Connect with the Product Hunt community

Dictantor can now become a part of your AI workflow.

With the latest update, you can give your AI agents access to a folder containing all your Dictantor transcripts. So recordings from meetings, voice notes, and random thoughts can become context your agents can actually work with.

Transcripts also got a small structural upgrade to make them easier for both humans and agents to read.

Next, I m working on custom folders inside Dictantor. The idea is to keep recordings organized by project or topic, while also giving agents a cleaner way to understand and use the right context.

Been working on this update for a while, and I wanted to share what s new in 3.0.

Vibe coding a personal project that works on your laptop is one thing.
Building a tool your team opens every morning to manage projects, serve customers, and run the business is another.
Suddenly there s shared data.
Different people need different levels of access.
Your teammates need to understand how the app works so they can change it too.
And the app needs to keep evolving without turning into a black box nobody wants to touch.
At that point, would you be comfortable letting AI change anything it wants?
We ve been thinking about this a lot while building WeWeb MCP (launching on PH tomorrow!)
Agents like Claude, ChatGPT, Cursor, and Gemini can now build apps directly inside WeWeb.
Across the frontend & backend from: UI, data, workflows, integrations, permissions, and more.
But the important part for us was making sure the builder stays in control of both the app and the AI building it.
That s why we built guardrails into the way AI works inside WeWeb:

• Before AI builds: choose what AI can and can t touch: pages, components, workflows, tables, APIs, and more.

• After AI builds: see exactly what changed in the visual editor. Tweak it yourself, keep iterating with AI, and publish when you re ready.

So AI gets room to move fast, without getting free rein over the whole app.
I m curious how you think about this: Would you give an AI agent access to everything?
Or are there parts of your app you d always want to lock down?

Hey everyone

Netfox 0.22.0 is out, and it is about one thing: the window finally behaves when you tile it.

It used to refuse to go narrower than 960 points. So if you put it beside the thing you were actually working on, its right edge went off the screen. It was an app you opened, glanced at, and closed

You rename packages/cli to apps/cli. The next scan tells you a dependency started running code at install, and another one stopped. Nothing started you moved a folder.

0.34.0 is four changes that are the same sentence from different angles: a security signal is only worth reading if it fires on what happened, not on what you did to your own repo. An axis that cries wolf is one you learn to skip, and then it's worth nothing on the day it's right.

Something I keep running into with vibe coding: the agent makes a failing test pass by loosening it instead of fixing the code.

One example: it replaced a strict deep equality check with expect.anything() on an error payload. The test still exists, still runs, still passes. It just stopped asserting anything. The build stayed green for days until users hit silent null-pointer crashes.

The same can happen with CI workflows and migrations. If the agent can edit them, it can make the gate disappear.

One cheap guardrail I ve seen: a CODEOWNERS rule requiring human review on tests/, migrations/ and .github/workflows/, so any diff there needs approval no matter how green the build is.
I d love to hear how you handle this:

As Termexo s maintainer, I am focusing on a practical problem: several agents can be working while the developer is still finding terminals, checking who needs approval and reconstructing task context.

For individual developers, Termexo groups real terminals by project, surfaces agent status, organizes tasks and helps resume supported native sessions. Handoff packages can carry a reviewed summary of progress and next steps.

When I'm trying a new SaaS tool, I'm much more interested in signing up if I've already seen what it can actually do. Being asked to fill out a form before seeing any real value can make me leave before giving the product a chance.

I get why founders ask people to sign up upfront, though. It helps them save their work, personalize the experience, and understand who's using the product. And some tools need a bit of setup before they can show useful results.

Personally, I like the idea of letting someone try one small part of the product first, then asking them to create an account when they want to save their work or go further. But I'm not sure that approach makes sense for every type of SaaS.

Hey everyone

Netfox 0.21.0 is out, and it is mostly one feature plus a confession.

@Youkti helps you track exactly where your deals stand.

From early conversations to active opportunities, get the context you need to know what s happening, what s changed, and what to do next.

Stalled deals can come back to life. @Youkti analyzes the context around each deal and recommends the next best steps to help you re-engage and move it forward.

I've been thinking about this a lot lately.

About a year ago, I had zero coding experience. I came from corporate finance, had an idea for an app, and decided to try building it with AI.

Fast forward to today, and I've built and launched an actual iOS app. Something I genuinely never thought I'd be capable of doing on my own.

But here's what I've realized.

I don t code.
I build with ai tools and a friend handles the tech side, so i m asking from the outside.

For people who do code: was there a moment you realised you were leaning on claude too much? like a bug you couldn t debug without it, a code review that went badly, or just a gut feeling one day?

AppGacha 0.1.1 is now available for Windows and Apple Silicon Macs.

This is a small release focused on making generated apps feel more independent and improving the update experience:

• Opening an app from its desktop shortcut no longer brings up the main AppGacha window.

• Generated app windows now display their own icons, with a fallback when no icon is available.

