PH热榜 | 2026-03-07

Hi everyone! I'm Yunhao, CEO and co-founder of TestSprite 👋

Thrilled to share what's new in TestSprite 2.1!

When we launched 2.0, the response blew us away. Thousands of teams told us the same thing: "This is the missing piece — we were shipping AI-generated code and just hoping it worked." That feedback pushed us to go even further.

2.1 is about making the MCP workflow faster, more precise, and enforcing it all the way to production.

🔌 Quick recap: TestSprite MCP

At the core of TestSprite is the MCP Server — it connects directly to your AI coding environment (like Cursor), reads your spec, and autonomously generates a complete test plan, writes all the test cases, executes them, and sends fix instructions back to your coding agent. AI validating AI — no manual testing, no prompting.

2.1 builds on top of that with three major upgrades:

⚡ 4–5x Faster Testing Engine

We rebuilt the AI testing engine from the ground up. What used to take 25 minutes now runs in about 5. Verification finally keeps up with the speed of AI code generation.

🖱️ Visual Test Modification Interface

If the MCP generates a test step that's slightly off, you no longer restart from scratch. Click into the step, see a live snapshot of exactly that moment in the test run, and fix it in seconds — swap the locator, change the interaction type, redirect the flow. Point and click. No code.

🔗 GitHub Integration — Automatic PR Testing

Once your MCP-generated tests are committed to your repo, the GitHub integration takes over. Connect via the TestSprite GitHub App (no workflow files needed) or GitHub Actions for custom CI/CD pipelines. On every pull request, TestSprite automatically runs your full test suite against the preview deployment — on Vercel, Netlify, Render, Railway, Fly.io — and the TestSprite bot posts a detailed pass/fail summary directly on the PR. Enable merge blocking to make sure failing code never reaches production.

The full loop: MCP generates and validates your tests locally → GitHub enforces them on every merge. Automatically.

If your team is building with AI, TestSprite 2.1 makes sure every PR is production-ready — without lifting a finger.

Free tier available. Give it a try and let us know what you think! 👉 https://testsprite.com

Thank you for the incredible support since 1.0 — we're building this with you, and we can't wait to hear your feedback 🙌

Excited to see TestSprite 2.1 live on Product Hunt! 🚀

Automating end-to-end testing is such a big pain point for many teams, and the idea of generating tests 10x faster with an AI agent is really compelling. The visual editor for adjusting test flows in plain English is a nice touch too—much easier than maintaining complex scripts. The new GitHub Actions integration also makes a lot of sense for CI workflows.

Congrats on the 2.1 launch, @jiao_yunhao @rui_li6 and team! 🚀

Really exciting to see the MCP workflow maturing—especially the focus on the verification layer. As more teams ship AI-generated code, automated test generation + enforcement in the PR workflow feels less like a “nice to have” and more like a necessary safety net. The 4–5x speed improvement is particularly impressive since verification often becomes the bottleneck when coding agents get faster.

One question I’m curious about: how do you see TestSprite evolving as AI agents move from generating code to operating longer-running autonomous workflows? Do you envision MCP extending beyond code verification into broader agent behavior validation?

Looking forward to trying 2.1 and seeing where you take this next. 👏

Strong launch — the MCP→PR merge-blocking loop is exactly the part most teams miss. One thing I’d love to see: a “flake profile” per repo (dynamic selectors, async data, auth redirects) so teams can separate true regressions from environment noise before blocking merges. If you already track this internally, exposing it in PR comments would be a killer differentiator for AI-native QA.

Congratulations, looks great!

The 'missing layer' framing is exactly right — the real unlock isn't test generation speed, it's closing the 'AI codes but nobody validates it actually works' loop at the workflow level. The hardest part of AI-native testing is the semantic oracle problem: when AI generates code, the ground truth for 'correct' is often fuzzy. Does TestSprite handle purely structural/functional assertions (output matches expected), or does it also do semantic validation — e.g. 'this UI flow still makes sense for the user'? That gap is where most AI testing pipelines quietly break down.

