Connect with the Product Hunt community

What happens after a suspicious session is diverted? It needs somewhere consistent to keep working.

DungeonQ's new reproducible walkthrough follows that path: write a world record through MCP, read it through HTTP, use a world-only Wrong Ticket, then restart the services and check that the state survives. A separately scoped operator grant permits a bounded follow-up record; independent evidence checks what happened.

You can run the walkthrough locally without connecting a model or production system. From the public repository, with Node.js 24.15.0+:

npm ci --ignore-scripts

One of the most requested MinkNote features here on Product Hunt was for MinkNote to automatically detect when notes are changed by other apps or AI tools.

It was actually the #1 request I received from the Product Hunt community.

Mac Duo 1.2 is out, with a broader visual refresh focused on making the whole app feel more at home on the Mac.

What's new:

There are two lines on OpenAI s GPT-5.6 model page that are easy to miss if you call the model by its alias.

The first: the plain gpt-5.6 alias routes to GPT-5.6 Sol. If your code asks for gpt-5.6 because it looked like the sensible default, you re paying Sol rates: $4 per million input tokens, $20 per million output.

The second: Sol s pricing is labelled promotional, available at least through November 21, 2026 . The page doesn t say what it becomes after that.

Take an ordinary feature: 100,000 requests a month at 1,200 tokens in and 300 out. That s $480 of input plus $600 of output, so $1,080 a month, on a price that s only promised until the 21st.

Ota lets one task declare native, container, or configured remote modes. The contract makes differences in context, dependencies, environment, command, and runtime explicit, while developers, CI, and AI agents discover the same task.

Tellie 1.6 is out. Liquid Glass chrome; it remembers which display you put it on (docked or floating); a coaching strip that tells you when a point landed and how much time you have left; and Take Coach, an after-take debrief written on-device (beta). Voice Follow is the same idea it always was: it matches the words you say, so pausing and ad-libbing don't lose your place. Free download, Pro $29 once, 10 days of Pro on every install. https://tellieapp.com/ph

Lately, it feels like even the smallest project comes with an unwritten rule that you need 15 dependencies, Docker containers, Tailwind plugins, and a complex deployment pipeline before you even render a <div>.

My recent experience was when I wanted to build a simple 3D browser runner game over the weekend. I started with the "recommended" modern web setup and immediately got buried in conflicting bundler configs, SSR hydration errors on canvas components, and package bloat. So I ended up tearing half of it out, switching to a lightweight editor (Zed), drafting the math directly in AI Studio, and keeping the game loop isolated in a barebones Vite canvas before wrapping it in Next.js just for hosting. The moment I stripped away the modern boilerplate, the project actually moved fast.

What s a part of modern web dev or startup tech stacks that you think has become completely over-engineered for 90% of use cases?

Your coding agent is still working on your desktop when you leave. Termexo v0.10.0 lets you reconnect to that same workbench from a phone or another computer, across networks, through a self-hosted relay.

The desktop opens an outbound tunnel, so it needs no public IP or router port forwarding. Open the relay address in a phone or another computer s browser to check terminal output, respond to your agent and keep working in the same workbench.

Self-host termexo-relay, enroll your desktop with a code or account, then open its address or QR code. The relay supports Linux, macOS and Windows, plus containers; the Termexo desktop app is for Windows 10/11 x64.

v2 session traffic uses AES-256-GCM after an authenticated handshake. Relay login and the desktop token are independent access checks. You need a reachable HTTPS relay and a desktop that stays on with Termexo connected.

I'd never considered packs until a friend complained about the "default" words they just weren't doing it for him. "Now, if you had a fantasy pack ". And thus fantasy, sci-fi, supers. I find the fantasy pack the most challenging, but on a good day that really works. :)

I pulled the numbers from Forg across 700 public projects, 1.5k+ indie hackers, and a mountain of company data to answer three questions I keep arguing about with other founders:

1. What are indie hackers actually shipping?

2. What stack are they running?

3. How are they charging for it?

The patterns turned out sharper than I expected. Here's the honest read.

AI didn't win. It lapped the field.

It is live.

If you run client work, you are the last pair of eyes on everything. Every report, every update, every deck. That is what caps how many clients you can carry. It gets worse as you grow.

I just released BiBimba 1.1.0!

Ever remembered which page you copied something from, but not the words you copied? You can now search your clipboard history by the source page title or URL too.

