PH热榜 | 2026-04-14

Figma opened the canvas to agents.

What is it: Figma's use_figma MCP tool lets AI agents create and edit designs directly in Figma, working with your actual components, variables, and auto layout not against them.

The problem: Every AI-generated design has the same tell: it doesn't look like your product. Components are invented. Spacing is arbitrary. The output is technically a UI, but it's nobody's design system. So designers throw it out and start over.

The solution: Skills are markdown files that encode your team's design conventions. Agents read them before touching the canvas. Combined with use_figma, agents now have both access and context they know how to work in Figma and they know how to work in your Figma.

What you can do with it:

• 🏗️ Generate component libraries from a codebase

• 🔗 Sync design tokens between code and Figma variables, with drift detection

• ♿ Auto-generate screen reader specs from UI designs

• 🔄 Run parallel workflows across multiple agents

Who it's for: Product and design-engineering teams that use Figma as the shared source of truth and want their AI agent workflows to stay connected to it. Heavy users of Claude Code, Codex, Cursor, and Copilot will feel this immediately.

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

This is exactly what multi-agent platforms need. We're building Kepion — an AI company builder with 31 specialized agents, including Maya (Designer) and Kai (Frontend Dev). Right now Maya outputs design tokens and Kai codes them into React components. But there's a gap: Maya can't "see" or "touch" actual design files.

Figma for Agents closes that gap. If Maya could create and edit directly in Figma using this MCP tool, then hand off real Figma components to Kai for implementation — the design-to-code pipeline becomes seamless. No more translating between "design spec as text" and "actual visual design."

Two questions: does use_figma support reading existing design systems (variables, component libraries) so an agent can stay on-brand? And is there a way to export generated designs directly to code (React/Tailwind)?

Following this closely. The future of AI-generated products isn't just code — it's code that looks good.

Good to note that Figma is also innovating forward to stay competitive in the AI landscape. Congrats on the launch! Looking forward to trying it.

How does it handle the conflict when the code variables in Figma and the code base diverge? Congrats on the launch.

the screen reader spec generation is the most underrated part. a11y annotations are always manual, always late, and quietly ignored in code review anyway. 

agents generating aria specs from actual design system components — if that's real, it's the first time accessibility sits upstream of the handoff, not downstream.

AI that doesn't treat Auto -Layout like a suggestion! Look forward to it!

Crazy. Would be happier if it works out great for multiple LLMs

This feels like a missing layer finally getting solved.

Most AI design tools create “usable UI”, but not your UI, which kills real adoption in teams. Connecting agents directly to the design system is the right direction.

If this actually maintains consistency at scale, it's not just a design tool upgrade; it changes how design and engineering collaborate.

👋 Hey Product Hunt,

I'm Mahdi, co-founder of CatDoes. Building this with Nafis. Big thanks to @thisiskp_ for hunting us today.

This is our third time here. v3 was "a team of AI agents that builds your mobile app." v4 is where that idea grew up.

The core of v4 is Compose, our new AI agent. It replaces the old specialist-team setup with a single autonomous agent that works more like an engineer you can hand things off to than a copilot that sits next to you while you code.

A few things that are actually different about Compose:

- It runs in the cloud. You describe what you want, close the tab, go to sleep, come back. Compose has been working the whole time.

- It has its own computer. So it can install packages, run scripts, try a build, read the output, and try again if something fails. It doesn't suggest code changes, it writes them, runs them, and fixes them when they break.

- It spawns subagents. If your request has three independent parts, Compose runs them in parallel instead of serially.

- It decides what to do next. Plans the work, writes the code, runs the tests, ships the build, reads the errors, fixes them.

What Compose builds

Mobile apps and websites. You can start from scratch or import a GitHub repo. Paste a URL (Lovable, WordPress, Shopify, whatever) and Compose will turn that website into a mobile app. Fork any of your projects to ship the same product under a different brand. It also monitors your app's errors via CatDoes Watch (Built-in error monitoring) and will fix them when you ask.

Every plan ships with a full backend: CatDoes Cloud. Database, auth, storage, real-time, edge functions. US and EU regions. No separate setup, no extra vendor.

What's new since v3

- Compose Agent: autonomous, cloud-native, has its own computer

- CatDoes Cloud: backend included on every plan

- Websites + custom domains

- Import from GitHub

- Monorepos with a preview size switcher (mobile, desktop, iPad, phone sizes)

- Website to Mobile App convertor

- File Browser (edit your code from the dashboard)

- Multi-page Canvas (see every screen at once)

- CatDoes Watch (built-in error monitoring, dev and prod)

- Env Manager (store secrets the AI can request but never read)

- Fork Projects

- Principal Agent tier, our smartest agent yet

Who it's for

- Founders going from idea to a real app in days

- Designers whose prototypes should actually run

- Small businesses that finally need a mobile and web presence

- Engineers who want to ship 10x faster

- Anyone who's been waiting for tech to catch up with their idea

Product Hunt perks

- 20% off your first paid month with code V4_PH_20, valid this week

- Drop your idea in the comments and I'll reply with how Compose would tackle it. If it's interesting, we'll spin it up and share what Compose built.

Thanks for taking a look. Feedback from people who showed up for v2 and v3 is a big reason this version exists. 🙌

Mahdi and Nafis

Co-founders, CatDoes

I created my habit tracker app with catdoes, and I love it! Congrats Guys.

Hey PH fam 👋

Super excited to hunt CatDoes v4 today — and this one is a meaningful step forward for AI-assisted development.

I know Mehdi well. He and his co-founder (Nafis) have been heads down building for a long time. Third PH launch. Best one yet.

Here’s what’s always frustrated me about AI coding tools: They write the code. Then they stop.

The moment something breaks — a failed build, a missing dependency, a cryptic error log — it’s back on you. You’re still the debugger. You’re still the one closing the loop.

CatDoes v4 changes that dynamic.