Great idea. Tools that focus on testing and debugging are going to be essential as AI-generated code keeps growing.
Great developers write code.
Elite developers debug reality.

Testing is the most boring part for me, but also the most valuable! It validates your core functionality; I love the idea :)

Looks like you're tackling a real issue with AI coding tools. Generating code is easy now, but validating that it actually works is still messy.

Really like the idea of automatically generating tests and enforcing them in PRs. The visual test editor also seems like a smart addition instead of constantly rewriting tests.

Curious if most teams use TestSprite mainly with AI coding tools like Cursor, or also in more traditional development workflows?

Looks good, but damn, I did the exact same thing 9 months ago: https://www.producthunt.com/products/testr-qa-solution-for-vibecoders/launches/testr - it even had an MCP server.

I have 0 clients, 1 comments, 8 upvotes.

I guess its distribution issue? Sad still

How does TestSprite handle complex dependencies and state management when automatically generating and executing tests across both frontend and backend environments?

Definitely going to check this out. The test generation can be a HUGE time saver.

hey,

right now Testsprite looks like an AI testing tool, but the stronger narrative might actually be AI verifying AI-generated code.

That framing could make the use case much clearer for teams building with AI coding agents.

The GitHub integration that auto-runs tests on every PR and blocks merges on failure is a game-changer for teams moving fast. That's the exact guardrail most small teams skip because setting it up feels like overhead, and then they regret it in production.

The visual test editor where you click a step to see a live snapshot is a really smart UX decision too. Debugging test failures has always been painful because you're reading logs blind, this closes that loop nicely.

Congrats on #1 today! Rooting for you.

Do you folks support mobile apps?

The GitHub PR blocking is the feature that makes this actually useful in a real team environment — without that gate it's just a nice to have that gets skipped when deadlines hit. Curious how it handles flaky tests that fail intermittently on the live preview environment due to timing or network issues rather than actual code problems. That's the thing that erodes trust in automated test suites fastest — if it blocks a merge for a false positive once, developers start ignoring it permanently.

I've been testing my AI pipeline by hand, and honestly, I'm over it. 4 different providers, each returning slightly different JSON structures; writing test cases for each takes forever. Does TestSprite handle non-deterministic outputs well? That's what always gets me with TubeSpark's generation flows: the same prompt gives you different shapes every time.

The MCP-to-GitHub loop is the real unlock here. Most teams I've seen still treat testing as a separate step after shipping, not something baked into the PR itself. The merge blocking on failed tests alone would save hours of debugging in production. Curious how the visual editor handles dynamic content that changes between test runs.

I stop trusting agent loops when the test result and fix step drift apart. TestSprite MCP reading the spec, generating the test plan, then sending fix instructions back to the coding agent feels useful, especially if each round keeps a visible trace you can audit.

Have you implemented any sort of caching or parallelization strategies within the Testing Engine to mitigate the increased load and potential slowdowns that might occur when dealing with large-scale, complex AI-generated codebases?

After dealing with those npm postinstall attacks lately, seeing an agent that actually validates findings is a massive relief. Most tools just spam false positives until you kill the notifications. Does this catch transitive dependency stuff too, or just first-party code?

Hey PH! I’m Brianna, co-founder of Copperlane (YC W26)!

The Problem
Lenders spend $11,800 to originate a mortgage, and the majority of that cost is burned during intake due to missing documents, back-and-forth, and costly errors. This manual process breaks down on almost every loan, across 8 million loans a year.

Existing tools fall into two approaches:

❌ Legacy portals – These systems are decades old, and loan officers need to chase down borrowers for docs manually.

❌ AI built by older mortgage teams – Our competitors are all mortgage people trying to build AI. We’re the only AI people learning about and building for mortgage. :)

My co-founder and I both come from mortgage families (Freddie, Fannie, FHA), so the space is personal to us and we built Copperlane to fix it.

