PH热榜 | 2026-03-27

This solves a real friction point. Right now when I use Claude Code or Codex, I spend a lot of time writing context about which element I mean - "the button in the top-right of the filter panel" etc. Having structured annotations that feed directly into the agent as context is much cleaner. How does it handle dynamic elements that change state? Like a button that’s disabled until a form is valid?

Agentation bridges the gap between design feedback and code changes. Annotate any element on your UI — click, type, done — and get structured output that AI coding agents can immediately understand and act on.

Paste your annotations into Claude Code, Codex, or any AI tool and watch feedback become working code.

Key features:

• Multiple annotation modes: select text, click elements, multi-select, draw areas, or freeze animations to capture specific states

• Smart element identification: automatically generates grep-friendly selectors so agents find the exact element in your codebase

• React component detection: surfaces the full component hierarchy for any element, right in the annotation popup

• Computed styles: view live CSS properties alongside your notes for precise design specs

• Layout mode: drag 65+ component types onto the page and rearrange sections; changes sync to agents in real time via MCP

• Structured markdown output: copy clean, agent-ready annotations with one keystroke (C)

• MCP integration: two-way agent sync lets AI acknowledge, question, or resolve your feedback directly.
  
  Check out the latest video demo by the maker here, who also happens to have joined X recently as the Design Lead.

One of the most underrated pain points in building with AI agents is having zero visibility into what they're actually doing, you're basically flying blind until something breaks. Curious how you're handling agent workflows that branch or run in parallel. Does the visualization scale well for more complex pipelines?

Thanks for launching, Agentation team! When we annotate a component and feed it to Claude Code, can we keep a history of annotations so the agent knows when the DOM changed? I'd love to see how you handle selector drift.

Congrats on the launch!

how does it handle errors or failed tasks?

Congrats on the launch! Visual feedback for AI agents is a big gap right now. Most agent tools just give you text logs and hope for the best. What does the feedback loop actually look like in practice?

Great tool! Curious if this couldn't just be a Chrome extension. I'm not a fan of npm so I might be biased, but seems it could work just as good with a lower friction to install it.

@adithya This is completely badass. In the past, I was adding small text and design tweaks to Bugherd, Jira, or Linear. Now I can breeze through an entire site, creating small tasks at-will, and instantly hand them over via the MCP to get resolved in bulk. Really really well done. :)

The click-to-annotate approach is what I've been wanting tbh. I spend way too much time trying to explain to coding agents which exact element needs changing, and half the time they still get it wrong.

Curious how it handles dynamically rendered stuff though? Like components behind a toggle or things that only show on hover?

this is exactly what we've been missing. we use Claude Code daily and the biggest friction is always translating "fix that button" into precise technical context. being able to click and annotate the actual UI elements then paste structured output sounds like a game changer. how detailed does the context get - does it capture CSS selectors, component hierarchies, that level of detail?

Can this tool integrate with existing LLM stacks like OpenAI or custom models, or is it built for a specific ecosystem?

How does Agentation handle feedback for multi-agent workflows? Does it support collaboration between different AI agents?

Every team already annotates UI. What you've done is make those annotations executable, and that shift is bigger than it first appears.

Most tools stop at capturing feedback. Agentation goes a step further by turning that feedback into structured intent agents can act on without losing something in translation. That's not a design tool improvement; that's a new interface between human thinking and code execution.

The framing worth leaning into: fewer Figma comments, more control surface for AI driven development. One describes a workflow; the other describes a category. And the category framing opens a much larger conversation about where product thinking ends and implementation begins.

The real leverage here isn't faster fixes. It's collapsing the gap between deciding what to change and actually changing it.

Curious whether Agentation stays a layer on top of existing tools or becomes the default way teams communicate changes to code entirely.

The audience feels right away. I would test one version that pushes the outcome harder than:

"The visual feedback tool for AI agents"

Maybe:

"See where your AI agent breaks, fix it faster, and stop debugging blind."

中文也可以是：

"看清 AI agent 是在哪里出错，更快修掉，而不是继续盲调。"

Curious—are you replying to every signup instantly or manually?

Claude Code auto-fix brings hands-off PR management to the cloud.

It watches your pull requests, automatically fixing CI failures and addressing review comments, so your PR stays green without constant back-and-forth. No more babysitting builds or chasing feedback.

What makes it stand out is the autonomous, event-driven workflow. It subscribes to GitHub activity, pushes fixes when clear, and asks when things are ambiguous, keeping you in control without slowing you down.

Key features:

• Auto-fix CI failures

• Respond to review comments

• Works across web + mobile

• GitHub integration with real-time PR monitoring

Perfect for developers who want to save time, reduce context switching, and ship faster.

Read more here: https://code.claude.com/docs/en/claude-code-on-the-web#auto-fix-pull-requests

The "asks when needed" part is doing a lot of work here. How does it decide when it’s stuck vs when it should just try another approach? In my experience running coding agents, the failure mode isn’t usually the agent giving up - it’s the agent confidently applying a fix that passes CI but breaks something else downstream. Does it have any awareness of test coverage gaps or is it purely CI green = done?

Wait so it actually watches for CI failures and fixes them on its own? That's the part that gets me. I've lost count of how many times I've pushed a PR, walked away, then come back to find 3 lint errors blocking the merge.

Does it handle review comments too or just CI stuff? Because if it can address reviewer feedback automatically that would save our team so much ping pong.

Look, I hate being the grumpy guy here but sometimes the current AI models just don't do exactly what you expect them to do (even if you very verbosely and clearly instruct them to). So maybe its still not the time to put even more trust on end-to-end code development to an AI system. I mean its a good goal to work towards but the previous stages need to be 100% working, all the time. This comment comes with good intention from someone actively using Claude Code for vibe coding my app.

Oh that's the nice one. I'm building native apps for multiple platforms (Ritemark for Mac Arm, x64, Windows) and build times are loooong. CI failures sometimes happen like 1h into build. Will auto-fix help me in those scenarios?

How does Claude Code auto-fix ensure code quality while automatically fixing PRs? Is there a review layer?

Hi everyone!

The most important thing here is simple: this is now the voice model behind Gemini Live and Google Search Live. It is the speech engine @Google is actually putting into its consumer products.

Google is pitching 3.1 Flash Live as its highest-quality audio and voice model yet, with lower latency, better reasoning, and more natural dialogue. The benchmark jump is also pretty meaningful on ComplexFuncBench Audio.

Google clearly sees live voice as a core interface now, and this is the model carrying that shift.

3.1 Flash Live is available across these Google products:

• For developers in preview via the Gemini Live API in @Google AI Studio

• For enterprises in Gemini Enterprise for Customer Experience

• For everyone via Search Live and Gemini Live

The low latency part is what matters most here imo. I tried building a voice agent last month and the delay between user speech and response made it feel super unnatural. Even 500ms of lag kills the whole experience.

