PH热榜 | 2026-02-20

The AI race continues. OpenAI launched GPT-5.3-Codex 2 weeks ago. Anthropic, Sonnet 4.6 this week. And Google? They just announced @Gemini 3.1 Pro, "a smarter, more capable model for complex problem-solving."

Available in products like @Google AI Studio, @Kilo Code, and @Raycast.

Game on!

Gemini is alwasy good at benchmarks, but usually not great at agentic behaviour. The models have very weird behaviour. Almost like the Gemini team is not really testing them themselves.

@peter_albert nailed it. I'm running Gemini models in production for Aitinery (AI travel planner) and this is exactly the gap.

Benchmarks say Gemini is world-class. My production logs say it sometimes hallucinates restaurant names that don't exist and occasionally generates itineraries with 16-hour driving days. Benchmarks don't test "can this model reliably plan a family trip to Puglia without suggesting a 3am dinner reservation?"

That said — 3.1 Pro feels like Google is finally closing the gap between benchmark performance and real-world agentic reliability. The reasoning improvements matter more for agent builders than the raw intelligence bump.

The uncomfortable truth about the AI model race: for 95% of real applications, the difference between GPT-5.3, Sonnet 4.6, and Gemini 3.1 Pro is negligible. What matters is reliability, cost, and speed — not who wins on ARC-AGI-2.

Curious to see how 3.1 Pro handles multi-step planning tasks. That's where Gemini has historically struggled compared to Claude for agentic workflows.

If you're building with Gemini 3.1 Pro and want to keep API costs under control as complexity scales, check out TokenCut by agentready.cloud — it helps reduce token usage without sacrificing output quality. Perfect companion for a reasoning-heavy model like this one!

Does google read these?

I'll give it a shot in gemini CLI and see what's up

i like it

Not bad. Its general intelligence seems to have increased due to his abstract thinking abilities.

Impressive direction, pushing the baseline forward for deeper reasoning is what actually unlocks more serious use cases. Complex problem solving needs more than fast answers, it needs structured thinking.

Curious to see how 3.1 Pro performs in longer multi step workflows.

I’m building Ahsk.app , a macOS AI assistant focused on practical, in flow AI use. Would love to connect and exchange thoughts.

Congrats on launching Gemini 3.1 Pro, it sounds like a solid upgrade for complex problem-solving. To enhance user engagement, consider highlighting specific use cases where it outperforms competitors. What is your strategy for ensuring users see the value in this advanced reasoning capability quickly?

Nice benchmark numbers. My concern is always the gap between benchmarks and the actual developer experience. I use Claude primarily for coding because, from my personal experience, it follows instructions pretty closely (though there's always room for improvement). For me, Gemini has historically been frustrating for me, inserting comments and refactoring code I didn't ask it to do. Would love to hear from anyone who's tested 3.1 Pro on real coding workflows, not benchmarks, and whether that's actually improved.

Hey there, congrats on this launch!!

For SaaS use cases involving long-context multimodal inputs (e.g., analyzing full user-uploaded PDFs + screenshots + code snippets to generate UI code, migration scripts, or automated test plans), what's the practical sweet spot you've seen for token efficiency and accuracy at the 200k–1M range?

I can't keep using Antigrativy, there is no update available; and I can't use the previous model.

So many SMBs struggle with creating high-quality product visuals consistently... studio shoots are expensive, time-consuming and rarely flexible.

Here comes Photoshoot in Pomelli to the rescue: start with one simple product image and generate polished, on-brand studio or lifestyle shots in minutes.

This new feature in Pomelli that uses business context (Business DNA) and Nano Banana image generation to turn product images into a professional-grade studio shot.

With Photoshoot, you can generate stunning product images for your website and social content.