How Copperlane is Different 🚀
Copperlane is an AI-native loan origination system powered by our agent, Penny, who behaves like a real loan officer.

🔸 Penny handles intake proactively – She pulls borrower docs, reads them, and checks the borrower’s loan eligibility.
🔸 Guides borrowers through the application – Penny answers questions and proactively helps borrowers complete the loan application correctly.
🔸 Proactively fixes issues – Is the borrower missing a paystub? Or reported conflicting income? Penny flags it immediately and follows up with the borrower.
🔸 Delivers clean files to lenders – Loan officers receive clear, organized files and can focus on bringing in new loans.

Who this is for
If you’re a lender or loan officer dealing with slow intake, Penny helps you close loans faster without extra headcount.

📎 Get started today
We’d love to show you what Penny can do! You can book a demo at here.

Carpe Diem,

– Brianna & Athan

Congrats @athanzhang @brianna_lin ! Great work!

This is pretty damn cool. How did you guys land on this idea?

How does Penny handle complex or non-standard borrower scenarios, such as self-employment income or inconsistent documentation, while ensuring full compliance with mortgage lending regulations?

this is awesome! i came from linkedin :) i was wondering whats the most exciting thing you’re looking forward to regarding copperlane’s launch?

The document verification piece is what catches my attention — that's historically been the biggest bottleneck in loan processing, not the rate calculation. Curious how Penny handles edge cases like inconsistent income documentation for self-employed borrowers, that's usually where automated systems fall apart and a human has to step back in. If you've solved that reliably this could be genuinely transformative for smaller lenders who can't afford large underwriting teams.

The "AI people learning mortgage" vs "mortgage people trying to build AI" framing is sharp and honest. Having Penny proactively flag issues like missing paystubs instead of waiting for a loan officer to catch it manually is where the real time savings happen. Congrats on YC W26, Brianna and Athan.

What usually kills these flows is the resend-this-doc loop. Copperlane looks strongest where it flags edge cases before underwriting, because W-2 auto-fill matters less than giving ops one clean place to clear missing docs and big deposits.

It solves a real pain point. I am curious to know how your underlying system is going to manage the high amount of context so that it does not hallucinate. Amazing work!

@GitHub UI is cluttered. @rhyssullivan just made it simplifier and faster. LFG

A read-optimized UI that never hits the GitHub API on page load makes FasterGH more than a visual cleanup. Does it keep that speed once review threads and check runs pile up? That is where GitHub usually starts to feel bloated.

cool

One thing stood out: it’s positioned as a component registry, but the product actually looks closer to AI product infrastructure (ui + agents + sdk)

Framing it that way might resonate more with teams building AI apps

Had a few shadcn installs turn into local forks fast. 21st pulling components, blocks, and hooks into one stack feels useful, and a really clear diff or version flow would make that npx shadcn path hold up once teams start customizing.

This is very handy for vibe coding and I imagine there will be many others like this around going forwards. Great job!

Great idea! On my screen a lot of the scrolling boxes are empty. Is that just a loading issue on my end, or is it just that way as more and more content gets populated?

How does this tool handle constrained briefs and existing brand systems when generating variations, rather than just open-ended exploration?

Love the "show don't tell" angle here — most design tools are still stuck in the prompt-and-pray loop, so just scrolling through generated directions feels like a much more natural way to find a vibe. I could see this being especially powerful early in a project when you don't yet have the vocabulary to describe what you want. Curious how it handles more constrained briefs though, like when you already have a brand system and just need variations within it rather than truly open-ended exploration.

very promising, with improved performance, it could be my go-to for inspiration !

Variant turning idea generation into scroll instead of prompt rewrite is smart. A pin and compare flow would make it stick, because endless options only help if you can narrow fast and carry one direction into HTML or React cleanly.

What can we design with it?

Thats interesting. What dod you use to build it?

