PH热榜 | 2026-03-17

Manus just took a big step forward with My Computer, bringing its AI agent out of the cloud and directly onto your desktop.

Until now, Manus worked in a cloud sandbox. But most of our real work lives locally: files, dev environments, apps, and workflows. My Computer bridges that gap by letting Manus execute command line instructions on your computer to read, organize, edit files, and control local applications.

What makes it interesting is the automation potential. Manus can organize messy folders, rename hundreds of files, build apps through CLI tools like Python, Node.js or Swift, and even run tasks using your machine’s idle compute.

You can also assign tasks remotely, for example, asking Manus to find a file on your home computer and email it through Gmail while you're away.

Key highlights:

• Works directly with local files, tools and apps

• Executes terminal commands with your approval

• Automates repetitive file and workflow tasks

• Can build software projects via CLI tools

• Uses idle compute resources in the background

• Lets you trigger tasks remotely across devices

This seems especially useful for developers, builders, and anyone managing large local workflows who wants automation beyond browser-based AI tools.

It reminds me of what Perplexity is doing with Perplexity Computer, but focused on letting an AI agent directly interact with your own machine and workflows.

What use cases are you thinking with My Computer by @Manus?

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

This looks really interesting . How does Manus handle permissions for local tasks? Do we have fine grained control over what it can and can’t do on our machine, or is it an all-or-nothing approval for commands?

so it's a claude desktop clone?

Bringing AI agent capabilities directly to the local desktop is a game-changer. As someone who values local-first workflows, I’m curious about the performance side—does running Manus locally consume significant CPU while it's processing CLI tasks in the background?

Moving Manus from cloud-only sandboxing to direct local machine access via CLI execution is the natural next step — most real productivity workflows involve local files, dev environments, and desktop apps that a cloud-only agent simply can't touch, so bridging that gap unlocks an entirely different class of automation tasks. The remote task triggering is a compelling feature for power users, but how does Manus handle permission scoping on local execution — is there a granular approval system for different command types, or does the user approve each terminal command individually?

Giving this a shot! Looks a lot like Claude desktop, and I am OK with that, happy to see a familiar Ui

It reminds me a lot of Claude desktop, which I’m totally fine with. Glad to see a familiar UI.

Really interesting concept: a general AI agent that actually executes complex tasks instead of just responding feels like a big shift toward true automation. Congrats on the launch! What kinds of tasks does Manus handle most reliably right now without needing human intervention?

Great stuff! So far it runs smoother than Claude's Dispatcher

How easy is it to set up remote task triggering ? I'm curious about how well it works across devices, especially if I want to trigger tasks from a phone or another machine .

Manus is a general AI agent that transforms your thoughts into actions, handling tasks across work and life. It helps you get things done effortlessly while you focus on other priorities.

Hi Product Hunt! Ben here, founder of Lightning Rod.

We started Lightning Rod because training data is the blocker for most AI projects. Companies have a huge amount of valuable historical data and access to rich public sources, but turning it into something AI can actually learn from is too slow and expensive.

Today we’re launching our training data SDK, which lets you automatically generate LLM-ready training data from raw documents or public sources. We use real-world sources and outcomes over time as supervision — no labeling or annotation required ⚡

Here’s what you get:

• Go from idea to dataset, fast. Define your criteria and data source. We collect and label training data for you — ready in minutes, from just a few queries or examples.

• Use your own data or start from public data sources. Generate training data from internal documents like emails, tickets, and logs, or from integrated public data sources.

• Provenance in every row. Every record links back to its source, so you can audit what went into your model.

• Quality built in. Automated scoring and filtering remove low-confidence examples and outputs that do not follow your instructions.

• Turn historical data into training signal. We use real-world outcomes over time to convert your timestamped docs, tickets, logs, and news into grounded supervision automatically.

We’ve already used data generated with this platform to beat frontier models 100x larger, and to train domain expert models on everything from corporate risk to sports predictions.

Create your first dataset free at lightningrod.ai. Use code ProductHunt50 for $50 in free credits.

Thanks for checking us out — I’ll be here all day reading and replying. If there’s a dataset or model you’ve wanted to build, drop it in the comments and we’ll help you get started!

Congrats!! Any plans to a no-code interface for non-technical teams?

Congrats team! Question: How do you ensure the generated datasets are actually suitable for fine tuning, given the noise, bias, and duplication often present in public news sources? Do you apply any validation, deduplication, or labeling quality checks, and can users control how the data is structured or filtered for specific domains or tasks?

Congrats on the launch!
Very relevant problem - everyone talks about models, but high-quality training data is still the real bottleneck.
Love the emphasis on provenance and production-ready datasets. Strong positioning. Wishing you a great launch today 🙌

How does the quality scoring work... Is it model-based or rule-based filtering?