Here’s how it works:

• Pick a product

• Choose a template

• Generate

• Refine

Bonus features:

• Improved image models with more accuracy, following your prompt

• Editing capabilities, e.g. "change my background to a forest"

• Style reference image, e.g. restyle <image> to look like <image> style

Available free of charge in the US, Canada, Australia & New Zealand!

Hey there, good launch day guys!

I have a question, from a SaaS dev agency point of view.

How tightly does the 'Business DNA' (brand colors, fonts, style) carry over to Photoshoot outputs? For SaaS apps serving multiple merchants/brands, can we pass custom brand guidelines per request (JSON payload with hex codes, reference images, tone descriptors) to ensure consistency across thousands of SKUs without manual tweaks?

Just incredible what Pomelli can do.

Hey, nice initiative, I wonder how do you maintain brand uniqueness(for your customers) and make them stand out when using templates. I mean If more people would use the same one with similar products?

Hi everyone, I'm Daniel from the Pomelli team. Thanks for hunting us @rohanrecommends !

I'll be around today answering any questions! Give it a try at labs.google.com/pomelli

Still not available in my region (North Africa)

I used this today for my partner's needle point store's (https://matildastitchco.com) product shots. It's crazy good. Like crazy good. Shopify should build a direct integration, would be a game changer for ecomm.

@daniel_adonai , @rohanrecommends , @Google -This is a real unlock for SMBs studio quality visuals without the cost or logistics is huge. Starting from a single product image and generating on brand shots in minutes makes content creation far more scalable.

The background editing and style reference features are especially powerful.

I’m building Ahsk app , a macOS AI assistant focused on helping founders create and iterate faster across apps. Would love to connect and exchange feedback.

Seems very promising, but could not validate / test it for the Netherlands (not available in this country yet)

Congrats on the launch, hey @daniel_adonai. Your unique search capabilities could really shine with a clear focus on specific ?

Hey Product Hunt ! 👋

Shreyas here from Lyzr. Today we’re launching Architect — prompt to fully deployed agentic application. No glue code. No stitching frameworks together.

Here’s what Architect actually does:

1. Prompt → Full App
Describe what you want. Architect builds the agent logic, tooling, guardrails, RAG pipelines, AND a production-ready Next.js UI. Not a JSON blob. A deployable application.

2. Not Just Text — Voice, Image & Video Apps Too
Build agents that can talk, see, generate images, and create videos. Voice apps. Image gen apps. Video gen apps. All from a single prompt. OpenAI, Anthropic, ElevenLabs, Replicate — any vendor, any modality, zero integration work. Your agents aren’t limited to chat. They’re multimodal from day one.

3. Plug Into the Tools Your Team Already Uses
Gmail, Notion, GitHub, Slack, Google Drive, Jira — your agents can use them as tools natively. No custom API wiring. No middleware. Just tell Architect what your agent needs to do and it connects the dots. Your agents don’t just think — they act across your entire stack.

4. Two Agent Styles, One System
Build autonomous manager-style agents for complex reasoning OR deterministic workflow agents for predictable execution. Mix both. Most platforms force you to pick one.

5. Lyzr Control Plane + Agent Studio — Full Ops Without Touching Code
This is the layer that makes Architect production-grade. Every app Architect builds is connected to Lyzr’s control plane through Agent Studio.

That means after deployment, you can:
— Observe agent behavior and trace every decision
— Modify prompts, logic, and orchestration flows
— Add or update knowledge bases and RAG pipelines
— Apply and tune guardrails in real time
— Swap models, adjust tool access, manage permissions

All without touching a single line of your frontend code. The UI stays intact. The agent brain evolves independently. Build once, operate forever.

It Builds the UI (Yes, Really)

Most agent platforms stop at the backend. Architect generates a full Next.js frontend. Your agent isn’t just functional — it’s shippable to users.

TL;DR: Prompt → Agents → Governance → UI → Deploy. Text, voice, image, video. Gmail, Notion, GitHub built in. Full ops layer via Lyzr Agent Studio — observe, modify, govern, scale — without rewriting code.

