PH热榜 | 2026-03-30

@Notion MCP brings your AI tools directly into your Notion workspace acting as a real-time bridge for tools like @Claude by Anthropic, @ChatGPT by OpenAI, and @Cursor.

It solves the gap between scattered data and generic AI outputs by making AI context-aware with your actual notes, docs, and workflows. Unlike traditional integrations, MCP is built specifically for AI agents... fast, seamless, and no complex setup.

Key features include real-time read/write access to Notion pages, support for multiple AI tools, admin controls, and workspace-level governance.

Perfect for developers, product teams, researchers, creators, and personal productivity - use it for documentation, roadmaps, research organization, content planning, and more.

Get started:

• https://www.notion.com/help/notion-mcp

• https://developers.notion.com/guides/mcp/mcp

• https://github.com/makenotion/notion-mcp-server

• https://developers.notion.com/guides/mcp/get-started-with-mcp

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

Congrats on the launch! My team uses Notion as our main knowledge base, so the idea of having AI actually read and write to it in real time is pretty compelling - instead of copy and pasting context every time you want something done. How does it handle permissions across a team workspace and can you control which pages or databases each AI tool actually has access to?

Perfect timing! We launched our MCP and Skills today as well, and this evolving ecosystem is making it possible for agents to collaborate through MCPs to deliver intelligent, multimodal content localization. Thank you to the Notion team for introducing official MCP support!

Interesting direction.

We’re starting to see AI agents move from tools to orchestration layers.

We’re building something similar in travel an AI agent coordinating real-world services like transfers, restaurants, and experiences.

Feels like this is where AI is heading.

Does the MCP structure materially change how many Claude AI credits are used for the average user (who uses primarily Chat and Cowork, not Code)? Or is it more determined by the context window?

Good news: For Claude connector with Notion, it seems they now have the following write/delete permissions that one can set to always approve, rather than requiring approval from the user.

It might be too trusting to enable them all at first, but I think that might solve some of the frustration shared in the video!

The launch of Notion MCP feels a bit late.

As another core productivity SaaS, Google Workspace already introduced the GWS CLI earlier this month.

That said, I’m a loyal Notion user, and I’ll definitely consider using MCP.

Been using MCP servers since they came out and it's wild how fast the ecosystem is growing. The fact that Notion went official with this instead of leaving it to third-party wrappers is a good sign. Curious — does the read/write access work at the database property level or is it more page-level? For task management setups with lots of relations and rollups, that distinction matters a lot.

Real read/write MCP access to Notion databases is actually useful for sprint planning workflows - I can see agents updating task status, pulling blockers from a database, writing standup summaries back to a page. The "turning scattered data into actionable workflows" bit is doing a lot of work though. What's the write latency like in practice? Notion's API has rate limits that can make write-heavy agent loops painful.

How does it differ from actual Notion MCP?

We knew this was coming. Does this handle large posts, earlier we had to go the chunking API route, does MCP have something similar?

Already using from Codex, highly appreciate for the support of that, a lot of work has been simplified within documentation and task tracking and updates across different systems in our ecosystem. We are using it to update the metadata of our datasets, provide columns description, buckets/s3/db paths, special params, previously all the hard work was done using manual source-to-table porting, but now its all done by AI and we are happy with the results

didn't Notion already have an MCP for a while? a bit confused what's new with this launch?

Upvoted and downloaded! Will see how it goes.

Do you see the possibility of integrating this with other task managers? One bane of them is not having something "universal" that connects through systems/platforms to keep things truly organized.

Hey @lilhadi -- as someone who has tried just about every to-do app, I really dig this. Really simple + the natural language support is killer. Two requests/comments: Thought about adding a system wide shortcut to initiate the menu bar pop down? Also, TAB support on the pop down to switch between fields and CMD + Enter to add new a new task? Feel like that would make it even faster to use. Nice work!

this is so clean. curious about Focus Mode: what does it actually do? does it hide everything and just show the one task you're on?

U know I maintain the tasks in notepad just because its always available quickly, but this will be a game changer for sure in long term!

Looks solid for quick task entry, but I'm curious about how well it handles context. What happens with tasks that require follow-ups or details?

