re-record just that line. Editing happens while you film.

→ Voice-aware scripting — we don't just prompt an LLM with "write like a founder." We index every transcript you've ever made on Jupitrr and ground generation in your actual phrasing.

→ One graph, many platforms — LinkedIn, YouTube, TikTok, IG each have different aspect ratios, hook windows, caption styles. We render once, adapt automatically.

If you're a builder, I'd love feedback on the recording flow specifically — that's the piece I think changes the game, and

the piece I most want to get right. Roast it. 🙏

Built with Harris and Lee — and yes, a lot of Claude. 😄

Editing videos takes such a long time and it feels like such a time sink as a founder. I know that video content is what's trending on LinkedIn but having to spend time editing feels like time that could be better spent on product or sales. When generating scripts, are you able to feed it context to better tailor the ideas to a specific topic?

Watch the whole product walk through. This was brilliant. Really well done, mate.

The calendar, the B-roll, the research, the integration with the stock images, the teleprompter. It's a great solution. Maybe feels a tad bit manual. But I'm not necessarily sure how much more you could automate there. I guess it's just the back and forth between mobile and desktop. But it looks good nevertheless.

Congrats on the launch! The mini pivot looks amazing.

The "idea → published video" gap is the real bottleneck for niche technical creators — and it's underrated how much friction sits in the script + edit loop, not the recording itself. I run Mod3Loop, a financial modeling YouTube channel (https://www.youtube.com/@Mod3Loop), and the trending-topics + AI-script pairing is exactly the workflow I've been hand-stitching across five tools. Curious whether the script generator handles technical jargon (terms like LBO / IRR / DSCR) without flattening them into generic finance copy — that's where most AI script tools fall down for finance content.

Awesome product! Congrats!

Congrats, guys! Finally, a tool that doesn't make me bounce between five different apps to post one video, you rock it

Looks awesome. Congrats on the launch. Gotta try this asap.

Congratulations on the launch!

Hi Product Hunt community 👋
I'm Hahnbee, co-founder of Mintlify, and we're excited to introduce the new Mintlify Editor - a new way for every team to build and maintain knowledge.

What is it?
A git-synced, WYSIWYG, live-collaborative, and AI-native editor built for cross-functional collaboration.

Why now?
Fragmented knowledge isn't just a collaborative problem anymore, it’s an agent problem. As organizations scale, knowledge sprawls. from different teams, tools, and sources of truth. AI agents need accurate, unified knowledge to participate effectively in your workflows.

What's included?
- WYSIWYG editing with slash commands, custom components, and navigation management. No markdown or yaml required.
- Bi-directional git sync: CLI and IDE-native for developers, web-native for everyone else
- Live collaboration so your whole team can jump in and ship together in real time
- AI-native support so agents can collaborate on your docs alongside your team

Let us know what you think and any feedback!

Excited to put this out! A huge unblock for how teams create knowledge bases 🚀

just wow.

i recently swapped our vibecoded docs site to mintlify and wish we had done it a long time ago. mintlify is unquestionably the best solution for docs - this release makes that even more crystal clear.

huge congrats @hahnbeelee @han_wang6 @shawn_lestage1 and team!

Knowledge fragmentation is the same problem in finance, just dressed differently — every deal team rebuilds the same project finance scaffolding because the templates and assumptions live in a senior modeler's local Excel folder, not in a shared, queryable layer. The git-synced + AI-native bit is what actually makes this an agent-friendly knowledge base instead of just a fancier Notion. Adjacent concept: I publish a library of project finance and valuation templates (renewables, M&A, LBO) on Eloquens (https://www.eloquens.com/channel/samir-asadov-cfa) and the structured-once-then-reused pattern collapses time-to-first-IC by weeks. Curious if Mintlify's roadmap includes structured-data blocks (tables, formulas) for non-prose knowledge?

we have been using mintlify from a long time! its a great product.

Very cool, is it a webapp or desktop app? and can i do it from mobile

This looks awesome!