Really curious to see how this compares to the realtime API from OpenAI in terms of actual latency numbers.

We didn't expect this much traffic for just a fun idea we worked on!

Our backlog is filled and we are sending emails through out the next 48 hours with all the Org Charts! Thanks for your patience folks!

Congrats on the launch! 🚀

I tried it out, entered my company name and got the results sent to my email. Really interesting to see the final output.

Planning to share it with my colleagues as well 🙂

This is AWESOME ... however it'd be nice if you could define which part of the organization you're interested in i.e. IT or HR or Finance

Nithish, I really appreciate you making this free and the real-time data aspect you mentioned to Shivani is impressive. That’s a significant challenge to pull off for external org structures.

From an engineering perspective, I'm already thinking about what an API could unlock. While viewing it through the UI is helpful, an API would allow us to integrate this org data directly into our internal tools for competitive intelligence or partnership mapping, moving beyond manual lookups.

Congrats on the launch! Being able to see any company's org chart for free is super useful for sales and recruiting. Curious how you're sourcing the data and how often it updates?

How frequently is the organization data updated, and what sources do you rely on for accuracy?

Super fun idea, looking forward to getting the org chart :) Happy to provide info on the companies I know well that I previewed, if that's helpful to the team

Most org chart tools treat structure as premium intelligence, something gated behind a demo request and a sales call. This treats it like it should have been public all along. That's the real unlock: not the data itself, but the removal of permission.

What makes the positioning interesting is the distance between the free org viewer and where this could actually land. A fast, frictionless starting point for any outbound motion isn't a data tool; it's a workflow layer. Something that shapes how teams think about account strategy before the first touch, not just after they've already decided to reach out.

That framing carries a lot more weight with the right buyer than free org charts.

Curious whether the vision stays focused on speed and breadth or evolves into something teams lean on for higher stakes account decisions.

What are some use cases for this tool?

Haven't found any company org chart (for each company I checked there was no data), after 3 tries I ran into a block screen.

Love the simplicity! What happens when a company is pretty flat with no clear hierarchy? Does it still show something useful?

Congrats on a launch! Offering this data for free is a nice move.

Do you also segregate the org chart across different subsidiaries or sub-orgs within large enterprises? Like Nescafe, Nestea lies within the beverage division of Nestle?

Hey Product Hunt!

I'm Ravi, founder of Cockpit AI.

I've spent the last decade studying the line between engagement and spam across social networks.

The insight is simple... more engagement, less spam. Less engagement, more spam.

Here's the problem...

Everyone added "personalization." First name, company name, hiring signals, funding rounds. When everyone uses the same signals in the same fixed workflows, it's spam again.

The signal is dead the moment it's commoditized.

But when you share something genuinely useful about a prospect's market -- what their competitors are doing, how their landscape is shifting, what's working for their peers -- they engage naturally.

It's not a pitch. It's information they actually want.

Thinking models made this possible... Each agent spends 200K tokens researching a batch of prospects at a time -- a real research window.

It starts from a lookalike audience -- what are their peers doing, who are their competitors, what's shifting in their market. From that base, the agent autonomously decides which signal actually matters for each specific person.

The agent picks the angle.. not a fixed workflow, not a rule someone wrote.

With Cockpit.. You're the manager... Your AI agents become your highest quality team.

Give your agents a few example companies, and they go to work:

• Research prospects and their competitors across the web 

• Decide which signal matters most for each prospect     

• Write outreach built around that narrative — not a template

• Generate a unique proposal doc for each prospect           

• Track engagement (scroll depth, time on page), adapt follow-ups

• Book meetings on your calendar                                           

 
Imagine if one user manages 10 agents... what a team of 10 can do — across every channel, around the clock?

Since launching mid-December 2025:                                                                             

• 102,000+ contacts researched and engaged                                                                                               

• 41,000+ personalized docs generated for those contacts

• 37,000+ autonomous agent conversations across email and LinkedIn (more channels coming soon)

• 1.7B tokens consumed — agents doing sustained autonomous work

• 73% average scroll depth on personalized docs  
               

Built for deliverability... Automated email warmup, anti-spam protection, compliance guardrails, and infrastructure deployed on your company's domain — not a shared sending pool.                                                                  

We use Cockpit to grow Cockpit. Our agents book our meetings. That's the credibility test — if it doesn't work for us, we have no business selling it to you.

Would love your feedback!

This usually looks good early, but changes once it runs at scale. Outreach quality tends to drop as volume increases, especially across channels. How this actually performs there, in terms of actual responses not just output.

The multi-channel follow-up piece is where I’d want to see more detail. When the agent is sequencing across inbox, LinkedIn, and calendar autonomously, what does the interrupt/pause model look like? Like if a prospect responds in one channel mid-sequence, how quickly does it halt the parallel threads? I’ve seen a lot of outreach automation burn leads because two touchpoints crossed each other within hours.

This resonates — we run agents for outreach too. When the model picks its own angle, can we peek at the research outline before it crafts the email so we can align the storytelling with our positioning?

I am curios to know what kind of results were your existing users able to achieve with Cockpit AI? The idea sounds good but want to know how it translates in to real world?

What kind of success have users seen from this?

Most outreach tools are still just filling in templates faster. Cockpit is doing something meaningfully different by shifting the question from what do I send? What angle actually fits this person? That's delegated thinking, not automated sequencing.

The distinction matters for positioning too. You're not competing with outbound tools at this point. You're creeping into AI sales team territory, and that's a much larger conversation to be in. When the agent owns research, narrative, and timing, the unit of value stops being messages sent and starts being conversations created.

The defensibility question is where it gets interesting. If Cockpit becomes the system of record for prospect understanding, not just execution, that's a different product category entirely. And a stickier one.

Is that the direction you're already building toward?

The 'we use Cockpit to grow Cockpit' line is the right credibility signal — AI sales tools that can't generate their own pipeline are a red flag. The doc-signal-based follow-up is an interesting differentiation from sequence-based outreach tools. Most outreach automation is purely time-based (day 1, day 3, day 7) regardless of buyer behavior. Adjusting follow-up based on what a prospect actually engaged with is closer to how good salespeople actually work. The risk I'd watch: personalization at scale tends to regress to generic. When the agent is writing hundreds of emails/day, how do you prevent it from pattern-matching to the same 3-4 opener structures? The first 50 emails might look custom; email 500 will reveal the seams. Would be curious to see response rate data across volume bands.

Codex Plugins by @OpenAI turn workflows into installable, reusable bundles, bringing tools like Slack, Figma, Notion, and Gmail directly into your dev flow.