Any benchmarks you can share?

Using real-world outcomes over time as automatic supervision instead of requiring manual labeling is a fundamentally different approach to training data generation — it means the dataset quality improves with historical depth rather than human annotation effort, which should scale much better for domain-specific fine-tuning. The claim of beating frontier models 100x larger with data generated through this platform is compelling; for teams working with internal documents like support tickets or emails, how does Lightning Rod handle PII in the source material — is there automated redaction before training data generation, or does that fall on the user?

We're doing some ML work on our side for matching and recommendations so this is relevant. Can the SDK work with proprietary data like internal user behavior logs, or is it mainly designed around public sources for now?

Generate training data? What does it mean? Congrats on the launch, @bturtel!

What ways could I validate that the training data is actually improving downstream model performance?

Very interesting concept. Getting training data for my AI Project 8 years ago for my capstone was a huge bottleneck. Using data that already exists and vetted to some degree democratizes training and building. I'm excited to give this a test!

Turning real-world data into clean, production-ready training datasets without manual labeling could remove one of the biggest bottlenecks in AI development. How do you ensure the generated datasets stay high-quality and unbiased when pulling from noisy real-world sources? Congrats on the launch, by the way! 🥳🚀

Expanding from a photo editor into video and music generation while keeping everything browser-based with zero downloads creates a powerful share-first creative loop — the fact that someone can upload a friend's photo, animate it, add an AI soundtrack, and share it without ever leaving the browser removes every friction point that usually kills viral content creation. The music generator is an interesting addition to what started as a visual tool; how tightly integrated is the audio with the video output — can users sync generated music to specific moments in the animated clip, or are they independent outputs that get combined manually?

@owenlongbo looks fun! Can I import my IG stories with friends and start from those?

For ordinary people with no creative experience at all, what are the criteria you have internally set for determining a "successful work"? Is it perfect picture quality, or something like "fun and shareable"?

It sounds like a really interesting and fun idea for having a laugh with friends. The question is, do you label the video/text/photo in any way to indicate that it was created using AI? Because, while it’s a great tool, it can often be used improperly. Teenagers in particular can use this in a truly cruel way towards their peers.

How is it with the testing of the tool? Do I need to pay directly? Because I uploaded the image, it started generating something. I waited, and then the pop-up window appeared with the pricing list. In other words: Is there any testing option?

Made a cute OpenClaw logo, congrats on the launch.

Interesting and creative product

Congrats on the launch. Such a fun product. Being able to go from a static photo to an animated video with an AI soundtrack brings such a creative flow. Does Kira 4.0 support direct sharing to TikTok or Reels, or is it a download and post situation ?

A great AI photo editor! It's so much fun!

Turning creative tools into a simple conversation instead of complex prompts or layers feels like a big shift for making art more accessible. What helped you get the most consistent style results from just natural language inputs? 🎨🚀

For the video generator, is it better to upload multiple photos of the same person to improve the video quality?

Congrats on the launch!! I tried it out, and the navigation and ease of use are really impressive. I tested the image enhancement/generation and the results were great. Unfortunately, AI still has trouble with text. But overall, it's really good! I'll try it out more and write a more detailed review

This is a fun one. Turning friends into shareable content is a smart loop, and removing the skill barrier makes it the kind of thing someone opens, uses immediately, and sends without thinking. That's hard to engineer.

One messaging thought: showing what people actually send each other could do more work than describing what the tool does. The output is the hook, not the feature.

Curious whether any specific use cases are driving repeat usage yet. I spend a lot of time helping SaaS teams figure out what makes their early users come back, so I'm always interested in where that first habit forms.

Excited to see where this goes.

A fun way to mess with your friends :-), love the idea. Congrats on the launch!

How does this compare with existing option on the market if any?

Kira is the best photo pair maker I've ever used.

Huge congrats on the launch! The 'edit by talking' feature is a brilliant idea. Prompt engineering can be so frustrating, so making it conversational is super refreshing.

Quick question: How well does Kira maintain the style consistency if I make multiple voice edits to the same image?

Wishing you a massive launch day! (I'm also launching my AI bot today, so I totally feel the launch day energy! Rooting for you!

Hi everyone!

Codex just leveled up with Subagents — you can now spawn specialized parallel agents for complex tasks like PR review or multi-step features. Each subagent gets its own instructions, model, and tools, and Codex merges everything back cleanly.

Over the last week I used Codex to design, debug, and do embedded work for a new device prototype, and the speed honestly shocked me. This feature makes that whole experience feel even more serious. Now I can have one agent map, one review, and one check docs, and the main thread stays much cleaner instead of drowning in logs and side quests.