I'VE BEEN WAITING FOR THIS

the "knowledge fragmentation is an agent problem now" framing is the right one. agents inheriting whatever sprawl ur org already has is the actual pain, not just humans not finding things.

curious if the editor surfaces conflicts when two sources of truth disagree, or if it just serves whichever one the agent hits first..

We have been using the editor every day and it's an absolute game changer!!!

I love this product and I think it's one of the leading editor experiences I have tried. Also, the ability to publish customer facing docs is just amazing.

The "knowledge fragmentation is an agent problem now" framing clicked immediately. We have a similar issue where agents picking up context from stale internal docs make worse decisions than agents with no context at all. Garbage in, garbage out, but at inference speed. Does the editor version-gate content in a way agents can reason about, or is it still just "latest wins"?

Have been using Mintlify at Helicone for a long time! The web editor is the thing that finally gets non-engineers contributing more to docs. No more "can you push this typo fix for me" pings.

We're mostly a Git user with Mintlify but super pumped to try out the editor!

Hey PH fam 👋

Thrilled to hunt Wonder on Product Hunt today. This one hits close to home.

I've watched the design-to-code handoff break teams for years. Designers craft something beautiful in Figma. Developers rebuild it from scratch. And somewhere in between, the original intent gets lost.

Aibek and Boris lived that pain firsthand building Superflex. Then they asked the right question:

What if design and code never had to be separate in the first place?

That's Wonder.

It's an AI design agent that lives directly on the canvas. Not a generator you prompt and wait on. An actual agent you collaborate with in real time.

→ Describe what you want. It designs it.

→ Click any element. Refine it on the spot.

→ Nothing is a static image. Everything is editable and ready to ship.

You can build websites, app UI, marketing graphics, pitch decks, and UI components — all from one canvas.

And here's the part that made me stop:

Wonder ships with an MCP server. Connect it to Cursor, Claude Code, or Codex and your coding agent pulls the design straight into your codebase. No screenshots. No rebuilding. No lost craft.

The gap between "I designed it" and "it's in production" just got a lot smaller.

Still early — public alpha, free to try with credits included. But the vision is clear and the team in Serbia has been shipping nonstop to get here.

Big congrats to Aibek, Boris, and the whole Wonder team 🙌

Drop your questions and feedback below 👇

Hey Aibek! Love the idea. How is the business model? Is it like a subscription? Wondering if brands or founders that need one shot designs gonna take it if so. Anyway it looks awesome and I wish you all the best here!

Nice work!

Over the past few months, a few tools have launched in this space.. how do you guys compare against them? (e.g. https://pencil.dev/, https://paper.design/, http://figr.design/, Claude Design, etc.)

Great work!

MCP handoff to Cursor/Claude Code is the part I'd watch — most design-to-code tools fall apart on token consistency once you hit a real codebase. Curious what the round-trip looks like: do design tweaks in Wonder reflect back from code, or is it one-way generation?

lets goo!

Deep Research Max by Google DeepMind is a powerful autonomous research agent built on Gemini 3.1 Pro 🚀

It tackles the biggest problem in research today... time-consuming, fragmented, and shallow analysis by automating deep, multi-source research workflows into fully cited, high-quality reports.

What stands out is its ability to combine open web + proprietary data (via MCP), generate native charts/infographics, and iteratively refine insights for expert-grade output.

Key features:

• Autonomous long-horizon research workflows

• MCP support for custom data integration

• Native visualizations (charts & infographics)

• Multimodal inputs (PDFs, CSVs, media)

• Real-time reasoning + collaborative planning

Perfect for analysts, enterprises, and teams in finance, life sciences, and market research who need fast, reliable, and deep insights.

If you're building or scaling with AI agents, this is worth exploring!

ALready got my Claude Code on this!

MCP-native research agent in the Gemini API is a smart positioning play vs hosted-only Deep Research alternatives. Two q's: when sources contradict, does it surface the disagreement in citations or just pick one? And any quota visibility for async DR Max jobs before they blow past budgets?

Very keen to see how it stacks against Parallel in my workflow:)

Hey Builders! 👋 I'm Tessa, founding [technical] GTM at Tabstack.

Tabstack is a web data and automation API with intelligence built into every call. You don't get raw content back to parse, clean, or run through another LLM. You get the output your product or agent needs, already done.