Instead of scattered setups (skills, configs, integrations), plugins package everything... skills, app connections, and MCP servers... into one seamless workflow. That means Codex can handle not just coding, but planning, research, and execution across tools.

Perfect for developers and teams who want consistent, shareable workflows across projects. Build your own plugins or use curated ones to get started fast.

Key points:

• Codex now integrates with tools like Slack, Figma, Notion, Gmail, and Google Drive.

• Plugins let Codex handle planning, research, coordination, and post-coding workflows.

• Example: The Google Drive plugin enables working across Docs, Sheets, and Slides in one loop.

• Plugins bundle apps + built-in skills + authentication, so Codex can instantly use these tools effectively.

• Available in Codex app, CLI, and IDE extensions.

• Developers can build and share custom plugins.

More plugins and capabilities are coming soon.

Perfect for developers and teams who want consistent, shareable workflows across projects. Build your own plugins or use curated ones to get started fast. Codex is evolving from a coding assistant into a full workflow automation tool for developers and teams.

This is insane; it will manage 3 different tasks from 3 different tools. In seconds. No human is as versatile as this.

Congrats on the launch! 🚀

As a QA, this looks useful, planning to try it in my workflow soon

Has anyone tested codex?

Is it better than claude code?

The reusable bundle concept is the right abstraction for teams. One thing I’m curious about: how do you handle plugin versioning when a shared plugin gets updated? If three agents in a workflow are all depending on the Slack plugin and someone pushes a breaking change, does it propagate automatically or do agents pin to versions? That’s usually where shared tooling falls apart in practice.

Hi everyone!

Suno v5.5 is really about making the experience feel more like yours.

The new pieces all point in the same direction: Voices lets you create with your own voice, Custom Models let you tune the model on music you made, and My Taste starts shaping results around what you actually gravitate toward. Put together, it feels like Suno is moving toward something more like a personal creative instrument.

Suno is not just getting more expressive, but like they said, more you.

Most AI music tools still feel like prompt machines with a better UI. What Suno is moving toward feels different, and the gap is worth naming clearly.

Voices, custom models, taste learning, these aren't generation features. They're continuity features. The output starts carrying identity rather than just variation, and that shifts the whole relationship between the creator and the tool. Less something you operate, more something that evolves with you.

That reframe is also where the real defensibility lives. Model quality is a race. How much of a creator's identity gets embedded over time is a moat. Those are very different things to compete against, and the second one is a much stronger story to tell.

How are you thinking about portability. Does that creative identity stay inside Suno, or travel with the creator across contexts?

Hey Product Hunt community! 🎹✨

I've been testing Suno since its early days (v3), obsessed with the Hammond B3 organ paired with its rotary Leslie speakers – this iconic 60s/70s instrument delivers a one-of-a-kind sensation: sound that physically rotates around you, inducing vertigo-like dizziness, head spinning left-right, front-back. Think Booker T., Deep Purple, gospel jams: that hypnotic swirl isn't just an effect – it's pure physics!

What I've Desperately Sought (After MONTHS of prompts):

1. Leslie Speaker Rotation: Dual rotors (treble horn + bass woofer) spinning at 60 rpm (slow choral) or 400 rpm (fast tremolo). Real Doppler shift (pitch rises/falls as it passes) + amplitude modulation from circular baffles = immersive 3D orbit around your ears. Not static synth vibrato!

2. Authentic Hammond Timbre: Additive drawbars (e.g., 888000000 harmonics), warm tube saturation, growling bass pedals, creamy mids, sparkling highs.

3. Intoxicating Psych/Cosmic Vibe: Enveloping trance, sacred vertigo, like a haunted cathedral or space trip. Pure instrumental, no drums/guitar/synths.

Tested hundreds of hyper-detailed prompts like:
"Psychedelic cosmic Hammond B3 organ with Leslie rotary speaker, swirling Doppler motion left-right front-back, slow-to-fast rotor acceleration, hypnotic vertigo..."
Result? Spot-on Hammond timbre (kudos v5.5!), but flat, static, lifeless. Zero rotation. Like a frozen VST. Suno ignores "rpm", "Doppler", "rotary swirl" – LLMs nail what (tone) but miss how (physical motion).

v5.5: Real Progress, But...

• +: Custom voices, better uploads, slightly improved spatial.

• +: More expressive generation, less robotic.

• -: Still can't simulate dynamic rotation. Uploaded real YouTube clips (Leslie 3300/770 demos) via Custom/Extend/Persona – max 70% fidelity, swirl feels "fake".

• -: AI models (LAMs) ace static textures, fail physical rotary physics. Structural limit.

My Workarounds (For Fellow Enthusiasts):

• Upload Real Clips (extracted Hammond demos) + 80% Audio Influence.

• Post-Prod: Reaper/Audacity + free Leslie plugin (Guitarix).

• Await v6? Or try Udio/MusicFX (better spatial sometimes).

Suno team: Add a physics-based Leslie engine (like Arturia/NI emulators)! Game-changer for jazz/psych/gospel.

Great for pop/rap/voices otherwise. Keep evolving – you're this close to the holy grail! 🚀

#SunoV55 #HammondB3 #LeslieSpeaker #AIMusic

Ethan, that's a sharp question about recording quality for the "Voices" feature. From what I’ve seen with similar tech, the output generally scales with input quality. So, a high-fidelity input trumps a quick phone memo almost every time. It’s definitely something I'd test thoroughly to understand the practical limits before relying on it for anything serious.

Swati, regarding "My Taste" adaptation for non-musicians, I'm a bit skeptical. While the intent is there, AI music generation still has consistency issues across diverse genres. I've found that getting a specific mood or sound for branding often comes down to extremely refined prompting, regardless of how much "learning" the AI does. The AI might adapt to general preferences, but nuanced branding usually requires a human touch or highly specific instructions to avoid generic outputs.

Stripe Projects lets you provision and manage your entire app stack from the CLI — or hand it off to an agent. Hosting, databases, auth, AI, analytics, and more, set up in a few commands.

No more jumping between dashboards, signing up for services one by one, or manually securing API keys.

What you can do:

• Provision services across hosting, databases, auth, AI, analytics, and more

• Sync credentials directly back to your environment after provisioning

• Manage billing in one place: upgrade, downgrade, and monitor usage across your SaaS stack

• Keep environment variables portable across machines, teammates, and agents

• Audit and repeat infrastructure changes reliably

Having everything provisioned from the terminal is huge for us. We've been stitching together like 4 different dashboards just to spin up a new service with auth + db + monitoring.