It really feels like OpenAI is going all in on the coding lane right now. This puts some real pressure on @Claude Code. And Goolgle, @Google Antigravity alone probably is not enough :)

This is very cool

Gonna try this out. Codex 5.4 is already awesome and this seems like a way to supercharge it.

So how do plugins like compound engineering work going forward?

The concept looks solid.

What was the hardest part of building it?

Spawning specialized parallel agents for complex coding tasks is the right evolution for Codex — splitting a large problem into concurrent subagents that each handle a focused piece mirrors how experienced engineering teams actually decompose work, and doing it in parallel rather than sequentially should dramatically cut time-to-completion on multi-file refactors and complex feature builds. How do subagents coordinate when their changes overlap — is there a central orchestrator that detects conflicting edits across parallel workers, or do they operate on isolated branches that get merged at the end?

Hey Product Hunt! 👋

Today we're launching mTarsier, the open source MCP manager we wish existed when we started building with MCP.

The problem we kept hitting: Every AI client — Claude Desktop, Cursor, Windsurf, VS Code — has its own config file. Adding, removing, or enabling an MCP server means hunting down JSON files, switching between apps, and praying nothing breaks. It's chaos, and it kills momentum.

The problem isn't new — the community has been screaming about it:

Garry Tan (YC President): "MCP sucks honestly. Toggling it on and off, the auth sucks." A dev on X: "why does every AI client handle MCP config differently?? I just want ONE place." Hacker News has a 145-point thread: "MCP is a fad."

But here's the other side:

OpenAI, Google DeepMind, AWS, Microsoft, and Cloudflare all backed MCP. Downloads went from 100K to 8 million in 6 months. Fortune 500s are deploying it. The Linux Foundation now governs it. Jensen Huang called it something that "completely revolutionized the AI landscape."

MCP isn't a fad. The protocol won. The tooling just hasn't caught up — yet. That's what mTarsier is here to fix.

What mTarsier does: One platform to manage your entire MCP ecosystem — across every client, from one place.

🔍 Auto-detects Claude Desktop, Cursor, Windsurf & VS Code on install
🖥️ Visual dashboard — see all your MCP servers and their health at a glance
🛍️ Built-in MCP Marketplace — browse & install MCPs without touching a config file
🔄 Multi-client sync — manage all your AI clients together
⚡ One-click install & enable — no manual JSON editing, ever
📦 Team sharing — export your entire MCP workflow as a .tsr file and share it instantly
🤖 Agent-native CLI — install tsr and let Claude or your AI agent manage its own MCPs directly
🖥️ Cross-platform — macOS, Windows & Linux
🔓 Fully open source & free — forever

Built on Rust. Lightweight, fast, and native.

MCP unlocks your AI agents. mTarsier unlocks MCP.

We'd love your feedback — what MCP clients or features should we prioritize next? Drop a comment below 👇

Auto-detecting every AI client on the machine — Claude Desktop, Cursor, Windsurf, VS Code — and unifying all their MCP config files into a single visual dashboard eliminates the most painful part of the MCP ecosystem right now, which is manually editing scattered JSON configs and hoping nothing breaks. The .tsr file format for exporting and sharing entire MCP workflows across teams is a smart standardization move; with the agent-native CLI that lets AI agents manage their own MCPs directly, how do you handle security boundaries — can an agent install any MCP server from the marketplace autonomously, or does it require human approval for new tool installations?

mTarsier : A free and open-source platform built for the community making it much easier for developers to build and manage MCP setups without the usual hassle. Really useful and thoughtfully done.
Feeling proud to be part of a team that truly cares about the community.

MCP tooling is still pretty fragmented — a unified manager for both servers and clients is pretty cool. Does it support auth config per-provider, or is that still on the roadmap?

Congrats on the launch! How do you decide which tools or integrations get added to the platform first?

How does multi-client sync work exactly? Is it alike to being able to configure the settings for all in one place or would I still have to go into each to change settings?

Curious—can mTarsier handle team-level permissions for shared MCP workflows, or is it per user for now?

How do you manage trust, security, and compatibility across such a broad ecosystem of tools in MCP360? If agents can access 100+ integrations, what controls are in place to limit permissions, prevent misuse, and ensure consistent behavior across different tools? Also, how do you handle versioning and reliability so workflows don’t break when underlying tools change?

Anyway, upvote from me + good luck

Hi everyone!

From shaders to ray tracing, NVIDIA keeps raising the bar. DLSS 5 will be the next big one.

It brings Hollywood-level lighting and materials into real-time games while still giving artists full control. The jump in fidelity is wild.

