Connect with the Product Hunt community

Dani here We're about halfway through launch day and the questions have been great. A few keep coming up, so here are the answers in one place for anyone just arriving.

"If someone copies my Managed widget onto their site, does it use my credits?"
No. The widget checks which site it's on before it loads. If that site isn't on your allowed list, it doesn't appear at all, so there are no messages and no credits used.

"Does it keep the conversation if a visitor comes back?"
Yes. History is saved in the visitor's browser, so it survives reloads and return visits on the same device. If you'd rather start fresh each time, there's a setting for that.

"Can it know who the logged-in user is?"
Yes. Pass the user's details as metadata and your n8n workflow gets them with every message.

Hey PH

Small release this week!

Improving a follow-up prompt in ChatGPT, now uses the whole conversation as context.

We're Dani and Ross, the owners of N8N Chat UI, and we launched on Product Hunt today.

One thing we're curious about from people building with n8n: what makes a chat widget useful on a site, beyond sitting in the corner waiting to be clicked?

The widget has a small JavaScript API, so you can start a chat from almost anything.

Hey PH
One use case keeps coming up with agencies, so we wrote it up properly so others can reuse the template.
The problem: an agency runs 10 clients through ChatGPT, Claude and Gemini. Every writer pastes a different version of the style guide, two clients end up in the same thread, and freelancers take their prompts with them when they leave. The result is voice bleed: Client A's copy starts sounding like Client B.
How teams are solving it with AI Context Flow:
One memory bucket per client: voice rules, audience, banned terms, on-brand and rejected examples, compliance limits
Shared buckets: contributor and viewer roles, an activity log of who changed what, and version restore
The MCP server: the same client voice in Claude, ChatGPT, Cursor, Copilot, Windsurf and headless agents. The AI pulls only the relevant context and writes client feedback back to the bucket
Full playbook, including a copyable voice-bucket checklist: https://plurality.network/blogs/...
If you run an agency or a multi-brand content team, I'd love to know how you handle client voices today, and what's still breaking, maybe we could make a custom built setup done-for-you. lmk below

Quick build-in-public update from Sponkly (real-time voice translation for chat, calls and meetings in 32 languages).

This week, as a solo bootstrapped founder:

• NVIDIA Inception: accepted

• AssemblyAI Startup Program: accepted

• AWS Activate: $10,000 in credits

• LiveKit Startup Program: approved today, $17,000 in credits + a year of the Scale plan

What I will use it for: Sponkly's calls and meetings are peer to peer today (WebRTC mesh). Great for 1:1, heavy beyond 5 6 people. Moving meetings to LiveKit rooms means one upload per person, so translated meetings can grow, and the live interpreter can join as a LiveKit agent that hears each speaker directly and translates once per language instead of once per listener. 1:1 calls stay peer to peer (cheapest, end-to-end).

When a customer hits cancel, most workflows park at an approval step. The question nobody asks: has anything already been cancelled by then?

We ran a rehearsal on our synthetic Acme demo, wiring FetchSandbox (fetchsandbox.com/mcp) across Stripe, Zendesk, Square, and Slack twins. Three boundaries we actually checked: nothing fires while parked at approval, rejected approval leaves the subscription unchanged, and the workflow survives an app restart while parked then cancels exactly once after approval. 27/27 passed on October 18.

Scope caveat: synthetic provider twins, not live traffic. But that's the point, catching a premature cancellation in rehearsal is cheaper than finding it after a real customer emails you.

AI-assisted code moves fast enough that these intermediate states are genuinely easy to miss. You end up testing the happy path and calling it done.

Hi everyone, Ashish here from Atlaso.

This week GitHub shipped Copilot Fusion, which routes one coding task across several models and lets a second model critique the first. Gemini 4 Carbon is also near the top of Product Hunt today. Switching models is quickly becoming the default.

Was at the Indie Hackers x Cognition meetup in Dubai last night. @marclou showed the list of boring stuff every SaaS has to build.

Right next to Stripe and auth: Emails: DNS records, avoid spam.

It stuck with me becuase it's exactly the pattern. you set SPF, DKIM and DMARC once at launch, then never look again. Until the welcome email lands in spam and nobody knows why.

A Semrush analysis of 1,094 US commercial categories found that only 15.2% had a clear brand leader in ChatGPT recommendations.

The study also found that 89.3% of estimated search demand belonged to categories where no single brand consistently led recommendations.

For me it's when it forgets everything I told it 10 minutes ago...

I explain the repo.

It says "got it".

Then it rewrites the same file twice hehe

From 1 October, leaving Apple In-App Purchase drops a Small Business Program developer from 15% to 10% in the EU. On a 4.99 euro subscription, a processor's fixed per-transaction fee can eat most of that on its own - and your choice locks for 12 months.

