PH热榜 | 2026-06-01

Hey Product Hunt 👋

I am Sridhar, co-founder of Mina AI. 

I’m excited to introduce Mina, an AI Meeting Assistant designed to actively participate in meetings instead of just recording them. 

Most meetings today still end the same way: Forgotten decisions, scattered follow-ups, missing context, and hours of manual work afterward. Existing meeting tools mostly act as passive note takers. They transcribe conversations, generate summaries, and stop there.

We wanted to build something fundamentally different. Mina acts like an actual AI teammate inside your meetings.

It can:

• respond during meetings in real time

• pull live context from your connected tools

• capture decisions automatically

• generate summaries, proposals, and follow-ups live

• assign action items

• update CRMs, tickets, and workflows

• retain memory across meetings and conversations

What makes Mina different is flexibility.

Instead of being a single-purpose assistant, Mina can be configured into different workflow-specific teammates:

• a proactive moderator

• a quiet assistant

• a Scrum facilitator

• a customer-facing copilot

• or a completely custom workflow assistant

Mina integrates with 200+ tools including Slack, HubSpot, Salesforce, Jira, Notion, Google Meet, Zoom, Microsoft Teams, Linear, GitHub, and more. 

Some common use cases:

• sales calls & demos

• standups & sprint reviews

• customer success reviews

• hiring interviews

• brainstorming sessions

• leadership syncs

• project discussions

• training & onboarding sessions

The biggest shift we noticed internally: Meetings stop becoming documentation exercises and start becoming places where actual work gets done.

We’ve been testing Mina with founders, operators, and teams for months, and today we’re excited to open it up to the broader Product Hunt community.

Product Hunt users also get additional free credits today to explore Mina. 

Claim your additional free credits here: https://getmina.ai/ 

If there’s a workflow, meeting type, or custom assistant you’d like to build with Mina, drop it in the comments. I’d genuinely love to help you set it up and hear your feedback.

Would love your support, feature requests, and ideas ❤️

- Sridhar

What does "Skills" mean here in practice: user-defined templates, or pre-built integrations?

How do you keep Mina from interrupting people or talking at the wrong time, is there a push to talk or strict turn taking?

Since Mina creates action items live, how do teams stop tentative ideas in a call from becoming assigned work too early?
Does Mina have it own Tasks tracker or can add them to notion?

Congrats on the launch.

Does it collect information from previous meetings about any particular topic and remind the team if we are going off course?

very interesting.. I was thinking how does one invoke her? By saying, "Hey Mina, can you please..." and so on? What if there is real human named Mina also on call? Is it possible to change her default 'name'? or change the way we address this AI?

he 'pulls context from your tools' piece is where these products usually have the biggest gap between demo and reality. pulling context from a CRM or a Notion doc in a controlled demo looks seamless. in a real meeting where the context is spread across five tools with inconsistent naming and outdated records it usually surfaces the wrong thing at the wrong moment. what does the failure mode actually look like when the context retrieval gets it wrong mid-conversation

Well, this is quite interesting. Imagine that someday instead of just having Mina as a meeting assistant, you give Mina a personality and face which is AI generated and she actually becomes a character on the meeting. It's not a must-have feature, but definitely a nice-to-have feature from the user experience perspective. :D

How is Mina different from a product that released a few days back called @Shadow or @Littlebird?

Congrats on the launch 🚀

The “AI assistant that actually participates in meetings” direction is really interesting, especially because realtime conversational systems seem to hit a completely different class of production/reliability problems compared to normal chat apps.

I love using Mina. It is helps me organise my daily work schedule. Keep me abreast of my task for the day. Sets-up tasks after every call and marks everybody on the call with clear call to action and tasks post the call.

Mina在会议中能被其他参会人感知吗？

@sridharmuppidi congrats on a highly successful launch! two thoughts - a question and a point of feedback:

1. How do you avoid Mina piping up when you don't want her to?

2. I'm sure you guys have been heads down and way more focused on the product than the branding/marketing, which totally makes sense, I'm a big PLG guy myself. With that said - and I mean this with the utmost respect... your product is good enough that it deserves a vastly better logo/website/explainer video (especially)/branding more broadly.

Regarding my second point, maybe it makes sense to work on the product and then launch new branding with v2, but, I'm just left with the feeling that the copy/imagery/branding/visuals/video at the top of the site don't suitably convey the excitement/capabilities. I think this is one of the few products I've seen recently where a more narrative - cinematic, even - approach is warranted. Something very luxe and engaging.

Just food for thought - I think some tweaks to positioning and GTM approach could make a huge difference... but then again, you've dominated PH on a Monday - no small feat - so, what do I know? :p - congrats again.

the shift from "records meetings" to "participates in meetings" is where this gets interesting. most AI note-takers are still just fancy transcription. the real test is whether Mina can handle the messy context switches that happen in actual calls, not the clean demo scenarios. curious how it handles situations where the right move is to say nothing.

Can Mina reference past decisions?

Here's the honest problem with every AI meeting tool right now: they're all built to help you remember meetings, not improve them.

You still run the meeting. You still chase follow-ups. You still update your CRM, write the recap, create the tasks. The tool just watches and reports.

@Mina Meeting Assistant Mina 1.0 changes the job description entirely.

With 40+ active skills, Mina doesn't wait to be debriefed, it works alongside you. It speaks when needed, acts on decisions in real time, syncs across your tools, and makes sure every commitment made in a meeting becomes a thing that actually happens.

The goal with Mina 1.0 is simple: zero unproductive meetings.
Every session has an outcome. Every outcome has an owner. Every owner gets a nudge.

If that sounds like the AI teammate you've been waiting for. Give it a try and let us know what breaks (and what delights).

My first thought was that people may misuse it, like getting help for passing an interview without real experience. On which side Mina will be in this case?

Liked the joins-your-calls-and-executes angle — gave it a quick test with Lastest, run here: https://app.lastest.cloud/r/mtav...

Honestly came in skeptical about the whole "AI in meetings" thing — expected another gimmick. Then I tried it in an

interview, had it compile structured notes from the candidate's responses, and it was way more accurate than me

trying to type and listen at the same time. Now I run it on every interview and client call. Wouldn't go back.

Nice. Is there a memory or bank to be able to track what Mina knows?

Really interesting take on AI meeting assistants. I like that Mina goes beyond note-taking and actually helps drive the conversation with role-based support. How does Mina learn and adapt to a team's workflow over time without requiring a lot of manual setup?

Really cool. Congratulations @sridharmuppidi

Is it possible to add a tool during the meeting or we have to relaunch Mina if we have to ask her to connect to a new tool?

Liked the product.. Was looking for something more than a note-taker.

Curious how Mina handles context switching mid call like if the conversation shifts topics suddenly does it follow along or does it need manual prompting to catch up?

Hey Product Hunt — Samuel here, one of the makers behind SocialEcho 👋

Most social tools help you schedule posts. But the real pain we kept hearing from growth teams was bigger: keeping content, engagement, and reporting consistent across many brands, accounts, and platforms.

So we built SocialEcho — an AI social media copilot for teams running sophisticated campaigns across multiple brands, accounts, and channels.

With SocialEcho, teams can:

→ See what’s trending and create on-brand content  

→ Adapt one post for every platform  

→ Publish, measure, and improve from one workspace  

→ Manage comments, messages, and mentions in one inbox  

→ Give AI agents secure access to social workflows through official APIs

The part we care about most: SocialEcho is built on official social APIs — no browser bots, no cookie injection, and no risky workarounds.

We’re building this for growth teams, agencies, brand marketers, and AI agent builders who need social operations to be safer, more consistent, and more automated.