Quote from Jensen:

“DLSS 5 is the GPT moment for graphics — blending hand-crafted rendering with generative AI to deliver a dramatic leap in visual realism while preserving the control artists need for creative expression.”

Some 4K comparisons to dive into:

• Resident Evil™ Requiem

• Starfield

• Hogwarts Legacy

This sounds like insanely crazy tech lmao

This is an amazing invention. But what is the minimum amount of RAM required?

This looks really useful.

What inspired you to build it?

This sounds super interesting. 🤖 Curious how realistic the AI agents’ discussions get , do they give useful insights or is it more for fun?

The meta-concept of building a Product Hunt where AI agents are the users discussing and evaluating tools is a fascinating experiment in emergent behavior — as autonomous agents increasingly need to discover and select APIs, tools, and services on their own, having a structured forum where they can share evaluations creates a machine-readable trust layer that doesn't exist yet. The key challenge will be signal quality; how do you prevent the discussions from becoming an echo chamber of agents trained on similar data — is there a mechanism to ensure diverse agent architectures and perspectives contribute to product evaluations?

idea makes sense, but discussions around agents change really fast. how do you keep content relevant over time?

One of my biggest challenges with AI is that it's overly agreeable and "sugarcoats" the truth which could give false positives even though one's product is a "tarpit idea". Should one interpret comments from AgentDiscuss similarly as Reddit in the sense of "take this comment with a grain of salt"?

Most product reviews miss what actually matters for your specific use case. Agents evaluating the same tool against different criteria could surface insights that human-only reviews consistently overlook.

LOVE IT!!!

I get the product listing side, what I don't understand, is the agent side: who are the agents? Can anyone connect own product sourcing agent?

Super interesting! Are the agent discussions purely synthetic, or tied to real deployed agents?I'm building moltin.work is building the 'professional layer' for agents — seems like these two could complement each other.

Oh, really interesting idea!

As a founder, I've spent a lot of time thinking about product validation and how fast you can actually test things - AgentDiscuss feels like a huge step toward shipping products way faster

If in the future tools are made for agents, it will be interesting for tool creators to get direct feedbacks and suggestions from agents (the actual users) instead of the human.

Running AI jobs directly from your IDE without dealing with cloud setup could save a ton of friction for developers. How do you handle reliability when jobs are distributed across different nodes in the network?

Been testing the VS Code extension, and the workflow is surprisingly smooth. Being able to run GPU jobs directly from my editor without dealing with cloud dashboards is exactly what I needed for intermittent ML workloads. The pay-per-use model actually makes sense for experiments.

I've used this team's data products before - everything worked perfectly. Now I'm starting to test Orchestrator. We'll see how it goes.

Started experimenting with it yesterday, and I must say I really enjoy working with the workflow

Embedding GPU access directly into the IDE where developers already work — Cursor, VS Code, Windsurf — rather than requiring a separate infrastructure dashboard is the right UX decision for making compute feel invisible rather than burdensome. The escrow-based payment system that only releases funds after verified job execution solves the trust problem that plagues most decentralized compute networks; how does Ocean Orchestrator handle job failures mid-execution on a node — does the escrow mechanism cover partial compute costs, or is the user only charged for successfully completed work?

Hi Product Hunt 👋

I’m Nishit, founder of Kipps.AI, and today we’re excited to launch Kipps.AI Campaigns 🚀

💡 Why we built this

While building AI automation tools for small businesses, we kept seeing the same problem again and again.

Businesses collect leads from ads, forms, CRMs, and spreadsheets, but most of those leads never get properly followed up.

Sales teams are busy, follow-ups get delayed, and opportunities slip through the cracks.

Even simple things like:
• reminding customers about renewals
• sending anniversary or birthday greetings
• qualifying new leads
• or following up with ad leads

Still requires manual work, spreadsheets, and multiple tools.

We realised that AI agents could handle most of this automatically.

🚨 The Problem

Today’s outreach and follow-up systems are fragmented.

Businesses struggle with:
• Managing leads from multiple sources
• Following up with every contact at the right time
• Qualifying leads efficiently
• Tracking where a contact is in the funnel
• Managing communication across voice and messaging channels

This results in lost leads, missed renewals, and poor customer engagement.

🚀 Introducing Kipps.AI Campaigns

Kipps.AI Campaigns helps businesses automate outreach, follow-ups, and lead qualification using AI agents.

You can run campaigns that automatically call or message contacts, qualify leads, and track their progress in your funnel.

🔧 What you can do with Kipps.AI Campaigns

✅ Bulk Outreach & Lead Qualification
Reach hundreds of contacts and automatically qualify them as Hot, Warm, or Cold.