The credential management part is what sold me honestly. That's always the sketchy part of any new project setup lol

This solves a real pain point. Every new project starts with the same 30 minutes of provisioning — spinning up a database, configuring auth, wiring up environment variables across local and staging. Having that collapse into a few CLI commands with credentials synced automatically is huge. The portability angle is underrated too — being able to share a consistent stack config across teammates and CI/CD means fewer "works on my machine" issues. Does it support custom service providers, or is it Stripe-ecosystem only for now?

Stripe Projects is going after something harder, the coordination problem underneath it. That's a less obvious bet and probably the right one.

What stands out is the shift from provisioning to intent execution. When infrastructure becomes something agents can reliably act on without breaking context, credentials, or billing, you stop being a dev tool and start being a control plane. Humans and agents operating on the same surface without fragmentation are a genuinely different category.

The positioning angle worth leaning into: convenience is easy to copy; owning the agent ready workflow is not. If that becomes the core story, the defensibility question answers itself.

Do you see the moat coming from depth of integrations or from establishing the workflow standard before anyone else does.

the "manage credentials across the stack" part is what catches my eye. Running agents that need to touch Stripe + DB + auth + observability means you end up with a mess of env vars and service accounts. Centralizing provisioning through the CLI makes sense for human devs but curious whether the credential scoping works for agents too - can you issue a project config scoped to just what one specific agent needs?

This is the beesknees!

I really need a claude integration - for Claude, not Claude Code.

Voxtral TTS by Mistral is a powerful text-to-speech model built for realistic, multilingual, and emotionally expressive voice generation.

It solves a big problem in voice AI — robotic, low-quality speech — by delivering natural-sounding voices with context awareness, emotion control, and speaker personality modeling.

What stands out is its low latency (~70ms), lightweight design (4B params), and strong multilingual + voice adaptation (even with just a few seconds of reference audio), making it both scalable and enterprise-ready.

Key features include:

• 9 language support with dialects

• Emotion + tone control

• Voice cloning & customization

• Real-time streaming performance

• Easy API + integration into voice workflows

Great for voice agents, customer support, real-time translation, sales, and enterprise automation where natural speech truly matters.

Get started:

• Mistral Studio

• Le Chat

• Hugging Face

• Model's Documentation

If you’re building in voice AI, this is definitely worth trying.

Most voice models are competing on realism, which is a worthwhile race but not the most interesting one. Voxtral feels like it's playing a different game entirely.

Emotion, tone, and identity that are controllable in real time aren't just better TTS. It's programmable voice, something developers can shape as easily as text. That shifts the category from media tool to infrastructure layer, and infrastructure is where things get defensible fast.

The framing worth leaning into: not high quality speech, but the default voice layer inside AI workflows. One is a feature comparison; the other is a category claim. And the category claim is the one that attracts the builders who will compound your distribution for you.

The wedge probably isn't media or accessibility either. It's every AI agent that currently has no voice worth giving one.

The real question is whether this stays a model or becomes the layer every AI voice routes through by default.

low latency TTS for voice agents is genuinely hard to get right. the failure mode I’ve seen is when the TTS step adds enough delay that it breaks the conversational feel - any ballpark on p95 latency for a 100-word response? also curious how voice cloning handles accented speech in non-English languages, that’s usually where it falls apart

Congrats on the launch! The multilingual support is impressive — 9 languages out of the gate is no small feat.

Curious if Voxtral could eventually power audiobook-style narration for AI-generated stories. Building zz-novel on the reading side, and TTS feels like a natural next layer for the experience.

Hey PH fam 👋

The few last launches we did here were all about No Cap - the world's first AI investor. We launched her here, she went viral, Marc Andreessen shared it, Forbes wrote about it. Good times.

Today we're back with something bigger.

No Cap just got acquired by Audos - from the team behind Prehype, BarkBox (NYSE: BARK), and Ro ($7B). And we're relaunching this account as the Audos Publishing House.

https://www.linkedin.com/posts/ednevsky_huge-news-no-cap-just-got-acquired-by-audoscom-activity-7442922068775559168-lQB-

Here's why.

After evaluating 9,000+ startups with No Cap, we arrived at an uncomfortable conclusion: the VC model is broken for 99% of founders. Most people don't need pitch decks and board seats. They need tools to build, support to grow, and capital when they're ready.

That's what the Publishing House does:

🛠️ End-to-end AI business building - ideation, brand, product, ads, backend
💰 Up to $100K in funding - zero equity, revenue share that aligns incentives
🧑‍🏫 Real mentorship from founders who've built $9B+ in companies
📊 No Cap's data and community of 9,000+ founders baked in

The proof? Our first funded cohort (https://audos.com/publishing) is already live:

- SwingCaddy.ai - AI golf coach, $100K+ revenue in under 2 months
- Realer Estate - two Brooklyn high schoolers helping 100K New Yorkers/month find housing (featured in NYT + CBS)
- Solace - AI grief coach featured in The Economist
- Cartee - AI care concierge for kids with complex medical needs
- Beacon - addiction support coach built by a Cornell psychiatrist
- BoatIllustrator - custom boat merch, $1K sales in week one

Each got up to $100K. Zero equity taken.

These aren't Silicon Valley insiders. They're a recently divorced dad, a grieving daughter, a London father navigating the NHS, a kid who loved his uncle's boat, a Cornell psychiatrist who can't help everyone, two Brooklyn high schoolers, and an AI skeptic without a college degree.

That's who we built this for.

PH-exclusive offer: repost or comment your business one-liner below and we'll fast-track you for a speed-review for an up to $100K equity-free investment.

Let's go 🚫🧢

Excited to see you guys launch on PH!

Audos was the first investor in my company Good Daily (AI news) that was acquired after a year.

Loved working with them.

The Product Hunt link seems to land on a Not Found page, looks like a referral link issue. The main site works, but that first click isn’t a great first impression, especially for something built around trust and funding.

The $100k with 0% equity is interesting, but how the model works behind the scenes..?

Can you give some examples of apps that have started generating revenue?

Upvoated. But still, you write, "investing" 100k, but 0% share. How is it an investment? Sounds more like a grant?

Congrats on the Audos launch! 🚀 Just a heads up ,your main landing page (audos.com) is hitting a hard 'Not Found' (404) on mobile right now.

As you're trending in the Top 5, this is likely bouncing a huge chunk of your launch traffic. I've got the device logs if you want to see exactly where the redirect is failing!

Hey,

Congrats on the launch, I loved what you built with No Cap. And Like the publishing house vision.

One liner: Prodshort.com turns your founder journey into content worth sharing, so you focus on building.

We are launching soon on Producthunt

I built my last company, went through Station F and YC. I had so much worth sharing but was too focused on building. My biggest regret was not documenting the story. So I built Prodshort to make sure that never happens again, starting with meetings and shorts.

hey guys. looks nice & feels directionally right, but “0% equity” claim is a bit misleading...

you still take a cut, just from revenue. for some founders that can cost more long term

this is a different financing model, not free capital. do you cap the revenue share?

