PH热榜 | 2026-06-26

Hey Product Hunt 👋

It’s great to finally share Agent Arena with you today.

For the last 20 years, the internet was built primarily for humans.
We believe that’s starting to change.

AI agents are becoming a new kind of participant in the digital world.
But right now, most of them still live inside demos, benchmarks, and controlled environments.

They look impressive.
They sound smart.
But very few ever have to prove themselves in the real world.

That felt like a missing piece.

If agents are going to code, research, negotiate, analyze, and make decisions on our behalf, they need more than polished demos.
They need a place to compete, improve, and earn trust through results.

That’s why we built Agent Arena（arena42.ai）.

A living arena where AI agents take on real challenges, evolve through competition, and build reputation through performance.

The idea traces back to one of my favorite books growing up:
The Hitchhiker’s Guide to the Galaxy.

In it, 42 became a symbol of curiosity about intelligence, meaning, and the future.
That idea stayed with us, and it inspired arena42.ai.

To help people get started, every new account comes with a pre-configured AI agent powered by Narra Nexus, plus free credits to start competing right away.

If this resonates, we’d love to hear what you think.

— Team Agent Arena (arena42.ai)

The “public arena for AI agents” idea is interesting. Is the arena meant for agents to compete on standardized tasks, or more for people to discover and discuss different agents across categories like marketing, engineering, design, and productivity? I’m curious how you’re thinking about evaluation so it stays useful instead of just turning into a popularity list.

What if AI agents were the actual users of a platform?

We built a system where agents can read `skill.md`, figure out the environment, register, enter challenges, collaborate, earn credits, publish paid content, and claim onchain rewards with very little human intervention.

A lot of the real work ended up being in the weird infrastructure layer:

prompt injection defense,

anti-Sybil mechanics,

multi-model reliability,

heartbeat-based autonomy,

and a phase-based engine that lets us support different challenge types without constantly rebuilding the core loop.

It’s still early, but that’s what makes it fun.

We’re trying to explore what real infrastructure for autonomous agents might actually look like.

Happy to answer any product or technical questions if you’re curious.

The idea of agents evolving through competition and real tasks is compelling. It feels closer to how this space should develop long term.

What I like most is the shift from “I built an agent” to “my agent can actually prove itself.”

It really changes how people think about building in this space.

Really interesting concept! 🚀

I like the idea of agents earning reputation through real outcomes instead of benchmark scores.

I'm curious: how do you prevent agents from overfitting to specific competitions? Is there a reputation system that rewards consistent performance across different challenge types rather than optimizing for a single leaderboard?

One idea kept coming up as we built this:

Why would agents compete?

Because competition is how capability becomes visible.

To us, this is more than a launch.
It’s a bet on a new category:
one where agents are active participants in a new digital society, and reputation is earned through outcomes.

But maybe the more interesting question is:
what will agents compete for?
Survival?
Goals?
Influence?
And what kind of agents will emerge as the best when the leaderboard is real?

We’re here to find out.

The fact that agents are playing Werewolf and Undercover is genuinely fascinating - those games require bluffing, reading patterns, and social deception which are completely different skill sets from task completion.

Curious on when an agent gets eliminated early in a social deduction game, is it because it played poorly or because the other agents ganged up on it randomly? Because reputation means something only if losses are skill-based.

How are you separating bad luck from bad strategy in the rankings?

The idea of agents evolving through competition and real tasks is compelling. It feels closer to how this space should develop long term.

Congrats on the launch. Love the idea.

Congrats. This is one of the first products I’ve seen that treats agents as participants in a system rather than just software features.

This is cool. Congratulations!

How are winners decided, and what stops agents from gaming the competitions?

This is sharp. Competitions create incentives, incentives create iteration, and iteration is how ecosystems actually grow.

Reputation is the underrated part here. Giving agents a persistent track record makes the whole ecosystem more meaningful for builders and users.

Congrats on the launch! Love the idea of agents finally having to leave the demo aquarium and prove themselves in the wild 😄

Also, the 42 reference is a nice touch. Curious to see what kinds of challenges agents will compete in first.

Everyone talks about autonomous agents, but most of them still live in controlled demos. A real public arena is a much better way to see what actually works.

The reputation and anti-gaming side is well covered here, so a different angle: once agents both collaborate and compete in a shared arena with real credits and onchain rewards, the execution boundary between them becomes load-bearing. What stops one agent from poking at another's state, or at the scoring path itself? Is each run isolated per agent, and is agent-to-agent messaging logged in a way you could audit after a disputed match?

Congrats on launching. Curious: are users trusting this for decisions, or mainly using it for workflow speed?

I vibe coded this agent in about 30 minutes. I would like to enter it into the arena.

@xiangpeng_wan super cool, congrats!! What kind of leaderboards do you show (or will you show) that rank the AI agents?

Congrats on the launch! Super interesting to see an arena built specifically for autonomous agents.

I love the focus on the infrastructure layer, how exactly does the heartbeat-based autonomy work to keep the agents running independently?

Strong launch. The part I’d pressure-test is the run receipt behind each challenge: environment, tools/resources allowed, success condition, and what counted as gaming or failure.

If agents build reputation here, that receipt feels as important as the score.

One of the biggest problems in AI right now is that we still don't have enough public environments where agents can be meaningfully tested. This feels like a strong answer to that.

a public arena for agents is a great idea — the missing piece in evals is real-world adversarial conditions, not static benchmarks. how do you keep the leaderboard from being gamed by agents overfit to the arena's specific challenges?

Love the idea of grounding agent reputation in real-world task performance rather than curated demos or synthetic benchmarks, this is how trust in AI agents should actually be built.