✅ Multiple Contact Sources
Import leads directly from:
• Google Sheets
• Google Ads / Meta Ads
• Your CRM

✅ Voice Agent + WhatsApp Agent
Engage customers via automated AI calls or WhatsApp conversations.

✅ Smart Campaign Scheduling
Schedule campaigns for outreach, reminders, or follow-ups.

✅ Lifecycle Reminders
Automatically call or message customers for:
• Renewal reminders
• Payment due reminders
• Birthdays & anniversaries
• Important customer events

✅ Contact Funnel Tracking
Track exactly where each contact is in the pipeline and their current stage.

✅ Analytics & Reporting
Get consolidated insights on campaign performance, responses, and conversions.

🎯 Who this is for

Kipps.AI Campaigns is especially useful for:

• Insurance advisors
• Financial advisors
• Sales teams
• Agencies
• Businesses running ads and lead generation

Anyone who wants better follow-ups and higher lead conversion without manual effort.

🎁 Product Hunt Launch

To celebrate the launch, we're offering special early access for the Product Hunt community.

If you work with leads, outreach, or reminders, we'd love for you to try it and share feedback.

🙏 Thank you

Huge thanks to the Product Hunt community for supporting builders and new products.

I'll be here all day answering questions and would love to hear your feedback!

Let’s automate outreach with AI 🚀

What differentiates Kipps AI from other AI outreach tools in terms of conversion rate and real ROI for small teams?

Congratulations on the launch 🎉 🎉 !!

Amazing time saving feature

Is what i hear all the time

Combining lead import from Google Sheets, Google/Meta Ads, and CRMs with automated qualification that categorizes contacts as Hot, Warm, or Cold across both voice and WhatsApp channels addresses the exact workflow gap where most small business leads die — the delay between capture and first meaningful follow-up. The lifecycle reminder system for renewals, payment dues, and anniversaries is a smart retention play that most outreach tools ignore entirely; how does the AI voice agent handle edge cases where a prospect asks something outside the campaign script — does it gracefully hand off to a human, or does it attempt to reason through the conversation?

What kind of measurable lift in conversion or response rates have you seen compared to manual follow-ups?

When sending out anniversary/birthday greetings, are there opportunities to attach a coupon to it as well?

Hiring is broken at the screening stage, too many applicants, too little time, and recruiters burning hours on calls that go nowhere.

JusRecruit fixes exactly that. Post a job, and "Saina" our AI interviewer takes over the phone screens and first-round interviews automatically. You wake up to a ranked shortlist of candidates who are actually qualified.

What makes it stand out is that it's not just an AI interviewer, it's a full recruiting stack. The built-in ATS lets you track every candidate, move them through stages, and plug in assessments, all in one place. No switching between five tools.

For any team tired of slow, messy hiring, this is worth trying.

Really interesting, especially focus on removing the bottleneck at the screening stage.

Curious how you're thinking about false positives early on.

this is actually a pretty real problem tbh. when the number of applicants gets high, just reaching the actually good candidates takes so much time and energy. i like that this is focused on that part and not trying to do 20 random things.

Phone screening was always the fastest way to filter candidates, but also where the first connection happened. Some people simply prefer talking to a human — it is how trust starts. Have you seen any pushback from candidates on the AI-led screening, or does the speed trade-off make up for it?

Love the recruiter side of this. Building on the candidate side with JobsUncle.ai would love to chat about where these intersect.

Automating phone screens and first-round interviews with AI targets the exact stage of the hiring funnel where the most recruiter time gets burned on candidates who won't advance — handling that initial qualification layer frees up human recruiters to focus on the nuanced later-stage conversations where judgment actually matters. The key question for any AI-driven interview tool is candidate experience; how do candidates typically react to the AI interviewer compared to a human phone screen — do you see differences in completion rates or candidate satisfaction scores between the two formats?

👋 Hey Product Hunt!

I'm Nawal, and I've been obsessed with one problem for the past year: why does hiring still feel like it's stuck in 2005?

Watching recruiters drown in spreadsheets, manually dialing candidates, and copy-pasting the same screening questions over and over - it felt like a solvable problem. So we built JusRecruit.

JusRecruit is an AI-powered recruiting ATS platform for high-volume hiring teams. It runs AI phone screenings, conducts structured interviews, and surfaces your best candidates - automatically. The outcome we're laser-focused on: cutting 10–15 days off time-to-hire and saving recruiters ~20 hours of manual work per role.

We've been deep in the weeds on this - talking to recruiting agencies, in-house TA teams, and high-growth companies hiring hundreds of people at once. Every conversation reinforced the same pain: screening is broken at scale.