Five endpoints:

• /extract/json — pass a URL and a schema, get back JSON that matches it

• /extract/markdown — clean markdown from any URL

• /generate/json — custom instructions, structured output back

• /research — multi-source research with citations, one call, no orchestration

• /automate — managed browser agent for JS-heavy pages, forms, multi-step flows

No scraper to maintain. No pipeline to build. No Monday morning incident because a site changed its data structure.

I joined this team because Mozilla has always believed the web should stay open and your data should stay yours. Ephemeral data, zero model training, robots.txt compliant. That's not a feature—it's the foundation.

Add it to Claude, Cursor, or Claude Code via MCP in 30 seconds. Check out the docs →

What use case are you reaching for first? I vastly improved a messy data parsing pipeline the first time I tried it.

A few other things I've built since joining Tabstack just 4 weeks ago:

• Rival — open source competitive intelligence tool powered by Tabstack. Tracks competitors daily, detects changes across their site, pricing, docs, jobs and social, and surfaces live intel via MCP whenever you need it for strategy. Uses all five Tabstack endpoints.

• LocalPlate — open source self-hosted meal planner. Imports recipes from any URL using Tabstack's extraction and automation endpoints.

• Scout — prospect intelligence, signal feed and CRM. Uses Tabstack to enrich prospects with structured profile data, synthesize ICP fit scores and outreach briefs, and run deep-dive research — all automated.

We're using Tabstack internally with some of our products b/c it out-performs many of the others we have tried against sites that have traditionally been harder to extract structured data from.

Congrats on the launch!

Interesting, does it handle different site format?

Also dynamic sites?

congrats on the launch! Does it require any integrations or extensions for Mozilla specifically or browser agnostic?

The schema-as-contract model is the right call. Most scraping tools dump raw content and make you figure out the mess. Curious what happens when a field can't be populated though, does the call fail hard or return a partial result with nulls? That distinction matters a lot for pipelines that chain multiple calls.

"Pass a URL and a schema, get back JSON that matches every time" — what does "every time" mean in practice? Does the schema act as a strict contract where the call fails if a field can't be populated, or does it return partial data with nulls? Would change how I'd design error handling around it.

Also The "cited answers" from research calls is the detail I keep coming back to. Are the citations actual source URLs pulled from the pages it visited, or more like attribution to the top-level domain? Big difference if you're building something where the downstream user needs to verify the source.

So sick!

Hi everyone!

Gemini App keeps choosing to go broad.

This update brings the output much closer to everyday workflows: Docs, Sheets, Slides, PDFs, Word, Excel, CSV, Markdown, LaTeX, and more can now be generated directly inside the chat. No more copying a good answer, pasting it elsewhere, and fixing the formatting by hand.

That says a lot about Gemini app’s direction. It is trying to be a wider productivity surface that fits into the apps and file formats people already use every day.

The LaTeX support is the interesting one here. Generating a `.tex` file is easy enough, but does it come pre-structured with proper packages declared in the preamble, or is it more of a raw content dump that still needs a working template wrapped around it before it'll actually compile?

Hey Hunters!

I'm the author and project lead of Quarkdown. What started as a simple university research project eventually evolved into a full-fledged typesetting system for writers, developers, and anyone who cares about control.

I've been building open source software for the past 10 years. Making things is what I love the most, and I'm driven by passion for what I do.

If you try it out, please share your feedback! And if not, a star goes a long way ⭐️

Hey all, thanks for the support!

I’m aware of the typo in the subtitle: a Product Hunt moderator changed my original description right before launch and left a typo there that I’m not able to edit out! I’m waiting for support to edit it for me

Hey Giorgio, congrats on the launch! markdown + latex in one system is an achievement by itself, but the syntactic ambiguity is brutal. How are you resolving the underscore problem (italic in markdown vs subscript in latex)? mode-switch on $..$ delimiters, or some smarter context inference? anyway, best of luck!