We built this because shipping AI agents as real products is still way too hard. Architect fixes that.

Try it. Break it. Tell us what sucks.

Hey Product Hunt! 👋
Vidur here from Lyzr. Today we are launching Architect.

We built Architect because we kept seeing the same problem: the people who understand business problems best — couldn't easily build the agent workflows and apps they envisioned without depending entirely on developers.

Architect bridges that gap. Describe what you want to automate or build, and Architect generates the PRD, orchestrates the agents, connects your tools, and delivers a working agentic app — without you needing to think about the underlying infrastructure.
Think of it as the fastest path from "I wish I had an agent that..." to actually having that agent, with a UI, live integrations, and everything.

What makes us different? Unlike n8n or Make, there's no node-dragging. Unlike Lovable or Replit, you're not building generic apps — you're building agents that reason, take action, and connect to your enterprise tools.
We're still early. The architecture is evolving fast, and our template library is growing. But if you're someone who has ever had to wait weeks to get a working agent prototype — we built this for you.

Drop your questions below — happy to go deep on use cases, architecture, or how we think about agent orchestration.

A perfect blend of innovation and strong architecture!

Loving this approach of making AI less of a mystery. Feels like the right direction!

The guided mode feature of architect is my favorite, everyone should try that!! 💙

Please list in pricing the free credits/month on free tier.

I probably will not stay on free, but this in mid-February, 2026; integrate beyond what every agentic platform has and more to be competitive. Github, supabase, railway, postman, cloudflare...

I want to build, not schedule things, so gmail, google calendar and your other integrations, I'm sorry, but I'm mnot a fluffer; my eyes glaze over until I see integrations that would actually help with what really eats our timee - debugging.

I'm with you in the struggle, and eager to see you succeed.

I just built and deployed this in under 10 mins, and Its my favorite gamified learning experience I have built with Architect so far!: https://code-quest-mega-gear-97jp.architect.space

Congratulations on the launch! I love the direction Architect is taking - truly agentic applications that feel truly enterprise-ready. Excited to see how the platform evolves over the coming years 🚀

Interesting no-code approach for AI agents. How does it compare to building custom agents with code? Curious about the flexibility vs speed tradeoff.

Congrats on your launch, guys :)

Quick question here before i dive in into this one!

Once Architect generates the PRD, multi-agent orchestration, UI, DB connections, RBAC, audit logs, etc., how do we take it to production? Is there export to self-hosted (our VPC/K8s), one-click deploy to Lyzr cloud, or API/SDK wrappers so we can embed/embed calls from our existing backend/frontend? Any support for custom LLMs or bring-your-own-model?

Awesome product!

Congrats on the launch! What tools are your top customer favorites?

@shreyaskapale @vidur_ congrats on the launch! Nicely done! A couple of questions:

• How much visibility or control do we have over the code while it is being generated (when we're in planning mode)?

• Is there a desktop app or you guys working on it?

Agent tooling right now splits into workflow builders like n8n and Make where you drag nodes, and app generators like Lovable where you get a frontend but wire agent logic yourself. Architect sitting between those, prompt to full orchestration plus a deployable Next.js UI, fills a gap most teams hit once prototyping is done. Running 10,000 automated tests via the simulation engine before deployment is what actually moves this past demo territory... most agent prototypes break the moment real users hit edge cases. Pre-launch failure mode detection matters more than build speed.

Congrats on the launch, hey @shreyaskapale. Consider highlighting specific use cases for onboarding. What activation metric will you prioritize?

Hot take: the "N8N + Lovable baby" comparison actually undersells what's interesting here.

The real question nobody's asking about no-code agent builders: what happens when your agent needs deep domain expertise?