The new agent runs on its own cloud machine. It doesn’t just generate — it:

→ Installs dependencies on its own

→ Runs the build

→ Reads the error logs

→ Fixes what broke. In a loop.

You tell it what to build, close the tab, and come back when it’s done.

That’s not a code generator. That’s agent autonomy.

And it’s the closest thing I’ve seen to a tireless dev running in the background while you focus on what actually matters.

Big congrats to Mehdi and Nafis and and the CatDoes team 🙌 Three launches in and they keep raising the bar.

Check it out and drop your questions and feedback below 👇

Is it good enough to deploy to public for non software engineer? Or is it only for vibe coding?

Congrats — really excited about this release! I built a couple of portfolio apps with CatDoes v3, and even without diving deep yet, this already looks super impressive.

This product is really amazing, and I've been watching their progress. I really like the quality of the product.

I am seriously impressed.

I had a few design ideas for my business and they were laying down in my Figma untested. I used Catdoes v4 and tada 🪄. Kudos team!!!

does it support like all Apple platforms?

👋 Hey Product Hunt community,

2025 was the era of shiny demos. 2026 is the era of useful business software your business runs on.

If you've ever tried to build a custom business tool - a client portal, a vendor management system, an internal CRM - you know the options aren't great. Off-the-shelf software almost fits, but not quite. Hiring a developer takes months and costs more than expected. And AI or vibe coding tools? Great for prototypes, but the moment real users and real data get involved, things fall apart fast.

Today, we're bringing Softr’s AI Co-Builder to the Product Hunt community: the first AI platform for building real business software. Not prototypes. Not demos. Software that works, every single time.

Here's what makes it different:

🏗️ Generate real business software from a prompt: Describe what you need, and Softr’s new AI Co-Builder instantly creates the database, application, and business logic - already connected, secure, and ready for real users.

🔁 Refine with AI or edit visually: Switch between AI prompting and visual editing at any time. You build it, own it, and iterate on it - no black box, no developer required.
Need something custom? Use the vibe coding block to build your own design and logic exactly the way you want.

🔒 Secure and fully-functional by default: Logins, roles, permissions, and security are built in from day one, so you can actually launch it to your team or clients without it falling apart.

🧩 Connect your data and tools: Your data, your apps, and your workflows all live in Softr. Connect your existing tools, automate processes, and manage your complete business operations without switching context.

👉 Try it free at https://softr.io/

Over 1 million builders worldwide already use Softr. We can't wait to see what the next million builds!

Let us know what you think - we'll be in the comments all day to answer questions.

– Mariam and the Softr Team

I'm very happy to see app building became even faster and easier in Softr. Shout out to the team!

Really proud of what the team has shipped here! Excited for what people will build with it 💪🏻

So proud of this one!

Real software building for non technical people, where they can reason visually versus getting stuck in code editing

Really game changer. Now the only thing which limits you in Softr is your imagination.

This looks seriously powerful <3

Building with AI app builders, you bump into the same issues: you start from scratch. I've tried the AI app building capabilities of Softr and it actually delivers what it promises: within a ridiculous short amount of time, you have an app you can actually use for your team, business or a client. The crazy part is that this is only the beginning!

Big congrats to the @Softr team--you've built something people love.

I've been building apps with Softr for 4 years and I must say that this AI cobuilder is actually useful, and actually works.

It implemented very nice layouts that I never thought about and it's super nice to get this extra inspiration and support as I'm building.

Manually configuring 12 blocks in an app was a bit repetitive and long (couple hours, nothing dramatic when building full stack business apps) so I'm now skipping some parts thanks to the co-builder.

I recommend anyone to test this prompt "Add a new page to my app to show 5 analytics charts you think are relevant".

1 prompt away from adding a nice analytics page to any app.

I’ve tried quite a few tools that promise to generate complete apps, and most of them tend to produce very similar, static, and somewhat monotonous results. With Softr, though, the experience felt different from the start, even the building process itself is genuinely delightful.

The output was surprisingly good and stood out compared to other tools I’ve tested. I especially liked how easy it is to preview the project across different devices, the responsive view is very handy.

If I could suggest one improvement, it would be adding an option like “be creative” (or something similar) when choosing a theme. That could give the AI more freedom and bring a fresh, unexpected touch to the generated designs.

That said, congratulations on the launch! 🚀

Back in 2023, not long after chatGPT came out, Softr was probably the first company to have launched an AI app builder (you can see it in our past launches 😄)

It felt like the most cutting-edge thing at the time.

...and our new AI co-builder just made that old version look like a toy.

I know the first thing on your mind is probably "How is Softr different from {another vibe coding tool}"?

Here is a TL;DR:

1. User auth, database, and automation out-of-box with their own UI. There's no need to wrangle multiple systems together or pay for multiple subscriptions.

2. You can build by both prompting AND visual editing. From our learnings over the past 5 years, combining both modes is truly the best UX. Many vibe coding tools get non-technical 80% there, and stay there. We let you actually finish the job with confidence.

3. Business apps require reliability – permissions, security, and custom logic...things that most people don't realize how complex it is (because most SaaS took care of it). We provide all the tools for any non-technical users to configure them without having to verify whether LLM-generated code does what they want.

4. We have highly flexible blocks, fields, actions, integrations, and features that are constantly being improved and maintained by our team. And if you need something more unique, you have the option to vibe code your own and connect to any API with secure auth. So you get the best of both worlds!

I know these claims are probably not enough to convince you, so you just have to try it out :)

The real test will be how well it handles complex business logic not just simple portals.

The quality of this launch, and especially the video, reminds me of how AI has improved because your outputs are on a very professional level. Nice job.

Hey @Softr and team, just wanted to say it's a great option out of all the other presents in the market, I have built some custom projects for my project-specific requirements and they are working fine with daily task....

all the best!!

This is a complete package now. Everything handled very well. Even the Auth, which I used to struggle a lot with. I now make apps while in between my meals.