The Markdown-syntax-with-LaTeX-power angle is the right pitch — Typst nails the latter but loses the muscle memory most writers already have. Curious how the same source compiles for academic PDFs (figure floats, refs, two-column) vs slide decks — same renderer, or separate backends?

Markdown + LaTeX in one toolchain solves a real pain for anyone teaching technical material — the eternal trade-off between "easy to write" and "professional output" usually forces a Word doc or a Notion page that can't render formulas properly. From a curriculum design angle, this could be a meaningful unlock: I teach Excel for Financial Modelling on Udemy (https://www.udemy.com/course/excel-for-financial-modelling/) and the supporting handouts (DCF formulas, IRR derivations, NPV calc walkthroughs) are exactly the case where a Markdown-first authoring flow with proper math rendering would have saved me weeks. Quick question — does Quarkdown export to printable PDF with selectable formulas, or are equations rasterized?

Symphony – Open-source spec for orchestrating coding agents 🚀

What it is: Symphony is an open-source specification from OpenAI for orchestrating coding agents, turning your issue tracker into an always-on execution engine.

Problem → Solution: Managing multiple coding agents creates context-switching overhead. Symphony solves this by assigning agents directly to tasks, automating execution without constant human supervision.

What makes it different: Instead of managing sessions, Symphony uses your task tracker (like Linear) as the control plane, agents continuously pick up and execute work in parallel.

Key features:

• Agent-per-task orchestration

• Continuous execution + auto-retries

• Workspace isolation per issue

• Built-in observability & logging

• Scales parallel work via DAG-based execution

Benefits:

• Up to 500% increase in shipped PRs

• Reduced cognitive load for engineers

• Faster experimentation & iteration

Who it’s for: Engineering teams, AI-native dev workflows, and builders leveraging coding agents at scale

Use cases:

• Automating feature development

• Large-scale refactoring

• Parallel task execution across repos

• AI-driven product development

If you're building with AI agents, this is a glimpse into the future of software workflows.

Chatgpt-5.5 has came back after a while of being on the DL. This is now doing more of my work than any other model, quite nice to use.

Mistral just shipped their most capable model yet, and it runs self-hosted on four GPUs.

What it is: Mistral Medium 3.5 is a 128B dense model that merges instruction-following, reasoning, and coding into a single set of weights, with a 256k context window and configurable reasoning effort per request.

Most frontier-class models either require massive infrastructure to self-host or lock you into proprietary APIs.

Mistral Medium 3.5 sits in an interesting position: it scores 77.6% on SWE-Bench Verified, ahead of models like Qwen3.5 397B A17B, while running on as few as four GPUs.

The reasoning effort is configurable per call, so you're not paying or waiting for deep reasoning on a simple reply, but the same model can handle a multi-step agentic run.

What makes it different: This is Mistral's first "merged" flagship model, meaning instruction-following, reasoning, and coding live in one set of weights rather than being split across specialised variants.

The open weights are released under a modified MIT license on Hugging Face, and it's already the default model in both Mistral Vibe and Le Chat.

The vision encoder was trained from scratch to handle variable image sizes and aspect ratios.

Key features:

• 128B dense model, 256k context window

• Configurable reasoning effort per request

• 77.6% on SWE-Bench Verified

• Open weights on Hugging Face under a modified MIT license

• Self-hostable on 4 GPUs

• API at $1.5/M input tokens and $7.5/M output tokens

• Powers Vibe remote coding agents and Le Chat Work mode (Pro/Team/Enterprise plans)

• Available on NVIDIA build.nvidia.com and as an NIM container

Benefits:

• Run a frontier-class model on your own infrastructure without a large GPU cluster

• Tune reasoning depth at the API level, useful for cost-sensitive agentic pipelines

• Single model handles the full range from quick chat replies to long-horizon coding tasks

• Open weights means fine-tuning, auditing, and on-prem deployment are all on the table

Who it's for: Backend and ML engineers evaluating open-weight alternatives to proprietary frontier models for agentic pipelines, coding tools, or self-hosted inference.

The interesting design choice here is the merged weights architecture.

Most labs at this capability tier still ship separate reasoning and instruction models.

Collapsing them with configurable effort per call is a practical tradeoff that's worth watching as other labs respond.