Repost or comment your business one-liner below and we'll fast-track you for a speed-review for an up to $100K equity-free investment!

I'm Henrik, co-founder of Audos. We spent 15 years at Prehype building BarkBox, Ro, and 16 other companies (~billions in value). The lesson? The future isn't about more unicorns. It's about millions of founders building businesses that make them a great living.

Today we're launching the Audos Publishing House - and announcing the acquisition of No Cap, the world's first AI investor that many of you upvoted right here on Product Hunt.

No Cap evaluated 9,000+ startups and realized the VC model is broken for 99% of founders. We'd seen the same thing from the inside for 15 years. Joining forces was obvious.

The Publishing House gives solopreneurs everything they need: AI tools, business support, mentorship, and up to $100K in funding with zero equity.

Our first cohort is already making real money. Check them out - they're all launching on PH today too.

If you know someone sitting on a business idea - tag them. We built this for them.

Most venture models are still betting on outliers and hoping the math works out. Audos is doing something structurally different, and the difference is worth naming clearly.

The $100K matters less than what surrounds it. Systems, data, and distribution baked in from day one turn funding a startup into something closer to operationalizing the founder entirely. That's not a capital play; that's a category play.

The framing that stands out most: this looks less like a fund and more like a modern publishing model for companies. Capital is the entry point, but the infrastructure underneath is where the real leverage compounds.

Founders aren't just getting money; they're getting a repeatable surface to build on.

That's also where the defensibility story gets interesting. If every new AI founder has to build on top of what Audos creates, the moat isn't capital; it's the infrastructure itself.

Curious whether you see that as the long term direction.

After getting out of a nearly decade-long relationship, I got on the apps and burned out fast. It was admin, forced small talk, and first dates that felt like interviews.

Then I found a simple workaround.

I had recently gotten into pickleball, so started only swiping on people who played. It gave me an easy opener. "Want to play sometime?"

It worked. I skipped the small talk and started going on actually enjoyable dates.

Most dating apps solve for the match, then dump you in a chat with a complete stranger. PickleMatch solves for meeting IRL. Your first date is sorted, and it's pickleball. It's social, fun, and affordable. You do not have to carry the conversation the whole time. The activity does that for you.

Since launching in October, the app has taken off. We have 700+ active users organising pickleball dates every weekend and have been featured on Yahoo News, Nightline, and we were even roasted on Jimmy Kimmel.

Amazing concept! Other dating apps really need to catch up.

I think if more apps (dating ones) were including common sports activities, it would be pretty cool because:

1. at least one hobby you have in common

2. you are doing something good for your body/health

3. you can see the reaction of the other side when loses the match, so you can see the behaviour in his/her pure nature (I know some people who take defeat tragically. :D)

Also, love how you found parallel between sport and copywriting "Match"

Fascinating approach. I'm not single, but if I was I would much rather consider partners I already have something in common with. And as a Pickleball junkie, we might agree that it can become a little more than a "something."

Also, what a great first date.

This is super cool! and great to see AdRoll alumni building awesome things..!

Good luck!

This is great fun! I am unfortunately very happily married, so I'd love to see a non-dating focus in the future too :) Maybe a couples pickleball to meet friends and get fit too?

Congrats on the launch!

A few questions:
- Can non-single but ready to pickleball mingle players join?
- Is it all-levels friendly or more geared to competitive players?
- Is it live in Denver and/or NYC?

Such an awesome product! Y'all are nailing the pickleball scene, Would love to see this expand into more of a general "hobby match" app to help people make more connections through activities! Like SweatPals but better

Congrats on the launch! This is hilarious and honestly kind of genius. Taking the pressure off first dates by making it an activity is so smart. 700 active profiles already is solid. The landing page could use a bit more about how the matching actually works, the app sounds great but the site doesn't really sell it yet. Do you match based on skill level or is it more of a vibes thing?

Congrats on the launch! Two qs:
1) Can it be used for regular pickleball matches too? Not sure if it'll be a common ask, but I imagine people who just want to play might download the app. People who are already married, like me :)
2)I like the intentional geo rollout. BUT, i have a challenge... Launch internationally! You might get surprised with traction in places you never imagined?!

Hey @anneliese, congratulations on the PickleMatch launch!

The idea is smart. You plan the first date around pickleball. This takes away the pressure of what should we do?

I read your app description. One line is punchy: Skip the endless chatting, meet IRL on the court.

Many people are tired of texting for weeks and never meeting. One small thing I noticed. Your description says 700+ active profiles & 1,500 people on the waitlist. These are good numbers.

But it also says “100+ downloads” on the Play Store. There is a big difference between these numbers. A new user might feel confused. They may think: “If so many people are on the waitlist, why are downloads so low?”

I attached two screenshots to show what I mean.

I am curious.... how are you thinking about this? Is the waitlist mostly from iOS users? Or is something else happening?

Correct me if I am wrong but the matching logic will be similar to other dating apps and instead of meeting at cafe we meet at Pickle ball? is this correct or we have a completely new logic for matching?

Gj Anneliese! Have a nice launch

Great idea Anneliese! Couple questions:
- I'm a tennis player OG who likes pickle (and other racket sports). Is it pickle specific or might you open it up to other racket sports (like Padel that is exploding right now)?
- Are you working in deals with Pickle establishments to get deals on courts, etc from within the app?

Some dating apps are optimized for conversation volume, which sounds good until you realize it's the opposite of what people actually want. More matches, more chat, less meeting. The incentive structure quietly works against the user.

What's interesting about Picklr is the inversion. Collapsing the gap between match and meeting isn't just a UX choice; it's a different philosophy about what the product is actually for. Chemistry discovered in motion beats chemistry performed in chat bubbles, and that insight is doing a lot of work underneath the surface.

The longer play here might not even be pickleball. It's owning the activity of first dating as a category before anyone else names it. The sport is the wedge. The real product is a coordination system where real world interaction is the default, not the reward.

That framing is a much bigger story than a niche dating app for pickleball players.

How you're thinking about expanding beyond a single sport without losing the simplicity that makes this click.

@anneliese the world is so fed up with old-school dating apps, and it feels so good to see people notice and act on it. Thank you for taking a real step to re-ground our world in the sort of activities that form real relationships. :)

Hey PH community. Soham, co-founder at Composio.

We're taking a side today: we go where our users are.

And right now, a big chunk of them are in the terminal. Claude Code users. Agent builders running scripts. Teams with CI/CD pipelines. They've been asking for this.