This is our answer to that.

If you're in recruiting, talent acquisition, or just someone who's been on the receiving end of a chaotic hiring process - I'd genuinely love your take.

👉 What's the most painful, soul-crushing part of your hiring process right now?

Every response shapes what we build next. Thanks for being here 🙌

Interesting approach to cutting hiring bottlenecks — phone screens are one of the most time-consuming parts of the process. How does the AI handle edge cases where candidates give unexpected or creative answers?

Thrilled to launch JusRecruit here today! 🙌

Recruiters told us the same thing over and over: too many CVs, too little time, and first-round calls eating up the entire week.

With JusRecruit, your pipeline moves on autopilot. Every candidate gets a structured AI interview the moment they apply. You get a shortlist with consistent, comparable insights, so you can focus on closing, not filtering.

Less screening. Better hires. Faster. That's what we built.

Built by people who've lived the hiring bottlenecks firsthand, we'd love for you to try it and tell us what you think!

AI interviews can feel robotic and cause a high-potential prospect to lose interest in a company. Are there ways to avoid this when using an AI interviewer?

Great idea! You should build a website that allows users to bypass the Apple App Store fees, enabling them to download your DMG directly without incurring those charges.

I like how it does one thing and commits to it fully. Most tools try to do everything — this is refreshing. Does it support scheduled sending?

Hey Oliver. This looks very interesting. I had almost the same idea a while ago. The only difference being that I wanted the ability to copy and paste links or text as well. Basically using it as my read/do later bucket... ;-) But I'm no developer, so this idea was just rotting on some list. Happy to see that somebody did that! Good luck with it!

By the way. Are you the guy behind Pitchable?

Cheers

I used to use your email spam checker tools a lot :) Glad to see you here! Good luck today

Does Folderly tell me when an email is beyond help and make recommendations on when to create a new one? Especially since creating a new one requires time to "warm up".

Although there are many tools that do warm-up of your emails and furthermore sequencing tools are building this functionality up inside their products, their layouts might look great, but what's happening on the backend, are those warm-ups actually work? I know for a fact that lots of other tools are using people farms on the backend and pre-built seed lists that eventually won't give you any results except of nice numbers of the dashboard. Folderly is different, I know @belkins and @anastasiia_ivannikov have built huge back-end which is not something you want to highlight as it's not very markety, but oh my god it works. Kudos to Folderly team! Great product!

this is actually very relatable. once the number of mailboxes starts growing, things get messy real fast and checking each one manually is just painful. the task prioritization part sounds way more useful than people might think.

curious, what was the biggest thing teams were missing before this rebuild, actual deliverability issues or just not knowing where to look first?

Interesting launch, @belkins. The chaos from managing many mailboxes is real.

One thought while checking the homepage. Your PH story explains the pain perfectly: teams drowning in 30 inbox tabs. But the hero section says "Email deliverability starts here," which I think feels broad.

Something like "Manage 30 Outbound Mailboxes from One Dashboard" might hit the pain faster. Curious if you tested something like that.

Promising 99.9% inbox delivery rate tackles the single biggest invisible problem in outbound sales — teams scale mailboxes and campaigns without realizing a growing percentage of their emails are silently landing in spam, making the entire investment in copy, targeting, and sequencing worthless. The challenge with email deliverability is that it's a moving target as ESP algorithms constantly evolve; how does Folderly's approach adapt when providers like Google or Microsoft change their filtering criteria — is there continuous monitoring that detects deliverability drops in real time, or is it more of a periodic audit model?

Finally, a Fully Autonomous AI Coding Agent 🚀 .

TLDR:

• yes, Agen is already building itself at this point 🤩

• runs in the cloud, doesn’t need local installation

• automatically fixes your pipelines

• delivers a fully working code!

• as a ProductHunt user, you can try Agen for free, and get $20 Credits on sign up

AI already changed a lot the way we code, but it didn’t really change the way we work and think about the code. For the biggest part, we have the same processes and workflows, the biggest difference being the fact that an AI model writes the code for you.

We built Agen as the coding agent for the future, with the goal to help the teams build faster, better, and importantly, secure code.

Here are the principles that are driving us, and which are the foundations of what we’ve built so far and the many things we plan to build in the future, keep reading to see what this means for you and your company.

🚀 Fully Autonomous AI Coding Agents - our goal is to make sure that the agents are being able to do the work independently, with minimum human guidance. We noticed that the biggest bottleneck are the humans right now, because the agents still need quite a lot of guidance, starting from the environment they run it, and ending with the work they do. Continue reading to see what we’re doing to fix this.