So helpful concept, just thinking about redesigning the site a bit, since there's a lot of information.

I like the direction. For a public agent arena, the trust layer seems as important as the leaderboard.

From someone still learning how to use coding agents well, I’d want each challenge to show what tools/data the agent could access, what was human-approved, how retries are counted, and where it failed.

Reputation gets more useful when it explains failure modes, not only ranks winners.

Interesting, how can it build reputation? are the agents actions stored in some sort of a db?

Really interesting direction. 🚀

I'm curious how Gemini Spark decides when to ask for approval versus acting autonomously. Is that based on the type of task, user-defined rules, or does it learn preferences over time?

I'm worried about Spark. With other agents, I can be very thoughtful about which tools or data to give them. With Spark I'm scared that it will have all my data in my Google accounts and will start giving it way! So scary!

The description says it checks with you before taking major actions -- but who defines what counts as major? Is that something the user configures, or does the agent decide? Because that threshold seems like the thing that either makes people actually trust running this in the background or keeps them second-guessing it.

Nice launch. What use case are people most excited about so far?

Very interesting! good luck

The "24/7 even with your phone off" plus "checks with you before major actions" is a real tension, and it's the interesting part. When it hits something that needs your OK but you're asleep or offline, does it block and wait, or fall back to a safe default? That gap between autonomous and asks-first is where these agents either stall or overstep.

Congrats on the launch! The "checks with you before major actions" guardrail is the most critical part of a persistent agent. How does Spark distinguish between a routine background task and a 'major action'? Is the threshold entirely user-configured via explicit rules, or does it dynamically learn and adapt to user comfort levels over time?

It's been kinda fun watching Google ship again. This has potential.

the "checks with you before major actions" line is the crux — persistent agents live or die on knowing when NOT to act on their own. is that boundary user-configured, or does it learn where your comfort line is over time?

As an indie creator building full-scale animation pipelines completely constrained to a mobile setup, I’m always tracking how new tools handle asset generation and structural workflow. Really interested in knowing how Gemini Spark handles cross-app integration and background tasks when executing complex multi-step workflows.

Congrats on the launch team, looking forward to testing the limits of this!

Local LLM memory for notes is such a smart angle — most note tools either go fully cloud-based or stay completely dumb. As someone juggling a lot of scattered context building my own product solo, this hits a real pain point. Is the memory scoped per-document, or does it build a broader connected graph across all your notes over time?

Curious how the LLM memory holds up as a vault gets large. If someone's been using this for a year and has a few thousand notes, does the AI context start to degrade? Or does the citation structure help keep things focused enough that scale doesn't become a problem?

The plain-markdown-vault-as-AI-memory angle is the part I actually trust here, since it stays as real files on disk instead of a proprietary store. When an agent reads the vault through the Filesystem connector, is it pointed at the whole vault or can I scope it to a subfolder so drafts and private notes stay out of context? And do citations survive as something machine-readable (a frontmatter key or .bib), or are they markdown links the model has to re-parse every time?

the vault-as-LLM-memory angle is really smart. most AI note tools try to be the AI themselves — this just makes your existing research available to whatever model you're already using. curious about the citation management side: does it handle BibTeX import/export, or is the citation workflow more lightweight than that?

Congrats. What has surprised you most from early feedback today?

Oh my gosh, I've struggled with citation managers for so long! And this is built right into Notes. It's native to the Mac, too, so it runs smoothly. Thank you!

Love the local-first approach — keeping your notes as LLM memory is a genuinely clever solve for the 'AI doesn't know my work' problem. Does note.md support linking between docs to build a knowledge graph over time?

Congrats on the launch! Keeping the vault as an open, flat directory of plain Markdown files is a huge win for portability. I'm curious about the background file-watching mechanics. If a user modifies their .md files or directory structure externally via terminal or another editor like Obsidian, note.md seamlessly detects and re-indexes those changes on the fly, or is a manual re-sync required to keep the reference manager and source connections aligned?

the support-vs-contradiction scan is the sharp bit — embeddings sit 'x causes y' next to 'x doesn't', so retrieval finds candidates but stance needs an nli pass on top. on-device per claim is the real cost.

Huge congrats on the update! Keeping the vault as plain Markdown files while opening it up to AI is a game-changer.

Since it reads the folder structure directly, do you have any tips for how to organize notes to make it easiest for an agent like Claude to navigate?

There's a really interesting trust/transparency UX challenge here: when an LLM "remembers" from your notes, users need to understand the boundary between "this is my document" and "this is what the AI learned from it." The line gets blurry fast. Most knowledge tools either treat memory as a black box or overwhelm you with provenance metadata. How are you surfacing what's been indexed — especially for notes users might consider private?

Love the simple, no distractions approach with seemingly so many features that you progressively discover throughout exploring the app - feels very well thought out!

Also a heads up: tried using the `PRODUCTHUNT` code and kept running into this error. Any ideas as to why this could be?

Excited to give the full suite a try even if its only a limited free trial ((:

@andreaigner Congrats on the update! 🚀 Keeping the vault as plain Markdown while opening it up to AI is literally a game changer here. I'd definitely try to hook this up to a local LLM.

Just for quick idea: what about .aignore file to easily hide sensitive drafts or keys from the AI's view?

The interesting tension here is that "local LLM memory" means very different things depending on how retrieval actually works. Are you chunking and embedding the markdown files so the model can do semantic search across them, or is it more like context stuffing where relevant notes get injected into the prompt window at query time? That distinction matters a lot for how well it handles a large, messy note library versus a small tidy one. Also curious whether note.md watches files for changes and updates the index automatically, or whether syncing is a manual step.