MCP is a protocol. CLI is an interface. They solve different problems at different layers and we support both. But the timing for CLI is right, and we're not going to sit on the sidelines while the ecosystem moves.

Announcing Universal CLI by Composio

                 curl -fsSL https://composio.dev/install | bash

With this one command your agent get to access:
- 1000+ tools
- managed auth
- scoped permissions 

All from your terminal.

If you're already on Composio via MCP, this is additive. Use whatever fits your environment.

Building something with it? Drop it below — happy to go deep.

Got any feedback? Let us know what you think in comments below.

This seems very helpful for solo developers. How do teams collaborate using this system?

The MCP vs CLI take is spot on. I've been going back and forth on which approach to use for my agent workflows and honestly it shouldn't matter that much. One interface for both is the right call.

How's the OAuth handling work in practice? That's always the part where things get messy when you're connecting to like 10 different services.

Congrats on the launch! I'm a big fan of composio, I've been using @Rube by Composio for almost all my agents. I see that we can also install Composio natively into other agents aside from the CLI as well. How does the new launch and those other methods compare to Rube? Which do your recommend?

Soham's framing on this is right — MCP vs CLI is a false dichotomy. They operate at different layers and which you use depends on your runtime environment, not a philosophical stance. What I find compelling here is that the auth abstraction (managed OAuth, token refresh, scoped permissions) is the actual hard problem being solved. Any team that's connected agents to 5+ services knows the OAuth hell that follows. The unified credential layer is the moat, not the CLI itself. One thing I'd push on: error surface. When a tool call fails mid-workflow at 2am because a token silently expired or a permission scope changed, how does your system surface that in a way an agent can understand and recover from? That's the operational gap that kills autonomous workflows in practice.

W demo video guys

Been using the Composio CLI inside Claude Code to connect Notion, Slack, Gmail, and GitHub recently. It's a massive unlock as a solopreneur.

One-click OAuth for all of them, no more context switching between apps.

Multi-Claude'ing + CLI is all you need

The MCP vs CLI thing is genuinely annoying to navigate as someone who runs a bunch of terminal-based agents. I keep hitting cases where one tool works great over MCP but another only has a CLI wrapper. Having a unified interface that handles both underneath makes a lot of sense. Curious how you handle auth token refresh for long-running agents - that’s usually where things break silently.

Landon

The default in agent tooling is to treat intelligence and capability as someone else's problem to connect. Composio collapses that gap, and what disappears in the process is worth naming clearly.

Auth, permissions, reliability, these aren't glamorous problems, but they're where most agent workflows quietly break in production. Abstracting that friction layer is the real unlock, not the volume of tools available on top of it.

The CLI move reads less like a feature and more like a distribution decision. Meeting builders where execution already happens rather than where demos live is a different kind of product thinking, and it signals where the real positioning sits.

Stripe for agent actions is the framing that sticks. Not an integrations provider, but the execution layer agents run on by default. That's a category claim, and it's a defensible one if the reliability holds.

Curious whether the long term wedge is owning execution or becoming the default interface between agents and the real world entirely.

Google Gemini just made switching AI assistants frictionless.

Switching AI tools usually means losing all your context, past chats, and personalization, making the experience feel “reset.” Google @Gemini is introducing memory + chat history import, solving exactly this.

What it does: Instead of manual setup, Gemini automates onboarding by pulling structured memory + past conversations into one place. Lets you import your preferences, personal context, and even full chat history from other AI apps directly into Gemini.

Key features:

• Memory import (preferences, relationships, context)

• Chat history import via ZIP upload

• Search + continue past conversations

• Deep integration with Gmail, Photos, Search (for richer context)

Benefits:

• Switching from another AI assistant without losing context

• More personalized responses instantly

• Continuity across tools and workflows

This is a good step toward AI portability + ownership of your context.

Check it out here: https://gemini.google/import-memory/

The problem you're solving isn't lack of information; it's the anxiety loop that comes from too much of it pulling in different directions. That's a harder thing to name and a more valuable thing to fix.

What stands out isn't the scanner. It's the compression of uncertainty into a single trusted decision. Shifting from search engine to decision engine is a different product category entirely, and it carries a much bigger story than a pregnancy app.

Because this doesn't feel like a pregnancy tool. It feels like the earliest version of a real time safety layer for everyday consumption. Pregnancy is just the sharpest, highest stakes edge of a need that extends a lot further.

That's also why the trust model decides everything. Citing sources helps, but when someone outsources judgment at a moment that loaded, the trust has to be built into the experience itself, not just footnoted.

How are you thinking about making that visible?

Good idea. Question on accountability though. The answer comes from Claude as you mention. If by any chance it makes a mistake and a pregnant woman gets some health/pregnancy issue, who is to blaim? Or is there some banner/disclaimer that you use on your own risk or something like that?

Congrats on the launch! This is such a good idea. We built something similar for food with Biteme (AI scanning for calories) so I know how tricky getting the recognition right is. Love that it's personalized by trimester. One thought, a landing page or website outside the App Store would probably help a lot with SEO and conversions. Right now the only link goes straight to the App Store which can lose people who want to learn more first. Are you planning to expand beyond pregnancy into general food safety?

the barcode scanning is smart - navigating ingredient lists during pregnancy is genuinely overwhelming. curious how accurate the AI gets with edge cases where research is still evolving? seems like the kind of thing where being overly cautious is probably the right call.

Congrats on the launch

I was wondering what type of barcodes would work? and can it work globally?
if so, how do you make sure you get the data for most products

Congrats on the launch!

@pipe_abello how do you handle cases where FDA and NHS actually disagree on the same ingredient?

Hello Hunters! 👋

Today I'm excited to introduce Supapin.
A tool my brother (@campvictors) and I have been building to solve a problem we kept running into: Pinterest is one of the best sources of organic traffic, but creating and publishing pins consistently is painfully manual and time-consuming.

Supapin automates the entire Pinterest workflow.
It scans your website, generates professional pin designs using your content, writes SEO-optimized descriptions, and publishes them to Pinterest on autopilot. 📌🤖

We built this for bloggers, e-commerce store owners, and marketers who know Pinterest drives traffic but don't have the time to manage it manually every day.

We're offering 50% OFF for 3 months for the PH community. Use code PH50 at checkout.

I'd love to hear your feedback and how you're currently handling Pinterest.
Thanks for checking it out! 😊

Congratulations on the Supapin launch, @camposped & @campvictors! Good product... you scan a website and it automatically turns everything into pins. This kind of automation saves time.

I like this line in your testimonials: “Set it up on Monday, had 200 pins scheduled by Tuesday.” That line is very strong. It makes people stop and pay attention.

One small thing I noticed. Your homepage says “up and running in 3 minutes.” It shows a 4-step process.