I wish I'd had this when I was using my Mac! Excellent work. By the way, do you have plans to release a Windows version?

One comments, I always got some picture with schedule, it will be great if I can paste the picture and popTask can OCR and find the schedule ,setup everything

The "just type whatever and it figures it out" approach is underrated. I've been building a focus timer app and the biggest UX lesson I keep learning is that fewer inputs = more people actually use it. Does the on-device AI handle stuff like "call dentist sometime next week" or is it more structured than that?

Since the task data is just a local JSON file on the Mac, have you thought about documenting the format so people can pipe tasks in from scripts? Something like `echo task | poptask add` from the terminal. Keeps the no-cloud thing going and opens it up to the developer crowd without touching the main UI.

the "mtg w team mon wed fri 9am" example sold me, that's exactly the friction point, most task tools make you format your thoughts before they'll accept them.

does it handle recurring patterns from freeform input? like if i type "standup every weekday 9am" does it set up a repeat automatically?

Appreciate it.

Will This be available for mobile platforms? Syncing with mobile will gain you more benefits.

See am commenting using phone 😊

Is it connected with Google Calendar or Calendly?

This is going to be so handy to manage tasks. I'll try it out.

Congrats on the launch 🎉

Hey PH! I'm Jeff. I built Goals because every productivity app I tried gave me more work to manage instead of less. I wanted something where I could just open the app, see what to do today, and get on with it.

The idea is simple: you type in a goal, AI breaks it into a realistic plan, and each day you see one action. You can still check in and see your lists, but the focus is on action over Type-A nonsense, just the next thing.

I'd love your feedback on what's working and what's not. Try it and let me know what goal you set! 🚡

The one-action model sounds clean, but I’m curious who decides what that action is? If I type “get healthier,” the AI has to make a lot of assumptions about what I actually mean. How much does the output depend on how well someone writes their goal? And what happens when the goal shifts over time? Does the AI course-correct? Congrats on the launch!

Hey Jeff. I really like the idea and the minimalism here! I think this will turn those long "planning sessions" into a simple "do this today", which is what actually drives results.

One thing I noticed: The pricing structure isn't clear yet. I don't see it either here or on the website. The app and play stores say it's free with in-app purchases, which is typical of subscription models with a free trial. Is that what you're going for? It would be great to see this upfront, and it will definitely prevent user suprises and churn.

Congrats on the launch!

This is a great tool to accomplish tasks.

All the best on your launch!

Any plans to attempt to integrate this with established task managers?

HI Jeff @jeffalgera Happy to be one of Goals early users! the UI design is really friendly to people like me who need schedule my goals specific and detailed. I used to waste a lot of time to sub-schedule my goals which is the pain point for me and Goals deal！
Before goals I used Rosebud which actucally is a pretty journal app which record、communicate as well as plan the goals. Now I think I have a better choice to arrange my daily routine and goals. Thanks you guys🌹

It's a pretty good product.

this would be so useful if I can create a separate work workspace, it would read tasks from my cal for this week, help me make progress and optionally create small achievable milestones by blocking time in my cal. then it would notify me of progress or required reschedules. And then I'd like to do the same for my personal cal and personal tasks. Looks great nifty tool !

The idea of AI turning goals into to-do lists sounds great, but the official website looks too basic, which is a bit confusing.

Hey Product Hunt!

I'm Dawson, co-founder @ Letterbook.

We grew our last product, Martin, to 100k users. At one point, we'd get 50 support requests every day:

• "How do I use this feature?"

• "Can I get a refund?"

• "I can't log into my account."

I had to hire a virtual assistant in the Philippines to reply to users and solve billing issues.

I had to use Zendesk, which was built 20 years ago and designed for a VP of Support, not a founder.

None of this made sense, so we built our own modern, AI-native support platform. We integrate with your database, knowledge base, and Stripe, and our AI auto-prepares resolutions to every ticket.

We’ve already replaced Intercom, Zendesk, Zohodesk, Helpscout, and Freshdesk for many fast-growing companies in YC.