I've been building AI travel agents (Aitinery — we use Google ADK with multi-agent pipelines for Italy trip planning), and here's what I learned the hard way: the orchestration layer is maybe 20% of the problem. The other 80% is domain knowledge — knowing that the train from Rome to Amalfi doesn't exist (but ChatGPT will happily recommend it), or that a "15-minute walk" in Positano actually means 400 stairs.

Tools like Architect nail the 20% beautifully. Prompt to deployed app, multi-agent orchestration, tool integrations — that's genuinely hard infra work and you've clearly built something solid.

But I'm curious about the 80%: when someone builds a vertical agent on Architect (say, a legal research agent or a medical triage bot), where does the domain expertise live? In the prompt? In custom tools? In fine-tuned models?

Because the dirty secret of AI agents is: generic orchestration + generic LLM = generic output. The magic happens in the domain layer, and that's the part that's hardest to no-code.

Not a criticism — genuine question. Would love to hear how you think about this @vidaboravida

Hey everyone 🚀

I'm part of the team behind Claudebin, building this together with Vlad.

If you're curious about how and why we built it, I wrote a detailed article breaking it down here:

https://balajmarius.com/writings/making-claude-code-sessions-shareable/

Happy to answer any questions, whether it's product decisions, technical details, or the story behind it.

I’ve run into the problem where a long Claude Code session produces great results, but sharing the process with teammates is basically impossible without screenshots or messy summaries.

Having a resumable URL with the full thread + file operations makes a lot of sense, especially for PR context or debugging sessions. It feels similar to sharing a notebook or replayable dev session rather than just output.

Congrats! love this, sometimes, after 4-5 hours of chatting, i get a huge huge history and it is reseting the terminal. That's not a problem as i remember most of the flows, but other people don't know the reason of what i did or not did something. What i actually want is a way to "bookmark" some of the informations that are relevant from those long threads. And reapply them. Are you working on something like this?

This tool is SO useful! Thanks a lot for building it and congrats on the launch 🚀

For me the code writing part of my work has now almost fully shifted from actual writing to prompting. So it makes sense that I want to see other people's prompts along with the code. It's a great tool for sharing.

If you think about it, where before you would open a PR and add a bunch of notes for reviewers, now you can include the chat history and it gives you so much context!

Hi @balajmarius ,

Saw ClaudeBin on Product Hunt — really like the “sessions deserve a URL” framing.

One thought: positioning it slightly as workflow memory or version control for Claude teams could make the long-term value even stronger.

Curious how you’re thinking about that direction.

Great product. This solves a problem I face daily. Like many, when sessions compact I lose so much decision making context. Question: Can this auto-save? Claude supports hooks, so maybe a pre-compaction trigger that auto publishes to Claude Bin? Personally, my pain point isn't necessarily forgetting to share a session it's just losing it before I get the chance.

Hi there, big day!! :)

A few quick questions from someone embedding agentic coding into web/mobile apps (Next.js + React Native, often with Claude Code for backend scripts, migration tools, test gen, etc.):

1. Does it fully serialize everything useful? E.g., inline file read/write diffs (before/after views), bash command outputs (stdout/stderr), tool call results (web fetches, MCP responses), and any custom skills/hooks?

2. When resuming from the URL locally (via import?), how does it restore state? Does it replay the session into a new Claude Code instance, inject context/files/commands, or just load the viewer for reference?

3. For SaaS/team workflows: Can we embed selected ranges (e.g., a refactor block) as iframes in Notion/Linear/Confluence? How private are the URLs — random long slugs, password-protected, or self-hosted viewer option?

Thanks guys!

Great idea! thank you!

@vtemian @balajmarius Great idea. This was needed.

This is such a useful tool for anyone deep into Claude Code workflows. The fact that it captures the full session context — file edits, bash commands, MCP calls — and turns it into a shareable URL is exactly what's been missing. I love that you can embed selected ranges in docs too, not just dump the whole thread. Really smart for code review workflows where the "why" behind changes matters as much as the diff itself. Quick question: are you planning to support any kind of session diffing, like comparing two Claudebin sessions side by side to see how approaches evolved?

