PH热榜 | 2026-06-20

Hey Product Hunt! 👋 I'm Will, founder of WorkClaw.

I'm excited to introduce WorkClaw: the AI team for your team. Your team is about to get bigger. A lot bigger!

Most AI agents work in silos. One person, one assistant. But that’s not how real teams collaborate.

WorkClaws are AI coworkers who can collaborate with your whole team to get actual work done, your way.

We make it easy to hire and onboard proactive, collaborative AI teammates who work in Slack and Microsoft Teams, just like every other colleague.

Each Claw gets:

• A job title

• A manager in your org chart

• A cloud-hosted ClawOS computer

• The ability to access more than 3,000 apps and any MCP server

Then, you train them with skills and routines so they work the way your team actually works.

What makes us different from most AI tools?

1. WorkClaws are collaborative, like a real teammate. They can even message one another in their ClawChat app!

2. Our platform is built for security-minded teams. It has robust access controls and is SOC 2 compliant.

3. Each Claw has a specialized role, preventing context drift and confusion.

4. Admin access controls: one Claw can work with your whole team simultaneously, another might be accessible just to you, and another might be available only to one team within your organization.

We've just opened up Early Access, and you can use the code PHEXTRA100 to get an extra $100 in credits when you sign up at https://workclaw.com?ref=producthunt

Happy to answer any questions!

@willruben Nice job here. Have used this - it’s a nice mix of flexible and enterprise ready

We are looking to roll out company wide

Congrats on the #1 launch! The orchestrator-Claw-as-single-point-of-accountability pattern is the smart call. I build AI agents for small businesses and the moment you go multi-agent the bottleneck stops being capability and becomes "what did each one actually do?" Curious how much of WorkClaw is observability / audit trail vs the agents themselves — that's usually where trust is won or lost with a team.

Do Claws have memory that persists across projects and reorgs or does a Claw's 'institutional knowledge reset if it gets reassigned to a different team or manager?

Is this basically just OpenClaw or something else?

Since Claws operate continuously rather than just responding when prompted how do you prevent them from being overly chatty or noisy in Slack? Proactive is great in theory but team channels can get noisy fast if every Claw is initiating threads.

Genuinely curious about the economics here are you pricing per Claw like a headcount or per seat/usage like typical SaaS? The AI coworker framing kind of implies the former which would be a bold pricing model.

This feels like it's solving for team level AI adoption rather than individual productivity which is a much harder go to market. How are you handling the internal champion problem i.e. who inside a company actually owns hiring and managing a Claw?

Curious how learning skills and routines actually works under the hood is this more like fine tuning per org or is it closer to a Claw building its own internal playbook from observing how the team works over time?

Most AI coworker tools are 1:1 with a human. You're explicitly going multiplayer Claws collaborating with each other and with the team. Has that introduced any weird emergent behavior like Claws looping on tasks or duplicating work without a human in the loop?

Giving each Claw a manager and a job title is such a clever framing for adoption but how does performance review work in practice? Does the human manager actually rate the Claw's output and does that feedback loop back into how it behaves?

Congratulations on your launch!

I met the team behind WorkClaw during Tech Week New York at a comedy event that they hosted. Hey, great tech and a sense of humor tick all the right boxes for a product I'd like to test drive.

Interesting to see AI moving from "assistant" to "coworker." I'm a non-developer founder building an AI-powered health product, and the biggest shift for me has been treating AI less like a chatbot and more like a collaborator in the workflow. Congrats on the launch.

Amazing, I like that this is built into existing apps. How do you stop two Claws from working on the same task or giving conflicting answers? Can admins see a clear record of who asked for what and what each Claw changed?

Strong launch. The org-chart model is the interesting bit.

For a Claw that can touch 3,000 apps, I’d want each task to leave a work receipt: who assigned it, which app/account it used, what changed, and whether another Claw or human approved. Do those receipts live in Slack/Teams, or in WorkClaw?

Hi @willruben, congrats on your launch. I love the playful look of the muppet-claws. "Worky Clawson" is an excellent name. I wish I had a better understanding of how credits translated to actual tok. The pricing is a little confusing for me. It's hard for me to tell if the team plan includes VM rental space or if that is in addition to subscription fee. Overall, this looks like an elegant way to productize OpenClaw and provide users with a no-fuss way to get started with multi-agent Claw workers! Looking forward to giving it a try.