Love that this runs fully local with no per-generation pricing. The open-source NLE + on-device AI combo is exactly what creators need to experiment freely without worrying about cloud costs.

How are you thinking about LTX Desktop on sub-32GB GPUs, especially now that the pitch is fully local and open source? That support path feels like the difference between a cool demo and a real daily tool for a much bigger group of creators.

Hey Hunters, I am excited to hunt LTX Desktop today! 🚀

If an engine is truly powerful, it should enable real products to be built on top of it — and that’s exactly the idea behind LTX-2.3.

LTX Desktop is a fully local, open-source video editor running directly on the LTX engine, optimized for NVIDIA GPUs and compatible hardware. That means:

• No mandatory cloud dependency
• No per-generation pricing
• Your data stays on your own device

The engine is the hard part — the interface shouldn’t be. And now it’s yours, free.

LTX Desktop is a great demonstration of what the LTX engine can already power, and more importantly, what developers and creators can build with it.

Curious to hear what the community thinks and what you would build on top of this engine. 👇

The on-device AI generation with no per-gen pricing is compelling, but curious about the minimum hardware sweet spot for smooth workflows, What's the min GPU spec you'd recommend for this?

There is an increasing population of users with Apple devices with those specs and your listing mentions "runs locally on your machine" Apple(MacOS) Excluding. You may want to mention that.

What specific SDK or API capabilities does the LTX-2.3 engine provide for developers who want to build their own specialized video tools or integrate the engine into existing creative workflows?

Doodle explainer videos are a solid format for breaking down complex topics but they usually require a designer or a studio. How much can the output from Thinking Line be adjusted to match a specific topic and brand?

Getting doodle video and vector export to stay in the same style is the hard part here. Does Thinking Line keep strokes and scene pieces editable after generation, or is the first win speed? That edit loop is what would make it useful beyond one-off assets.

It looks fantastic! Congratulations and best of luck with the project! I truly adore the concept and the finished video! Just so you know, the pricing section's "Get Started" button is broken

How does Thinking Line ensure that the generated doodle-style vectors maintain clean, layered paths for seamless editing in professional design software?

The use case for explainer videos and SVG generation from prompts is really compelling, especially for content creators and educators who need visual assets fast without design skills.

The doodle style is a smart differentiator too. It's warm and approachable in a way that polished AI renders aren't, and that actually makes it better for social content and tutorials.

Would love to know: how does it handle more abstract or conceptual prompts where there isn't an obvious visual metaphor? That's usually where text-to-visual tools struggle the most.

Congrats on the launch!

Interesting concept

One observation: the product looks positioned as a doodle generator, but the bigger opportunity might actually be AI for explainer storytelling

That framing could make the use case much clearer

Unique endpoints that push form data straight into Google Sheets make sense for Webflow and Carrd. What would make this really stick for lead gen is built in spam control and source capture, so the sheet stays usable once real traffic starts.

No login required + clean Tailwind CSS output is such a smart positioning decision. The number of times I've needed a quick form for a landing page or static site and didn't want to set up a backend or deal with an embedded third-party widget is genuinely too many.

Supporting React and Vue export alongside plain HTML gives this real legs for developer use cases too, not just no-code folks.

One question: does it support custom validation rules or is it purely visual/structural for now?

Solid launch, congrats!

Interesting concept

One thing that stood out: the product is described as a spreadsheet API, but the stronger angle might actually be rapid backend for prototypes and internal tools.

That positioning might resonate more with developers.

I've got to say, I love the creativity. When I get a macbook again i'll hopefully remember it but bookmarked for now.

How are you thinking about app exclusions and auto-expiry for NotchPad's clipboard history, especially around password managers and one time codes? That part decides whether a local, encrypted utility feels safe enough to leave running all day.

Congrats on launch 🚀
Curious how you’re managing internal workflows as traction picks up?

Hi everyone!

AI2’s new Olmo Hybrid is the first hybrid 7B that clearly beats a pure transformer baseline (Olmo 3) in a fair fight.