For new text copies, BiBimba saves the page or window title and source URL when the app exposes them. No browser extension needed. Your copied text stays unchanged, and the source details stay on your Mac. You can turn this off in Settings History.

This applies to new copies after updating; it doesn't add source details to older entries.

A migration can pass with the new build and still break a worker from the previous release that's finishing a queued job. Rename a column in one deploy, and that worker may keep writing the old name after the database has moved on.

Adding the new field first lets both versions coexist. Then you can backfill, switch reads, and remove the old field in a later deploy. The dangerous bit is declaring the old writers gone too early.

What do you check before that cleanup: running worker versions, drained queues, or logs showing no writes to the old field? If you run a small team, what's the smallest check you'd trust?

Thank you all for being part of the launch yesterday, all the people I got to meet, all the experiences throughout the day, the pitches, the competition... that's all part of it, and I couldn't be happier

I also got some really nice insights about Toone from the community. A lot of people showing genuine interest in the tool and in what comes next.

I don't want to spoil too much, but we have a mobile application coming for iOS that will allow you to control Toone, dispatch and watch routines, and talk to an agent as if it were running on your machine. Apple just gotta approve.

We've also already implemented a sharing functionality for routines, allowing people to make their expertise and workflows available to the public.

Hey everyone! We just shipped v1.11.0 for macOS.

No more password prompt every session

This is the big quality-of-life fix. Approve SideDisplay once under System Settings > General > Login Items & Extensions (under Background App Activity), and starting and stopping just works from then on. No more entering your password every time you connect. (If you'd rather not approve it, nothing changes and you'll be asked for the password as before.)

This release is about giving you more than two folders at once, and about folders that keep themselves tidy.

Up to six panes in one window

Split a window into one to six panes, each with its own folder, its own view mode and its own terminal if you use that. Pick the arrangement from the toolbar, or press Control Command and the number of panes you want.

been going back and forth on this. I send about 20 linkedin dms a day and I've tried both ways: a short note mentioning something they posted, and a blank request with a message once they accept. blank feels lazier but I'm honestly not convinced it does worse, people seem to judge on the profile more than the note.

what do others do? and has anyone actually tracked it properly, or is everyone going on gut feel like me?

Hi everyone maker here. Thanks for taking a look at Holon during our Product Hunt launch.

Holon is a local-first workbench for agents that keep following a task beyond a single conversation. They work in your repositories, track progress in durable WorkItems, and wait for events such as CI results or incoming messages before continuing.

Since launch, we've shipped v0.43.0, v0.44.0, and v0.44.1. The v0.44.1 update includes improvements to memory indexing, idle scheduling, and file links across workspaces.

If you're wondering what that looks like in practice, we've also shared three longer reads on X:

Watch on YouTube:

Hey Product Hunt! I m Yasser, one of the makers of ApplySeed

My co-founders and I wanted to raise money, so we started looking at accelerators and funding programs. We opened applications, started filling out forms and quickly noticed we were answering the same questions over and over again: describe your startup, what problem are you solving, who are your customers...

So we built ApplySeed.

We ended up building an agent that finds opportunities that fit your startup, prepares the answers from your profile, and fills out the applications for you. You review everything and send.
Let us know what you think !! https://www.producthunt.com/prod...

Termexo v0.10.4 is live. Grok Build joins Claude Code, Codex, OpenCode, and Antigravity in our Windows terminal workbench.

With Grok Build CLI installed and signed in, you can launch it in a workspace, list and resume sessions, and use it on the task board. The usage panel now recognizes an exhausted free allowance, while an unknown percentage stays unavailable instead of appearing as 0%.

This release also fixes a frustrating long-session issue: the session change count could fall to zero and the Git panel go blank after the app had been open for a while. Changed files and diffs now keep refreshing.

Terminals now show a startup animation instead of the launch command. Claude Code accepts Ctrl+V, workspaces can be reordered by dragging, and the duplicate task-board Settings button is removed.

1. Wrote to 105 immigrants.

2. Got about 20 replies.

3. 10 people gave interviews.

Update from launch week.

The publisher path no longer waits on a review. You create a key, drop the SDK, send a message. If the auction is empty you still get a house ad. Same day.

If you tried the SDK last Tuesday and saw nothing, try again.

I want the boring reports:

Over the entire time of its existence, AI has improved a lot, which is problematic.
The amount of dishonest creation and scams is on the rise.

How do you protect yourself and distinguish AI from reality?