If you’re building a B2C or self-serve B2B product, give it a try or book a demo at:
https://letterbook.ai

P.S. Unlike Zendesk, which takes days to set up (a friend literally hired a consultant to teach him), Letterbook takes 5 mins. And if you need help, you better believe we have top-notch customer support!

Went through the signup flow. The 15 minutes to first resolved ticket claim on the landing page is the right hook, that's exactly what a founder dealing with support volume wants to hear.

The gap is that Signup asks for the organization name before showing any evidence that the 15 minutes is real. For a founder evaluating whether to commit, that's a small friction at exactly the wrong moment. The stronger move would be showing one resolved ticket in a demo state before asking for anything, let the claim prove itself before the founder has to name their organization. What does drop-off look like between landing page and completed signup?

Dawson this is so real. We're two founders building WTMF and support has been one of those things that just quietly eats your time. Never felt like Zendesk was made for teams like ours.

The auto-prepared resolutions thing is what caught my eye. Going to try it out. Congrats on the launch and honestly 100k on Martin is a tough act to follow, excited to see where Letterbook goes.

Congrats on the launch! Support is one of those things that sneaks up on you, and can eat away at your hours. Once you connect Stripe, how much customer context does the AI actually pull in when it's drafting a reply?

Support tooling for founders is usually an afterthought - either paying for Intercom before you have revenue, or handling everything manually through email threads. The AI angle makes sense here. What does Letterbook do when it doesn't know the answer - does it hand off to you, or does it try to figure it out?

Good one @darweenist I want to ask if there is a provision for seamless migration from existing support platforms to Letterbook. This will be a game changer for companies with large user database on their existing platforms instead of starting over. The homepage could also do with some relevant FAQs. Good luck with the launch.

Letterbook is the best support platform ever. We love it at Respan. Congrats on the launch Dawson and team!!

Congrats on the launch! Happy to be a paid customer. Letterbook is a game-changer (and we've tried and paid for zohodesk, help scout and freshdesk in the past)!

I'm exactly your ICP (as the founder who is replying to every single CS ticket)!

I actually chose Zendesk because it also supports a Help Center. Does Letterbox also come with a Help Center? If not, what are the usual ways you see your users setting them up and connecting to Letterbox?

The Stripe + database integration is what sells this for me. Running a B2B SaaS with self-serve billing, and 60% of our support tickets are "where's my invoice" or "cancel my subscription" — exactly the kind of thing AI should handle autonomously. How does the AI handle edge cases where it's not confident? Does it escalate to a human automatically or does it attempt a response anyway?

So cool. Is it possible to close the loop from tickets -> code?

Congrats! Trying it now...

Congrats on the Launch, this is one of those products where I wish I knew about it way sooner!!

Zendesk is v bad

Congrats Dawson!! That's incredible that it can even process the refunds on stripe as well. Excited to try it out!!

Congrats on the launch, Dawson!

Hi everyone!

I only got into FreeCAD recently, and what surprised me most is how powerful it already feels for a tool that is, quite literally, free.

FreeCAD 1.1 is a HUGE release. There is a lot of real workflow polish here: transparent Part Design previews, interactive draggers for tools like Fillet and Chamfer, three-point lighting, Clarify Selection, Assembly and FEM improvements, and a completely new CAM tool library system.

I am also used to the SolidWorks style of interaction, so it was great to see FreeCAD support that navigation style directly.

For a free and open-source tool spanning CAD, CAM, and FEM, this honestly no longer feels meaningfully behind expensive commercial software in the ways that matter most day to day.

Saw this update last week, hadn't checked it out yet! Congrats on bumping that number haha.

congrats on the 1.1 release! the combination of transparent previews, interactive draggers, and the new CAM tool library in one update is a big deal. open source CAD closing the gap with SolidWorks genuinely matters for hardware startups who can't afford the $$$ licenses :)

the board and sandbox together make sense, but after a few iterations things usually start drifting a bit, do people keep updating the board or does it get ignored over time?

This is exceptional, I was using clickup and the claude code as well as the Xcalidraw to manage my projects, an all in one solution was much needed thanks man!