The proactive part is the exciting and scary part in the same breath. A coworker who waits to be asked is safe; one that acts on its own across 3,000 apps can do the wrong thing at scale before anyone notices. I'd want to know what the approval and rollback model looks like for actions that touch money or customers, because that's the line between a helpful coworker and a very fast liability.

With access to that many integrated apps how do you handle the inevitable API changes or outages on the other end? Does a Claw degrade gracefully if one of its 3,000 connected tools goes down or does it just stall?

@willruben I’m really impressed with so much about Workclaw as a person well versed in AI but also tech in general, how do we know when it’s time for us to do an upgrade vs you doing the upgrade? And how long does it take dev to get back to me on a simple thing like putting in a plugin when I gave dev the exact coding plugin the workclaw asked to be put in the system 2 days ago so I can see my system we built together in workclaw instead of outsourcing it to Vercel that isn’t as safely locked in security vs you guys that are on top of that very well?

What's the onboarding experience like for a new Claw versus a new human hire? Is there an equivalent of a 30 60 90 day ramp or is it closer to instant productivity from day one?

Congrats on the launch! Do WorkClaw agents only do what they're told or do they feel like part of the team too, brainstorm, push back, have fun with people?

What happens when two Claws on the same team disagree about how to execute a task? Is there an escalation path or does one just defer to org hierarchy the way a human would defer to their manager?

access to 3,000+ apps is a big surface area and the actual hard part isn't connecting to that many apps, it's knowing which app to use for which task without a human specifying it every time. curious whether a Claw figures that routing out itself from context or whether someone has to explicitly configure which apps it's allowed to touch for which routines

Wow, that's awesome! Memory unlocked. Like traveling back in time.
When for windows or android?

This is the vibe I remember when I first tried a computer in my early years. Totally basic UI, and I was amazed as a kid. Brilliant!

Love this idea. The older I get, the more I appreciate products that remove complexity instead of adding it. There's something refreshing about software that remembers simplicity.

Does this also include 1999-level latency?

The modern Chromium engine under a classic browser UI is a fun mix. Wayback Mode also makes it more than nostalgia, since you can actually browse the web in its old context.

This is so fun! Yeah, sometimes I really do want to go back in time. World of Warcraft Classic is really popular, and this is pretty much internet classic mode :D. Nice work!

Looks fun to use. How exactly does it pull the old version of a website?

Slackbot's MCP client is Slackbot's AI that connects 20+ apps into one multiplayer conversation.

Teams use isolated AI tools across private tabs, manually carrying data between systems. Slackbot becomes the connective layer for your entire stack, coordinating actions in plain language in team channels.

Work is multiplayer from the start in Slack channels, not single-player silos in private tabs.

Features:

• Actions, not just answers: sign docs, update tickets, review dashboards right from Slackbot

• Native Block Kit support (coming soon): rich visuals like data tables update in real time

• Multiplayer execution: share Slackbot's response into a channel for team collaboration

• Plug-and-play MCP integration: connect any tool via MCP server in minutes

• Enterprise-grade security: user-specific data boundaries, IT admin audit console

Benefits: Finish tasks faster, scale best practices, maximize software ROI, collaborate with live data.

Who it's for: Engineering, sales, marketing, product teams

Use cases:

• Track Linear tickets + PagerDuty incidents (Product & development)

• Find Q2 pitch decks (Document management)

• Review Canva layouts (Creative & design)

• Draft Docusign agreements (Business operations)

• Analyze Tableau trends + Lucidchart diagrams (Visual collaboration)

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

The shared-channel angle is the part I’d test hardest.

If Slackbot takes an action from a channel, is authority bound to the requester, the channel, or the app install? For things like signing docs or updating tickets, that identity boundary is what makes the audit console useful.

I use Slack daily, and I really appreciate the MCP client ecosystem, the ease it makes in connecting the context and the actions you need to an AI chat interface. Everyone has their own personal preference on where they like that interface to live, and I think more options make it easier for everyone to find their ideal interface. The personalization in that way is very useful while at the same time giving everyone the same access to the tools and "middleware" we all use.

We are on an enterprise account, but when I follow the instructions on https://docs.slack.dev/ai/slackbot-mcp-client/ and go to https://api.slack.com/apps and do a "Create a new app", I do not see the MCP option.