Same size as Olmo 3, trains at the same speed, but matches its accuracy with half the data and crushes long-context evals. The 3:1 RNN+attention mix just works.

Super clean weights on HF. And you can even run the full model 100% locally in your browser on WebGPU!

The spend analytics caught my attention more than the invoice collection itself — so many companies only find out what they are actually paying when someone sits down and goes through bank statements. Does GetBeel work only with Gmail or are you planning Outlook too?

This is pretty cool. Anytime I used e.g. Pliant, I had to do things manually, which took time.

How does it pair invoices with the right transactions?

This is great idea! The "invoice hunting" problem is surprisingly universal – especially for small teams using dozens of tools. 🤯 Do you see GetBeel staying primarily an automation tool for accountants, or evolving into more of a financial visibility layer for founders?

Right now Getbeel reads like an invoice extraction tool, but the stronger angle might actually be automating month-end financial ops for startups

that framing could make the value clearer for founders

Tried wiring this kind of flow once, and the ugly part wasn't extraction. Getbeel tying Gmail capture to exact, fuzzy, and unmatched bank statement review feels right, because accountants trust the exception queue, not the happy path.

Love the idea of having a real-time AI wingman during calls.

High-stakes conversations are exactly where people forget key questions or lose context. This feels like a very practical use of AI. Congrats on the launch!

The positioning as a “wingman for conversations” is great.

Calls are often where the most important decisions happen, and having contextual prompts in the moment could make a huge difference. Nice work on the launch!

Huge congratulations on the launch of Shadow!

​The 'AI Winman' concept is such a refreshing and highly personalized take on meeting tech. While most tools focus on what happened before/after the call, having an assistant that provides real-time nudges and intelligent suggestions during the conversation is a massive step forward. It really feels like having a digital partner in the room helping you stay on track and hit your goals in the moment.

​Love the focus on making conversations smarter rather than just documenting them. Can't wait to see how this evolves, definitely a game-changer for anyone who wants to level up their communication!

Just checked this out and the core idea makes a lot of sense, tools I use now kick in after the meeting with summaries and transcripts. Great that you're tackling the harder problem of helping during the conversation itself.

Congrats to the team on the launch.

This will be incredibly useful for scheduling multiple meetings and will save a significant amount of time. I definitely want to start using this.

Multi lingual support is a great idea

Great idea! What's the pricing?

I tried it for my meetings, it's giving really good results.

The fact check is specifically helpful for me.

@shubham16180 @mayank_gupta40 Are you also going to launch it for enterprises or is going to be for the personal use only ?

I can already imagine this quietly running in the background of all our team syncs 😄

Kudos on the launch @shubham16180

Hey everyone, excited to launch Shadow today.

We built Shadow because important conversations often go wrong in small ways: you forget one critical question, miss context in the moment, or leave the call without a clear next step.

Shadow acts like a real-time wingman during conversations. It helps you stay sharp while the call is happening, so you can ask better questions, capture what matters, and walk away with clearer outcomes.

We’re especially looking for feedback on this:
What is the one thing Shadow should do extremely well for you during a call?

If you try it out, I’d love to hear where it felt useful and where it broke.

Really excited to launch TicketX on Product Hunt today 🎟️

Hidden fees at checkout are one of the most frustrating parts of buying tickets, which is why we built TicketX to show the all-in price upfront.

We’d love to hear your thoughts and feedback. Thanks so much for the support!

Hi, Product Hunt 👋

We’re excited to finally share TicketX with you today!

# Motivation
Buying tickets online often feels frustrating. You find what looks like a great deal, only to be surprised by hidden fees at checkout. On top of that, it’s especially hard to find tickets for Japanese concerts and sports events on global resale platforms.
We wanted to fix both problems.