The Boards + dependency mapping before agents start building is a legitimately different take. Most coding agents are just "here's a prompt, generate code" - having explicit task ordering baked into the IDE means the agent isn't deciding what to tackle next based on vibes. How does it handle when a dependency is partially built? Does the dependent task queue or does the agent try to work around it?

the "draw dependency arrows, hit Build" approach is interesting, most agentic IDEs just hand the whole task list to the model at once and hope it figures out the order.

does the dep graph actually affect parallelism, like does it run independent branches concurrently or is it always sequential?

This looks awesome, congrats! Any plans for team collaboration features or is it mainly solo for now?

I like the visualization on dependencies and data flow. This is personally something I plan out before rolling out a feature, and it takes time. Looks interesting and useful.

Visual planning boards for agentic coding is a good call. The hard part isn't getting agents to write code - it's defining what to build clearly enough that they build the right thing. Multi-model support across Claude, OpenAI, Google, xAI, and Ollama is a lot to maintain but the flexibility is real. If you want to catch bugs while launch traffic is up, we built a free community stress-testing tool called VibeFix Playground. Post your URL, people try to break it, reports come in with screenshots.

Hey Product Hunt! 👋

We've been quietly building something we believe the developer community has been missing: A proper localization layer for the agentic era

Why we built this:

Every AI agent, every app, every workflow is still English-first by default. Getting to 240+ languages means stitching together 5+ APIs, managing file conversions, handling dubbing, subtitles, and i18n files separately, all while keeping quality consistent. It's a mess 🤯.

🎁 What Ollang MCP Skills API & SDK does:

→ One API call to localize any file type — video, audio, DOCX, PDF, SRT, JSON etc.

→ Native MCP/SKILLS integration — Claude Code, Cursor, Cline, Codex and 15+ agents can localize files directly from their workflow

Begin like this now:

npx skills add ollang/skills

or like this:

claude mcp add --transport http ollang https://mcp.ollang.com/mcp

What we'd love your feedback on:

- Which agent integrations matter most to you?

- What file types are critical for your localization workflow?

- Would you use this for a personal project, startup, or enterprise?

We're answering every question today. Drop your hardest localization challenge below, and we'd love to solve it with you.

Get started free → ollang.com

Hey Product Hunt 👋

AI agents are everywhere, but localization is still fragmented, manual, and painfully multi-step.

We built Ollang MCP, Skills, and SDK to fix that.

→ One API to localize any file type
→ Works directly inside your agent workflows
→ Built for real-world complexity (video, audio, docs, i18n — all in one flow)

Curious, what’s the most painful part of localization in your current stack?

Localization is one of those things that gets shoved to the end of every sprint and done badly. The MCP + SKILLS approach for agent-driven workflows is interesting - what does a typical localization workflow look like when the agent runs it end-to-end? I'm curious how it handles context (strings that need different translations depending on UI placement) vs just raw key-value substitution.

Localization is one of those things that always gets pushed to "later" and then becomes a nightmare when you finally need it. How does it handle context-dependent translations? Like in our app, the word "network" means something very specific - does it learn domain-specific terminology or do you need to manually define a glossary?

This is exactly the kind of product that makes you think 'why didn't this exist before?' Multimodal localization with MCP + Skills + SDK in one platform is genuinely elegant architecture. The developer experience looks clean, and the enterprise angle is well thought out. Congrats on the launch @mazula95 — the future of AI-native localization is in good hands. 🚀

Congrats on the launch, Aziz!

Localization has always been one of those things that feels like it should be a one-liner but ends up being 5 different tools duct-taped together. Can't wait to finally having something work natively inside agent workflows instead of as a separate step is a big deal! Nice work 🚀

Long time Ollang user! Congrats on the launch @mazula95 !

I’m curious, how do you handle iteration and feedback loops? Like when translations get revised multiple times (by PMs, local teams, etc.), does the agent learn from those edits over time or is each run stateless?

the "english-first by default" framing is spot on. every ai workflow i've seen just assumes english and offloads the rest to a separate team or pipeline.

curious how the SKILLS layer works in practice across different agents, is it framework-agnostic or does each one need its own wrapper?

So proud of you for shipping this! 🎉 Honestly, it is such a clever move. To answer your questions: Cursor and Claude Code are definitely the most critical agent integrations for my workflow right now. As for file types, handling JSON for i18n directly within the workflow without breaking the structure is a lifesaver. Congratulations! 🚀

Love this! feels like something that should’ve existed already.