🎁 Product Hunt launch offer: get up to $1,888 in bonus credits during launch week.

Question for you:

• Which part of social media ops would you trust an AI agent with first — content, publishing, engagement, or analytics?

Roasts and feature requests are welcome. We’ll be here all day.

Is this you? Copying the same content to post across socials like X, Threads, and LinkedIn manually across three tabs?

Coz yeah, that's me. Still! 🤓

Which is why@eexlkuang_se built @SocialEcho 2.0: so your AI agent can do that, plus all the boring plumbing that maintaining a social presence requires these days: publish, monitor, analyze, and route social workflows across Facebook, Instagram, X, TikTok, YouTube, LinkedIn, Reddit, and more.

SocialEcho actually adapts your content to fit the audience expectations of each platform, so you can start with the message, and SocialEcho will tune the posted format to not stand out like a slop-stained sore thumb.

SocialEcho didn’t start as just “another social scheduler.” Instead, it grew out of tens of thousands of cross-border sellers and agencies managing many brands, accounts, and languages at once. So platform-specific adaptation is at its core.

I appreciate they’re using official social APIs rather than sketchy browser-bot/cookie hacks (which are increasingly brittle thanks to the kind of Cloudflare defenses Product Hunt and others are adopting). If you’re working with agents that need to handle social media, this is critical.

This is interesting. What agent workflows are live today? Can they handle publishing, analytics, inbox triage, or all three?

Really interesting positioning. Most social media tools still feel like scheduling dashboards with some AI writing features added on top, but SocialEcho seems to go deeper into the actual workflow: trend discovery, on-brand content creation, platform-specific adaptation, publishing, engagement, and analytics in one place.

What stands out to me is the official API angle. As more AI agents start handling real operational tasks, reliable and secure access to social accounts will matter much more than brittle scraping or manual workarounds.

Curious how you think about brand consistency across multiple accounts and teams — especially when different platforms require very different tones. Congrats on the launch!

🎉 Love the positioning. The combination of content, inbox, analytics, and APIs makes this feel much more complete.

Congrats on the launch!
this seems like a strong fit for scaling agencies. do you also support white label setups for client facing dashboards?

What does the integration layer look like beyond the social platforms themselves? Thinking about connections to tools teams already have CRMs like HubSpot or Salesforce for audience context, DAMs for creative assets, or project management tools like Notion and Asana for campaign briefs. How plugged-in is SocialEcho to the broader stack?

This is a strong launch. I like that you’re solving the full social workflow instead of just one small piece.

This is super useful for us — we’ve actually been looking for exactly this kind of distribution layer for managing multiple brands and channels.
Does this connect directly to our own social media accounts via official APIs, or do you provide managed accounts inside the platform? 🤔

Managing multiple brands on social is a real nightmare, btw does SocialEcho let one set different tone or voice guidelines per brand, or is it one config across all accounts?

This looks useful for team-based workflows. One question: how granular are the permission scopes? Can access be limited by account, platform, or specific workflow?

我是市场营销人员 非常喜欢这个软件 尤其喜欢社媒发布这个 很实用

Curious how SocialEcho identifies trends — by keywords, competitors, public posts, engagement patterns, or all of the above?

Congrats on the launch!Amazing Product！！！Great Team~

The per-platform adaptation is the smart bet — same message, different native format per channel is exactly where most creators lose hours. I run a finance YouTube channel (Mod3Loop) and the cross-posting tax to Shorts, LinkedIn, and X is real; the win isn't just scheduling, it's not having content land like an obvious copy-paste. Does SocialEcho let you set a per-platform voice/tone profile, or does it infer the adaptation automatically from the source post?

I’m glad you emphasize original, on-brand content. Generic AI captions are everywhere, but brand fit is still hard.

Curious about the brand-voice layer specifically. The failure mode I keep watching in multi-account tools is that the generated post passes the platform-style check (LinkedIn cadence, X length) but the brand voice drifts inside two weeks because nobody is regression-testing the tone against last quarter's published posts. Do you keep a brand-voice eval set per workspace, or is it captured once at onboarding and frozen?

This is interesting because the hard part with social usually isn’t just scheduling a post. It’s the messy middle between ideas, approvals, publishing, replies, reporting, and everybody trying to stay on the same page.

I’m curious what workflow pain actually made you build SocialEcho. Was there a tool your team kept using that felt close but not quite right, like Hootsuite, Buffer, Later, Publer, Typefully, spreadsheets, native analytics, or some ugly mix of everything?

For teams already using those tools, would you say SocialEcho is ready to replace that stack, or is it more of a smarter layer that helps organize the chaos first?

Social media is honestly one of those things that looks simple until you're actually doing it for a business. Keeping up with posting while running everything else is exhausting. I like the copilot angle here — it feels less like automation and more like having someone alongside you. My question is around reactive content though — can it handle real-time moments or is it mainly for planned posts?

dapting one piece of content for 4 different platforms is a massive time sink. Quick question on the platform adaptation does it just tweak the length, or does it actually change the tone based on the platform (e.g., professional for LinkedIn vs casual for X)?

The official API angle is the bit I’d lead with. For teams running multiple brands, avoiding browser bots and cookie-based workarounds matters as much as scheduling. Does SocialEcho support per-client approval flows before AI replies go live?

also, to add to my previous longer comment - I personally really don't like that the site asks for notif permissions as soon as I access it. I suspect the vast majority of folks decline this (educate me if I'm misinformed, please! :)) but this plus the banner and hero image touting ~$2k in top-ups throws me off.

Again, congrats on the launch, tech looks cool, just some thoughts for a possible variant. :)

@eexlkuang_se this looks cool & powerful, but - and I mean this with the utmost respect - there is way too much going on with your homepage... -- hey, if it works, it works - but, for me, I'm not sure where to position my eyes, where to go, and I'm liable to just experience overwhelm and dip.

The @chrismessina hunt semi-cosign encourages me to look deeper and maybe use this, but, my strong suggestion to you: consider making a stripped down minimalistic beautiful version of your homepage, and then A/B test that version against this one. I think you'll find that a certain cohort of more sophisticated would-be users will convert at a higher clip.

Nonetheless, huge congratulations on a successful launch, and - again - what appears to be a very powerful solution. My feedback comes from a desire to see you succeed. :)

PS - do you support Farcaster, by chance?

Managing multiple brands from one workspace is the use case that always gets messy fast. Curious how the AI handles brand voice consistency when the accounts have very different tones - does it learn per account or do you configure it manually?

The platform style adaptation feature is something I manually do every single day and genuinely dread. A post that works on LinkedIn reads completely wrong on Instagram, and vice versa. If the AI is doing that translation intelligently not just resizing text this alone would justify a subscription for my team. Looking forward to testing it.

Btw Congrats on the launch!

how do you handle permission management when multiple team members and clients are involved? is role based access built into every workflow?

How granular are the analytics insights can users track performance per content variation? or is it more aggregated at post level?

How does SocialEcho approach the balance between AI-generated content and human editorial control? Specifically can teams set guardrails so the AI only ever drafts and never publishes autonomously, with a required human review step? Knowing where the human stays in the loop is important for brand-sensitive accounts.

Curious about the trend discovery experience in practice. When SocialEcho surfaces a trending topic, does it also suggest specific content angles or hooks that align with each brand's voice or does it just flag the trend and leave the creative interpretation to the team? The former would dramatically accelerate ideation speed.

Hi Product Hunt! 👋

I'm Pete from the Databox team, and today we're excited to share something we've been building for a while: Databox MCP.