I'm also very fond of this one, and I think it's actually a pretty new view into the software development using AI agents. This doesn't do bare code development, but rather reuses components that are battle-hard and sure that they work. It's much better than just writing code that you cannot look into its details because you're lacking expertise or anything similar.

Love this team! Excited to try this out on a meaty business case.

I had used Softr when the widget-based building was the norm. Any particular capability that differentiates Softr from the Replits and Emergents of the world @arturmkrtchyan

Our new AI co-builder is absolutely amazing for building business apps at scale and with speed! I've literally seen customers now build secure, scalable, and powerful apps in hours instead of weeks and months like with other vibe-coding tools.

I loved seeing the whole Softr team bring this one to life! And amazing that we can finally share it with everyone. :)

Great fresh energy in the vibe coding space, giving users control over all aspects of their full-stack app!

This one is really a game changer! I really like how it reduces the barrier to entry — you can just use chat and ask for new things instead of building everything from scratch.

Excited for this one to be out there. Real business software, generated by AI, that you can actually put in front of users!

The "2026 is the era of useful business software" framing is right. No-code tools that fail when real users and real data get involved have been the consistent gap.

I built StoryRoute (https://storyroute.netlify.app/) as a side project — an interactive travel app where users explore cities through curated story-driven routes. It's a niche use case, but shipping something with real interactivity without a full dev team is what made it viable. The tooling is finally at a point where the idea-to-working-app gap is genuinely closable for solo builders. Softr's focus on apps that run real operations rather than just demo well is the right priority. Curious how the AI Co-Builder handles complex relational data (the point where most no-code platforms start breaking down).

The gap between "working prototype" and "software your business actually runs on" is where most no-code tools quietly give up. That's the right problem to go after. Netflix and Google as reference customers is a bold flex though — would love to see some small business examples too. 👀

Congrats on the launch! Excited to see the evolution of Softr over the years.

Really interesting direction. The idea of generating not just UI but also database and logic in one go feels like a big step forward. How are teams typically iterating on apps after the initial AI build?

congrats on the launch!!!

Excited to see Softr pushing in this direction. Congrats on the launch!

@mariam_hakobyan5 the product speaks for itself. Softr is the only AI platform in the space that can take you from painful workflow to complete end-to-end REAL app that solves that workflow, in a few minutes. AI + No code is the future. True autonomy in the workplace as a non-technical team is the best way to go faster, and the hundreds of thousands of users building portals and internal tooling have already proven this is a reality.

Hey Product Hunt 👋 Mikita here, founder of Ovren.

We built Ovren because most AI coding tools still optimize for assistance.
We think the bigger opportunity is backlog execution.

Every team has engineering work that never makes it into a sprint:
bug fixes, refactors, UI changes, integrations, tests, cleanup, and all the repetitive tasks that pile up.

Ovren helps teams move through that backlog faster.

Today, teams can assign scoped tasks to AI frontend and backend engineers that work inside a real codebase and return reviewable code updates, not just suggestions.

We’re focused on well-scoped backlog automation first, then expanding toward deeper repo understanding, stronger multi-task execution, more autonomous task pickup, and AI QA automation as one of the next major layers.

What backlog tasks would you already trust AI to fully execute today inside a real repo?

Would love your honest take 🙌

A lot of products in this space are still one general coding agent, and then you prompt it to “act like a frontend engineer”.

@Ovren is taking a more concrete route by turning the roles, responsibilities, and input/output boundaries into actual product structure: FE handles UI features, component refactors, and visual bugs; BE handles APIs, services, migrations, and tests; QA is coming next.

That makes the whole “AI engineering department” idea much easier to understand inside a real backlog workflow.

The real challenge will be ensuring AI understands repo specific architecture and conventions deeply.

Biggest value here is not writing new code but cleaning up the engineering debt that teams ignore.

The scoped task approach is smart it reduce risk compared to fully autonomous coding agents.

Interesting! Congrats on a launch. How does Ovren integrate with other tools and existing workflows like Claude? Is it a web platform? Does it has CLI/skills to plug in?

This is a really interesting direction.

The idea of “AI working through the backlog” sounds great, but in practice that’s usually where all the messy, ambiguous tasks live 😅

In our experience, the hard part isn’t writing the code, it’s understanding context, edge cases, and intent behind old tickets.

Curious. What kind of tasks are actually working well for you right now?

More clearly scoped things (bugs, small features), or are you seeing success with more ambiguous work too?

Guys, congrats on your launch day, and I love the positioning.

Backlog is one of those problems - painful, but somehow still unsolved. What about your target audience right now? Whether there are solo founders, small teams, or larger engineer teams?

Kirill here — I’m focused on the data and intelligence side of Ovren.

For me, backlog automation gets interesting when it moves beyond code generation and into real context understanding.

To be genuinely useful, the system has to make sensible decisions inside messy repos and return changes a team can actually trust.

We’re starting with well-scoped tasks first, then pushing toward deeper automation layers like QA.

Would love to hear where people think AI becomes truly useful first in the software delivery workflow.

Maxim here, CTO at Ovren

What I like most about this space is that the real challenge isn’t code generation — it’s making AI reliable inside real codebases, where architecture, conventions, and reviewability matter.

That’s why we’re focused on well-scoped backlog work first: practical trust, strong repo context, and clear outputs instead of black-box automation.

Really curious what engineering tasks people here would actually automate first.

There are so many different solutions of this kind on the market, but what sets this one apart, I would say, is the sensible and meaningful usage of AI and the nice UI that orchestrates it all together.

I wish the team all the luck and best success in this. This Product Hunt launch is just the first step in their journey, and I'm excited to see where this leads them.

Good luck!

Is the pricing model affordable for small startups ?