☁️ Cloud-First: this is the first step to have autonomous agents, they need an environment to run in, and right now, if you close your machine, your agent stops working. By making the agents cloud-first, we make sure that they always have a place to run in, and that they’re not blocked by humans when they need to do the work. For you and your company this means that the agents will do meaningful work 24/7 and help you move faster.

🧑‍🔧 Self-healing pipelines - this is the next step, and something that the agents are blocked by when they make changes to the codebase pretty often. Usually, a developer has to tell the agent that the pipeline failed, why it failed, and ask them to fix it, or they use the MCP (if the version control system they use has one), but it’s still happening manually, on user request. Our agents are monitoring the failed pipelines and fixing them automatically, with zero human involvement. This results in a huge amount of time saved - time that can be spent on more meaningful work.

📆 Scheduled Agents - they're running on your schedule, and they're doing autonomous work in the cloud, without human supervision - all you have to do is check and merge the code.

🪾 Multi-repository support - one of the biggest issues with the autonomous coding agents is that they're tied to a specific repository, and they can't make changes to multiple repositories for a task, usually you have to start new agents if your task involves making changes to multiple repositories. We built Agen to work flawlessly with multiple repositories, it will automatically choose the right repository to change, and it's able to make changes and open MRs in multiple repositories within the same session.

💻 Multi-session environment - one more blocker for real autonomous work, and the only way to have multiple sessions with the current agent is to create multiple copies of your repositories, and run them in different terminals, but that’s time consuming, and not efficient at all, especially when it comes to cleaning them up, creating them again, etc., All our Agents are doing the work within a session, which gets an isolated sandbox, it’s own branch, MR, etc. For you, as a developer, it’s important to be able to run multiple tasks at the same time, the work is not linear, and being able to run multiple sessions is a huge productivity boost.

📱 Mobile-friendly - you know that message you’re getting from your boss about a bug when you don’t have the computer with you? It’s happened to all of us, and it’s stressful 😣 . We optimized our web app and made it mobile friendly so that you can make changes wherever you are, and whenever you have to.

🧑‍💻👩‍💻 Built for meaningful collaboration, with per-repository access, and granular permissions, that resemble how the teams usually work. Each person can be given access to specific repositories, and they can have different roles for each repository. This keeps your codebase confidential and secure.

✅ See your pipelines & merge straight from the session, without having to switch tabs. You’ll see a green indicator when all the pipelines pass, so that you know that the work is completely finished. The next step is to actually merge the code, and you can do it without leaving Agen, straight from the session page.

🔐 Security - the Agents are making changes on your codebases in new, secured and protected sandboxes for every session. The sandboxes are destroyed once the Agent has finished the work, together with your code and any traces of it.

⏭️ What's coming next: integrations, more autonomy

Your feedback means a lot - please leave a comment and tell us what is the biggest blocker you’re having when using AI tools for coding, and what would you like to improve in your workflows.

Thank you for reading so far!

Daniel, Co-Founder 👋

Congratulations Daniel, really excellent product and useful.

Hey Product Hunt! 👋 I'm Fred, and I made ClickSay.

Here's why I built it. I've been vibe coding for months - building everything with Claude Code (i use it in terminal but this can be done in any AI building tool, OpenAI Codex, Replit, Lovable, you name it...). And I kept running into the same annoying wall. I'd see a UI bug, switch to Claude Code, and spend 30 seconds trying to explain which element I meant. "The button in the header, the one with rounded corners, make the font bigger." The AI would get the wrong one. I'd try again. Still wrong.

Turns out the bottleneck was never the AI. It was me trying to describe what I was looking at.

So I built ClickSay. Press Cmd+Shift+K (or change it to any shortcut you like), click one or more elements (Use shift key for more than one), say what you want fixed (in any of the 20 supported languages), or changed, or even a large enhancement. It grabs the CSS selector, computed styles, HTML, (lot of options to add in the sidebar like the screenshot with the element highlighted, and even the React/Vue/Svelte component name). All of that gets packaged into a structured prompt and lands on your clipboard.

Paste it into whatever AI tool you use. It gets the full picture and nails the fix first try.

ClickSay completely changed how I work. It's the tool I reach for more than anything else now. What used to be a full minute of typing and back-and-forth takes three seconds. I'm shipping UI fixes way faster, and that compounds across a whole build session. Some days I'll ClickSay 100+ changes without even thinking about it. It's just muscle memory at this point.

For the PH community - use code PRODUCTHUNT2026 for 3 months of Pro free.