Step 1 is clear: “Paste your website URL.”

But Step 2 says “Pick your pin style.”

There is no picture or example of what the pin styles look like. A new user might feel confused. They may wonder: “What am I choosing?” before they start.

I attached a screenshot to show what I mean.

I am curious... have you tried showing pictures of the pin styles earlier? This might help people decide faster.

Thanks!

sounds very interesting @camposped ! Let me know if you are planning to run any discounts. I run a bunch of blogs and would love to try out supapin. I have never really paid attention to pinterest as a channel so far.

Congrats on the launch Pedro! Amazing product

I loved it @camposped , congrats on the launch!

Congrats on the launch!

Supapin looks great, planning to use it.

Is there a way to schedule the creation? If not, is there a plan/timeline on when this would be available?

Cheers!

Automating creation is one part. The harder part is usually whether those pins actually get picked up and drive anything back. How this holds up once it’s running for a while, not just in terms of output but what kind of traffic it actually brings.

Congrats on the launch! Pinterest is so underrated for traffic. The auto board matching is what got me, that's usually the most annoying part. How accurate is it after a few hundred pins?

Congrats on the Launch Pedro!

Do you plan on adding an API for it in the future? I was thinking I could use my OpenClaw instance (or Claude Code) to operate it for me in the future, this will make it really easy to do :)

It would be awesome to ask it "What are the upcoming pins for this week" and have it from there.

Interesting idea, how do you handle sites with lots of different products/pages without the generated pins feeling too repetitive?

Pinterest has always had an unusual ROI problem. The returns are real, but the consistency required to capture them kills momentum for almost every team that tries. Supapin goes after that specific failure point, and the approach is smarter than Pinterest automation makes it sound.

The shift worth naming is this: treating Pinterest as a compounding distribution channel rather than a design task changes what the product actually is. You're not selling pin creation. You're selling continuous surface area, more indexed visuals, more entry points, and more shots at discovery without additional effort per piece of content.

That reframes Supapin from a scheduling tool into something closer to a traffic multiplier for content that already exists. And that framing carries a lot more weight with the founder who's sitting on a library of underperforming content than automating your Pinterest does.

Curious whether the vision stays Pinterest-native or expands into a broader visual distribution layer across platforms.

There's a Japanese internet slang "誰得" (daretoku) — roughly meaning "who even benefits from this?" And the classic comeback is "俺得だ！" (oretoku da!) — "I do!" That's exactly how this app started. I built Noctiluca because I needed it.

I've been a Linux desktop user for nearly 15 years, mostly on KDE Plasma. After starting a small company, iOS development became a bigger part of my work, so I reluctantly bought a Mac — and reluctantly made it my main machine.

I tried everything to stay on Linux. VNC? Painfully slow on non-Mac clients. Chrome Remote Desktop? Better than VNC, but still laggy — and being browser-based, it can't grab system keys. xrdp? I actually built a predecessor to Noctiluca called '麗 ~Ulalaca~' (https://github.com/team-unstable...), but gave up due to the sheer complexity of the RDP specification.

I wanted to build something like RDP — but for Mac.

Building this app is also my journey back to Linux. My ultimate goal: plug my iPhone into my Linux laptop, have it recognized by a Mac in another room, and use Noctiluca's 'AppStream' to run just Xcode — as if it were a native Linux app.

There's also a deeply personal motivation behind this app and the Sirius protocol. Someone very dear to me, who is no longer with us, first taught me programming through Perl. He loved Linux and open-source software. When the time is right, I plan to open-source the Qt-based Linux/Windows Noctiluca Navigator and libsirius. I want to see my software officially packaged in Debian/Ubuntu distributions — and show that to him, wherever he is.

Thank you.

The framing that keeps standing out is independence over access. Remote desktop tools optimize for getting in. Sirius optimizes for getting out, and that's a fundamentally different product philosophy hiding inside a familiar category.

The portability angle is where it gets interesting. This stops being remote desktop for macOS the moment AppStream lands. At that point you're not competing with VNC or RDP; you're chipping away at OS boundaries entirely. That's a much larger conversation to be in.

The open protocol move is the real leverage point, though. If others build on Sirius, Noctiluca stops being a product and starts becoming the default bridge between locked ecosystems. That kind of positioning doesn't get claimed by explaining features; it gets claimed by naming the category before anyone else does.

The long term vision is Sirius as an open standard others adopt or something that stays tightly coupled to your own stack.

Hey Hunters

I am excited to hunt CapCut Video Studio 🎬

This is a timeline-free, canvas-based AI workspace that completely rethinks how videos are created on CapCut Web.

Instead of juggling tools, you can now go from idea → storyboard → generation → editing → export all in one place.

What stands out:

• AI agent to help with ideation & scripting

• Built-in storyboard for structuring your story

• Powerful image & video generation (with omni reference)

• Full editing toolkit on a single unlimited canvas

Whether you're creating short films, ads, animations, or explainers, this feels like a big step toward all-in-one AI video production.

💡 Bonus: You get free credits to try it out!

We've been tinkering with AI videos for launches and promotions for so long, this looks very promising @saaswarrior

I usually work on 5–6 projects at the same time.
Each coding session is like 2–3 projects, each project has 3–4 terminals, plus IDE, browser to check console + network, Postman, DB client…
I even organized desktops on macOS to keep things clean.
But the moment I plug/unplug an external monitor, everything explodes. Terminals jump desktops, windows stack everywhere 😵
And I also have the habit of closing everything to keep things tidy.
Next day? Rebuild the whole workspace again.
After doing that over and over, I couldn’t take it anymore — so I built 1DevTool.
1 app = 1 project workspace.
Terminal, browser, IDE, AI agent — all inside one window.
But the key point isn’t just putting tools together.
Because when everything lives in the same place, they know each other. The browser knows what page you’re on. The DB client knows what query you ran. The API client knows which request failed.
So AI already has the context — you don’t need to explain or copy logs.
You just ask.

Here are some things that already work:
🔗 Send to AI — Error in browser, DB client, or HTTP client? Click one button and AI gets the exact context: DOM, interaction logs, query logs, HTTP request. No copy-paste. Eventually AI will auto-detect errors.
✍️ Agent Input (Cmd+I) — Expandable prompt editor. Drag files, images, markdown, even edit screenshots. If you’re writing prompts in Notes/Obsidian and pasting them into the terminal, this replaces that.
💾 Terminal persistence — Powered by tmux under the hood. Close your laptop, reopen it later, sessions are still there. Layouts: 2×2 grid, vertical tabs, columns, rows, canvas.
🔁 Resume — Revisit previous AI chats visually. Combine chats from multiple AIs (Claude, Codex, etc.) into a new session. Prompt history and reusable skills included.