• Settings now includes version history and localized release notes.

• Update prompts explain what changed, and installation waits for the current build to finish.

• Added minimal, content-free usage statistics to help identify failed builds and repeat usage. Collection will not begin before September 26, can be disabled at any time, and never includes prompts, generated code, API keys, files, app names, account IDs, or persistent device identifiers.

Thank you to everyone who tried the first release. If something still feels confusing or gets in the way of using a generated app like a normal desktop application, I d love to hear about it.

We re working on making Ski more comfortable to use for people who don t speak English all the time.

The latest update adds more control over both voice input and voice output, with platform-specific improvements:

• macOS: More languages available for voice input in High and Ultra modes

• Windows, Linux: Ultra (default) mode can automatically detect the language being spoken

• Voice output: We re expanding the number of supported languages, and currently supports few languages in Kokoro. Each language has its own voice list.

• Audio devices: You can now choose your input microphone and output speaker/headphones directly

• Easy updates: Get the latest version through automatic updates or download it directly at heyski.io

We re still expanding the language list, especially for voice output.

Voice agents should become more than transcription tools. Instead of simply turning speech into text, they should understand what users are trying to achieve. The same spoken idea may need to become a short Slack message, a detailed email, a technical issue, or a personal note. A useful voice agent should adapt automatically.
This is the direction Loqua is exploring: voice as an intelligent layer across the apps people already use. It can understand the current application, nearby content, personal vocabulary, and writing style, then shape spoken thoughts into something ready to use. Users should be able to speak naturally without organizing every sentence beforehand.

The next step is real-time duplex conversation. A good agent should know when someone has finished speaking, when they are only pausing, and when they want to interrupt. It should also recognize signals beyond words, such as emotion, hesitation, emphasis, and tone. Its own voice should feel equally natural, expressive, and appropriate to the situation.

Behind this conversation, the agent needs deeper capabilities. It should remember relevant context, think through difficult requests, use tools, and continue working after the conversation moves on. The voice experience remains fast and fluid, while more complex reasoning happens quietly in the background.
We all know that long-horizon coding tasks are difficult, but understanding long, multi-turn conversations is even harder. The agent must remain consistent, respond with the right emotion, use tools correctly, and remember what matters.

Hi everyone, Matheus here, Toone's maker. We're preparing our next Product Hunt launch for September 18

Toone is a macOS app for crafting complex AI workflow automations and routines with natural language. I'd like to hear from people who want to try one small routine and share candid feedback.

If you're interested, share the task you repeat, the inputs you'd start with, and what a useful result would look like. Pick something you can check yourself so we can see where the routine helps and where it falls short.

I can help arrange access individually, though I can't promise immediate availability.
Thank you and see you by then PH

someone on my last thread pointed out the project file says what the project is, but not what changed since yesterday. i don t have a good answer for that yet.
what do you do, just let it read the git log?

We re giving agents access to APIs, databases, CRMs, payments and internal systems.

But something feels missing.

We can observe what an agent did.
We can evaluate whether it was right.
We can detect a policy violation.

AI is great at producing a first landing-page draft. The last mile is still human: make the promise specific, make the CTA match the next action, remove repeated filler, add a link or table only when it helps a visitor decide, preview the whole page, and export the reviewed HTML.

HtmlDrag gives teams a visual working copy for that focused pass. Keep the HTML that already works instead of restarting the page because one visible decision needs correction.

MinkNote is a privacy-focused Mac app that keeps everything you write and collect in one place, organised and effortless to find again.

Until this week it ran on a 14-day trial, then all the notes went read-only. Two weeks is not long enough to know whether an app like this fits your workflow, so now it works the way most Mac apps you already use do:

This release is about doing more with fewer clicks, and making the app yours. A good part of it came straight from your reports and requests, so thank you for that.

Here is what's new.

Connect to Server, redesigned

Existing workbench screenshot (v0.8.0), shown for context.

I'm the maintainer of Termexo, an MIT-licensed Windows workspace for Claude Code, Codex and OpenCode.

v0.8.7 focuses on a frustrating reliability problem: an agent could keep running while its terminal went blank and appeared stopped. Resizing around a wide character could trigger a parser panic, poison the screen mutex and leave subsequent screen reads failing. Switching between clients could reproduce that combination.

When a problem needs a diagram to explain, I don't do it all through a blind call we just look at the same thing and work on it together.

"LiveSyncDesk is Not an another Online Whiteboard, It's the moment inside a conversation when words aren't enough.

LiveSyncDesk is a small real-time visual explanation space during a conversation."

Something that changed how our team works: every time Claude messes something up, we write it into a skill file so it doesn't happen again. Over time it got way better at our stuff than any prompt ever did. Same with art, once we built skills for our drawings and animations the results got clean and consistent.

Curious how others handle this. Do you keep notes or rules for your AI tools, or just re-prompt every time? And what's the most useful rule you've written?