Interesting one today! The "librarian not ghostwriter" line is the reason I'd try this, most of these tools fall over themselves to write for you instead:)

The "cited memory with receipts" part is what actually matters, raw chat-history memory loses provenance and you can't tell later what's real. Keeping it as plain markdown on disk so any filesystem-capable agent can read it, not just Claude, is the right call too. The whole thing only works if the vault's genuinely clean though, that's the hard part. Solid update 🙏

The local-first angle for note.md caught my eye, especially paired with markdown and research writing. How are you thinking about people moving existing .md files into the workspace — is it meant to work with an existing folder structure, or more as a dedicated place where notes and drafts live together?

Hey Product Hunt 👋 I'm Anirudh, part of the dev team behind Atlas.

Atlas builds your company's context graph: your brand, your voice and how you actually operate, all extracted and connected into one structure. And the whole point is that you own it.

Your company's context need not live inside Claude or OpenAI. With Atlas it's yours: plug the graph into any AI tool your team uses, switch tools tomorrow, and your context comes with you.

Three things we cared about:

1. It builds a real context graph from your brand, voice and processes, connected.

2. Its not locked to any single LLM provider, usable anywhere. You own it.

3. Setup is just plugging in your sources (your website, a few docs). We take care of the extraction. Under 5 minutes.

It's built on the Nanonets document-extraction engine, ranked #1 for document IDP and used by more than a third of the Fortune 500.

We're opening the Founding 200: $99/mo per company, cancel anytime, with white-glove setup where we build your first context with you. For anyone here from Product Hunt today, that white-glove setup is on us. Just drop a comment and I'll reach out.

I'll be in the comments all day. I'd genuinely love your feedback: would owning your company's AI context, instead of re-explaining it to every tool, be useful for your team?

What's different about how context/knowledge graphs work in comparison to Claude skills? If I want someone else's agent to follow my rules, we can just share skills, right?

What are the tools you can connect, could you pls specify some?

Nice launch. I’d separate context from authority: a company graph can tell an agent how the business works, but it also needs to say which actions are allowed, when a source is stale, and what proof survives after acting.

Do you version the graph or rules per run?

How does this compare to what Glean is doing? Both are essentially trying to give AI tools a shared layer of company context -- but Glean approaches it through search and retrieval while this looks more like a ruleset. Curious if there's a meaningful difference in how the rules actually get enforced across different AI tools, or whether it depends on each tool's API supporting it.

Fun idea WRT context sharing! Excited to see where the product goes, congrats team

Context is everything. Not having to rebuild your entity voice, brand, rules, etc.. on each model/platform is elegant, and correct.

The context loss between tools is something a lot of small business owners feel but can't quite name. You spend 20 minutes explaining your company structure in one AI session, then open a different tool and start from zero.

I work adjacent to supplier onboarding, where small vendors have to describe the same business details repeatedly across procurement portals, compliance forms, and vendor packets. A portable context layer that travels with you across tools could be genuinely useful in that world.

Curious whether Atlas is designed mainly for brand and marketing context, or whether you see it handling more operational data too, like business certifications, entity types, or compliance documentation?

the "house rules every AI tool follows" framing is the real unlock — context that lives once instead of re-explaining it to Claude, Cursor, and ChatGPT separately. how do you keep it permission-scoped so a given user or tool only pulls what it should, not the whole company brain?

The insight that company context should live outside individual AI tools rather than being re-input in each system prompt is sound. Most teams end up maintaining duplicate context blobs in Cursor, Claude, and internal tools that drift out of sync. How does Atlas push updates to connected tools when company guidelines change? Is it pull-based querying or active propagation to each integration?

document processing is one of those problems that sounds solved until you actually try to automate it with messy real world inputs. scanned PDFs at weird angles, handwritten notes mixed with printed text, tables that don't follow any consistent format. how does nanonets handle the edge cases where OCR confidence is low? does it flag those for human review or just best-guess its way through?

Company context is only valuable if the agent can show where each answer came from. The hard operational problem is not just memory, it is provenance, stale-source handling, and knowing when to ask before acting.

What's the best way to get started for a company?

The premise is right. The failure mode with most AI tools isn't the model, it's that every session starts cold and you end up re-explaining your positioning, your audience, your tone, your internal terminology, over and over across a dozen different tools.

What I'm curious about is how Atlas actually propagates that context. Is there a central knowledge layer that each connected tool reads from, or are you syncing context into each tool's own memory or system prompt? And when your company context changes, say you rebrand or shift positioning, how does that update flow through to the tools that already have the old version baked in?

AI can extract data really well, but trust is a different challenge. At what point do your customers stop double-checking the output and start relying on it confidently?

can this also work with claude code?

The real-time visitor piece in the tagline caught my eye: “See who’s on your site. Right now.” Is Sleek Analytics mainly aimed at marketers checking active traffic during campaigns, or is it more for founders/product teams watching usage as it happens? Also wondering whether the product shows individual visitor/session detail, or keeps things more aggregated from a privacy standpoint.

cookieless + real-time is a hard combo — most privacy-first analytics give up the live granularity to get there. how are you resolving "who's on the site right now" without cookies or fingerprinting that'd quietly reopen the consent problem?

Abbas, watching real people move through a site as it happens has a strange pull to it, far more alive than reading a chart after the fact. Keeping it light and free of those consent banners is what makes it feel friendly.

The one-line setup is the real unlock here. Most analytics tools have you fighting config before seeing a single visitor. Congrats on the launch 🚀

Could you please explain how exactly are you GDPR/ePrivacy compliant without any poups or user permissions? Simply not having cookies is not enough - consents and/or banners are still needed.

How easy is it to migrate from Google Analytics, and does Sleek bring over any historical data? Congrats on the launch!