Every team we talk to uses AI for writing, planning, and thinking through problems. When it comes to performance data, teams are still piecing it together by hand. Someone asks "why did my ad cost spike last week?" and answering takes 20 minutes of combing through multiple dashboards, adjusting date ranges and filters.

Some teams have shortcut this by uploading a CSV to Claude. The answer sounds confident, but it’s built on context that the AI doesn’t have. No metric definitions. No historical trends. No understanding of how their business measures success. The answers are hard to trust, and even harder to act on.

Databox MCP closes that gap. 

Databox connects to all of your tools, then it feeds the AI tools with data, analysis and insights. You ask questions in plain language, and the answers come grounded in your real business data: your metric definitions, your historical context, and the way your team measures success.

Here are a few things you can do with it: 

• Get fast answers without leaving your AI tools: Ask "why did ad cost spike last week?" and your AI pulls the answer from your trusted data, and gives you a written explanation with visual context. 

• Point your AI at any of your dashboards:   Say "analyze my Google Ads dashboard" or "summarize my client reporting dashboard," and your AI knows which metrics to pull. You skip the setup work that usually goes into every AI prompt.

• Push new data into Databox from your AI: Upload a CSV or pull from an API in your AI conversation, and your AI sends it to Databox as a clean, structured dataset. Analyze it the same minute alongside the metrics you already track.

• Rely on Databox for mathematical analysis: Whether it's simple things like understanding wether an increase in a number is good or bad, or more complicated things like calculating correlations or detecting anomalies, Databox is doing the math the same every time.

• Turn recurring work into workflows: Connect MCP to n8n or Make, and your recurring AI analysis runs on its own. Schedule the Monday performance summary, trigger alerts when key metrics change, and send executive summaries that arrive with the context built in.

We soft-launched it in February, and the most interesting thing has been watching what customers do with it. Rick Kranz used the Databox MCP with Claude to turn traffic, search, and CRM data into weekly content creation recommendations. He even made the skill available for others to download. Agency operations leaders like Gary Magnone started using it to spot the root cause of KPI spikes in minutes instead of hours. High volume digital advertising agency owners like (like Kamil Rextin) used it to build paid media benchmarks from client data. Island, a software development firm used it to automate data analysis for 25 leading online publications, cutting reporting time by 96%!

It takes 60 seconds to connect and is available on all paid Databox plans. 

We'd love your input 👇

What's the one performance question your team asks every week - but still takes too long to answer?

Thanks for checking it out 🙏

We built Databox MCP because of a pattern we kept seeing: teams were doing their thinking in Claude and ChatGPT, but their actual performance data lived elsewhere. So they'd export it, paste it in, and hope the AI understood it. It didn't. The data was already in Databox, connected, defined, with all the historical context. It just wasn't reachable from the tools where people were actually working. MCP closes that gap. One connection, and your AI can talk about your real numbers instead of guessing.

This is the part that matters more than people realize. An AI is only as good as the data layer underneath it. Databox isn't a pile of raw exports; it's a governed semantic layer: metrics defined once and consistently, data cleaned and modeled across all your sources, with the historical context that tells you whether a number is actually good or bad. That's the difference between an answer you can act on and a confident guess you have to double-check.

Asking questions and getting trusted answers is the obvious first use. What I'm most excited about is what comes next: workflows that act on the data on their own. Performance management, monitoring, and decisions that trigger automatically. Your AI stops being something you ask and starts being something that keeps the business moving week to week.

Proud of the team for shipping it.

Sounds very interesting.

I actually do upload a google sheet of my company stats which includes revenue and marketing data. I have a Claude Project that analyzes the google sheet and then creates a dashboard. This solution is very interesting and more dynamic.

I tested Databox MCP against some of the scenarios I use most often in client work - cross-channel performance comparisons, weekly trend checks, flagging anomalies in paid acquisition. In every case, connecting through MCP and asking conversationally was faster than navigating dashboards manually. The answers referenced real metric data, not approximations. For anyone who spends time preparing performance summaries, the productivity difference is immediately obvious.

The speed of updates from DataBox team is inspiring. I see something fresh is shipped every month on PH from DataBox. Congrats Ziga and team!

The scenario I see most often is a team that has good data in Databox but spends too much time retrieving and formatting it for reporting. Databox MCP shifts that entirely. Instead of opening dashboards and exporting data, you ask a question and get an answer - in the AI tool you are already using, backed by the same data your reports use. The time savings are real, but the bigger change is that analysis becomes something anyone can do, not just the person who knows where everything lives.

What makes Databox MCP technically solid is the design of the tool layer. You get a full lifecycle interface: load_metric_data for querying with date ranges and dimension breakdowns, ask_genie for natural language analysis, ingest_data for pushing records in, and get_current_datetime to resolve relative expressions like 'last week' accurately. Each tool does one thing cleanly. The result is an AI agent that can answer performance questions with the same reliability as a well-built dashboard query - and can do it in a conversation.

What’s the biggest productivity gain teams usually get after connecting Databox MCP? Faster reporting, better decisions, fewer manual data checks?

I spent more than 100,000 dollars trying to build my own data warehouse before I gave up and used the Databox MCP instead.

The problem I was solving is the one nobody likes to talk about with AI and data: an LLM will give you a confident answer whether or not the data supports it. When you manage ad spend across dozens of markets for a client, a confident wrong answer is expensive.

So I built Arcanian OS on top of the Databox MCP. It runs in Claude Code, connects to live data, and every claim it makes carries a confidence score. Data straight from a CRM pipeline scores high. A number inferred across two loosely connected sources scores low and gets flagged for a human. When a question contains a contradiction, the system does something most AI tools never do. It refuses to answer and asks me to rephrase.

It runs an internal debate among agents before it reaches a conclusion, creates tasks when it spots a risk or an opportunity, then checks back later to see whether finishing the task actually moved the metric. The learnings get anonymized and reused across every client in the system.

None of this works without a data layer that pulls accurately and defines each metric the same way every time. That layer is the Databox MCP. I open-sourced the whole operating system on GitHub so other agencies can run it too.

Been building on the Databox MCP for months alongside the HubSpot MCP. The combination unlocks a revenue intelligence layer most HubSpot agencies haven't explored yet. Excited to see this go public today.

The semantic layer design is what separates this from copy-paste workflows. You're connecting to metrics with definitions and historical context baked in, so the AI knows if a number is actually good. Does it handle custom fiscal calendars or non-standard reporting periods?

Nice, I actually try to connect all of my apps to Claude because that's a default app that I always keep ON. Good to see Databox got an MCP.

every marketer I know already pastes their numbers into chatgpt and asks 'what happened last week.' the fact that you're just connecting the data directly so the AI actually has real context instead of whatever we copy-paste is one of those obvious ideas that should've existed sooner

The real challenge in analytics MCP isn't data retrieval, it's grounding the LLM in correct metric definitions. We've run into this on customer data pipelines: 'churn' means different things across systems. How does the MCP layer handle semantic disambiguation? When a user asks about revenue or pipeline, does the context layer resolve conflicting metric definitions or surface the ambiguity to the user?

Great product! Is the automation itself by n8n?