Curious about one thing: how do you handle quality consistency across very different modalities (e.g., subtitles vs. dubbed audio vs. structured JSON)? Is there a shared evaluation/QC layer, or does it vary per file type?

Impressive work on MCP Skills!

If I want to make my app available in whatever language users want, is this something Ollang can help with? I've noticed llms are good at some languages, and less fluent in others, which can make for a non-uniform experience for users

Hey Product Hunt! I'm Samuel, the developer behind Bluor.

The hardest part wasn't the ESP integrations or the AI pipeline. It was making the output genuinely look good. Most AI email tools are functional but visually mediocre. We didn't ship until we solved that.

We'd love your feedback — what ESPs or integrations would you like to see next? We currently support Klaviyo and Mailchimp, with more on the way.

Happy to answer anything technical.

Incredibly useful tool, but how does Bluor handle complex layouts or dynamic content? Sometimes one prompt oversimplifies the nuanced needs of diverse campaigns.

I just tried this. Works really well. The number of credits for the free plan we are all going to start with to test this out seems low. I created my email, had to make a couple of edits to correct the text chosen and I've run out of credits. There does not seem to be a way to simply edit the suggested text. I had to tell it exactly what to change. Burning through an unknown amount of credits. Pricing for the first monthly tier seems quite expensive if, based on my free test, I will probably only be able to generate 2-3 emails. Your Pro tier works out to about $31 Canadian. For casual users I'd suggest another tier between free and pro. This tool seems really, really great (I love what it did for my test) but like all tools that rely on a credit system, you never what that really gets you until they're gone. Now I'm not sure what to do with my first seemingly uneditable email that is 90% finished.

This is surely a pain point I've been experiencing. I even thought of building something like this, thanks for not making me go through that phase :) Looks super cool. Does this also handle weird email processing of older outlook versions and whatnot?

Good job on the new Bluor launch, @santifarre! Changing from drag-and-drop to prompt-based design is a good move. Email tools should have done this many years ago.

I visited your website. One thing caught my eye. Your homepage says: "Stop wasting hours on email design." Then it shows this example: [A welcome email for a new SaaS product launch.]

That is okay.

But the real power of your tool is not just speed. It is consistency.

You have a brand kit section. Users can upload their logo, colors, and fonts there. This makes emails match their brand without extra work.

Any AI can make an email. But not every AI can make an email that looks exactly like the brand. This brand kit is the big difference.

Right now, this feature is hidden. It is under the words [your brand in every email.]

A marketer who visits your page first sees the speed message. But what they really need is consistency for many emails.

Speed is just one feature. Brand consistency is the real value.

Also, your "works with your stack" part shows logos of Klaviyo, Mailchimp, HubSpot, Brevo, and SendGrid.

That is good. But it does not explain how the integration works.

Does it push with one click? Or do users still need to copy and paste HTML?

People want to know this to feel safe. I saw a few more small things that can make your message clearer. Happy to share if you want.

Wow! Looks great. Is the website is necessary or can I upload branding and general guidance?

This is interesting. Email design is one of those things that always takes longer than it should, especially if you want it to actually look good.

Getting something usable in seconds is a big shift if the quality holds up.

How close are the outputs to being ready to send without tweaking? That’s usually the sticking point.

Santi this is so well timed. We were literally prompting Claude, copying the output, cleaning it up, then manually pushing it into Mailchimp. Every single time. It worked but it was such a painful workflow.

Browsed the website and this solves that exact problem. The direct integrations are what got us. Congrats on shipping this, trying it out today.

"no editor. no templates. just describe and ship." santi nailed the positioning here. this is what email design should've been from the start :)

The idea is great, but i a missing a round button on the right corner for a feedback agent or chat. had some troubles while using the preview-button, that didn't work at all. The integration of www.klaviyo.com/ is a great idea and the results are good, but the synchronisation sometimes fails. Keep on optimizing the tool.

This is going to be a highly useful app for the diabetic.

All the best for your launch!