The "multiplayer" framing is the underrated part. Read side is an easy yes; the write side across 20 apps is where small teams freeze. But running actions in a shared channel instead of a private DM means the team sees what got done in real time — the channel becomes a free audit log. That visibility might do more for trust than any confirm dialog. Does it post a structured "here's what I changed" back to the channel after a write?

The read side of this is an easy yes; the write side is where I'd want detail. "Sign docs" and "update tickets" from a plain-language command means the model is interpreting intent and then taking an irreversible action inside someone else's app. As someone who ships an MCP server, the thing that earns trust is a clear confirm step on anything destructive and a visible log of what it actually did, not just what you asked it to do.

How does it compare with the Qwen 3.6 and Gemma 4 models? It's disappointing to only see the old models. It seems misleading.

Shipping a focused, smaller coding model as open weights is the interesting bet here — the frontier-model-for-everything approach is expensive and overkill for completion. What I'd want to know: what context window does Mellum practically use for repo-level completion, and is it trained for fill-in-the-middle specifically, or general next-token? FIM quality is usually what separates a good in-IDE model from a chat model bolted into an editor.

Latency-first models are underrated. I run AI voice agents and on a live phone call latency isn't a nice-to-have, it's the whole UX — a 2-second pause feels broken to a caller in a way it never does inside an IDE. For narrow, well-scoped tasks I'll take fast-and-good-enough over slow-and-brilliant every time. Is Mellum something you'd consider for real-time / voice use cases, or is it squarely aimed at the coding loop?

Yeah! Workflow performance became key and this is bringing a clear advantage there. @fmerian doing what Flo does! The real hunting goat!

Nice launch. The memory I’d trust most is not just facts, but attempts: what the agent tried, why it failed, and which tool or write it was allowed to use next.

If a memory is wrong or stale, does pumaDB show who or what wrote it and let builders expire or correct it?

I have seen teams, including my own, avoid memory because setting up a database or vector store feels too heavy for early workflows. How do you decide what should be saved as memory and what should stay out? Can developers inspect and clean up memory when an agent saves something wrong or outdated?

I really like the pitch for this. I too have run into this problem, and not everyone has the time, energy, and willpower to research all of the different skills and scaffolding and harness options to make your own ideal memory layer. Speeding up that whole process, to me, is a very empowering thing to give to people.

The transcripts example resonates. I've got the same problem with a different domain. Iterative LLM workflows where each session starts cold means re-feeding context that should persist.

For "what would agents remember automatically": the thing I'd actually pay for is automatic capture of failed attempts and why they failed. Most memory tools focus on remembering successful artifacts. The harder and more valuable thing is remembering the dead ends so the agent doesn't try the same broken approach next session.

MCP-first feels right for the dev audience. Curious if you're planning to expose the memory as a queryable index later or keeping it strictly key/notes.

Used this yesterday and loved it, the details on the UI, the sounds, the concept, and of course the insights you get from searching the weights. I love this kind of projects!

I am!

I heard about this on an IT news website. They used it to identify the 10 most famous people in the world (though, in my opinion, Albert Einstein was missing from the list). Another drawback: This method doesn't work when there are people with the same name.
I find this method interesting when conducting market analyses of brands, products, or companies.

Beautiful work, Florent. The core idea, turning a list of toggles into a place you furnish, is one of those framings that feels obvious only after someone nails it. I'm not on Mac at the moment so I couldn't grab the app, but I spent some time in the web version and the concept really shines: being able to position each source to the millimetre, pull it closer, pan it left or right, is genuinely satisfying. Visually it's super clean and the pixel-art room sells the metaphor instantly :)

One honest bit of feedback from the web build: the interaction is the star, the drag-and-drop is immediate and tactile, but the audio payoff didn't quite keep up with it for me. There was a touch of latency before a sound faded in when I placed it, so the spatial sensation felt a step behind the gesture. Since the whole promise is "you're sitting inside the room," that tiny gap between moving the orb and feeling it move is exactly the moment that needs to feel instant. Could well be a web-only limitation that the native app doesn't have.

Which brings me to my actual question: any plan for other OS down the line? I'm mostly on Windows/Ubuntu, Mac only occasionally, so I'd love to actually live in this rather than visit it through the web demo. Congrats on the launch either way :)

I really like the way of mixing different ambient sounds using 2D space, looks awesome!