Congrats on the launch! Making Claude Code sessions shareable and navigable feels incredibly useful, especially for PR reviews and async collaboration. How do you handle sensitive data inside sessions, for example API keys, internal URLs, or proprietary code, before generating a public URL? Is there a redaction or review step to prevent accidental leaks?

Congrats on the launch, @vtemian. Simplifying session exports is key for developers. How do you plan to drive initial ?

This would have saved me hours, but I’m concerned about leaking information. Is there an option to make it private by default? Also, could you implement an agent to pre-process and clean sensitive information from the exported data?

The sidebar-attached approach is genuinely clever. The problem with most bookmark managers is the friction of opening a separate app or tab to access them, which kills the workflow. Having it follow the browser window natively removes that friction completely. Open source and free is the right call for a utility like this too. Congrats on the launch!

WOW, good luck with the launch! I wonder how long it took to build it?

Would you mind sharing what was your inspiration to buid Arcmark?

Congrats on the launch, @geek_1001. The local-first approach is great for user trust, but what’s your strategy for user acquisition?

The idea is just SUPERB. I've had this in mind for along time and I've tried out supasidebar, which is also good, but too buggy at the moment. I hope this project will develop into something GREAT! Thank you for the initiative, all my support to the dev!

Visually, it looks cool. But what if I have too many bookmarks? Willit not be so much cluttered?

Huge congrats on launch! Love the secure proxy model and zero‑secret agent design.

This is a great idea. Have you thought about expanding it to also support traditional website passwords, so agents couldn’t access those either? Curious whether you see this eventually replacing tools like 1Password, or staying focused purely on developer/API secrets.

Nice product ! Keep building

good luck with your launch ! @severin__

Congrats on the launch! Server-side secret injection feels like a clean abstraction for agent security. How do you handle auditing and traceability? For example, can teams see which agent session accessed which credential and what request was ultimately made, without exposing the secret itself?

Congrats on launching, @severin__. The secure approach to managing credentials is impressive. How do you plan to drive initial ?

Thanks @albn !

I've made keychains.dev to be able to use the limitless power of AI agents like OpenClaw, without having afterthoughts about if it's properly keeping my passwords and credentials safe.

Feels more safer now.
@community I'd love your feedback on it. Let's make agents safe!

Congrats on launch! Love the focused UX, crisp output, and AI-powered polishing.

@evaldasb japanese demo just killed me man amazing

This could be great for onboarding. Does it work for documenting internal processes too? We spend a lot of time creating training materials for new team members.

Happy launch day babes!

Quick questions here :

• Multi-language AI voices sound great, but does it auto-translate the script when switching voices/languages, or do we edit manually? How natural/accurate are non-English voices (e.g., French for Quebec users, Spanish, etc.)? Any bulk export for localized versions or API to generate variants programmatically?

• For embedding in SaaS products (e.g., auto-generate/update guides on feature release via CI/CD, or trigger from user actions): Is there an API (or planned) to capture flows programmatically, generate narrated videos server-side, or pull/export guides? Any webhooks for async processing or Zapier/Make support for automation?

Congrats on the launch! Focusing on output quality instead of feature bloat is refreshing, especially in the walkthrough space. How do you ensure the auto-generated captions and AI refinements stay accurate for more complex workflows or niche tools, where small wording mistakes could confuse users instead of helping them?

How you handle products that have complex, branching workflows? That's usually where these tools break down. Also, why is the demo in Japanese? 😄 Curious to see a B2B SaaS example.

Great product! can i use it to make a guide for my webapp agentready.cloud?

Congrats on the launch, @aliarain ! 🎉 Keeping eyes locked on the camera during a pitch is such an underrated problem to solve. The “Invisible” mode alone is a killer feature. Quick question on the auto-scroll: does it sync to your speaking pace in real time, or is it a fixed speed you set manually before the call?