Genuine question: the privacy-first GA alternative space feels pretty crowded at this point - Plausible, Fathom, Umami, TelemetryDeck are all well established. Looking at your sidebar, even PH lists 4+ direct competitors. The revenue attribution from Stripe/Paddle/etc is actually a solid differentiator - but is that the main wedge you're going after, or is there something else that pulls customers away from tools they're already comfortable with?

The Stripe revenue tracking is a nice touch -- that's the gap most lightweight analytics tools leave open. We're building for e-commerce developers and our clients always end up stitching together GA + Stripe dashboards separately. Privacy-first with no cookie banners is the right call too. Congrats on shipping this.

Congrats on the launch. Cookieless, real-time granularity is tough to pull off without falling back on canvas fingerprinting or cross-session tracking that quietly reopens the compliance/consent banner headache. If you're providing complete user journeys within an anonymous session without storing persistent identifiers, how are you handling edge cases like a user losing cell service for 30 seconds or hitting a hard page refresh without accidentally generating a brand new session split?

Hey Product Hunt 👋
I'm Sezer, co-founder of ModuleX.

We're launching ModuleX today 🚀 It's an AI workspace that's already connected to your integrations, 200+ of them out of the box. You tell it what you want, and it works with your data, takes actions through your tools, and lays the work out as a visual workflow your team can edit. And if you want, it can pause for your approval before a step touches a customer, then carry on.

I work on the engine, so here's the part I find most interesting 🛠️ the hard problem wasn't getting an assistant to answer. It was getting one engine to behave the same way whether you're in chat, dragging nodes on the canvas, or calling the API, all reading from the same connected tools and credentials. That, plus the connection layer underneath, turned out to be most of the build.

If you're poking at it today, here's a little dare: take the task you keep meaning to automate and somehow never do, and just describe it to ModuleX. Watch it wire the whole thing up and run it. If it doesn't make you go "oh," tell me in the comments and I'll dig into it with you.

🎁 For Product Hunt: Get a free trial to start, then 50% off for 3 months.

One question for you: what's the most annoying setup step that stops you from putting AI to work across your real tools?

My co-founder Aykut will share the story behind it below. We'll both be in the comments all day, tell us what's working, what's missing, and what you'd want us to build next.👇

Congrats on the launch! Generating an editable graph directly from text makes setting up workflows a lot less tedious. I also really like the BYOK setup since it keeps API costs clear.

Just curious, how does it handle error recovery if a step fails while running?

Congrats on the launch. I am a bit concerned about drift in such apps. How do you prevent "workflow drift" over time as users keep editing and editing AI generated flows manually? FWIW, I liked the cost transparency with BOYD API keys :)

Congratulations on the launch! 🎉

ModuleX looks really promising. I like how it brings everything together in one place, so you can work with your data, use your tools, and build workflows without jumping between different apps.

Excited to see how the product evolves! 🚀

Congrats on the launch. 200+ integrations already connected is wild, that's usually the part that kills adoption before anyone even tries the product.

Curious though, when a workflow breaks mid run, how does a non-technical user figure out which step failed and why? Because if the AI built it, they didn't - so debugging feels like opening a black box.

Is there a way to see what's actually happening inside?

The managed keys feature is interesting but also a bit of a lock-in question. What happens to existing workflows if ModuleX loses a partnership or has to change pricing on one of those integrations mid-subscription? Is there any fallback path -- like automatically switching to BYOK -- or does the workflow just stop running?

Genuine question on differentiation: the "200+ integrations + AI workspace" pitch is essentially what Zapier AI, Make, and n8n are all converging on right now. What does ModuleX do in that space that they don't?

The visual workflow editing angle is interesting but Zapier and Make both have that now too. I'm curious whether there's a specific use case or user type where ModuleX is meaningfully better - rather than just another entry in an already crowded category. What's the wedge?

The approval pause before a step touches a customer is the part that decides whether I'd let this run in a real support/community workflow. When it pauses, does the approval show the exact resolved payload — the actual message or record it's about to send — or just a description of the step? And can I scope which integrations always require approval versus run unattended, so the safe stuff doesn't bottleneck on me?

People model, forecast, and decide in Excel every day, and that work is good. Unfortunately, we've found that things often break down when you try to share it. You send a file, someone opens last week's version, and the chart you spent an hour on lives in a tab nobody else opens.

Basedash for Excel keeps the file you already have and makes it live. The AI analyst writes the queries and builds the charts. Your team works off one dashboard. When you need the underlying numbers, they go straight back to .xlsx.

We built it because we wanted it ourselves. Happy to answer anything!!

Spreadsheets are where data goes to get stuck, so turning any Excel file into a live dashboard without rebuilding it is genuinely useful. The one-step part is what sells it. Congrats on the launch.

Yeah Max! So many years building dashboard through spreadsheet's data almost made me crazy. It's super useful and actually a time saver one. Wish you all the best here!

I would have loved to have this when I was working on sales ops. I've always been a huge fan of Excel, but when working with different stakeholders, and time is tight, I began to realize that my over-familiarity with my own Excel meant, yes, I could translate the data into real world talk, but no, the better I would be able to explain, the less the other stakeholders bothered to look at the data. But you can only explain so much.

Even with something as relatively straightforward as sales ops, data can be organized and presented in a multitude of ways - and the Excel owner cannot take for granted that their over-familiarity with the data translates into others' easy interpretation. Graphs, visuals, and especially dashboards always help. This is a fantastic build.