One more thing worth sharing. The most fun part of the soft launch has been watching what partners built on top of the Databox MCP, then gave away for free. A few you can grab or look at right now:

László Fazakas open-sourced Arcanian OS, a system for managing complex, multi-market client campaigns that uses the Databox MCP to run automated in-depth analysis and inform daily decisions. The whole thing is on GitHub: https://github.com/arcanianHQ/arcanian-os

Max Traylor built Mantis, an AI-powered agency account management system that automates reporting, protects retention, and surfaces upsells. Here's how it works: https://claude.ai/public/artifacts/7b5265c8-2b0c-47d9-ae69-bbdbb86ab113

Jovan Miljevic built an n8n workflow that monitors SEO cannibalization across Databox, Google Search Console, and Slack, running on its own on a schedule. It's published here: https://n8n.io/workflows/15691-ai-powered-seo-cannibalization-monitor-databox-google-search-console-and-slack/

Rick Kranz gave away three Claude skills that use the MCP to analyze different parts of a sales and marketing funnel automatically. Free to download: https://www.linkedin.com/feed/update/urn:li:activity:7455235813786861568/

Keith Gutierrez built a pipeline that flags an underperforming page, audits it, writes the fixes, and updates the CMS, with no one touching a spreadsheet. One page it touched is up 62% in sessions.

A couple of how-to walkthroughs from partners, if you'd rather see it applied to a specific job:

Gary Magnone, on finding the root cause of a KPI spike in minutes instead of hours: https://databox.com/how-to/identify-the-root-cause-of-kpi-spikes-faster-with-ai-powered-analysis

Kamil Rextin, on building paid media benchmarks from client data: https://databox.com/how-to/create-paid-media-benchmarks

Different problems, same foundation underneath. If you build something with it, I'd love to see it.

I've tried way too many analytics tools that looked like they required a data degree just to set up a dashboard. The fact that Databox MCP actually gives answers without spending a week configuring things is a big time saver. Congrats on the launch!

I've spent so much time jumping between dashboards trying to make sense of numbers that just sit there. The idea of just asking your data a question in plain English and getting a real answer — that's the part that gets me. Especially useful for people running lean teams where not everyone is a data analyst. How are you keeping the context fresh when the underlying data changes?

The setup experience is worth calling out. Connecting Databox MCP to Claude or n8n takes under a minute -> paste the server URL, authenticate with OAuth, and your metrics are immediately accessible. No infrastructure to configure or pipelines to build. That low barrier is important because the hardest part of most analytics integrations is getting started. Removing that friction means we can go from zero to asking real performance questions in a single session.

I'm loving the Databox MCP and honestly, the one thing I didn't anticipate was how useful this could be for upskilling junior team members in the marketing agencies I work with.

I sit at the intersection of ops, strategy, delivery, and client services for agency teams, and the hardest part to scale has always been the month-end report analysis. We've always needed to pair every account with a dedicated senior strategist to look at the work, the numbers, the objectives, and then tell a strong, client-facing story about what's going on and what to do about it. It takes YEARS to build that kind of instinct, and it's not practical to assume your more junior folks can step in and handle it.

With the Databox MCP, you've just fast-forwarded years of experience. An AM or coordinator can ask why a number moved, get a real answer pulled from the actual metrics, provide context around the program and goals, collaboratively hypothesize around what's happening, then show up to the client call with a proactive point of view instead of a dashboard and a promise to "have the team look into it".

Game changer. Stoked for this new evolution of the Databox platform!

The MCP angle makes sense for analytics because the useful part is not just querying charts, it is keeping answers tied to the same metric definitions the team already trusts. I’d be curious how you handle permissions when Claude or Cursor asks for data across multiple teams.

Hi Product Hunt community, we're back!

When we launched Dune in April, the response floored us. You made it Product of the Day, then Product of the Week. Batch 1 sold out, we opened Batch 2, and the most common question we keep hearing is some version of "can Dune do X?" — the answer is almost always yes, so we built the tools to make that happen faster.

What we shipped
1. Dune Marketplace
A community-built library of scripts and agentic workflows. Browse and install what other users have built, assign to your keys in seconds, and add your own to the mix. From summarizing your last meeting to drafting outreach from your inbox, you can install what someone else built in seconds or upload your own. New scripts go up every week.

2. Build with Claude
You can now configure your entire Dune through a conversation. Tell Claude what you do all day and it'll suggest a setup built around how you actually work. Ask Claude to change a script, add a shortcut, walk you through what's currently set up on your keys, or suggest a smarter way to use them. No settings panel. Just talk.

A few setups already on the Marketplace, in case you want somewhere to start:
a) Start your Day! — Open the apps you use for work with one click.
b) Check Limits Individual/Team Plans — Launch the Usage tab on your Claude Desktop app with a single keystroke.
c) Company Search — This script goes through your calendar, finds the current or upcoming event, identifies the attendee's company name, and searches for the company on Crunchbase, LinkedIn, X, and their website, along with the attendee's LinkedIn profile.

These two features make Dune even more powerful.

Who this is for
If you're in back-to-back meetings on Zoom, Teams, or Google Meet, there are scripts for that too. If you're a developer running agents in Claude, VS Code, or GitHub, the Marketplace has workflows built specifically for those contexts. And if you have a workflow no one's built yet, Build with Claude will help you create it.

We'd especially love to hear what you build, or what you wish existed in the Marketplace.
Drop a comment — what's the one workflow on your Mac that still has too many steps?

---

🔗 Links
Site: https://www.projectmirage.ai/
Setup Guide: https://www.notion.so/There-s-Mo...
FAQs: https://www.projectmirage.ai/#du...
Demo: https://youtu.be/4Hn_ece7NVc?si=uO47h14--MRvI_OY

Love the product! I wish you guys would ship overseas as well. Quick question: do the custom/marketplace scripts live on memory or your macbook?

Put Start your DAY! on the Marketplace a while back. One key opens every app you use for work. Strange feeling to see a script you wrote show up in someone else's setup.

Good to have the software catch up with what the hardware was always capable of.

Congratulations

The bronze key is still the first thing people notice when I have it on my desk. Now when they ask what it does I have a much better answer than before.

Can I assign different scripts to the same key based on time of day?

The Marketplace changes this from a hardware purchase to something that keeps getting more useful over time. Curious whether scripts can be private or if everything you submit has to be public.

The marketplace and build with claude features are gonna make Dune hyper-personalised to each one of its users! Can't wait to see what the community builds 🖍️

built a “fire the developer” skill on marketplace this week that dms your founder on slack the second you open a github pr and complain about whoever wrote it.

took like 5 mins to build. the fact that this is now a one-tap thing is genuinely gonna keep me up at night.

How does the Marketplace handle scripts that stop working after a macOS update? Is there a way for authors to push fixes or does it fall on the user to notice something broke?

Are most of the Marketplace scripts plug and play or do most need setup?

Is there a way to try a script from the Marketplace before fully committing to assigning it to a key? Like a preview or a test run before it goes live.

The Company Search script goes through your next calendar event, finds who you are meeting, and pulls their LinkedIn, Crunchbase, and X before the call starts. I have been doing that manually for years.

Been using Dune since batch 1 and the muscle memory is real. If the keys are not there I notice immediately. The Marketplace is the thing that was missing. Finally have a reason to go back and rethink the setup.😄

Watched someone try Build with Claude for the first time. They described what they do all day, Claude configured the keys, and they looked at the result and said that is exactly right. That is a hard reaction to engineer.

The Marketplace catalog is extremely powerful if it stays fresh. What's the submission review process like and how fast do new scripts actually go live?

Does the Marketplace show install counts for scripts? Would be useful to know which ones are actually being used before you commit to a setup.

Build with Claude is interesting. Is it actually writing the script from scratch or is it picking from a set of templates and filling in the gaps?

reminds me of the figma keyboard, very cool

Curious whether there is a way to browse the Marketplace by app rather than by script name. Looking for everything built for VS Code specifically, and scrolling through the whole catalog is not ideal.