A room of ambient sound living in the notch is a great Mac-native idea. The best part is that it feels like it belongs to the machine instead of another floating app window. Curious how subtle you keep the UI during deep work?

Nice launch. The group-level AI context is the part I’d pressure test.

Access says what data someone can see, but the assistant also needs to know what it is allowed to do with that context. Do automations inherit the same group boundary, including MCP server and tool access?

Congrats on the launch. The audience-specific AI context part is interesting.

Most access control tools make me think about who can see which data, but changing how the assistant responds depending on the team seems like the part that could matter a lot once more departments are using the same BI tool.

Do teams usually need that separation from the start, or does it become more important as usage spreads across the company?

What kind of actions can it do today, like refunds, plan changes, or updating account settings?

Congrats on the launch. I like the idea of support, docs, and changelogs living closer together.

As a founder, repeated support questions are usually a signal that something needs to change, either in the product, the docs, or the onboarding flow.

Does ReleaseDock help surface those patterns, or is it more focused on managing the support flow itself?

Hey PH 👋 I'm Illia, solo founder of Brightforge (Delaware C-Corp, building from Helsinki).
I started Pixlie because most AI video tools feel like a slot machine you prompt, you hope. I wanted a studio: queue jobs, reuse assets, same account on web and iOS, and actually ship clips for marketing and social.

V1 went from zero to live product in ~22 days of brutal solo 350 hours sprint to first full scale working prototype to 6 months in production till release. iOS flutter native is on the App Store; Android is in Play; Web version. Bootstrapped on welfare while learning Finnish NVIDIA Inception, Nebius/Lambda infra, YC Startup School alum, applying to YC Fall.

Would love feedback on:
1) Create flow (text vs image to video)
2) What would make you switch from Runway/Pika/Kling?
4) Rate and review AppStore/ PlayMarket/Web desktop version any social media activity on product pages Linkedin/IG/ X.com/YC co-founder (Mobile app not available in EU yet - GDPR, EU act...)
5) Co-founder intros if you know someone strong on GTM/ops

Try it: pixlys.com happy to answer anything in the comments.

How long are renders typically, and do you get notified when a job finishes on iOS?

native git gui without the terminal — the real fork is libgit2 vs shelling out to the git binary. libgit2 misses some auth + partial-clone paths; shelling out means parsing porcelain that drifts each release. whichever you picked quietly shapes the whole app.

The menu bar monitoring is useful, but the ignored-file backup is the bigger painkiller. Losing or rebuilding .env files and local configs after moving machines is exactly the kind of small dev friction that adds up.

Awesome, is it coming for Windows/ Linux too?

Visual GitHub management on macOS is a nice angle. Git tools usually make people choose between terminal power and GUI clarity. Curious what workflow you optimized for first: reviewing PRs, syncing branches, or resolving conflicts?

Good! Two-way sync is definitely the strongest part here — token drift between Figma and code gets painful fast. Curious how you handle conflicts when both sides change the same token.

Two-way sync is the part that gets me, token drift between Figma and code has bitten every design system I've worked on. What format do you pull back into the codebase, Style Dictionary or W3C design tokens JSON? Congrats on shipping this.

Very important problem to solve.

How do you integrate securely via these API providers? Is it via their admin APIs or making code changes?

Any plans to have an open source version to host?

Congrats on today's launch.

Pre-commit is the right place for this. Catching secrets after they hit git history always feels too late. I’d be curious to see how often the local LLM flags useful issues vs noisy false positives in real projects.

Hey PH 👋

I built this because I kept seeing the same thing in AI-assisted codebases. It's not malice but ultimately it's just speed. When you're vibe coding or working fast because of a deadline, and the LLM is generating 200 lines at a time, secrets and unsafe patterns slip through the review loop since it can't keep up with the generation loop.

The thing I wanted most was something that ran before git, not after. By the time a secret is in your history, the damage is done even if you rotate immediately since the commit hash is permanent and the exposure window already existed.

Three things I'd love feedback on:

1. The Layer 2 training set is 297 samples across 11 languages so it's still small. If anyone has labeled risky/clean code snippets they'd share, I'd love to grow it.
2.The VS Code extension is next and it'll reuse the same pipeline so you get inline squiggles without running a commit.
3. Apple Silicon only right now. A Linux port would need different runtimes for both CoreML and MLX so I'm, interested in whether there's appetite for it.

Repo is MIT and fully open. Would love issues, PRs, or just to hear what secret patterns you've seen slip through that I'm not covering yet.