One of the best usecases of tech I can see is in the medicine. I would wish to have more similar tools that help with quality of personal life :) Wishing a successful launch :)

Hi Dmitry,
I really love your idea! My girlfriend has diabetes.

So I will definitely need something like this. One thing I don't understand is:

she has a sensor on her arm, how can she connect it to the app?
For people with diabetes it's already annoying to have a sensor on their arm, so she will never enter data manually into your app.

I would love to help you build this app, how can I contribute?

I have one question, how does the tracking works, I mean we still need to have a glucose monitor attached right?

ok 'your post shouldn't wait until you're back at a desk' actually got me. the push live or open a PR straight from your phone is genuinely well thought out. well done matt.

This looks really cool, love the Git-as-CMS approach.

I’m building a docs framework called MDX Docs (mdxdocs.com) that uses MDX files in a repo and maps them directly to routes (React + Vite).

Curious if Git Blog supports MDX, or if it assumes plain Markdown today? If not, do you think MDX support would be feasible? I’d love to try wiring it up on my side.

Interesting ... but only for apple?

The image resize and optimise step is the part that would actually save me the most time-curious what formats and size targets you're working with, and whether that's configurable per site setup?

Very cool! I'm wondering how is this different from publishing a Notion site?

sent this to a couple of engineers i know who are building local agent setups. they've been duct-taping credential management together for months, this is right up their alley.

I use Claude Code and Cursor daily and I've learned that you can't just trust the output blindly - agents will tell you they implemented something and you'll find out later it was half done or the tests were never actually run. How does Vet handle cases where the agent made a reasonable interpretation of an ambiguous request? Does it flag those as potential issues or only catch clear deviations?

The credentials never showing up in logs or chat transcripts detail is the actually important thing here. I've seen agent setups where the auth storage is secure but the credential ends up in tool call output anyway - solved the wrong problem. Does token rotation work automatically? If a service refreshes the token mid-session does latchkey pick that up, or does the agent need to restart?

proof of work displayed on a profile makes it clear who is serious and who just talks about building

How about quality of design? So much vibe coding is simply templatized..

Streak and live URLs show who's consistent, but how do you tell quality apart? Tons of projects are technically "live" but held together with duct tape. Does the vibe score look at activity only, or does it factor in stuff like uptime or actual users?

Nice.

In the page Find Talent the field Required Tech Stack I am not sure if necessary since for Vibe coder the stack is less relevant and easy to adjust?

Does this mean, regular coders can't participate in this? and how do you define vibecoder unless they define themselves as vibecoder?

Much needed marketplace. All the best!

we've hired a few contractors for features.vote over the years and the evaluation process is always the same mess. going off portfolios and github stars with no real signal on whether someone will actually ship.

the streak + proof of work angle here is exactly the missing signal. bookmarked.

@kath_nguyen Seems like a really nice idea . Does it work with normal typing as well , not just via voice ? Sometimes I wished there was a popup that generates the translated version of my writing in place as I type .

Switching between apps to translate is always frustrating. This makes a lot of sense.
How do you make sure the final translation sounds truly human and not AI-generated?

This looks amazing. Congrats on your launch!

Seems like a really nice idea. does it work with normal typing as well, not just via voice? Sometimes I wished there was a popup that generates the translated version of my writing in-place as I type.

shared this with someone we collaborate with who writes in both english and spanish all day. the constant copy-paste loop in and out of google translate while messaging is exactly the kind of thing that sounds small but adds up fast. good timing on the launch.

Such a relevant pain point! Coming to PC soon too?

Voice typing in iOS is very limited, especially for Polish, so I totally understand of the need for such solutions... good luck!

Awesome idea! How does it do on Dutch?

Interesting approach going mobile-first with voice typing. The challenge I've seen is accuracy drops significantly with form fields vs free text — how does it handle structured input like emails, phone numbers, dates?

Congrats @gabriel_alonso !

How does your transcription accuracy compare to Apple's native dictation (this is literally pain in the ass apples feature) and models like Whisper in real-world scenarios? Also, is the processing done on-device or via cloud?

The pronunciation exercises are helpful because I can hear mistakes and correct them immediately.