Hello Mikita, congrats on the launch, i like the demo, one question though, do you consider letting user assigns those tasks on the phone using app or messenger? I would personally have value from that

My team and I use Orwin, and it was a fantastic product. We had a UI glitch. We made it visible, but it wasn't linked to the backend code, so there was a backlog in that. We had to fix that, and Orwin was there to speed things up, so it was a great help.

Congratulations guys! Such a cool product

I built a similar system for personal use — Velo, an agentic engineering team built on Claude Code. It comprises a full squad of specialised agents: Product Manager, Tech Lead, domain engineers, and reviewers across security and observability. The workflow is approval-gated at every stage — PRD before design, design before build, review before commit. Nothing reaches the codebase without explicit sign-off.

Really like this direction. Focusing on actual backlog execution instead of just suggestions feels like a meaningful shift. What kinds of tasks are teams trusting it with first?

Congrats on the launch @mikita_aliaksandrovich

Hey Product Hunt Community! I’m Saurav, Co-Founder & CEO of FuseAI.

Before this, I worked at Deel on GTM Strategy & Ops, where I saw how much effort actually goes into making outbound work at scale. It’s not just hiring great reps, it’s stitching together tools, data, workflows, and constantly maintaining everything behind the scenes.

Big companies have an enormous advantage on distribution. They can spend $1,500–$2,000/month per person on tooling and have an entire revenue operations team building outbound systems.

Meanwhile, startups and SMBs are stuck duct-taping together low quality tools, starting behind the eight-ball, and often failing on distribution before they ever get a real shot.

We built FuseAI to change that.

Our goal is simple: give every company in the world access to world-class sales infrastructure without the maintenance, overhead, or costs to empower the best product builders to win.

With Fuse, you don’t need 5 tools to run outbound or an internal team dedicated towards managing integrations.

It takes 10 minutes to setup instead of 10 weeks.

Your dialer, email, and LinkedIn workflows are all connected.
You can waterfall enrich data across several vendors.
You know exactly who is showing intent.
You get access to the same quality of data as companies with GTM engineers and multiple data providers.
AI agents will help execute an end-to-end outbound motion, in one place.

All for $159/month.

No surprises. No gatekeeping. No $20,000 entry level contracts. Just the system modern companies deserve.

We been working non-stop over the past year since YC to make this happen, so would love your feedback !!! 🙏🙏🙏

Super excited for what the team has build here! Finally feels like the beginning of the end of fragmented outbound tooling 🙌

Congrats! Sounds like very helpful tool, but how do you make sure, each part of your software doesn't just scrap the surface but provides data and feature depth? f. e. If you waterfall enrichment on par with clay and LinkedIn outreach on same level, as let's say, heyreach?

Hi Saurav,

I'm wondering if I can use Fuse.ai to create custom signals, such as:

• companies that were founded six months ago

• companies that are product-led growth mode

• companies listed on AppSumo

Is it possible to create such custom signals with Fuse.ai as well?

the "tool sprawl" problem is so real. we've been dealing with this in healthcare sales where teams juggle CRMs, enrichment tools, sequencing platforms, and more. curious how the AI workflows handle edge cases - like when prospects respond with objections or reschedule requests? does it adapt the sequence automatically or hand off to humans?

Hey Saurav! Fuse sounds awesome!

Does it work for every business size? I mean, can an early stage startup take the most of this?

Btw, wish you all the best here!

Julius taught Claude to talk like a caveman. 24.9K stars later, it's the most useful meme in developer tooling.

LLMs are verbose by default. Phrases like "I'd be happy to help you with that" and "Let me summarize what I just did" contribute nothing — but burn tokens, slow responses, and push you into usage limits faster. Caveman makes Claude skip the throat-clearing and go straight to the answer. Same fix. 75% less word. Brain still big.

What stands out:
🪨 ~75% output token reduction: Benchmark average 65%, range 22–87% across real coding tasks
⚡ ~3x faster responses: Less token to generate = speed go brrr
🎚️ Four intensity levels: Lite, Full, Ultra, and 文言文 (Classical Chinese) mode
📝 Caveman-commit: Terse commit messages, ≤50 char subject, why over what
🔍 Caveman-review: One-line PR comments: L42: 🔴 bug: user null. Add guard.
🗜️ Caveman-compress: Rewrites your CLAUDE.md into caveman-speak, cutting ~46% of input tokens every session
🔌 Works everywhere: Claude Code, Codex, Gemini CLI, Cursor, Windsurf, Cline, Copilot, and 40+ more
🆓 Free, MIT, one-line install

Before and after:
🗣️ Normal Claude (69 tokens): "The reason your React component is re-rendering is likely because you're creating a new object reference on each render cycle..."
🪨 Caveman Claude (19 tokens): "New object ref each render. Inline object prop = new ref = re-render. Wrap in useMemo."

Note: works best for coding tasks. Nuanced responses still need full Claude, and the system prompt loads as input tokens, so net savings vary per use case. A March 2026 paper found brevity constraints improved accuracy by 26 percentage points on certain benchmarks. Verbose not always better.

Perfect for developers hitting usage limits and anyone who wants their AI agent to do the work and shut up about it.

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

I'm using right now. Also I've added to AGENTS.md/CLAUDE.md to always load caveman skill on the first message.

input token is higher! but with cache saves tokens.

Love it! I've been using it for a while together with RTK, and I'm saving a bunch of tokens.

Heard about this before, looks awesome. Where'd you get the 75% metric from?

What I like about this is that it gives critical security warnings in full sentences, and the rest of the time it just saves tokens by talking just the essentials.

It is pretty cool indeed. Saw this going viral on Twitter.

What started as a simple Hacker News post from Paul in November 2022, "A tool to help you remember shit you are interested in," has, three years later, transformed from a summarizing tool into a platform that brings your knowledge to the forefront.

Recall 2.0 isn't just a feature launch. There has been a series of updates building to this point. What we've envisioned is a place where people intentionally engage with their saved content, learn from it, and bring that knowledge to the center.