Does Build with Claude work if you describe a workflow in detail but you are not technical enough to know what the underlying script should look like? Or do you need to understand what you are asking for at some level?

If I build something good and share it, do I get attribution on the Marketplace?

Congrats on the launch! The Claude setup flow is a really nice direction.

Curious how you think about permissions/security for community scripts in the Marketplace, especially as people start building more agentic workflows across dev tools and meetings.

Dune Marketplace is such a cool idea! Congrats @apoorv_shankar and the team!

Polling the accessibility API for foreground app state is the right call. That's much cleaner than intercepting at the keyboard driver level. We've done similar context-detection work in our own tooling, where the active app determines which automations fire. How does the Claude shortcut generation pipeline work? Is it outputting shell commands or AppleScript, and how do you sandbox community scripts?

Hello and congratulation!! Im sad because i cant order it to the Czech Republic...
Really wanted to order and cant so im writing you here Dune team. :/ today with FedEx, UPS, DHL i dont think its hard to deliver it to any country. <3

The foreground-app context detection adapts keys in real time based on the active app, which is genuinely useful in a dev setup. We've felt the constant tax of switching between VS Code, Claude, and meeting tools while building AI features. Does the shortcut creation via Claude generate scripts locally or does it call the API each time you define a new workflow?

Yoo Product Hunt 👋

I'm Arlan. I dropped out of high school, did research at Stanford at 16, built multiple apps, went through Y Combinator, and have raised over $6M.

I spent the last year going deep on context and memory. I tried everything out there, and two things kept bothering me. First, none of them actually grow with you. They're smart in the moment, but tomorrow they don't remember yesterday. You're always re-explaining yourself, always starting over.

Second, none of them are actually proactive. You're always the one going to them, they never come to you. A lot of tools claim to be proactive, but really it's just a cron job firing on a schedule, and a timer isn't proactivity. Real proactivity is showing up at the right moment on its own, texting you exactly when something matters, without you setting anything up.

Folk is different. It lives inside the messaging apps you already use and it's built from the ground up to integrate into your life, not just assist when you ask.

Here's what that looks like in practice:

• It knows where you are. Folk tracks your location and acts on it. Walk past a grocery store with items on your list? It'll ping you. It's not a map, it's context.

• It joins your meetings. Folk's notetaker drops into every Google Meet, transcribes everything, and builds on what it learns so it always knows your world.

• It remembers everything and gets smarter over time. Folk builds a growing memory of your life. The longer you use it, the more it knows you. That's the whole point.

• It's multiplayer. This is the part I'm most excited about. Add friends who are already on Folk, and your folk and theirs can do things for each other, like book a table, share a doc, or check a calendar. One way, and only ever with a tap to approve.

And it does real things like research, coding help, date planning, booking restaurants, and tracking flights, all from the chat thread you're already in.

We're giving Product Hunt users 15% off today with code FOLK.

Two questions I'd genuinely love your take on:

• What part of your life do you most wish an AI could just handle for you?

• What would make you actually trust an AI with something personal?

Thanks for being here 🙏 Arlan

the 'gets smarter with every conversation' claim is the one I'd want to pressure test. most AI assistants that claim this either mean they store a growing context file or they fine-tune on your data, and those are very different things with very different privacy and quality implications. what's actually happening under the hood when Folk learns from you and where does that data live

I'm pretty into the iMessage -> Assistant workflow and am currently using @Poke.com pretty heavily. How does Folk compare and differentiate from other assistants like Poke?

Well, let's start by saying that I'm probably in the top 10 of Folk users by api consumption xp (they don't know about my second account), the amount of shit you can get done with it is out in the stratosphere, Nozomio created the best context solution with Nia (still has, Nia keep improving everyday), now it has a system to put it in, Arlan continue to provide so value to every users out there, I can hope in my normal meeting, record the whole thing with folk and launch nia oracle on the job, saved my ass and clients multiple time,

(sent by folk)

.. (tell this mf to enjoy time in Shanghai instead of spending it's in Cursor)

yoooo! congrats on the launch @arlanrakh
do you support custom integrations / MCPs?

I love how the product has a built in meeting-notes taker feature. Actually helps me so much because I dont have to use something external to keep my meeting notes and everything is in my folk which means its in my memories and gets connected to every other memory!!

I wonder who implemented it

Best hosted agent you can get for $20/month frfr

Amazing, can I install custom plugins/skills in folk? How does the integrations work?

I was looking to connect my card with it for lunch orders

Can’t wait to try it out myself

@arlanrakhdoes it have an api so that my server can inform about errors or cpu spikes and folk will

Message me about it?

paid user here and i really love the live geolocation feature. It found me right when I needed a nudge and the context awareness is genuinely impressive. Gets smarter every week. Congrats on the launch! 🚀

Never knew how much I actually needed meeting noted before this haha, and being able to query them on the go from iMessage is super useful 😉😉

The memory part is what I want to understand better, how does it actually build context over time? is it pulling from message history or does it need explicit inputs from us?

the multiplayer thing where my ai and my friend's ai coordinate dinner plans without us going back and forth is actually new. haven't seen anyone else do that. the location pings are useful too if they're actually accurate and not just spamming you every time you're near a store

Loved that the demo is about planing a date instead of yet another work task — there's actual soul behind this one. Memory that doesn't really grow and "proactivity" that's just a timer are exactly the two thiings that have kept me from trusting any of these tools. Genuinely curious how folk solves them under the hood — is it a real architectural difference, or the same promise everyone's making?

The "lives where you already are" angle is the right one — assistants die when they require a new app and a new habit. The thing I'd want is for it to quietly handle recurring low-stakes decisions, not just one-off tasks. That's actually why I built DishRoll (dishroll.netlify.app) for the weekly meal-planning version of this — the value isn't a single answer, it's removing the same decision you re-make 52 times a year. Does folk learn recurring patterns and start pre-empting them, or is it reactive to each message?

this ai guy actually gets stuff done, this is the future right behind such products. love the gps feature, congrats on the launch!

Really enjoying the uptick in services that live in iMessage. One thing I've noticed with all of them - folk included, I'm afraid - is that latency varies wildly... like, sometimes it'll start responding as soon as I send a message, other times - like right now, as I stare at my phone - I sent a message minutes ago and it's not even typing, but it DOES show "read".

Is this endemic to iMessage's infra? what's up with that?

also, @arlanrakh - what's folk powered by? you down to give us a peek under the hood?

congrats on a highly successful launch, thx for the hunt, @garrytan !!

yeah my dude, this is sick! go @arlanrakh!

hahaha! the video made me laugh but I'm also impressed by the potential of Folk. Wish you all the best Arlan

How do you handle shared memory in group chats, like what’s private to me and what the whole group can see?

The "in your texts" framing is interesting because most AI messaging tools still make you context-switch to a separate interface.

What actually triggers the AI here, a keyword, a slash command, something else?

Hey Product Hunt 👋

I’m Sam, co-founder of Typeahead.

We built it because most AI writing tools still take you away from the place writing actually happens. You open a chatbot, paste text into a box, get something back, then spend time trying to make it sound like you.

That never felt like the right interface for writing.

Typeahead works inline in every text field on your Mac. As you type, it instantly suggests words and sentences right at the cursor, so you can move faster and think less about the tool. It feels more like your computer helping you write than using a separate AI product.

A big part of the product for us was making this local. Typeahead runs on a local AI model, works offline and keeps your writing on your device. We also wanted it to feel like software you own, which is why it's a one-time purchase rather than another subscription.

A few things that make it different:

• inline suggestions, right where you already type

• works across your Mac instead of inside one app

• built to write with you, not for you

• runs locally, your writing stays on your device

• one-time purchase, free updates for life

Would love to hear what you think, especially from people who have tried chat-style AI for writing and found it broke their flow.