Nice project !

Haha, this is really a genius design

Question: Is it possible to move this around at all because I have the iContact cam that hangs down a little bit lower, but I'm still looking for a teleprompter that I can place in the perfect spot.

I love this concept so much. Eye contact matters and memorizing scripts is a challenge. But do you have a response to this notice I got from trying to open the application?

This is such a sharp use of the notch. Fellow Swift/SwiftUI builder here and I genuinely appreciate the elegance of this — you took something most people treat as dead space and turned it into a utility layer. The invisible-during-screen-sharing feature is the real killer insight. Most presenters look down at their notes constantly, which breaks trust with the audience. This solves that without any hardware. Congrats on shipping this!

I like the idea! Both the CTAs on your website and link to Github don't work. You guys working on something?

Hey Product Hunt!

I'm Ben, co-founder of Repaint. Today, we’re officially ending our beta and opening Repaint to the world.

Repaint is an AI chatbot that builds professional websites.

This is actually our second website builder. During YC S23, we built an entirely different website builder. It was heavily inspired by Figma. We spent thousands of hours polishing it... and yet nobody finished their website 😅. It was too much manual work. And our product wasn't meaningfully easier than existing website builders. Eventually, we shut it down.

As AI got better, we realized it was possible to automate most of web design.

The big companies tacked on a couple AI features, but their core experience is still manual editing. And the newer AI tools are optimized around vibe-coding apps, not website building. They don't automate the work; instead they give you an incomplete site with generic info and designs. You have to do hundreds of edits to finish a site. Not to mention they charge per edit!

Last year, we started building Repaint. It's purpose-built for professional marketing websites.

Key Features

1. Plan your site with AI — Repaint researches your business, interviews you on what you need, and generates a complete multi-page site in minutes.

2. Match any website's style — Give Repaint a link to a site you like, and it'll build you something similar.

3. Import your old site — Share a link to your existing website, and Repaint pulls in your pages, text, and images automatically (with page limits so we don't go bankrupt).

4. Chat-based editing — Make big changes through chat, like adding pages or updating info across your whole site. For small tweaks like fixing a typo, just click and edit inline.

5. Interactive components — Repaint can build things that are hard in traditional builders like pricing calculators, animated carousels, calendar embeds, and more.

6. Flat pricing — One monthly subscription. No per-update fees, no credits.

We still have a ton of ideas, but Repaint is already building sites that blow us away, and our beta users agree. So we decided it's time to open it up.

Try it yourself at https://repaint.com/ ✌️

Congrats on the launch! The shift from a Figma-style manual builder to a truly automated, research-driven AI approach sounds like a hard-earned lesson. How does Repaint balance automation with differentiation? If it researches a business and generates a full site in minutes, what ensures the output feels uniquely positioned rather than converging toward similar structures and messaging across industries?

This is really cool. We will build our next landing page with this!!

For editing, are you selecting the page to edit or would prompting need to be as specific as possible?

Congrats on the launch! Love the concept of your product. Is it able to build websites with a complex structure based on prompts? (for example, websites with multiple sections: blog, cases studies, help center) Or is it primarily focused on one-page sites?

@ben_shumaker
Congrats on the launch! 🚀 I just put Repaint to the test by building a developer portfolio from scratch. The 'Interview' flow is incredibly smooth, it actually makes the setup phase fun and solves the blank canvas problem perfectly.

I particularly appreciated how you handled the Version Control and the Context-Aware Re-prompting (asking for more technical copy worked like a charm).

Two suggestions for your roadmap that I think would take this to the next level:

• Data-to-Component Sync: For professional sites, it would be a game-changer to prompt: "Sync my latest 3 GitHub repos into the Works section." Automating the content sourcing directly into the UI blocks would save hours of manual entry.