Everyone is looking to AI for answers. But with Recall, you can prioritize your own knowledge first and foremost. Choose to chat with your saved insights and invite the internet in as a supplement. That's the priority order we believe it should follow.

This update lets you get answers that no other AI can provide, as it is grounded in the sources you intentionally chose to save and the notes you specifically took. Ask questions, and Recall pulls from your curated knowledge base.

• "What did that sleep podcast say about melatonin timing? Find me the clip." [It pulls the exact moment where something was mentioned, and you can even play the clip inside Recall without breaking your flow.]

• "Condense my last six months of research and notes into references with timestamps, page numbers, and quotes. Also see if any new studies are out." [Combine your personal knowledge and invite the internet in to see what is new and what you might be missing.]

• "Pick a movie for tonight based on everything I've loved this year." [Just a fun one. Save all the movies you love to Recall, write your reviews, and then let Recall make a recommendation based on your own notes.]

You get to choose the frontier AI model you prefer, including GPT 5.4, Claude, Gemini, and more. Switch mid-conversation to compare outputs. You are not locked in. You get the best of all the AI models in one place.

Several other highly requested features are launching:

• API and MCP access: The community was clear about their needs here. Your knowledge can now be accessed from anywhere, whether you're building custom workflows or feeding it into your chatbot.

• Bulk actions for power users: With high content volume, you can now manage it all with just a few clicks. Generate summaries, tags, and connections for 100 pieces of content at a time.

• A UI overhaul: Everything you've seen in Recall from launch until now was homemade by our founders. We are finally getting closer to the intuitive, polished experience our users deserve. More to come.

Since December, we have shipped a series of major upgrades:

• Our rich text block editor lets you capture your notes and ideas with all the bells and whistles you need, including tables, to-do lists, and more.

• Graph View 2.0 lets you interact with your knowledge visually, discover patterns, and find new connections.

• Quiz 2.0 reinforces learning and retention. It is not just about saving and interacting It references the exact point where something was mentioned and links to that video inside Recall, which you can play inside Recall. It doesn't break your flow.with content. It is about learning from it. This is about intentional knowledge building replacing mindless doomscrolling, with a personalized spaced repetition schedule. One of the best parts is that you can challenge friends publicly. Call out the folks dropping hot takes without receipts.

We want to say a massive thank you to our Discord community. Your feedback, bug reports, and feature requests have shaped Recall 2.0. This is only the beginning, a step toward the vision we are relentlessly iterating on and delivering.

Super amazing product, so proud of the team to keep delivering the best product for learning and managing personal knowledge. Great job everyone!

Congratulations to our incredibly talented and passionate team. It's been a crazy few months working on this new launch, and I'm so proud to see how it's turned out. My personal favourite feature of this is the Agentic Chat. It makes it a lot easier for me rather than going back and tagging my content. I can just ask in the chat, which makes it a lot simpler.

An AI that remembers my research better than I do. I can finally stop pretending my Second Brain is a messy folder of 400 unread bookmarks and admit it's just Recall.

Wow, this is such a huge milestone for Recall. I can't believe how far we've come. The last few weeks have been a real grind getting this release out, and I'm so grateful for the team. We're usually fully remote, but for this launch we all flew out to Cape Town and rented a hacker house. It's been amazing spending time in person, getting to know everyone beyond the screen. Really excited for this new chapter; this feels like just the beginning as we have so much to more still to come.

Congrats on the launch!

I have been a user from the very start of Recall, and it has consistently improved over time. Lately, it seems the development team has gone into acceleration mode, and I consider it now the standard for doing research on the web. The ability to chat with my whole knowledge base and to be able to create flashcard reviews is awesome. And today it also got the academic stamp of approval from the infamous Andy Stapleton: https://youtu.be/wkwYcNu8yNY?si=FYQ5RCO89u1IFYnj They have a very active discord community and developers respond quickly when there is an issue.

The latest improvement, which lets me choose whether to chat only with my KB or include the web, is fabulous. If you are not into Recall yet, you owe it to yourself to explore it!

Experience: It's been being part of this awesome team producing this incredible app. I can't wait to see how it delights our users!

It’s been wild building Recall 2.0!

What excites me most about 2.0 isn’t a specific feature. It’s the shift in mindset. Your knowledge first, AI second. That’s what we believe in.

Really proud of how the team showed up for this, and grateful to the community for constantly pushing us to do better.

We’re just getting started.

Amazing!

Recall has always been the best summariser. I expect it to become the best... I don't know... AI-riser?

This feels like a big step forward for personal knowledge management. Being able to chat with your own sources instead of the open web is huge. How do you surface the most relevant sources in responses?

This is a very handy extension. I tested it with a video in Portuguese, and I was really curious to see how it would perform.

The results were fast and quite good, in just a few seconds, I had a clear summarized version of the content, which made it much easier to understand the video.

I did notice a few minor typos (for example, with words like “Claude”), but nothing that takes away from the overall experience.

Overall, it’s a solid tool with a lot of potential, especially for anyone looking to turn content into something more actionable and easy to revisit.

I've subscribed and paid the annual plan, but since couple of days application was logging me out, right now I am unable to login via Apple SSO

I built Ghost Pepper to be 100% private and run on local Huggingface models. I open-sourced it to get help from the community, little did I know Jesse Vincent, creator of Claude Superpowers would end up contributing more code than I (read: my Claude) did. I called it Ghost Pepper because all models run locally, no private data leaves your computer. And it's spicy to offer it open source.

I've always been a bit paranoid using cloud-based apps that collect super sensitive data. I expect more open-source, on-device apps like this will rise in popularity for that reason and the ability to modify to fit inside one's infra and workflows.