Hey Product Hunt 👋

We built Invite Only because WhatsApp was never meant to host events - we just never had anything better.

Until now.

One link. Your guests see a stunning invite, RSVP in 30 seconds, and pay via UPI - all without downloading anything. You get a real guest list, not a chat thread.

We shipped this in 2 weeks. Try it for your next event. It'll feel like cheating.

Would love your feedback 🙏

Does it work with phone numbers outside India? I tried to create an event but can't publish because I lack an indian phone number...

After a lot of sleepless nights, hundreds of commits, and more edge cases than I'd like to admit - Invite Only v2 is finally live. Built the entire stack from scratch - real-time RSVPs, UPI integration, glassmorphic UI that actually works on budget Android. Proud of what we shipped.

Can't wait to see it in the wild 🙏

We’re encountering an issue with event publishing. Even after marking an event as “Live,” its status continues to remain in “Draft.” Additionally, the event link results in a 404 error, indicating that the event is not being properly published or routed.

This suggests a possible gap between the publish action and the backend/state update or routing layer. Needs investigation on event status sync and URL mapping.

Congratulations on the launch! I was curious, how do you handle multi step flows with authentication like does it manage session state across or does each test start fresh?

Hey John! The 'in your browser' bit is the hard part I guess! universe-scale rendering means lod strategy is the make-or-break call. Are you streaming pre-built tiles for the deep zoom, or is it procedural extrapolation outward from real catalogs (gaia, sdss) with hand-curated near-sol detail? curious how you handle the seam between real data and the procedural fill. congrats on the launch, good luck!

Hey everyone,

Quick context on why I built AstroGrid : I wanted to make something for anyone who's curious about the universe, whether you're a student, a self-learner, a teacher, or just someone who likes staring at space at 2am. Every explainer I could find was either a flat textbook diagram or a YouTube video you'd zone out of in 30 seconds. I wanted something you could just grab, tilt, spin, and fly through.

It kept growing, and now it's a 3D atlas of the universe you can explore in your browser. No install, no signup. Built with classroom and self-learner use in mind.

A few things you can actually figure out by using it

How big things really are. Park Earth next to Jupiter, then next to the Sun. Numbers like "100x larger" don't really land until you see it. Same with distance: flying from Earth to Mars at a fixed speed feels very different from reading "225 million km."

Why Kepler's laws look the way they do. Crank the time speed up and watch Mercury whip around while Neptune barely moves. You stop needing to memorize the law and just see why it has to be true.

Why eclipses don't happen every month. The Moon's orbit is tilted, and that's almost impossible to picture from a 2D diagram. In 3D it's obvious in about five seconds.

Constellations aren't flat. Fly sideways out of the solar system and watch Orion fall apart. The "belt" is actually three stars at wildly different distances. This one tends to genuinely surprise people.

What a black hole does to light. There's a real general-relativity lensing effect around the black holes, and you can orbit one and watch the background stars bend around it. Hard to forget once you've seen it.

The scale of the universe. Zoom out from your street to the Solar System, to the Milky Way, to the Local Group, to the cosmic web. The sheer emptiness between things is the part that hits hardest.

Everything is based on real astronomical data (NASA JPL for the Solar System, the HYG catalog for stars, OpenNGC for deep-sky objects, LIGO data for gravitational wave events, etc.), so what you're seeing isn't decorative. It's where those things actually are.

If you teach, or you're just curious about space, I'd really love to hear what's confusing, what's missing, or what you wish you could do next. That feedback is what shapes the next update.

Thanks for taking a look.

The "flat textbook diagram → actually fly through it" reframe is exactly what's missing from most explanatory media. We've been overusing static visuals for things that only make intuitive sense in motion. Adjacent take from a different domain: I built StoryRoute (https://storyroute.netlify.app/), an interactive city-walk app that does the same thing for travel — instead of reading a top-10 list, you walk the city as a narrative. Same instinct: replace passive consumption with embodied exploration. Curious whether AstroGrid plans curated guided tours (Voyager probe trajectory, Kepler's discovery walkthrough), or stays open-ended exploration?

the constellations point is the one that actually changes how you think about them permanently. once you know Orion's belt stars are at completely different distances it's impossible to unsee. curious whether the classroom use case shaped any specific design decisions, or did that come later once you saw how people were actually using it?