First of all, congratulations on the launch!! This is a really impressive learning project. I’ve started using it and so far, the experience has been great!

One thing that genuinely stood out to me was the analysis after the assessments. I’ve tried quite a few language learning apps, and I can confidently say this is one of the best feedback systems I’ve seen, really insightful and helpful.

If I could suggest one improvement, it would be the textarea in the “Assessment with Lex” section. Since the input size is fixed, it becomes a bit difficult to review longer responses. I tend to write more detailed answers, and navigating through them isn’t the easiest after finishing.

I’m about to start the sessions next and will definitely share more feedback if anything else comes up. Congrats again on the launch 🚀

Seeing this at a really good time as I am trying to polish up my Swedish! Seems like an awesome application, congrats on the launch! What languages do you support atm?

nicely done! smart features.

@oldcarnewradio @Neuralingo Language Learning
Hi Julius! It's my first time to know AI product via makers' real story which's really cool and vivid. Here's my perspective from not professional but real language learner.
1. 🤩👍Good aesthetic taste which makes me have attention and intent to go deeper! THAT IS IMPORTANT FOR language learning App. You know what I mean.
2. Honestly I have left my Duolingo cuz I do not need a game app which only use one language explaining another. 😓What I need to learn is a vivid and like storytelling bridge between cultures not only languages. Neuraligo really delivers a new learning experience.👍
3. I wonder if you guys develop speak features it will be much more convenient.💬

Most language apps make things too easy so you feel good but don't actually learn. How many languages does it currently support, and does the pronunciation mode work well for non-Latin script languages?

I know that ProductHunt launches are overwhelmingly AI-based now, but Neurolingo is, after signing up, doing the assessment, and even paying for a plan(!) a really good example of why "vibecoding" and LLMs in general are not suited to language learning (at least not at the moment).

It's why I canceled and uninstalled DuoLingo when they said they were an "AI-first" company.

This review/comment is kinda long, so the TL;DR here, from a paying user who intends to cancel is: DIY flash cards are a much better use of your time and money.

(As a note: all images below are screenshots of the app, in browser, doing something frustrating or confounding.)

1. The session was maddening. First, Lex insisted on things like accents for French, even though those are not native keys on an English keyboard. I was able to move past this with Lex, but it's a stupid oversight unless we're specifically working on an accents exercise.
   

2. Lex would ask me if I wanted to explore a specific pain point further, or continue the lesson. Any response ("continue" or "let's work more on article-noun gender agreement") would comply...and then immediately jump into an unrelated lesson and/or ask me to translate Je voudrais un café et un croissant. When I pointed it out, it said it had some system confusion.
   

   

   
   In the post-lesson report card (mistake review), it marks responses to the user's confusion as ACTUAL MISTAKES, which is mind-bogglingly stupid.

   

   
   Another "mistake" was rightly evaluated as a mistake in the post-lesson, but in the lesson was marked as "ABSOLUTELY CORRECT!" which is confusing, frustrating, and just...bad.
   

   

3. The constant move back to Je voudrais un café et un croissant created a confounding interruption of the flow of the lesson, ensuring that almost nothing would stick. Sincerely, before the lesson self-ended (which...why?!) it had asked me to translate Je voudrais un café et un croissant no fewer than half a dozen times, completely out of context, and without any relevance to the lesson. That is probably the only thing I remember confidently, other than the sycophantic insistence that I was "absolutely correct to be frustrated" and "translated that perfectly!" (when I had not).
   
   

4. The pre-start evaluation put me at A1, but the session for some reason was in A1, and the pre-start analysis had me at 10, but this hour long session is saying I'm now at 11, even though it's a full half-level below where I should've started?? Also: This is a minor gripe, but I didn't spend 108 minutes in this lesson (it was about 50, and then the lesson auto-ended without warning).

   

5. Who is this little bit of "sage wisdom" supposed to be for? Why are you shipping a language app with a formulaic fortune cookie saying?

   

6. The mistake overview is actually just worthless, because half of these just were not mistakes, and many of the others were me attempting to troubleshoot Lex's condescension (which was especially high contrast when Lex was just dead wrong, e.g. The root cause of your mistake is....) even if the mistake was something like noun gender, which is learned through rote memorization, and cannot be divined by looking at the word. For example, none of the "mistakes" listed as mistakes were actually mistakes. The evaluation is just hallucinating.
   