This is the category I've been waiting for someone to take seriously. Every meeting-notes tool I've tried sends audio or transcripts to a cloud I don't control, and for anything under NDA that's a hard no. "100% local" being the headline (not a buried feature) tells me you understand the actual buyer. Question for the maker: what's the model running under the hood for the TTS side, and does it hold up on older Macs or is this an M-series-and-up product? Upvoted. Rooting for the local-first AI wave.

I think we can integrate the Gemma models also into this. One other thing is that I really want this for Windows too because right now I don't think we have any system which can work natively for Windows. Can you do that lab? That would be really helpful

The local-first approach resonates deeply. I built NexClip AI with the same philosophy — video stays on your Mac, only audio is sent for AI analysis when needed.

The OCR context for disambiguation is clever. We solved a similar challenge with audio RMS data — using silence detection and sentence boundaries to create precise segment cuts instead of relying purely on transcript text.

Curious: with the 2B Qwen model running locally, how much memory overhead are you seeing during a typical 60-min meeting transcription?

this is super refreshing

everything going cloud-first, while privacy is becoming a bigger concern

fully local voice + transcription is a strong angle

how’s the performance compared to cloud models right now?

Ran into this building something with voice input. Had to drop cloud STT because of data policies at a couple companies I was demoing to. Local first completely changes that equation. Curious how your models handle technical vocab like camel case and library names? That's been one of the hardest parts for us.

From @rauchg :

Today we're open sourcing http://open-agents.dev, a reference platform for cloud coding agents. You've heard that companies like Stripe (Minions), Ramp (Inspect), Spotify (Honk), Block (Goose), and others are building their own "AI software factories". Why?

1. On a technical level, off-the-shelf coding agents don't perform well with huge monorepos, don't have your institutional knowledge, integrations, and custom workflows.

2. On a business level, the moat of software companies will shift from 'the code they wrote', to the 'means of production' of that code. The alpha is in your factory.

Open Agents deploys to our agentic infrastructure: Fluid for running the agent's brain, Workflow for its long-running durability, Sandbox for secure code execution, AI Gateway for multi-model tokens.

(Because of our focus on Open SDKs and runtimes, this codebase is a gem even if you're not hosting on Vercel.)

TL;DR: if you're building an internal or user-facing agentic coding platform, deploy this:

https://vercel.com/templates/template/open-agents

love this

going from prompt to actual code changes without staying in the loop is where things get interesting

nice to see this as an open reference as well

how are you handling sandbox isolation for different tasks?

Every new tool is changing their focus to become CLI and Agent friendly this year, and the latest one to join the trend is HeyGen.

HeyGen CLI and their Developer platform brings video generation to the terminal. Your agent can now generate video, translate content, and create avatars — straight from the command line.

Every command returns structured JSON and works out of the box in scripts, CI pipelines, and agent workflows. Runs in human mode too. One binary, no runtime prerequisites, installs in seconds.

Built on the new HeyGen V3 API. Full docs and quickstart on their newly launched developers.heygen.com.

Can you create a digital twin avatar without being on an enterprise account?

This is exciting. I am using HeyGen right now in one my apps and even though my pipeline allows me to generate videos fast, this is will definitely improve my workflow.

Where does the output go? Is it visible and editable in terminal, or only on the heygen platform?

What is up?

Voice agents are moving fast into production. But most teams don't have a way to enforce what their agents should and shouldn't say, especially when users actively try to push past system instructions.

ElevenAgents Guardrails 2.0 is a redesigned safety layer that validates user inputs and agent responses in real time, before anything reaches the end user.

Agents drift in long conversations. System prompts don't hold under pressure. One wrong response in healthcare or banking breaks trust fast. Guardrails 2.0 gives teams three independent enforcement layers: system prompt hardening, user input validation, and agent response validation. Custom Guardrails let you define policies in natural language and enforce them automatically across every call.

What makes it interesting:

• 🔒 Custom rules run as an independent parallel check, not a filter on the main model

• 🎯 Pre-built protections for focus, content safety, and prompt injection

• ⚙️ Execution modes tuned for voice latency tradeoffs

• 🚪 Configurable exit strategies when a guardrail fires

• 📋 Conversation history redaction for compliance-sensitive deployments

Built for enterprise teams deploying voice agents in regulated industries: healthcare, banking, and retail.

Note: currently in alpha. Redaction and Zero Retention Mode are enterprise-tier.

If you're moving voice agents from pilot to production, this is the infrastructure layer that makes it viable.

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

Sometimes the voice agents can drift, and guardrails can help prevent that from happening. Congrats!!

Sounds like a must have tool, like a shield that saves u from a PR crisis

This solves a real problem. We're building Kepion — a multi-agent platform with 31 AI agents that can create and run businesses. Voice interface is on our roadmap (Telegram voice messages, morning briefings, phone calls via Twilio), and ElevenLabs is our primary TTS choice.

The guardrails angle is especially relevant for us because our agents handle business-critical tasks — market research, financial analysis, legal checks. When these go through voice, the stakes are higher: a hallucinated number spoken aloud sounds more authoritative than one typed in chat.

We already have a supervisor layer (Warden for quality, Sentinel for fact-checking) that catches issues in text outputs. Guardrails 2.0 adds the missing piece: catching problems in the voice layer itself — prompt injection through voice input, off-brand responses, policy violations.

Question: does Guardrails 2.0 work with custom voice agents that use ElevenLabs TTS via API, or only with ElevenAgents specifically? We'd want to apply these rules to our own agent pipeline that outputs through ElevenLabs voices.

Hi, I'm Jaleh, CEO and Co-Founder of Mutiny. I founded Mutiny in 2018 to end the web of dependencies that has slowed down GTM teams for years.

Today, we are launching the new Mutiny. Mutiny is your AI agent for creating anything customer-facing, on-brand and personalized in minutes to help you take your accounts from cold to closed.

Problem → GTM teams are a web of dependencies.