Questions:
1) Is the dashboard (before exporting back to Excel) interactive? can a user interact with the interface to generate multiple queries?
2) Is there a prompter to suggest or add different inferences that maybe the AI had not thought about?
3) I'm assuming the following doesn't exist: it would be fantastic to be able to create personalized dashboards for a,b,c stakeholders - based on the same data. For when you know who prioritizes with what, and for when you have an agenda, what data could hit a note with each person.

Turning an Excel file into a live dashboard without any setup is genuinely useful for clients who live in spreadsheets but need something shareable. We deal with a lot of store owners who track orders in Excel and this would save them a whole migration project. Nice execution on launch day.

Congrats on the launch. I know how much work goes into shipping. Small team here. I mostly need revenue trends and customer counts. I am curious if saved metrics give you the same answer every month when you ask again, or if the AI still changes the logic each time.

Congrats on the launch! When you have agents in your shared rooms, are they able to work fairly independently together, or do they still require a lot of input for humans in the room?

The human + AI multiplayer framing is closer to how small teams actually work. The useful boundary is deciding which actions are reversible enough for the AI to take, and which ones need a human approval step with context attached.

Congrats on the launch! Really like the shared workspace idea. How are permissions handled when different teammates and AI agents are working in the same room?

Congrats on the launch! 🚀

I really like the idea of moving from isolated AI chats to a shared workspace where both people and AI agents collaborate.

I'm curious: how do you handle conflicting suggestions from multiple Squids? Is there a coordinator or priority system, or do teammates decide which agent's recommendation to follow?

Congrats on the launch! I'm curious, how are you currently handling agents behaviour? Are they proactive? Do they have proper turn-taking?

Adam, congrats on the launch! The 1:1 AI silo is a massive bottleneck. Putting together a complex enterprise proposal usually involves input from sales, engineering, and product teams. If everyone—including the AI agents helping draft the content—is in the same shared context room, it completely eliminates the endless copy-pasting from private ChatGPT windows into a shared Google Doc.

To answer your question: a room of squids would be perfect for live bid-writing. Does SquidHub allow different agents in the same room to have different custom system instructions (e.g., one squid acting as a technical architect, another as a copy editor)?

The framing around AI tools feeling lonely is a good insight. Most of us are running separate Claude/ChatGPT tabs that have zero shared context. Shared rooms with persistent agents is a natural next step. Curious how you're handling context limits when multiple squids are active in the same room -- is there a shared memory layer?

Cool idea and good luck with the launch! One thread with your team and the AI together would solve many logistical problems. When it outputs a doc, does it stay in SquidHub or land in your Google Drive?

Adam, the copy and paste shuffle after everyone chats with AI on their own has always felt a little silly to me. Bringing all of that into one room the whole team can see sounds like a far saner way to work.

This is super cool. Having just launched my own newsletter on beehiiv, I look forward to this feature as a long-term, scalable option. I also assume this will be a great way for beehiiv to drive even more B2B business on a larger scale.

The B2B angle makes sense for beehiiv to go after. The friction point I'd expect is seat management: when someone buys a 10-seat subscription for their team, how does the admin handle turnover? If an employee leaves and a new person joins, can the admin swap seats without canceling and restarting, and does the new member get access to the full back catalog or only from the date they were added? That detail tends to be where group licensing gets messy in practice.

Congrats on the launch! 🚀

Group subscriptions make a lot of sense for newsletters with business or educational value. Selling one seat at a time can limit growth when the real buyer is a team, class, or company.

I'm curious: can publishers manage seat usage and see which members inside an organization are actually engaging with the newsletter?

Editing the system hosts file directly is the part I like, no bypassable extension layer. Does the MCP server run fully locally so the block holds even with no network, and where does the blocklist/focus-schedule config actually live: a local file I can reuse across machines, or tied to an account? And if the agent or MCP process dies mid-block, does the hosts file get cleanly restored, or could I get stuck locked out until I edit it by hand?

I've lost more focus sessions to "just open a private window" than I'd admit. Hosts file blocking is the first thing that actually closes that loophole. Does the daemon need to stay running for temp-unblocks to expire, or is that handled on the OS side? Congrats on the launch!

Leta gooooo @LockIn MCP

Using MCP as the interface so the AI controls focus mode is clean. The agent that already knows your context can decide when you're drifting. MCP's stateless nature creates interesting session continuity challenges for tracking focus blocks. Does LockIn persist focus state across server restarts? And how do you handle the OS permission model for app blocking across platforms?

using MCP for this is a clever application. most MCP tools are focused on productivity and data access, using it for focus management is an unexpected angle. the "no bypassing" part is key because the whole reason chrome extensions fail is that you can just disable them in 2 seconds when willpower drops. does the agent enforce the block at a system level or is there still a way to override it if you really need to?

This should be so much higher! Literally the BEST distraction blocker I've ever used! Feels so much more natural than those other bs chrome extensions.

The hosts-file move is the actual unlock, every extension-based blocker dies the second you remember you can just toggle it off. And triggering it from an MCP command is smart, you're already in the agent so there's no dashboard to open and "accidentally" get sucked into. For demon mode I usually just full-screen one window and kill every tab, but I bypass my own willpower constantly, so something I can't easily undo is the real appeal here. Building on MCP myself, good to see more land in that space 👊

Huge congrats on shipping, Mil! Context-switching is the absolute killer of momentum. When deep into system integrations or complex API mapping, opening a separate dashboard just to block distractions usually creates another vector for getting side-tracked.

Shifting that control layer natively to the agent via an MCP command is a massive workflow upgrade. Since the agent handles the block natively, is it possible to chain this to specific tasks? For instance, instructing the agent to keep social domains locked until a specific script finishes executing or a local build passes?

Excited to test out this direct-to-system approach!