   
   

I gotta say that it is deeply agitating for this to be the outcome here. I could've made this in ChatGPT back when OpenAI launched the "custom GPTs" feature in like, 2023, and it would've been just as if not more effective. I have very little confidence that NeuroLingo is ready for prime time in any sense of the word "ready," and, aside from the reasonably attractive UI, I'm not confident in anything about NeuroLingo.

Importantly here, I'm not confident that Lex/the product even has a good enough internal understanding of language or how languages are learned to believe that the lesson I just spent an hour in was accurate on a language/lexicon/vernacular level.

In almost any scenario I can imagine, I would've been much better off using Google Translate to make personal flash cards, and just memorizing them–and I'm saying that with the understanding that Google Translate isn't particularly great.

I would say "good luck" but I honestly feel like this product might be actively deleterious to users' attempts to learn a language.

Really clean UX on this. How are you handling rate limiting on the API side?

Good luck with your launch!

Learning a language properly through apps is really hard (I'm struggling to learn German with Duolingo actually), and I think your app can really solve this problem!

I'm curious to know more about the method you talk about in the website: did you do some deep research on neuroscience or are there neuoscientists in your team who helped you develop the app?

That's a cool idea. Congrats on the launch 🎉

Btw, will the music in the headphones change as per the environment?

Definitely will try it out as the automated deployments is the most interesting part right now in the era of vibe coding! Have a great launch!

replacing a full afternoon of ssl + ci/cd config with 3 clicks is a great pitch. congrats on shipping! :)

Congrats @snakajima and good hunt @gabe, the plugin architecture sounds really smart for letting developers build custom visual experiences. Can multiple people collaborate on the canvas simultaneously during voice conversations?

The diagrams were always a problem for me to draw. In the past, good diagram drawing tools were paid and expensive. And the free versions of them were so painful to use.

Can I use the tool to create a whole presentations with some visual animations? Like, moving the text, images fading out, etc?

spent way too long trying to get ai to generate a proper mind map and it always just spit out markdown. ended up copying it into miro manually every single time. this is what it should've looked like from the start.

Hey PH! Aditya here, co-founder of nCompass.

My team and I have spent years accelerating GPU kernels. It’s complex and time consuming, primarily because we spent most our time identifying which kernel to optimize and then identifying what the bottlenecks were in our new kernel. Using the dev tool we’re launching today, we implemented a Hopper GEMM kernel that outperformed NVIDIA's CUTLASS GEMMs by 3%, within a day - this took us months before.

Here's the problem today: if you profile a system like vLLM, you have to copy a giant trace file to your local machine just to view it. Then you spend hours identifying which GPU kernels to target. Then you profile the kernel, spend hours or days digging through .ncu traces that are massive data dumps. Then you identify your bottlenecks and formulate a plan. Running this loop till you have a performant kernel can take weeks, even months before you have a performant kernel.

All of this is going to change. Coding agents have gotten great at writing GPU kernels once they know what to target. What's missing is the tooling to identify these targets.

We built the tool we wish we had. With our VSCode extension, you can view traces natively — no more copying files around. Once our MCP is integrated, just "@" an execution trace into Claude Code / Cursor and our agent takes over: our agent analyzes the trace and gives you a prioritized list of GPU kernels to target. Using NCU, profile your kernel source code, "@" the .ncu-rep in your favorite coding agent, and the nCompass agent works together with your coding agent to develop the high performing kernel you were looking to build!

If you're an experienced GPU engineer — this removes the busywork. You stay in control, vet the agent's outputs, and only act on what makes sense.

If you're learning GPU optimization — you now have an expert pair-programmer you can always consult. It's the best way to learn how to optimize GPU code.

We'd love for you to try it and let us know how it works for you. We're building fast and your feedback directly shapes what ships next.

Try it → https://docs.ncompass.tech/quick...
Full feature list → https://docs.ncompass.tech

'outperformed nvidia's cutlass gemms by 3%, in a day' is the kind of before/after that sells itself. congrats on launching this :)