A decade of leading GTM teams taught me that dependencies are the biggest blocker to growth. Ambitious, talented people sitting on their hands. Not because they weren't good enough. Because the system wasn't built for speed. This led to too many great ideas dying waiting for resources:

- Sales needs a one-pager to close a deal. Marketing is slammed.
- Demand-gen has a killer campaign idea. Web can't get to it for six weeks.
- Product marketing has the content. Design is backed up.
- Everyone is waiting on everyone and your best ideas die in someone else’s queue.

Solution → Mutiny’s agent gives sellers and marketers the ability to create anything they need to take their accounts from cold to closed.

The reality is that every deal hits friction. A champion who can't get internal buy-in. A competitor who shows up late in the process. A prospect who goes dark after a great call.

Mutiny gives you something to deploy at every one of those moments.

- A polished business case that sells for you when you're not in the room.
- A tailored customer story that creates urgency.
- A battlecard that takes a competitor off the table.
- A deal room your champion is proud to share.
- and more.

One person with Mutiny can do more in an afternoon than a team in two weeks.

What sets Mutiny apart is the quality:

1. On-brand, every time.
Mutiny extracts your colors, fonts and visual style from your website to set the guardrails, so every asset looks and sounds like you. Just describe what you need and get a polished, on-brand asset in minutes.

2. Personalized, automatically.
Mutiny researches your prospect and pulls in your CRM and call transcripts, so every asset is tailored to the account. Their challenges, relevant metrics, case studies, industry messaging and logo are all personalized instantly.

3. Full visibility.
Most of the buying process happens behind closed doors. Your champion is forwarding decks & pitching you in meetings. Mutiny shows who opened your assets & what they read, so you can sell even when you're not in the room & engage new stakeholders w/ context.

4. Proven GTM best practices.
Our agent has analyzed top-performing GTM assets and frameworks and built templates for every deal stage. The agent applies your brand and context so you get a polished asset in minutes.

Customers → The best GTM teams in the world, like Rippling, Snowflake and Uber, are already using Mutiny to hit their goals.
Users report:
- Speed: 4.5X faster to create
- Quality: 100% said it meets or exceeds their design bar
- Impact: 4 out of 5 reps said they're more likely to hit quota with Mutiny
- Edge: 89% said it gives them an edge in competitive deals

Mutiny is now free to try. Sign up today.

We are super excited to be on Product Hunt today and will be around all day to hear about your experiences, any ideas and feedback you might have. :pray:

Mutiny is awesome! Helps our team create stunning assets for a prospect, personalized and on brand. It’s honestly a life saver and helps us close a ton of pipeline.

Creating personalized landing pages and deal rooms is nice, but knowing whether the VP you sent it to actually looked at it and which sections they spent time on - that's where the real sales intelligence is. We're building a B2C product but we also talk to investors and partners, and being able to create a personalized page for each conversation and track engagement would be really useful. Does Mutiny work well for smaller teams or is it mainly built for enterprise sales orgs with a big pipeline?

This feels super practical. Creating deal-specific assets on demand instead of waiting on other teams makes a lot of sense. What kinds of assets are seeing the most usage so far?

I used this to spin up a personalized landing page for one of my target accounts. It took about 2 minutes and was super easy to edit from there. I can see this having a massive impact on our outbound effectiveness and the overall trajectory at the company as we're just getting Marketing stood up and are about 50/50 with IB/OB. I also plan to use it for proposals and exec decks, but am just getting started in the role and don't have any late-stage deals just yet

The WhatsApp-to-yourself pipeline for saving links is painfully relatable — I have definitely done the same thing, along with emailing links to myself and leaving seventeen tabs open for weeks pretending I will read them later.

The auto-categorization without manual tagging is the part that gets me. Every other bookmarking tool eventually dies because maintaining the folder structure becomes a second job. Removing that friction entirely is the right call.

I have my coffee; I can't have my breakfast.

Curious whether you are planning any kind of collaborative or sharing layer down the road, or if the intentional direction to keep this strictly personal and private?

Would love to also have a dedicated email address that I could forward links and articles to, because that's what I do in my current workflow, especially if I'm on mobile. I email myself links for me to read or check out later. Excited to give this a shot!

Having a dedicated search layer just for saved links is going to save me so much time digging through my chaotic browser history. Do you have plans to build out an API so we can programmatically push URLs from other apps? I could easily see myself hooking this up to a script that automatically archives interesting repositories my team drops in Slack.

I took a look at your website. I don't see any pricing plan there, but it doesn't say it's completely free, so it's a bit confusing. I really liked the idea and want to use it, but I don't want to know what links are getting attached to this. Can you explain a bit more on that?

I've tried Raindrop, Pocket, Notion databases, even sending links to myself on WhatsApp (sounds familiar?). None of them stuck because the save action always had too many steps.

send/links eliminates that entirely. Alt+L and it's done - no tab switching, no naming, no categorizing. The auto-organization is surprisingly accurate.

What didn't work: Telegram bot integration is broken for me, which is frustrating because saving from mobile is where most "I'll read this later" links come from.

Genuinely one of the most useful free tools I've found this year. Just fix mobile and this becomes a daily driver for a lot more people.Amazing! 🔥👌

Google's Nest Doorbell just got a lot easier to own: battery-powered, no wiring required, and smarter than anything else at the door.

Most video doorbells either need an electrician to install or send you a flood of useless motion alerts every time a leaf blows past. Nest Doorbell runs on a built-in rechargeable battery and uses on-device AI to tell the difference between what actually matters and what doesn't.

What stands out:

• 🤖 AI detection: People, parcels, animals, vehicles

• 🔋 Battery-powered, optional hardwire

• 📐 3:4 view: Head-to-toe + parcels

• 📹 HD 30 FPS, HDR, night vision (3m)

• 🎙️ Two-way audio + quick replies

• 📱 Works with Google, Alexa, Nest, TV

• 💾 3h free video history

• 🔒 Local backup (1h offline)

• 🛡️ Theft protection + 5yr updates

• ♻️ 45% recycled plastic

Currently on sale at $129.99 at Best Buy and Walmart — down from full retail price. Perfect for homeowners, renters, and anyone who wants AI-powered front door security without the installation headache.