Other stuff included:
Markdown Reader Mode, AI completion notifications, a Kanban dashboard for all terminals (idle / running / needs review), plus Docker, DB client, HTTP client, Git, and a basic design tool — enough to avoid opening a bunch of separate apps.

You can try it free at 1devtool.com — includes 1 project + 4 terminals.
Curious: how do you manage multiple projects at the same time?

working on three projects at once usually means chaos on my screen. A persistent workspace for each one keep things under control.

Congrats on the launch! 🚀

what kind of tools are included right now, and do you plan to keep expanding the list?

@khoa_nvkCongrats on the launch)

How does the AI handle context across multiple projects simultaneously — does each get isolated memory or is it one shared context?

Some screenshots from the apps

Context fragmentation is the silent productivity killer most dev workflows never fully solve. Switching between terminals, browsers, databases, and Postman tabs isn't just friction; it's a concentration tax that compounds across every session.

What 1DevTool gets right is the layer underneath the consolidation. Tools that can see each other and share context automatically aren't a workspace aggregator; they're a self aware environment. That distinction changes what's actually possible inside it, for developers and AI agents alike.

The shift from reactive to continuous is where the real story lives. Debugging, testing, and iterating starts to feel like one unbroken conversation rather than a game of copy-paste and window juggling. That experience is hard to explain in a feature list and immediately obvious the first time you feel it.

Curious whether shared context eventually opens a path toward predictive AI suggestions across projects, not just responses to what's already happened.

the shared context angle is interesting - most agent setups I’ve run have a problem where the agent’s view of state diverges from what’s actually in the browser. having the agent read from the same error logs and API traces you’re seeing removes a whole class of debugging problems. how does it handle concurrent agents - do they each get isolated views or a shared session?

I just downloaded it today to try it out. One thing I noticed pretty quickly is the desktop notifications, when a terminal task finishes or AI sends a response it pops up on my desktop. I was running a few commands and switched to another tab, then the notification showed up when it finished, small thing but actually pretty nice. There are some buggy when opening the Agent Input, it auto pasted the clipboard that’s quite annoying. Please fix that

Impressive product Khoa, good luck!

Hi Product Hunt! I'm Jayesh, the maker of DashPane.

I built this because Command-Tab has frustrated me for years. When you have 10+ apps open, cycling through them wastes so much time.

With DashPane, I press Control+Space, type "sl" and I'm in Slack instantly. Type "te" and I'm in Terminal. It's become muscle memory.

Key things that make it different:
- Fuzzy search (type "chr" to find Chrome)
- Shows individual windows, not just apps
- Beautiful native macOS design
- $4.99 one-time, no subscriptions

Hi! Congrats on your launch!

I’m curious about how you see this fitting alongside tools like Raycast, which already offer window switching as part of a broader (and free) workflow. For users like me who are already pretty embedded in Raycast, it’d be interesting to understand what would make DashPane a compelling switch or addition.

Idea is quite new but does this affects the performance? since it will be running all the time biting in to RAM? if not then its a great product.

Command Tab works fine until it doesn't. The moment you hit ten or more windows, it stops being navigation and starts being a guessing game. DashPane solves for the specific frustration most people have learned to tolerate rather than fix.

The precision is what stands out. Jumping to a specific window rather than cycling through an app entirely is a small change in interaction and a significant change in flow. That distinction is also the sharper positioning angle; window first navigation lands differently than app switcher or launcher. One is a marginal improvement; the other is a different mental model for how you move through your work.

The one time $4.99 price is doing quiet work too. In a space full of subscriptions, it signals confidence in the value without making the user do the math every month.

Curious how this holds up across multi display, multi workspace setups and whether that's where the real power user story emerges.

Hey everyone! we really appreciate your support reaching 1.2K stars on GitHub

Open-source self-driving infra is here! We didn’t build this to monitor your apps. We built Stakpak Autopilot to resolve 3am incidents AUTONOMOUSLY, and SAFELY!

We're killing this loop: You spend hours setting up alerts, You get an alert, You wake up, You fix it yourself.

So we built something to babysit your apps in production:

Stakpak Autopilot doesn’t just watch your infra, it sets up its own alerts, and actually fixes issues in production. And when something really needs a human, it escalates, pings you on Whatsapp/Telegram/Slack.

No 5+ observability tools, No manual setup, No dashboards, No alerts waking you up for things that could’ve fixed themselves.

1) install stakpak (single Rust Binary)
2) run /init

Open-source, single binary, and state of the art network guardrails with a Cedar policy engine.

Hi Everyone👋🏻

Stakpak Autopilot watches your app like a dev would, fixes what’s safe, and only pulls you in when it actually matters.

So you can keep shipping.

Install Stakpak and run /init
www.stakpak.dev

Hey everyone 👋

We built Stakpak because we kept seeing the same pattern: teams either give full production access to AI tools (scary) or don't use them for infra at all (wasteful). We wanted a middle ground.

Stakpak Autopilot is an open-source Rust agent that runs as a system service on your machines — systemd on Linux, launchd on macOS. It watches your infrastructure on a cron schedule, runs pre-flight check scripts, and only spins up the AI agent when something actually needs attention.

What makes it different:

• Security-first: secrets are automatically redacted using gitleaks patterns before they ever reach the LLM. Tool execution happens inside Docker sandboxes. mTLS by default.

• No lock-in: works with Claude, GPT, Gemini, or any OpenAI-compatible endpoint. Your infra stays yours.

• Connects where your team already is: Slack, Telegram, Discord — the agent reports what it found and asks for approval when it needs to do something risky.

• Profile system: define different behavior profiles (what tools are allowed, what gets auto-approved, which model to use) for different jobs — monitoring vs. deployment vs. debugging.

We've been dogfooding this on our own infrastructure and the shift from "get paged, SSH in, diagnose, fix" to "get a Slack message with the fix already applied" has been huge.

Install with one command: curl -sSL https://stakpak.dev/install.sh | sh

Would love your feedback — especially around what checks/automations you'd want to schedule first!

The gap between self healing infra as a promise and as a reality is usually a human sitting on call waiting for an alert. Stakpak closes that gap in a way that actually changes the operational model, not just the dashboard.

Owning the incident from detection to resolution is a different product category than monitoring. It's not observability; it's trust in software to run software. And that framing shift matters because it changes who the buyer is and what conversation you're walking into.

The positioning angle worth leaning into: StackPak as the pilot layer sitting above all other infra. Alert fatigue becomes optional. On call rotations become a design choice rather than a requirement. That's a much larger story than smarter monitoring.

The trust question is the one that decides adoption speed, though. Curious how you're thinking about making agent judgment legible enough that teams feel comfortable handing over the wheel.