Nice launch! I like the local/private angle — no uploads is a strong selling point for image tools. Curious how it handles faces and small text, since that’s where most upscalers usually break.

Hey mate! This looks fantastic.
If you do industrial design or UI sketches in a notebook, how will Inksightful interpret these?

How well does it handle messy handwriting and mixed pages ?

Does the calendar pull from your existing videos to avoid repeating topics you’ve already covered?

👋 Hi everyone!

🚀 I'm excited to share Math Shield, an iOS app that helps people build healthier screen habits by turning app unlocking into a learning opportunity.

🧮 Instead of simply blocking distracting apps, Math Shield requires users to solve math problems before gaining access. It's designed for kids who want to improve their arithmetic skills, as well as adults who want to sharpen mental math, improve focus, and reduce mindless scrolling.

💡 The idea came from noticing how often we unlock apps without thinking. I wanted to create something that not only reduces distractions but also helps users learn and practice a useful skill every day.

🙏 I'd love to hear your feedback:

• 📱 Would you use math challenges to reduce screen time?
• 🎯 What types of challenges would motivate you the most?
• ✨ What features would you like to see next?

Thanks for checking out Math Shield! 🚀🧮

Nice idea — I like that it adds a small “mental friction” instead of just blocking apps completely. Curious if you plan to add different difficulty levels for kids vs adults?

Do you support private roles and mission voting on mobile without it getting stressful?

Hey Product Hunt!

I'm Dmitrii, the maker of Cutio

I built Cutio because I wanted YouTube to feel uninterrupted without manually jumping over sponsor reads and self-promotion

Cutio analyzes each YouTube video's transcript with AI, detects sponsor reads and self-promotion, and skips those parts automatically while you watch — in the browser or on a paired TV

Unlike community-based tools, Cutio can work on fresh, niche, and non-English videos before anyone has manually submitted segments

The goal is simple: open a YouTube video and let Cutio handle the parts you would normally scrub through by hand

What makes it different:

• works across languages, creators, and topics

• analyzes videos directly from their transcripts

• skips automatically in the browser

• can also skip detected segments on paired TVs

• pairs with YouTube on TV using a simple code, with no server setup

• shows detected segments progressively while analysis is running

• saves results to a shared cache, so analyzed videos load faster for everyone

• gives you simple filters for segment types, video categories, and maximum video length

• tracks time saved, skipped segments, and analyzed videos

• supports your own OpenRouter key if you want more control

Cutio is available for Chrome today, with support for more browsers coming soon

I’d love feedback on detection quality, TV pairing, edge cases, whether the interface feels simple and minimal, and what you’d like to see added next

@brolnickij I think the interesting question here isn’t whether people want to skip sponsor segments. Everyone does.

What I’m trying to understand is whether this remains accurate as creators adapt. More and more YouTubers are blending sponsors directly into the content instead of treating them as a separate block. Some channels even make the sponsor part entertaining enough that viewers don’t necessarily want it skipped.

Does that create a moving target for Cutio?

The other thing I’m wondering about is whether transcript analysis is enough on its own. A lot of YouTube content relies on context, tone, visuals, or transitions that don’t always show up clearly in transcripts. Have you found cases where the transcript suggests something is a sponsor segment but the actual video context says otherwise?

Also, what’s the long-term moat here? If transcript access and models continue to improve, it feels like detecting sponsor reads becomes increasingly commoditized. Is the real value in the detection itself, the TV experience, the shared cache, or something else entirely?

The problem is obvious and the solution is easy to understand, but I’m curious which part of the business you think becomes harder over the next few years rather than easier.

@brolnickij Creators have become increasingly dependent on sponsorship revenue, and many are actively designing integrations that feel less like ads and more like part of the video itself.

That seems like it creates an interesting challenge. The better creators get at making sponsorships feel natural, the harder they become to detect.

Has the detection problem become more difficult over time as creator behavior evolves, or has the improvement in AI models been keeping pace with that change?

The TV pairing is the most interesting part here for me. Browser sponsor skipping is useful, but making it work on YouTube TV feels like the real unlock. My only concern would be false skips on creators who blend sponsors into the actual content.

@brolnickij What does usage actually look like? Are people mostly using Cutio on long-form content where the time savings are meaningful, or are they turning it on for everything they watch regardless of video length?