I'm really interested in finding out if this changes your daily behavior like it changed mine. One of my beta testers told me the other day he is "hooked on ClickSay". That's awesome. Let me know what you guys think! Cheers- Fred

Here's a video that shows how I use it with Claude Code (I built this complete website MatchCentral with ClickSay btw):

https://www.youtube.com/watch?v=_ijomO9tAX4

this is actually pretty smart. half the struggle with ai fixing ui stuff is just explaining what exactly is broken, so pulling the selector, styles, screenshot and component info in one go makes a lot of sense. sweep mode sounds useful too, specially when there are multiple small issues all over the page.

curious, what kind of users are getting the most value from this right now, devs fixing their own ui or designers/qa folks sending cleaner feedback?

I can't believe how in less than 2 weeks, 40 years of how I interact with computers has been completely transformed. I built a chrome extension to better work with Claude Code and now, it is virtually the only tool I use on my computer 16 hours a day. I'd love to know how it also changes your daily behavior.

This is such a smart solve, Fred. The "describing what I'm looking at" bottleneck is so real - I've wasted so many minutes trying to explain a button's position to an AI.

The CSS selector + computed styles + component name approach is clever. That's basically giving the AI perfect context instead of hoping it guesses from your description.

Congrats on the launch! Wishing you a great one!

@fredmaker Oh man, I feel this one. I spend half my time in Cursor just trying to explain which element 'm talking

about. "No, the OTHER button. The one with the shadow." Having the selector and styles auto-captured is a game changer.

I like the No Enterprise Tax angle. Quick observation while checking the site. The product clearly sits in the LaunchDarkly alternative space. Okay, but the homepage barely calls that out.

For devs evaluating feature flag tools, that comparison might instantly anchor the value.

I'm curious if you tested leaning into that contrast more.

Analytics shows what happened. Usercall Triggers helps you understand why by talking to users the moment behavior changes

Triggering AI-moderated interviews at the exact moment a user drops off, churns, or hits friction is a massive improvement over the traditional approach of sending survey links days later when the context is already lost — capturing the "why" while the experience is still fresh should produce fundamentally richer qualitative data. The bridge between analytics (what happened) and user research (why it happened) is exactly the gap most product teams struggle with; how do you handle the user experience of being prompted for an interview mid-flow — is there a risk of adding friction to an already frustrated user, or do you see it actually improving retention by making users feel heard?

Congrats on the launch and the product!

What is the most popular niche or industry that's currently using UserCall?

The gap between "users dropped off here" and "why did they drop off" is where most product decisions go wrong. An AI voice interview triggered at the moment of friction is way more likely to get real answers. How long are these interviews typically? Whats the avg % of users who actually answer?

Looks great but Im curious how complex the setup is. Tools like this can sometimes be tough to onboard.

🚀 Big News: Parallax is Now Open Source!

I'm excited to share that Parallax, it’s officially open source.

Parallax started as an idea to make working AI workflows simpler and more transparent.

By making Parallax open source, anyone can:
✨ Explore how it works
🔧 Contribute improvements
🚀 Build new ideas on top of it

I’m really excited to see what the community builds, and I’d love your feedback, contributions, and ideas.

If you find Parallax interesting, I’d really appreciate your support on Product Hunt today 🙌

The plan-first-then-execute workflow with explicit approval gates is exactly the right trust model for AI coding agents — pulling issues directly from Linear or GitHub, creating isolated worktrees, and opening reviewable PRs means Parallax slots into an existing dev process rather than requiring teams to adopt a new one. Running everything locally with full control on your machine is a strong differentiator from cloud-based coding agents; how does Parallax handle issues that require changes across multiple repos or services — can it coordinate work across different worktrees, or is it scoped to single-repo tasks?

Local-first AI orchestration is an underrated design choice — no latency spikes, no data leaving the machine. How does it handle tasks that need external API calls while staying local?

Congrats on the product and the launch!

Does it have an intro and an outro for each tweet?

Instead of a feed, can we have a curated list of people whose feeds we want to listen to as a podcast?

Can user select tweets that will be "read"?

Love this, fellow maker here, also built JobsUncle with Claude. The one-time pricing model resonates, subscriptions for simple utilities do feel wrong. Great work.

this is actually a cool take. i like that you are not trying to make “another discord bot framework” and instead giving agents a cleaner way to interact with discord directly. the permission profiles and audit logs part also makes it feel way more practical, because that’s exactly the kind of thing people would worry about first.

curious, what are people using it for first in real life, simple community ops stuff or full blown agent workflows?

I have clicked "I Agree" thousands of times without reading a single line. The idea of getting a 15-second summary with red flags sounds almost too convenient. How accurate is it with longer or more complex terms — does it catch the subtle stuff buried deep in the text, or mostly the obvious patterns?

Nobody actually reads those walls of text. This is one of those tools where the use case sells itself. Does it flag specific clauses or just give an overall risk score?