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

Is there a way to disable it from sending my neighbours behaviour to a tech company, the US government and local law enforcement, so only I can access the data?

Cool and trendy.. i am just more curious about its "🛡️ Theft protection" feature..how does it work?

Cool, because it's difficult to get wiring done sometimes outside of home. A battery-operated doorbell with a camera and audio essentially can help prevent the wiring.

Watch a demo for Multi-User real time collab!

Link: https://x.com/mxnyapps/status/2043789858095710509?s=20

This feels super useful, especially for people preparing for system design interviews or thinking about resilience. How do you decide which failure scenarios to simulate?

The idea of having a dedicated space just to build and break code without wrecking my local environment is super appealing. I can definitely see myself using this as a sandbox to test out messy API integrations before wiring them up in my actual app. I would love to hear how you are handling state management under the hood.

The "make failure the teacher" approach is exactly how deep technical intuition gets built — and it's radically underused in professional education.

I've been thinking about this a lot designing my Excel for Financial Modelling course on Udemy (https://www.udemy.com/course/exc...). The courses that stick are the ones where you build a model that breaks — circular references that crash the file, linked workbooks that lose their source, sensitivity tables that return errors. Working through the failure is what turns someone from a user into someone who actually understands the tool.

Cascode is applying this to infrastructure in a really direct way. The live canvas + deliberate node killing is a clever mechanism. Most cloud certifications test whether you can recognize a diagram, not whether you understand why a system fails. This closes that gap. Congrats on the launch.

Hey 👋

Shuffle’s AI Website Redesign is getting a ton of traction. People are really pushing it to the edge.

So we shipped something new to remove one of the biggest blockers: redesigning local & auth-protected sites

New extensions:

• Chrome Extension

• Firefox Extension

Now you can redesign any website.

Just write a prompt and let leading AI models reimagine your site. Compare results side by side, pick your favorite, and export directly to Next.js, Laravel, WordPress, or grab a design.md for coding agents.

What you can do with it:
✓ Redesign outdated websites
✓ Pitch ideas to clients faster
✓ Generate fresh design inspiration

Curious to see what you’ll create 🚀

Hey Product Hunt 👋

Strix started as an open-source framework for autonomous pentesting.

Since launch, it’s grown to 80,000+ users, 15B+ LLM tokens processed daily, 1,300+ pentests per day, and 78,000+ vulnerabilities reported.

The demand became clear: teams wanted more than the framework. They wanted Strix running continuously across their repos, apps, and attack surface, with scheduling, validation history, auto-fix, integrations, and enterprise controls.

Why now? 🚀

• AI increased software shipping velocity

• security workflows mostly stayed the same

• periodic pentests and manual triage do not work when systems change every day
  

So today we’re launching the new Strix Platform:

• continuously pentest full-stack apps

• block vulnerable PRs before merge

• verify findings with proof-of-exploit

• generate merge-ready fixes

• retest automatically

• track security posture over time
  

Excited to hear what you think and answer any questions :)

Does this also work for mobile web apps? Or is this only mobile apps?

Hey Product Hunt! 👋

We built Amadeus because I kept running into the same wall:

- I had a piece of sheet music for a song I loved.
- I had the keyboard.
- I had...terrible sight reading skills.

We wanted something that would just listen - follow along with what I was playing, highlight where I was, and show me in real-time when I hit a wrong note. Existing apps either locked me into their library or didn't provide real-time feedback.

Amadeus supports any sheet music you already have: take a photo, upload a PDF, image, MusicXML, MIDI, or ABC file, and it becomes interactive.

Try it free at playamadeus.com or on iOS (3 free uploads). I'd love feedback: what sheet music format do you use most? Do you practice in notation mode or falling notes mode?

Congrats on your launch!
I like the design, and variety of options. Can it be connected via bluetooth ? Also does it has edit mode for the sheets ? I like to rewrite them from scratch in some apps so I can "understand" the song a bit :D

Hello Product Hunt! 🚀

I’m Alex, the founder of Anamap.

I’ve spent my career managing data platforms, and I kept seeing the same exhausting cycle: A metric dips, and a team of talented people spends the next three days in "investigation mode" instead of "growth mode."

We didn't need another dashboard to stare at. We needed a way to automate the reasoning that happens after the dashboard shows a red line.

That’s why we built Cartos.

It’s an AI agent designed to act as your proactive analytics co-worker. It doesn't just show you a chart; it investigates the "why" by cross-referencing your entire stack; from your data tools to your latest code releases.

🎁 Exclusive Launch Offer: We’ve made it simple for the community, no codes needed. Just head to https://anamaps.com/ph to claim your discount:

• The First 20 Hunters will see 50% OFF automatically applied.

• Everyone else gets 25% OFF for the rest of the launch.

⏳ Note: This offer is only valid until 11:59 PM PT on April 16th.

I’d love to hear from you:

1. What is the most frustrating "data mystery" you've had to solve recently?

2. How much time does your team spend manually auditing event triggers?

We’re excited to be here and can’t wait for your feedback!

I’ve been waiting for something like this. Every time my metrics dip, I end up digging through multiple tools and still feel unsure.

The "co-worker" framing is interesting because a good co-worker tells you when they're not sure. Does Cartos flag its own confidence level?

If this is a coworker, does that mean i can officialy fire my google analytics?:D

Love the concept of community-built “studios” for different creative needs. How do you see this evolving as more people start building their own workflows?

Congrats on your launch! As a marketer at a pre-launch startup, I'm constantly needing visuals but have little budget for a designer right now. The idea of getting studio-quality output without knowing how to prompt is honestly what I've been looking for. Can you keep a consistent brand style across multiple generators, or does each output start fresh every time?