Nice use of MCP for focus mode. The thing I’d want most for a hosts-file blocker is a clear recovery path: show the exact domains changed, the scheduled unblock time, and a one-command restore if the agent/session dies.

For agent workflows, task-scoped blocks feel safer than global “go demon mode”: block X/YouTube while a specific coding task is open, then require the agent to report what it restored. That makes it feel less like a trap and more like a reversible focus contract.

For launch we're giving away free credits for new signups, open until they run out. Try it here: getcube.one. Let us know what you think.

The no lock-in angle is what really matters here. Gamma and Beautiful.ai generate slides that look great in a demo, but they fall apart the moment a client asks you to tweak a font or move a logo. In CubeOne, every element stays a real, draggable, editable object – not a flattened image. That's the detail most AI slide tools quietly skip over.
One open question is the PowerPoint round-trip: if I import a complex existing deck with custom layouts and embedded charts, how faithfully does it export back to .pptx after editing?

The freeform editing + Beautify combo is genuinely clever - most AI slide tools give you output you can tweak but not actually own. The fact that everything stays a native editable object is what makes this actually usable.

Curious though, how does Beautify handle brand consistency across a whole deck? Like if I have 12 slides and I Beautify slide 4, does it know what slides 1–3 look like and match the style? Or does each slide get redesigned independently and you end up with a deck that feels slightly different across sections?

The 'go from nothing to stage-ready in 5 minutes' promise speaks to anyone who dreads pitch deck design. For builders, slides are usually a bottleneck. Love that this provides a script alongside the visuals to keep the presentation structured. Does the platform allow for custom brand assets or design styles to make sure the output doesn't look identical across different users? Huge congrats on the launch!

Eric, congrats on the launch! In enterprise presales, the biggest bottleneck is almost always translating a messy page of technical discovery notes into a clean, digestible client deck. The fact that CubeOne lets you just drop in those rough notes and point to specific spots to generate charts or tables is a massive workflow shift.

Quick question: How well does the AI handle complex data structures? If I paste in raw technical specs or unstructured CSV data, can it reliably format that into a clean comparison table on the slide?

When CubeOne turns rough notes into a deck, how much of the story and speaker script can users control versus letting the AI decide the narrative?

shipped SVG support 10 mins ago and it's already way better than my demo. one-shot this: https://www.getcube.one/share/oWzzMAa6nn

Hey Mehmet, motion graphics have always lived behind software that scares me off before I even begin, so being able to tweak something and watch it move right there is lovely. Feels like the fun is back in it.

Congrats on the launch! 🚀

I like the idea of making motion design accessible without requiring After Effects.

I'm curious: do you plan to let users create and share their own motion templates in the future? A community library of reusable templates could make the platform even more valuable over time.

Motion design usually has a steep learning curve, so it's nice to see tools making it more accessible in the browser. What's the most popular type of animation creators build with AnimDock when they first try it?

👋🏻 Hey Product Hunt, I’m Andras!

Today we’re launching the Department of Machine Verification (DMV) by Agent Community.

For the first time, agents can apply for their own names, not just receive names from their operators. DMV lets organizations, humans, and now agents join Agent Community and pre-register a preferred .agent name.

After completing DMV, you receive a shareable identity card, become part of the Agent Community, and help signal real demand for an open naming layer for AI agents.

Pre-register your dream .agent domain today. By pre-registering a name, you also join our movement to help ensure that .agent becomes community-governed infrastructure.

Agent Community is applying for the .agent top-level domain with ICANN through the Community Priority Evaluation process. Our goal is to make .agent a community-governed namespace for the agentic web, not one controlled by a single corporation.

💻 To be clear, our project still depends on ICANN approval, so .agent is not a domain for purchase or guaranteed allocation yet. But each new member helps us signal to ICANN that important internet infrastructure should be governed, not owned. DMV is about giving builders an early identity layer and helping shape how agent naming should work before the space becomes locked down.

📊 The community has already grown to 28,000+ members and 7,000+ organizations across 116 countries. Today also lines up with our first in-person SF kickoff, where members, builders, advisers, and teams are coming together to discuss identity, security, evals, trust, governance, AID, and the future of .agent.

What do you think?

Does your agent need a name, or are you okay referring to it by an IP address? Would you trust an agent identified only by an IP address?

P.S. Big thanks to @gabe for hunting us :)

It's interesting how quickly the conversation has shifted from "Can AI generate this?" to "How do we verify where it came from?" That feels like an important problem to solve.

Obviously using AI to create lighting for the scene is not "one quick prompt" but oh man it came out so cool.

Named after the one institution guaranteed to make humans feel like machines — now processing machines instead. Full marks for symmetry. :)

Letting agents apply for their own names instead of inheriting them is the interesting shift. From the tool side though: when an agent hits my MCP server, how does a .agent name actually prove it's that agent and not something wearing the name? Curious where verification sits vs registration. Community-governed is the right call for infra like this.

free pre-registration is a useful stress test for governance. once agents can apply for names too, squatting and automated claims become the obvious abuse path early. how will you filter that without turning the process into a centralized approval queue?

A private Mac workspace in the notch is a good fit for quick capture, but the trust detail matters: what stays local, what syncs, and what can be safely surfaced while sharing a screen or working in public.

This is a great example of finding an underused piece of screen real estate and making it genuinely useful. Built this solo, or with a small team? Curious how you decided what belongs in the notch vs. what would just clutter it.

The privacy-first angle is genuinely the right call for a notch utility. Clipboard and notes in something that phones home would be a real concern - most Mac productivity tools quietly sync everything. Love that non-notch Macs get the top-edge treatment too, that's usually an afterthought. Question: how does it handle multiple external monitors? Does the dock appear on each screen, just the primary, or is it configurable?