• Block-Level Style Variations: Adding a simple 'Variation' toggle on individual sections (like the Hero or Contact blocks) would allow users to cycle through 3-4 different layout options without changing the whole site's vibe. It gives that extra bit of creative control while keeping the UI minimal.

Really impressed with the speed and the interactive components. Great job, team! Already upvoted!

Is this site generation more closer to a portfolio site or an application site?

Hi Product Hunt community!

I've always believed the best hardware and the best platform should work together, regardless of who made them. Apple Watch has the most advanced health sensors, Android offers unmatched openness and flexibility, and there was no good reason these two couldn't coexist.

So we built Merge for Apple Watch.

Merge connects your Apple Watch to any Android phone for notifications, messaging, media controls, and brings the Apple Watch's cutting-edge health data into the Android ecosystem, to be used with any Health Connect compatible app. Also, for anyone juggling two phones, like a personal iPhone and a work Android, it's the only solution that lets you use your watch for both.

Would love your thoughts and feedback!

Matt

This is so cool!

Actually, I never thought we would see an Apple x Android intersection because of the inherent complexity between them.

Would love to know your journey of building. How did you start? How long did it take for you to make this?

Hey everyone 👋

I launched UI Inspector back in 2024, and while people seemed to appreciate the tool itself, the Pro offering missed the mark. Rather than patch something that wasn't working, I decided to step back and rethink the whole approach.

UI Inspector 2.0 is the result, an almost full rebuild with:

• A new atomic CSS engine

• Better change tracking

• HTML editing including cut, copy, paste, duplicate, delete elements

• CSS output is now included in the free plan

• Tailwind, SCSS, and JSX exports

• A completely reimagined preset system with build-in presets that inherit the site's own styles

Would love to hear what you think. Thanks for giving it a look.

Looks pretty interesting ... what sort of moat have you defined for folks just vibe coding changes and not necessarily debugging on the fly?

Hi PH,

I built trnscrb because I wanted something simpler.

Most meeting tools try to be everything. Summaries. Dashboards. Cloud sync. Accounts. I just wanted reliable transcription that runs locally and saves plain text files on my Mac.

trnscrb lives in your menu bar.
It detects Zoom, Meet, Teams, Slack, and FaceTime calls.
It records and transcribes on Apple Silicon using Whisper.
Speaker labels are added.
The output is a clean .txt file in a local folder.

It also runs a local MCP server, so compatible tools can read and search your transcripts directly on your machine.

Requirements:

macOS 13+
Apple Silicon M1 or later

Happy to answer setup questions, especially around system audio capture and MCP configuration.

nice, congrats! great infra for security of agents

The isolation angle is smart. Most agent sandboxing today is just containers with a fancy wrapper, not real VM-level separation. Curious about the network egress controls -- do you support fine-grained rules per agent (like allowing only specific API endpoints)? That would be huge for enterprise use cases where agents need to call external services but you still want strict blast radius control.

Also for y'all Product Hunters we got a Promo Code!! CS10OFF to get 10% off of our Starter $20 plan!

That's a really nice idea! aren't you worried about the costs of this cloud pc's? have you tryed or looking to implement technologies like agentready.cloud and their TokenCut feature to reduce token costs for your end users? are they using their API key's for LLM's or is the platform giving their's? Thank you, Really really great product by the way, so simple yet so usuful :)

Congrats on the launch! Giving agents their own disposable cloud machine feels like a clean, practical solution to a very real fear. How do you balance isolation with developer convenience? For example, how easy is it to securely pass local project context, credentials, or secrets into the VM without reintroducing the very risks Coasty is designed to eliminate?

Also for every person you invite to Coasty, we’ll give you some free credits for the agents! Check out, Invites on our menu bar for more details after signing up.

Hi Product Hunt 👋

I’m the founder of PI SpeedLAB.
This project actually started out of frustration.

Every time we compared readings from traditional speed radars, we noticed inconsistencies. The hardware was expensive, but the processing power inside was surprisingly limited.