Super pumped about this. The part that took the longest wasn't generating SQL, it was figuring out what "a thoughtful default" actually means.

When someone types "everything about new signups," what do they really want to see first? Which chart type carries the headline number vs. the breakdown? How many KPIs before it stops feeling useful? We rebuilt the layout logic a bunch of times, and the version that shipped is the one where dashboards started feeling like something a person made, not something a model generated. Which is super cool!

Would love to hear what kinds of dashboards you try it on... the weirder the prompt, the more useful the feedback!

This is so good. Feels like the result of a lot of hard work, product vision and full-stack polish coming together. This is the way to interact with data.

The "thoughtful defaults" piece is where I'd expect this to break down in practice. If I ask for "MRR and churn this quarter" and my data model has three different tables that could plausibly be "revenue" — does the agent ask me to clarify, or does it just pick one and hope? How does it handle ambiguous schema without turning into a back-and-forth chatbot?

Hi there! I’m Sacha, cofounder of Tinfoil. Excited to share what we’ve been building!

Tinfoil gives you a familiar AI chat (browser and iOS) and an inference API, featuring the latest open-source models like DeepSeek V4, Gemma 4, Kimi K2.6, and GLM 5.1. However, with Tinfoil all data is stored end-to-end encrypted and processed privately, even during inference.

Backstory: my cofounders and I have always been very aware of how important privacy is, especially with tools like AI chatbots that we use daily for personal discussions and to process our thoughts. We strongly believe nobody should be privy to these chats.

I did my PhD in cryptography and internet privacy, and was an early user of ChatGPT. My cofounders and I quickly realized that the amount of control we were giving up to get access to powerful AI like ChatGPT was simply unprecedented, and frankly creepy. We found ourselves hesitating when sharing certain things or wondering if our deleted conversations would end up in the next training cycle. Today, we’re all leaking our brains to AI labs. Tinfoil is the Flex Tape to stop that.

The latest NVIDIA GPUs have built-in support for secure enclaves. These are security mechanisms built into the hardware that allow running LLMs in a way that keeps data private during processing. Nobody, not even the operators of the GPUs, can see the data being processed. Secure enclaves also allow you to perform remote attestation, which means the chip can return a signed fingerprint of the code and security configuration currently running inside the enclave. This means you don’t have to take our word that your data is secure, you can actually check it yourself:

1. All the code running in the enclave is open source and it’s fingerprint is pinned to a transparency log, Sigstore. You can inspect this code yourself and verify that it’s secure.

2. When you connect to our chat or inference API, the client fetches the attestation report from the enclave it is connecting to.

3. The client checks that the fingerprints match. If they do, the server is running exactly the code that we claimed it would be running.

This whole process happens automatically, so you always know that you are connecting to a trustworthy service. If you’re curious, you can read more about that in our docs: https://docs.tinfoil.sh/verification/verification-in-tinfoil

Apple, Meta and others have been using secure enclaves to build private AI for their own apps and services, but Tinfoil’s goal is to give everyone else the ability to build verifiably private AI applications with state-of-the-art open source models. We put a lot of effort into removing the friction that security & privacy tends to introduce, so we’re excited to hear what you think!

Pricing:

• Chat: $20/month but you can try it out for free!

• API: $5 in free credits when you sign up.

At my last employer in the nuclear space, I remember how much effort there was put into security and privacy -- those concerns definitely put a damper on AI adoption that would have otherwise accelerated development. Glad to see that there is innovation in this space so that privacy is less of a concern (and provably so).

Love this. Really thoughtful product and the verification piece is especially compelling.

One question I had: for teams or developers building on the API, where have you seen the biggest tradeoff between verifiable privacy and product usability or performance, and how have you tried to minimise that?

Feels like that balance is probably where a lot of adoption decisions get made.