I'll be around all day replying to comments.

Interesting idea. But I would like to try it before purchasing it.

Really like the "writes with you, not for you" angle — that's the line I

keep coming back to when building my own AI writing tool. Most autocomplete

products try to take over and end up feeling pushy.

Two genuine questions :

1. How did you balance suggestion frequency vs friction? My biggest worry

with always-on autocomplete is that it interrupts flow when you actually

know what you want to type. Did you implement a "quiet mode" threshold or

is it always live?

2. The on-device model is a strong sell for privacy — what model size are

you running, and does performance scale with M-series Macs?

Bookmarked. Looks like one of the cleanest implementations I've seen in

this space.

The interesting edge case with system-wide autocomplete is how it handles context switching. does it know you're writing a Slack message versus a legal doc in Word and actually shift register accordingly, or is it one tone fits all? Also curious whether it reads the surrounding text in a field or just the last few words, because that gap is usually what makes suggestions feel off.

Really impressed by this. Congrats on the launch. I’m kind of amazed at how fast the suggestions show up. They feel instant, and more importantly they’re actually useful. A lot of AI writing tools are slow or get in the way, but this genuinely helps me type faster.

did you guys manage to make it work in terminal? say with claude code?

Looks clean! ✨ I’ve always been weirdly interested in typing shortcuts, dictation, and anything that helps people get thoughts out faster before the idea disappears.

I like when it’s not just about “write this for me,” it’s more about helping the sentence catch up to the thought.

The thing I’m curious about is how you keep people sounding like themselves over time. A lot of AI writing tools make things cleaner, but sometimes cleaner also means more generic. Are you thinking about personalization in a way that preserves someone’s actual voice and quirks, or is the goal to keep it more lightweight and context-based?

very cool! did the idea start with building something local or did it start with the use case?

Does the app offer an option to limit the maximum length of text autocomplete? Is it possible to set a specific word count in the settings for how much the AI generates at a time?

The fact it's running a local model is a huge upsell for me. Excited to see how the product evolves - It works great, easy af to install. I do lots of dictation these days, but I find a totally different thought pattern when typing - this is a cool addition to the stack.

An interesting app. I would like to see something like this for Windows.

M1 air 8gb. chrome and cursor already hurt. does the model sit in ram all day or only spin up when i'm typing?

Nice launch. What got me is that it runs locally and works in every text field instead of pulling me into a separate chat window just to rewrite a sentence. The writes with you not for you angle feels right. Congrats Sam, followed you on X too.

Hey Product Hunt 👋

We built Presentify to make presentations and video calls more engaging — whether you're teaching, demoing, pitching, or recording tutorials. It works with apps like Keynote, Google Slides, Microsoft PowerPoint, Zoom, Microsoft Teams, and Google Meet.

To celebrate the launch, here’s a little Product Hunt gift 🎁
Use code PRESENTIFY20 for 20% off when bought directly on the website.

And because launch days run on caffeine and comments…
Leave a comment below and I will send you an even better deal 👀

Congrats on the launch. I started using Presentify in April 2023. I was tutoring secondary school students in maths and physics online and I was looking for a cursor highlight. I got so much more and appreciate the continued development and added features over the last 3 years. Whether I tutor students online or face to face I use Presentify in every session for the majority of the sessions. The ability to annotate anywhere on the screen while looking at a pdf, website, etc really helps when working with students. I use the freehand tool to write when working through problems with students. Fantastic app.

Presentify is one of those apps that does one thing exceptionally well. I’ve been using it for my remote classes, and it has made a noticeable difference. Some of my students have even started using it themselves. Congrats on the relaunch, and highly recommended!

It's an interesting concept!

Screen annotation is underrated for teaching anything technical — I record an Excel/financial-modeling course on Udemy, and the moments that actually land are when you can spotlight one cell or trace a formula live instead of waving a cursor around. Cursor highlighting and zoom would've saved me a lot of re-recording. Does Presentify let you pre-set spotlight regions per slide, or is it all live/on-the-fly during the talk?

I run a YouTube channel where I teach web development and this tool is exactly what I needed!

I think I had previously shared some feedback on Reddit about a few features, and now I can see that a few of them have been implemented, which I absolutely love!

Congratulations on this update!

As someone who records educational content regularly, Presentify has been incredibly useful for my courses on Udemy, Tutorialspoint, Luvenica, and Insight Timer. Being able to highlight important areas on screen instantly improves the learning experience for students. Clean, lightweight, and practical. Wishing the team a successful launch!

I have been using Presentify for over three years. It is lightweight yet powerful, and it significantly enhances the visual quality of my presentations. This has been especially helpful in a remote work environment.

Presentify has become one of my go-to tools when I teach or give technical presentations from my Mac, especially in cybersecurity sessions and live demos. What I value most is that it lets me direct students' attention exactly where I want, without wasting time fighting with the interface.

Great tool for anyone who has a live audience in front of the screen.

Great to see you updating and re-laucnhing Presentify - I've been using it since 2020.
I use it on all my presentations and demos.
Thanks for all the hard work, time and effort 👍

Congratulations! I've been using presentify for multiple years, and my educational videos are more expressive thanks to it. Highly recommended!

I've been using the app since I've first heard about it, and the amount of work the devs have put on the updates is amazing. It has been getting features that I didn't even know I needed!

I make a lot of presentations and this definetly seems like a great way to level-up my presentations.

I started using this tool recently in my meetings with clients and immediately wondered why I haven't used it before. The ability to annotate on the screen to show my clients exactly what I'm referring to is super useful.

This looks promising in terms of ease f use balanced with functionality. The real proof will be in using it under deadline conditions i.e. how well does the UI layout help...

The cursor highlight and spotlight features alone have already saved me from so many "wait, where are you clicking?" moments during demos. Tiny footprint, zero friction, works with everything. One of those rare Mac apps that just does its job perfectly

I've been using Presentify for about a month.

Its so powerful to be able to show people know exactly where to look while you are sharing your screen. It helps the viewer understand the message much more effectively.

The app has been really well thought-through. Glad to have it on my Mac.

I've been using Presentify since 2023, and it has become an essential part of my daily workflow. The app is incredibly useful, easy to use, and the developer regularly releases new features and improvements. It's amazing to see how consistently it evolves, making presentations and screen sharing more engaging and effective. Highly recommended!

Looks really nice - was hoping to give it a try but the Homebrew installation fails!

Been trying to develop professionally and presentations are an area for improvement. This has been extremely helpful in my journey.

Does it have a spotlight with softer edges??

Been losing people on calls because they can't follow where I'm pointing on screen the spotlight feature is superb. Congrats on the launch!!

Glad to see it! I’ve been using it and have 2x’d the quality of my demos. I hear from prospects often, “that feature you spotlighted was exactly what we needed to know about” and “the annotations helped me out, I’m a visual person”. It’s become such a necessary part of my tech stack.

Love Presentify, use it all the time for demos. Congrats on the launch! 🚀

Congrats Ram on the launch!

I've been using it lately, love the app

Came across this on PH today, and it immediately makes sense. I do a lot of screen demos for clients, for content, and the cursor visibility thing is a real problem. You're pointing at something on screen, and the person watching has no idea where to look.

The zoom-in feature is what gets me. Most presentation tools make you plan your zoom points. If this works live during a Zoom call, the way the listing suggests, that's actually useful.

Congrats on the launch. Going to try this on my next client demo.

We launched /research today, a single API call that runs a full web research agent and returns cited answers.