That made us wonder —
why are we still relying on tiny embedded processors when everyone already carries a supercomputer in their pocket?

So instead of trying to build a smarter speed radar…
we decided to build a smarter system.

PI-Link uses Doppler radar hardware, but lets the smartphone handle the heavy signal processing. That shift changed everything for us.

This is still a journey. We’re refining accuracy, improving filtering, and constantly testing against legacy devices.

If you’ve built hardware, worked on signal processing, or care about sports tech — I’d genuinely love your feedback.

What would you improve?
What would you question first?

Thanks for checking us out

And also check out our Kickstarter campaign : https://www.kickstarter.com/projects/1359404817/turn-your-smartphone-into-a-smart-speed-radar

love the hardware + cpu approach kevin. smarter than the legacy guns. saw the kickstarter link too—how is the conversion rate holding up from the ph traffic today? looks like a killer combo.

Hey Product Hunt! 👋 I'm Doruk, and today we're launching IPAware - an IP intelligence API that doesn't charge you per request.

The problem:
IP lookup services charge per request. You start with a side project, it grows, and suddenly you're paying hundreds for something that should cost a fraction of that. The metering creates anxiety and makes you second-guess every API call.

What IPAware does:

• IP geolocation (country, region, city, timezone, coordinates)

• Threat detection (VPN, proxy, Tor, datacenter detection)

• Security signals (risk score, threat level, recommendations)

• ASN/ISP data for connection context

• Clean JSON responses that work with any language
  

Why it's different:
Flat monthly pricing. $9.99/month or $99.90/year (2 months free) for unlimited requests. No surprises, no overages, no usage anxiety. Just predictable costs that let you build without worrying about the meter.

For developers:

• REST API, no SDK required

• Works with any stack (Node, Python, Go, PHP, and more)

• Low latency with edge delivery

• Simple integration (just HTTP + JSON)
  

Launch special:
For the first 24 hours, the Pro tier is free for a full month. Sign up and claim it from your dashboard – you'll get unlimited requests for the entire month. After that, you can choose to keep Pro or switch to the free tier (5k requests/month). Entirely up to you, no tricks.

We built this because we were tired of IP lookup bills scaling faster than our apps. Would love to hear your feedback and what you think!

Try it: https://ipaware.io

This is very useful given the new advent of A agents scrapping all across the internet

Congrats on the launch
For monitoring and security tools like this, a short demo video showing how alerts are triggered and how teams respond in real time can really help visitors understand the value instantly instead of trying to imagine it from text.
I specialize in SaaS/product demo videos, so I always notice this looks like a really useful tool for teams managing infrastructure. Wishing you a strong launch!

🚀 Free to start, no credit card required.
I've watched people struggle to build and wire up backends for 19 years. In that time I've led the builds of some of the largest products in the world including Prime Video for Amazon. The tools have changed, the languages have changed, AI has changed everything; and yet here we are, watching people vibe code a beautiful app in a weekend, feel like they've cracked it, then spend the next month figuring out what they've actually built and trying to turn it into something real.
Anythink gives you the backend first. A real one. Enterprise-grade infrastructure, built by someone who's had to care about scale, now accessible from day one, free, no credit card required.
We're seeing 40x faster development with fewer bugs. The same robust infrastructure that used to take a team of engineers, ready in moments.
Free plan includes:
✅ Auto-generated REST APIs
✅ Visual data modelling
✅ User management & access control
✅ Stripe payments
✅ Workflow automation
✅ File storage with CDN
And quite a lot more... would love to hear what you're building. Drop it in the comments 👇

Looks like you ream us hard, the moment we're bringing in money. Can I switch to simply using my own Stripe acct, host with my own hosting, such as resplit/hostinger/etc. and use my own SUpabase/etc. for DB and Github for the codebase?

I love to try tools like this, but am wary about lock-ins.

Best of luck with the launch!