The NVIDIA GPU secure hardware angle is interesting. Are you using confidential computing features like TEEs (Trusted Execution Environments) on the H100s, or something else? And does the privacy model hold if someone at NVIDIA has access to firmware-level hardware behavior?

Secondly "Everyone else is locked out" — does that include Tinfoil itself? Like, can your team see conversation logs for debugging, abuse monitoring, or anything else? That's usually where the asterisk hides in these setups.

Amazing api! Literally just works, and they really ahead of the times :)

Will be huge

we love using tinfoil!! not enough people are thinking about security, and are sending all their private data to claude and openai. tinfoil solves that! happy customer here :)

I'm building proprietary systems, and didn't want my IP sitting in a mainstream AI provider's logs or training pipeline. Tinfoil has been a huge unlock for me.

I leaned on it heavily during design and prototyping to have a strong LLM thinking partner. Tinfoil has let me keep my existing LLM workflow without running a local model that would roast my laptop.

The architecture they're using- hardware enforced privacy via secure enclaves with client-side attestation of the CPU/GPU inference server is the best approach I've seen for capable and private AI.

A few specifics:

• In-session UX is comparable to mainstream non-private chat services.

• Passkey/recovery flow had a couple rough edges early on, but the team is actively improving this and Sacha was super responsive to help me.

• Also recommend the documentation if you're the type of person that wants a primer on secure enclaves and a detailed breakdown of their architecture. I had a fun time nerding out on it

ElevenLabs built a creator royalty model for music, and ElevenMusic is where it lives.

What it is: A music platform combining AI-assisted creation, remix tools, independent artist discovery, and an engagement-based earnings model, on web and iOS.

Problem and solution: Independent artists building audiences outside mainstream platforms have a distribution problem and a monetization problem at the same time.

ElevenMusic addresses both by giving creators a publishing surface with a built-in royalty mechanism modeled on ElevenLabs' voice library, which has already paid out over $11M.

Listeners on the platform can also remix tracks by shifting genre or tempo, which extends the reach of each original track and feeds the engagement metrics that drive creator earnings.

Earnings depend on listener engagement and eligibility thresholds, so this is performance-based rather than a guaranteed payout.

What makes it different: Most AI music tools are generation-first.

ElevenMusic is discovery-first, which changes who shows up. Starting with a catalog of 4,000+ independent artists and wrapping AI creation and remix tools around that catalog means the platform has content before the creators arrive.

The composition tools let you start from a lyric, melody, or mood and develop a full structured track, which is a lower barrier entry point than a DAW.

Key features:

• 4,000+ independent and emerging artist catalog

• Genre and tempo remix on any track

• AI composition from lyric, melody, or mood

• Publish and earn through engagement-based royalties

• Curated playlists, daily mixes, live sessions

• Web and iOS

Benefits:

• Single platform covers the full loop from listening to earning

• Remix mechanic turns any track into a creative starting point

• Royalty structure has a proven precedent in the ElevenLabs voice library

• Catalog skews toward emerging artists, not mainstream catalogs

Who it's for: Independent artists and producers looking for a single platform to publish, grow an audience, and earn from original tracks without managing separate distribution infrastructure.

The creator economy angle here is worth watching. If the engagement-based royalty model scales across music the way it did for voice, this becomes a meaningful alternative surface for independent artists. That's the bet.

The remix piece is what I want to understand better — when you remix an independent artist's track, how does the royalty split work? Does the original artist get a cut automatically, or is it opt-in on their end?

Hey PH 👋

I'm Aryan, one of the co-founders.

We built SuperMind because I was tired of being the last approval in every chain at my own company. Every follow-up, every client update, every internal decision, it all routed through me personally.

SuperMind is what I wished existed. A system that actually runs the business, not just a smarter way to write emails.

Happy to answer anything. Ask me the hardest question you have about whether this actually works. I'll answer honestly.

Hey ProductHunt 👋

I’m Damanjeet, one of the co-founders of SuperMind.

We built SuperMind to feel less like one AI chatbot and more like an AI team for your business - with agents for Growth, Sales, CTO, CS, Finance, HR, Legal, Product, Ops, Analytics, Recruiting, and more.

Would love your honest feedback.