# What makes TicketX different
TicketX is a no-hidden-fee ticket marketplace for concerts and sports across the US, Canada, and Japan. What you see is what you pay. No last-minute add-ons 🤝

We’re also backed by one of the largest ticket resale platforms in Japan, which allows us to offer a wider selection of Japanese tickets, including events that are often hard to find outside Japan 🇯🇵

# We’d love your feedback
TicketX is still evolving, and we’d love to hear your thoughts.
What’s the most frustrating part of buying tickets today?

If you read this far, I have a gift for you:

🎁 Coupon

🎟️Use PH10OFF for 10% OFF — valid until March 31st, 2026!

Thank you!
TicketX

It looks interesting! Is it similar to OpenClaw but for mobile or something more specific?

Hey Product Hunt! 👋

I'm Anoy, and I’m back with the team to show you how we’ve completely reimagined Deforge.

When we first launched, our goal was to make AI agent building accessible. But since then, we’ve listened to thousands of you and realized that even "no-code" can sometimes feel like "too much work." You shouldn't have to be a logic expert to build a genius agent.

Today, we’re launching the new and evolved Deforge, featuring a massive architecture overhaul and a UI that feels more like a conversation than a configuration panel.

What’s New in this Version:

1. Chat to Build 💬: Don't want to drag nodes? Just tell Deforge what you want. "Build me an agent that tracks AI news on Reddit and Slacks me a summary every morning." Our AI will architect the logic for you instantly.

2. Total UI/UX Redesign ✨: We’ve stripped away the clutter. The new interface is faster, cleaner, and built for speed, making it easier than ever to visualize your agent's "brain."

3. Major Architecture Update 🏗️: Under the hood, we’ve rebuilt our core to be more robust. Agents now run faster, handle more complex multi-step workflows, and integrate more deeply with your data.

4. Enhanced Multi-Model Support 🤖: Switch between GPT-5, Claude 4.6, and Gemini 3.1 on the fly within a single workflow to get the best performance for every specific task.

The team (Shamba, Gautam, and myself) will be here all day. We’ve poured our hearts into this update and can’t wait to hear your thoughts.

What’s the first agent you’re going to build?

Check it out and let us know: https://deforge.io/

Hey Product Hunt! 👋

I got ghosted after spending 10 hours building a mockup. That was the last time.

The #1 reason web design deals die isn't price — it's that clients can't visualize what they're buying. So I built Thyonix to solve that in a completely different way:

Here's what it does in ~5 minutes:

🔍 Finds businesses in your city that don't have a website yet

⚡ Builds them a custom AI demo site automatically

📩 Sends a personalized pitch via email or WhatsApp with their site attached

You show up to the conversation with the website already built. The deal practically closes itself.

In our beta:

1. 100+ sites generated

2. 10+ agencies and freelancers using it

3. Deals closing in under 24 hours

Try it: generate a free demo site for any local business in your city. Takes 3 minutes. You'll immediately get why this works.

Would love your feedback — what would make this a no-brainer for your workflow? 👇

Loved the flow, congratulations on PH launch! kudos to the team!

Love the 'family politics' angle—seating drama is definitely a real thing! Does Tablex allow us to color-code guests or tables to help visually separate those 'feuding aunties'?

@noumanbinsabir Congrats on the launch! I really like the idea of removing friction from calorie logging. Using the camera as the main interaction feels very natural for this kind of product.

From a UX perspective I'm also curious about the transparency of the AI estimation. Many people still log meals manually because they want to trust the numbers.

It could be interesting to show how the estimate was calculated, for example ingredient breakdown or portion assumptions, so users feel more confident about the result.

Also curious how the correction flow works if the AI estimate is slightly off, for example the dish or the portion size. Being able to quickly adjust the result seems really important for trust

Solo dev shipping a complete AI food scanner is impressive. The camera-first approach is right, logging meals manually is exactly why most people quit tracking within a week. Curious how it handles Indian food where one plate has 5 different items mixed together. Congrats on the launch, Nouman.

Excited to have everyone on board