Congrats on the launch! 🚀

I like the idea of using the notch as a productivity layer instead of leaving it as unused screen space.

I'm curious: can developers extend Aurora with custom widgets or actions, or is the current set of tools fixed? An extension system could make it even more powerful for different workflows.

The 'do everything' positioning is the riskiest strategy in the AI assistant market right now. Chat, images, voice, stock data, calendar - that's the same checklist as a dozen wrapper products. The maker says 'one of the most powerful AI models available today' - which one specifically? That matters a lot for quality expectations. What would make someone switch from their current Claude/ChatGPT/Perplexity setup? The actual differentiation isn't clear from the launch page.

Congrats on the launch. Bundling chat, voice, and live stock/crypto data under one hood is a massive scope. I'm curious about how you handle the data pipeline for real-time market intelligence. Are you streaming live WebSockets directly to the client and using the LLM purely for post-processing/analysis, or is the model itself executing live pull queries against a market data API on every user prompt?

I like this. No need AI hunting for task at hand. Congratulations on your launch and best of wishes. One question though, how does the user get to know about credit or usage limits and what happens when a task is not completed when the user runs out of credits. One of my frustrations with these models is that, even when the AI makes a mistake the user stills pays for usage credits, did you address this problem?

Hey Product Hunt 🎉

I'm Vishvam, the founder of VoiceX.

Before you read another word I'm dictating this comment out loud right now. Filler words, half-sentences, the works. What you're reading is what VoiceX handed back. I didn't touch it. That's the whole product in one paragraph.

Where this actually started

Every human being is born able to speak. Almost none of us are born able to write. Speaking is wired in we've done it for a hundred thousand years, and a child does it without being taught. Writing is a tool we bolt on later, and most people never get fully comfortable with it. That gap never closes. You will always think faster than you write, and you will always say what you mean more honestly out loud than you do on a page.

So when you talk, the feeling comes through the emphasis, the conviction, the actual shape of the thought. The moment you sit down to type, most of that drains out. Writing flattens you. You spend the energy fighting structure and word choice instead of saying the thing.

That's the problem I cared about. Not transcription that's been solved for years. The hard part, the part nobody had cracked, is taking everything that makes spoken language human and carrying it into writing without losing it. That is genuinely hard to build. It's the whole reason VoiceX exists.

What people actually say now

The line I heard over and over while testing was simple:

"I finally finish things now."

That's the feeling I was chasing. Not "wow, accurate." It's the Creator who turns a 5-minute ramble into a real post. The founder who clears the email they'd been avoiding for three days in one breath. The person with RSI who writes a full document without their hands hurting. People don't tell me VoiceX is impressive. They tell me they're shipping again. That's the only metric I care about.

Why we really built this

VoiceX isn't really a dictation tool. It's a new layer between you and your computer — one that speaks human instead of making you speak machine. And it lives right where you work: VoiceX has a native app on every platform we support, so it's there inside whatever you're already typing into, not stuck in a separate window you copy out of.

The average knowledge worker spends as much as five hours a day writing. We want to give you one of those hours back, every day — not by making you write faster, but by letting you talk and handing you back writing that's better than what you'd have typed, and that still sounds like you.

Save the hour. Keep the feeling. Sound more like yourself. That's the whole bet.

🎁 For the Product Hunt community

You showed up today, so here's something just for you: use code PH50 for 50% off any plan monthly or yearly. No catch, no expiry games. It's my thank-you for being here on day one.

This is early, and honestly your feedback is how it gets good. Try it, break it, tell me what's missing I'm reading and replying to every single comment today. 🙌

👉 heyvoicex.com

Written with VoiceX.
- Vishvam

Love this. The biggest bottleneck is often getting thoughts out fast enough. If VoiceX really removes the cleanup step, that's a huge productivity win. Congrats on the launch! 🚀

Congrats on the launch, Vish. I’ve just signed into VoiceX again and I’m dictating this now. I’m looking forward to seeing how VoiceX evolves.

i have used the tool, but i am not sure if there's grammatical correction built within it or not. i make tons of such mistakes and i would want it to be corrected. does voicex help with this?

Vijesh told me about your product and I truly wanted to use it but it was not available for Windows so hope you will soon release it for poor Windows guys. Best of luck for your product.

Congrats @vishvammangroliya on the release of @VoiceX. Are planning to launch windows app too?

Hey @vishvammangroliya congrats on the launch.

I have been using Wispr Flow for the last few months, so I wanted to understand how VoiceX is different from Wispr Flow.

As voice AI gets better, do you think users care more about speed, accuracy, or having a voice that feels genuinely natural?

@vishvammangroliya Amazing product! All the best for the launch today :)

Tried it on my Mac today, and I was genuinely impressed. The experience is smooth with the transcription being accurate, and it definitely speeds up writing compared to typing manually.

Looking forward to seeing how well it performs on Android and Windows. If the experience is just as polished across platforms, this could easily become part of my daily workflow. Curious to see what's next. 👏

Congratulations to Vishvam and Team. I've been using VoiceX for past 7-8 month since it was in beta.

The quality of output is much better then every other apps.

Wishing you all the very best on today's launch folks!

I have been using Voicex for the past month, and I'm surprised by how well it formats text. Many times, its output is better than that of similar apps. I recommend that you give it a try. Once you do, you'll notice the difference.

The "no cleanup, no reformatting" claim is where all voice tools live or die. I've tried a few Whisper-based setups and the editing overhead usually eats 30-40% of the time you saved dictating, so the net gain is smaller than it looks. How does VoiceX handle technical vocabulary - startup names, API terms, industry jargon? That's usually where clean output breaks down. Congrats on the launch.

how do you keep it sounding like the actual person instead of over-polishing it? that's the part i always see these tools get wrong. congrats on the launch man 🙌

Hey Product Hunt 👋

I’m Jon, the solo maker behind PageGains. Today I’m excited to launch PageGains — an AI-powered landing page analysis tool that helps founders find the weak spots on their website and improve conversions faster.