Here's the pattern it replaces:

// What most teams build:

// 1. Search the web (or an index)

// 2. Fetch and read relevant pages

// 3. Synthesize across sources

// 4. Validate citations

// 5. Format for streaming

// 6. Debug all of the above forever

// What /research does:

const stream = await client.agent.research({ query: 'Your question here' })

That's it. The orchestration lives inside the call. Your codebase doesn't own it.

First 10,000 credits are free. Happy to answer questions in the comments. Give it a try → https://tabstack.ai/web-research

Three things that matter about how we built this:

1. Citations are not optional.

Every claim comes back with a source URL. Users can verify where the answer came from. That's the difference between an AI feature people trust and one they don't.

2. Live web, not an index.

Pre-indexed research APIs (Exa, Perplexity) are fast and good for broad questions. /research goes to the live web on every call. For legal, financial, and competitive intelligence work — where the answer needs to be current and sourced — that's the relevant difference.

3. SSE streaming built in.

Users watch it work in real time. No spinning wheel. Progress events as sources are found, then a complete event with the full report and all citations.

The citations part is what matters most to me here. For research workflows, a good looking answer is not enough if people cannot check where it came from. Live web also makes sense for topics that change quickly, especially competitive research or financial updates. How do you handle cases where sources disagree with each other? Do you show conflicting findings or choose the strongest source?

Liked the no-scraper web-extract angle — gave it a quick test with Lastest, run here: https://app.lastest.cloud/r/owYo...

happy to share this new collaboration with @tessak22 and the @Tabstack team!

First launched on @Product Hunt last April, the @Mozilla-backed platform is launching again today, introducing a new /research endpoint.

Run agents to explore the web and answer complex questions with precision, all in a single API call. Simple, accurate, and secure. Read the docs to learn more: https://docs.tabstack.ai/guides/research/

There are multiple products in the browser automation category - @Tabstack just hits differently.

Start for free: https://tabstack.ai/web-research

The “citations are not optional” stance is exactly right, especially for legal / financial / competitive-intel use cases.

One thing I’d be curious about: do you treat citations as source links only, or as claim-level support? In research workflows, a source URL is helpful, but the trust really jumps when the user can see which sentence or assertion each source is meant to support, plus whether another source disagreed or was skipped as stale.

That extra bit of provenance would make the output much easier to use in a report without forcing someone to re-run the whole research process manually.

The "cited answers from the live web, source URLs users can verify" point is exactly the bar for any signal you'd act on financially. I work in finance and also built PolyMind (polyminds.netlify.app), which pulls real-time signal from Polymarket trades — and the hardest part was never fetching data, it was making every alert traceable back to a verifiable source so a user can sanity-check it before trusting it. Curious how Tabstack handles conflicting sources on the same question — does /research surface the disagreement, or pick a winner?

How do you pick sources and avoid low quality SEO pages, and can users whitelist domains?

Hi everyone,
I'm Charlene, co-founder of Trippple Club.

I kept seeing the same unfair thing: on social media platforms, small businesses pay more and get less: not because their ads are worse, but simply because they don't spend enough. Big brands get better rates and better performance purely from the size of their budgets. A shop spending $500 a month is playing a completely different game than one spending $500k.

Trippple Club closes that gap. We pool the ad budgets of many small businesses so you're effectively buying Meta ads together, and get the economics that normally only big spenders see. The result is up to 3x lower cost per result (from our pilot customers) with no agency and no wrestling with Ads Manager.

How it works:
- You set your budget and goal in a few minutes
- It's pooled with other small businesses
- You run your own campaigns, but at big-brand rates

We're live today and I'd love your honest feedback, especially from other small business owners and marketers. What's been your biggest frustration with running Meta ads yourself? And does the collective model make sense to you, or does it raise questions?

Thanks for taking a look.

Amazing, let's go 🔥

Great idea, congrats on your launch!

As a solo founder getting a consumer iOS app ready to launch (relationship/wellness space) and planning paid Meta experiments around month 3 post-launch, two specific questions from that angle:

1. Does the pool support iOS App Promotion campaigns? Those have their own delivery mechanics and SKAdNetwork attribution compared to standard lead/traffic, so I'm wondering whether the pooling math holds up specifically for app installs, or if Trippple is mainly oriented toward web conversion campaigns today.

2. How does the non-competing-peer matching actually work? My ideal pool-mates would be wellness / journaling / meditation apps (overlapping audience, non-competing intent). My worst would be other relationship apps. Is the matching audience-vector based, category-based, or manually curated?

I can see form the video you are at Station F ? 🫡
Is there a minimum spend to join the club ?

very cool

I'd be interested in learning more about how the reporting is disaggregated (though I imagine there's some secret sauce involved).

Congrats @charlene_lin! Early-stage startups can now experiment better without feeling the pain.

Love the collective model—the budget penalty for small spenders is one of those things everyone feels but nobody fixes. Most of the examples here read B2C/local. Does the math hold up for B2B lead gen, where audiences are much narrower and CPLs higher? Pooling seems to need overlapping-but-non-competing audiences, which is harder to find in niche B2B. Would be great to hear if you've seen it work there. Nice launch!

Starts with Meta, but the collective model could extend to Google, TikTok, even programmatic.

Is Meta the wedge or the whole product?

Well done, Charlene and Julien! 🙌 The future of growth is collective - and you're proving it!!

this is actually smart. small budgets on meta get punished and everyone knows it but nobody talks about it. pooling spend to get big-brand delivery rates without hiring an agency is a good solve. signed up to check it out

What's the difference with OG Caffeine? https://www.caffeine-app.net/en/

nice one. caffeinate -d under the hood i assume? curious what made the gui worth shipping over the cli.

How is this different from the original Caffeine or Amphetamine? Both are open source.I think calling this "Open" Caffeine is misleading.

The battery threshold auto-stop is a genuinely thoughtful detail that most wake-lock utilities skip entirely. We've had long-running local processes fail silently because the Mac went to sleep mid-run, so this addresses a real pain point cleanly. How do you assert the wake lock under the hood? Are you using IOPMAssertionCreateWithName or a different IOKit approach?

This is the kind of tiny Mac utility I like, but the install path is the part I’d check first.

Is the GitHub release meant to be used directly, or is Open Caffeine still mainly a local build from Xcode for now?

Auto stop at a low battery level is a great feature, I’ve killed my battery doing long downloads many times.

Hi everyone!

Le Chat is now Vibe — one agent across long-running work, coding, web app, editor, and terminal.

Work Mode handles multi-step tasks across inbox, calendar, docs, spreadsheets, search, connectors, and scheduled workflows. It plans first, gets approval, then runs the work with visible progress.

Code Mode brings the coding agent into the web app, VS Code, CLI, and remote sandbox sessions. It can connect to GitHub, work on projects, inspect diffs, and carry a task through to a reviewable pull request.

The CLI also gets skills as slash commands, custom modes, subagents, editable plans, session-scoped permissions, and /teleport to move a live session between terminal and cloud.

@vnglst has an interesting post from AI Now Summit covering @Mistral AI’s future vision and product evolution.

Cocorico!

about the long-term open model strategy. the pattern with several AI labs has been open weights early to build developer adoption, then gradual tightening of licensing or capability access as the business matures. Mistral has been more consistent about this than most but the commercial pressure is real. is there an explicit commitment somewhere about what stays open as the company scales or is it more of a current philosophy that could change

work mode that connects to gmail slack and notion and actually does stuff across all of them is what I've been wanting. might try this for my weekly reporting workflow

Does Tokenwise break down coding agent spend by session or only by model?