Where would you want an AI agent to help first?

Hey PH

I’m Yuvraj, one of the co-founders of SuperMind.

I built this because I was exhausted from being the final bottleneck in my own company.

Every client update.
Every follow-up.
Every internal decision.
Every approval.

Somehow, it all ended up on my plate.

SuperMind is the system I wish I had back then not just another AI tool that helps you write faster, but one that helps your business actually move without everything depending on you.

It follows up, coordinates, tracks decisions, and keeps work moving like an operator inside your company.

Happy to answer anything.

Ask me the hardest question you have about whether SuperMind actually works, where it breaks, or why it’s different.

I’ll answer honestly.

Hey Damanjeeet, How do you make sure the agents don't hallucinate and are alligned with business goals?

Hey @damanjeet_singhh Is it possible to have multiple projects running in parallel?
Can the customers select the models?

Hey Damanjeeet! How do you get the initial context about the user's company? Cool Supermind builds memory and understands business better & better, but how to get it up & running quickly?

"One tap approvals" is the thing I keep coming back to. When a Legal agent flags a contract issue AND a Finance agent flags a cash flow concern on the same morning briefing — do they stay siloed, or does SuperMind surface the connection between the two? Feels like that cross-agent reasoning is where this either becomes genuinely useful or just a fancier notification inbox.

Does it offer any help in terms of thinking about content that shoud be added to these announcements ?

Congrats on launch! Quick one: How do you measure whether a feature truly gets adopted vs just seen?

Hey Alexandre! Looks cool & I appreciate the pricing structure!

I'm curious - how easy is it to implement? I feel like this pricing favours Adoptly for vibe-coders etc, who need such updates (they ship a lot, they don't have much $ to pay for it). + does it have API/MCP?

thanks!

the mac dock has been the same shape forever, so adding real grouping is a meaningful shift. quick question on the interaction model: when you group, are the apps stacked inline like macos stacks for downloads, or is it a separate flyout panel on click? big difference for muscle memory once you've got 20+ pins.

The app doesn’t seem to be working properly on my side. In addition, it appears to have affected system behavior, specifically automatic Dock hiding has stopped working.

I also tried fully removing the app, but the issue persists.

Hope this can be resolved soon, happy to provide more details if needed.

MacOS 26.4.1

Hey Product Hunt! 👋

I'm Abhishek, CEO at KushoAI. I'll be honest about why we built this.

Playwright is great. The problem isn't the tool. The problem is that writing comprehensive tests by hand takes forever. You write a test, a selector breaks, you rewrite it, the flow changes, you do it again. You end up shipping with partial coverage because you're out of time.

Here's what we built:

Record your user flow once using Playwright's browser recorder. Close the browser, and the generated script opens in your terminal editor. Review it, tweak it, save it.

Then our TUI orchestrates LLM calls to expand that single recording into comprehensive test coverage: different input combinations, boundary conditions, error states, edge cases. You can guide it with plain-English instructions ("add error handling tests", "test with special characters").

The whole workflow stays in your terminal. No tab-switching to ChatGPT, no copy-pasting context, no manual prompt engineering. Record in browser, generate in terminal, run tests, get HTML reports.

Key details:

• Fully open-source (MIT licensed)

• Runs entirely local, your API keys stay on your machine

• Supports OpenAI, Claude, and Gemini (bring your own key)

• Interactive TUI menu if you prefer guided workflows over typing commands

Why we're building this:

KushoAI started with API testing. Developers kept asking about UI testing. Both problems share the same root cause: writing exhaustive tests manually doesn't scale. We're betting that AI-native testing infrastructure is how modern teams actually achieve real coverage.

I'd genuinely love feedback on:

1. Is the current setup too complex? (Considering making it a single npm install command)

2. What frameworks beyond Playwright? (Python/pytest is the most requested)

3. What's the first flow you'd want to test on your app?

Star the repo if this is useful, and drop your toughest UI testing problem in the comments. I'm here all day.

GitHub: https://github.com/kusho-co/kusho-ui-testing-tui
Docs: https://resources.kusho.ai/kusho-ai-ui-testing-tui