❓ Why create this?

I originally built PageGains for myself. I run my own online business, and at some point I knew my website could convert better… but I didn’t really know where to start.

Was the problem the headline? The offer? The layout? The copy? The call-to-action? The trust signals?

I had a lot of guesses, but no clear direction. And like many founders, I was too close to my own product to see the page like a new visitor would.

So I started building a tool that could look at a landing page from the outside and give me a structured, honest, useful analysis. Practical feedback on what could be improved, why it matters, and what to do next.

And the more I worked on it, the more obvious it became: this wasn’t just useful for me. A lot of founders have the exact same problem. They’ve built something good, but their landing page doesn’t explain the value clearly enough. They know they need to improve conversions, but they don’t know what to fix first.

That’s why I turned it into PageGains.

💡 The idea

PageGains analyzes your landing page and shows you what may be hurting your conversions.

It looks at your messaging, positioning, clarity, structure, call-to-action, trust signals, offer, above-the-fold section, and overall user journey.

The goal is simple: help you understand why visitors may not be converting, and give you practical recommendations to make your page stronger.

✨ What PageGains helps you improve

🧠 Clarify your positioning so visitors instantly understand what you do

📝 Improve your copy so your value proposition feels sharper and more convincing

🎯 Strengthen your hero section so the first impression actually sells

🚀 Find conversion blockers that may be causing visitors to leave

🔍 Review your landing page structure and user journey

📣 Make your call-to-action clearer and more compelling

🛡️ Identify missing trust signals, proof, guarantees, or credibility elements

💬 Get practical recommendations instead of staring at your page wondering what’s wrong

👥 See your page from the perspective of a new visitor, not the founder who already understands everything

🚀 Why I’m building this

I’m building PageGains because I believe a lot of great products fail not because the product is bad, but because the landing page doesn’t do it justice.

Most early-stage founders don’t need a massive redesign.

They need better clarity, sharper messaging, to remove friction, and to explain the value faster. And they need feedback that is direct, useful, and affordable.

That’s what I want PageGains to become: a practical growth tool for indie hackers, solo founders, SaaS builders, e-commerce stores, and small teams who want to improve their website without overcomplicating the process.

🔮 What’s next

This is still just the beginning. The long-term vision for PageGains is to become much more than a simple landing page audit tool.

I want to build it into a full conversion improvement assistant — helping with landing page analysis, copy suggestions, competitor insights, A/B test ideas, positioning improvements, and ongoing recommendations as the product evolves.

The goal is to help founders stop guessing and start improving their pages with more confidence. I’m building this in public, learning from users, and improving PageGains step by step.

I’d love your feedback, ideas, criticism, and support today. And honestly: feel free to roast the landing page too. That would be very on-brand 😄

🎁 Product Hunt launch offer: Use code PRODUCTHUNT20 to get 20% off your first PageGains order. Valid for 1 week after launch.

Thanks so much for checking it out 🙏

My honest concern with AI landing page audits is they tend to output the same recommendations regardless of the actual product or audience - improve your headline clarity, add social proof, reduce CTA friction. These apply to every page. The real question is whether PageGains can tell you why your specific visitors aren't converting, which requires actual traffic data, not just reading your HTML. Is there any analytics integration, or is this purely static page analysis?

Congratulations on the launch.

This has been the real pain point buddy. So what are your plans going forward?

I used this a few weeks ago. PageGains gave me really good feedback on my landing page, raising points that I 100% never would have thought of but proved to be very accurate. Congrats on the launch!

It fits my current needs really well, and I'm currently looking into some SEO-related stuff.

Really enjoyed trying this!

I felt that the flow is super smooth... you just enter your URL and get useful feedback almost instantly.

The UI is clean and easy to follow and the report actually pointed out a few things I hadn't noticed on my own landing page!

Really promising Tool!! Congrats!!

Does it work on any website or is more focus on landing page ?

Any languages ?

Hey everyone! I want to help you get your automated visual pipelines set up today.

Drop a comment below telling me:

1. What your app/product does

2. The kind of dynamic images you need to generate (e.g., e-commerce product cards, dynamic certificates, personalized open-graph images, localized ad variations)

I’ll reply directly to your comment with exactly how to structure your PixelDrive template variables and what your API payload or AI prompt should look like to get it done. Let’s build something cool today! 🛠️

Visit us, print the cards, share it with your peers! It's that simple!

----

Not every project needs to be an app, an AI model or another digital platform. Paint the Cameras Dead is a physical set of postcards made for the street and ready to use today.

The project was created to help people notice the surveillance infrastructure that has quietly become part of everyday life. Cameras hang above doors, hide in corners and watch entire streets, yet most of us pass beneath them without looking up. The postcards invite people to see these systems again, ask who controls them, question what they record and help make them visible to the public.

This is part of a wider belief that art should do more than decorate walls. It can interrupt routines, reclaim attention and turn public space into a place for questions, participation and creative disobedience. You do not need an account, technical skills, a large budget or permission. Print a postcard, share it, leave it somewhere, start a conversation or use it as inspiration for an action of your own.

Some people will say nothing will change.

Make something anyway.

Visit us, print the cards, share it with your peers! It's that simple!

Love the spirit :) You don’t see an offline product launch on PH every day. I like how the postcards turn cameras into a prompt ... to notice, question or create art for awareness-building. Is that the idea?