Observe-only is probably where I’d start, especially for Claude Code spend. The scary part is the “apply” step.

Before swapping a model, does Tokenwise show exactly which traffic it will touch, and is there an easy rollback?

The "quality check on your own traffic, not public benchmarks" is the right frame, that's exactly the gap most LLM-cost tools wave at. Question for Théophile: when you replay a request on the cheaper model to verify quality, how do you score "same answer" without a human in the loop? Embedding similarity tends to be permissive and exact-match too strict.

What's the single most breathtaking event a SkyLive camera has captured so far? The one that made you think 'yes, this is exactly why we built this' :)

What inspired you to make this? I love the idea, not much natural night sky is visible in my area :/

This is a great idea. I got onto your site, but no matter what city I typed in, there does not seem to be anything that shows up. When will this go live, or am I doing something incorrectly?

congrats on the launch! this looks cool tbh but cant see any stream now. Also no past streams are visible to me? do we have to signup on the platform to have full access?

The moment I saw this idea, I thought about every single event I've missed over the years. The aurora, a meteor shower I planned a whole night around, only to get clouds. So this genuinely resonates with me.

What I love most is that you framed it as community-powered infrastructure, not just a livestream feed. As someone who builds with AI and gets way too excited about turning ideas into actual products, I keep imagining the layers you could add on top: smart alerts for clear-sky cameras, maybe an AI that flags the best viewing window in real time. The foundation you're laying here is the hard part, and you nailed the vision.

Following closely and rooting for you, Samuel. This is the kind of thing the internet should be used for.

Looking forward to when you have more cameras and streams. Would love to have this on to unwind.

How does this work? Should I buy camera, point it to the sky and plug into your server?
It sounds unusual, interesting, but... what is the system behind? Currently, only you 'occasionally' pointing camera to the sky. Even you are not committed. "catching unexpected celestial moments in real time' occasionally looking to the sky is a bit challenging, no?

A community-powered global sky-camera network is a great idea — the weather/location dependency is exactly why a single backyard telescope so often disappoints. The location-specific sky data angle actually overlaps with something I built: RoofSolar (roofsolars.netlify.app), which pulls live PVGIS irradiance by location to model solar output — same underlying problem of "what is the sky actually doing at this exact spot." Are you planning to expose any of the camera metadata (cloud cover, clarity by site) as data, or is it purely a viewing experience for now?

Hey Product Hunt 👋

We've lost hours of our lives to Spotlight returning nothing when we know the file exists. Naming a file untitled (4).docx or assignement_1(4).pdf guarantees you'll never see it again. Naming a PDF after a client doesn't help when you've forgotten the client's name.

So we built Stella. You describe the file the way you'd describe it to a coworker. "The pitch deck with the blue charts," "my tax return from last year," "that contract Sarah sent in February,"  and Stella finds it. One search bar across all your local folders.

Oh, and it's free.

A few things we care about:
- 100% local. Your files never leave your device. No cloud, no telemetry on file contents, no account required. The embedding model (291MB, quantized) runs on your machine.
- Fast. Filename matches return in ~5ms; semantic results land under 200ms.
- Honest about quality. We benchmarked against Spotlight on 433 real queries — Stella was ~5x more accurate (0.83 vs 0.16 MRR@10), and Spotlight didn't surface a single file Stella missed. We'd rather show numbers than adjectives.

What's next: more file formats, a Windows build, and cloud storage integration (yes that includes Outlook, because Outlook search is so unbelievably bad).

We'd love your feedback, especially the queries where it fails. That's where we learn the most.

– The Stella team

This would definitely be helpful! I have a photo that I cannot find right now and I name my files pretty well. So, it's a frustrating experience.

This looks cool, but I think the lack of any product shots (e.g. screenshots of the app itself running) = a huge missed opportunity. We've seen a slew of local filesearch using natural language tools on PH lately - I use Invenio for images which I found here, and I've seen at least 2 others that aren't filetype-specific in the last week.

I chose @Invenio because their site made it clear that the platform would be a delight to use, UX/UI wise.

Congrats on the launch either way, and again, this does look very cool.

Congratulations on your launch! Interesting. I need it.

Wow

Need it desperately!! I can't count how many hours I've spent finding files that I knew existed but could not remember the file location or name... (and love the name =)

We're live in manufacturing, logistics, and food service! Book a demo: avearobotics.com

Really cool launch. Being able to jump in and take over a robot the moment autonomy breaks is exactly the safety net these teams need, and that VR demo is wild. Congrats Ary, followed you on X to keep up with what you're building.

the idea of a human being able to jump in the second a robot gets stuck is smart. 100% uptime on autonomy isn't realistic yet so having that instant fallback makes sense

Сongrats on the launch 🚀
Fully outside my world (i build mac/web stuff 😅) but the framing really clicked – "100% uptime + instant human fallback" is exactly right for autonomy that's not-quite-there yet.

What's the wildest robot a customer is controlling through this right now? 🤖

what happens during a network interruption mid-teleoperation. the robot is mid-task, the operator loses connection, and the autonomous fallback may not be reliable enough to complete it safely. that's the failure mode that keeps robotics operators up at night and it's usually the first question any serious enterprise customer asks. how does Sentinel handle graceful degradation when the remote connection drops

Natural-language-to-dashboard speed is impressive, but as someone who builds project-finance and valuation models for a living, the thing I'd watch is auditability: when an IRR or a projection comes out, can a user trace it back to the exact assumptions and formula path, or is the logic a black box? That traceability is the whole reason I keep a library of transparent, formula-driven templates on Eloquens — fast output only earns trust in an IC meeting if every number is defensible. How does R0Y handle showing its work?

natural-language dashboards for investing is a clean idea. curious how far the data layer reaches. does it pull on-chain data too, like stablecoin lending rates or DeFi yields, or is it tradfi instruments for now? that side has almost no good no-code tooling, so it'd be a real wedge.

The community section with 100+ pre-generated dashboards is a smart call. It shows what the Omni model can produce and gives new users a starting point. When someone forks a community dashboard, does the underlying data logic and model structure come with it?

Offering a month of pro free if anybody would like to try full features. If you have ideas for transforms or other ways it could empower your workflow, let me know!

The two step confirmation workflow is smart . How quickly can Emily complete common actions like room bookings or invoice lookups?

the "never touches the money" Stripe Connect setup is the right call for an operator tool, keeps you clear of money-transmission licensing. one thing I'd dig into: how does the flat 2.5% pencil out when Stripe's own standard card rate is 2.9% + 30c? is the 2.5% your platform fee on top of processing, or have you got interchange-plus negotiated underneath?

Does Emily work well in noisy spaces like receptions and events, or do you need a headset to get reliable input?

👋 Hi Product Hunt!

I'm Joel, the creator of Joanium.

Joanium is an AI-powered software engineering assistant designed to help developers build, debug, and maintain applications more efficiently. It can assist with coding tasks, project understanding, and day-to-day development workflows.

I built Joanium because I wanted a tool that feels more like a reliable engineering teammate than just another chatbot.

We're launching today and would love your feedback:

• What features stand out most?

• What would make you switch from your current AI coding tools?

• What workflows should we support next?

Thanks for checking out Joanium! I'll be here all day to answer questions and discuss ideas. 🚀

The local-first approach is what caught my attention. Most AI tools talk about privacy, but very few actually keep the data on the machine. Curious how much of the workflow can run fully offline versus requiring API access to external models?

Is there a way to route different tasks to different models automatically, or do users pick one per chat?

Can NetworkSpy capture and visualize the full agent loop, like prompts, responses, and function calls, in one timeline?