Apple updated the Developer Program License Agreement on 18 August, and the new EU business terms go into effect on 1 October. The coverage I've read frames it as Apple finally opening up. That's true in a structural sense and mostly irrelevant to anyone running a small subscription app, because the interesting question isn't whether you can leave Apple's payment system. It's whether the arithmetic works. For most of us it doesn't, and it's worth being able to say why rather than just having a feeling about it.

We ran a Lovable checkout app through FetchSandbox. Six checks passed. Two stayed not checked.

We delayed an email response, but the webhook still returned success. We hadn't shown it recovering after a failed callback. So the overall run stayed incomplete.

That's the thing I want us to get right as we build this. Did the test cause the failure, or did the app just succeed? Same question for a background job that might crash halfway through.

What evidence do you ask for before trusting a recovery test?

I understand that hunting has traditionally been seen as a man's thing.

But Product Hunting is a completely different thing =)

There are so many women makers, marketers, designers, and developers on PH. I ve met and interacted with quite a few of them.

But when I look at the top hunters

Since launch, your bots and agents can check a change before they apply it.

Aperture is now an MCP server. Grok Bot, Claude Code, Cursor or any MCP client can call check_change before it applies a change. It checks that the files parse, the imports resolve, and the TypeScript compiler finds no new errors, including in files the agent never opened. It also flags a change that skips or cuts short the tests. Nothing the agent sends is run or kept. Setup takes three steps at https://aperturesais.grok.me/agents

Aperture Agent Check runs the same checks on every pull request, as a CLI or a GitHub Action. Red lines get annotations right on the diff. If a failure was already there in the base branch, it shows as amber, not blamed on the PR. Your code never leaves the runner. Aperture's own PRs go through it.

Honest numbers: in the benchmark, the editor catches 27 of 33 bad edits, with 1 false alarm in 15 good ones. check_change stops 20 of the 33 without running anything. The other 7 only show up when tests run, so keep running them. Every miss is listed at https://aperturesais.grok.me/ben...

A customer offers to pay for a feature. The money would help, but the feature only fits their workflow.

Saying yes could fund the next few months. It could also leave you maintaining a second product inside the first one.

I'm Shun, the maker of Ownfeed.

We launched here today, and I want to use this thread to learn from you, not to pitch.

Every launch I see is about agents, orchestration, reasoning chains, autonomous workflows, then I open my own chat history w/ claude and it s all about chores (task support), and almost nothing regards agentic workflows. it is mostly small text chores I could describe in one sentence.

I don t think that s a failure of imagination on my part, and I don t think I m unusual. People (including I) use HUUGE models to do irrelevant tasks, that's like asking stephen hawking to sort out your monthly expenses' PDF.

Hi everyone,

today I want to share some exciting news with you about Basket Score something that will completely revolutionize the way games are streamed.

Introducing the new AI-powered scoreboard reader.

There s a growing tendency to flag AI-generated text. There s even a term for low-quality AI-generated content - AI slop.

And honestly, I understand why.

Communication should feel real. If every comment, post, or message is generated by AI, the picture becomes blurry.

But then I have a question.

As search becomes a core part of AI workflows, things like data coverage, QPS, and cost start to matter, especially when your agent needs to search repeatedly across different sources.

That's why we've launched AnySearch Pro , built for developers and teams running AI agents and search-heavy applications.

We launched here last Monday. One number from our prep keeps coming up in conversations, so a question.

We ran every stable version of 14 of the top 30 npm MCP servers from March to September in a throwaway environment, tools/list only. 86 version-to-version transitions. 40 of them changed what the tools do or accept, 47%. The largest category was not added tools. It was rewritten parameter descriptions, 92 of them. The agent reads that text and acts on it. The person approved the server once and never reads it again.

mcpgawk sits on that gap: baseline at approval, refuse the call when the surface changes until a human looks. https://mcp.gawk.dev/ph if you want to try it.

So: when you add an MCP server to Claude Code, Cursor or Codex, what do you do at its next version? Pin it, re-read the tools, or trust the update?

Hey everyone! Ramiro here, CEO of Rambox.

One of the newest things in Rambox v3 is MCP support (Settings Remote Control). It lets AI assistants like Claude talk directly to Rambox and the apps you have open in it.

TL;DR: every fact in our graph is its own node, with properties on it. The vocabulary behind those nodes is RDF/OWL. Neither camp would call this clean, and Wikidata is not strictly one or the other either.

Here is why we ended up there.

What forced it

Our facts need to carry things that are not part of the fact: where it came from, how confident we are, whether a human has reviewed it, and when it was true. Not this document mentions X , but this specific assertion came from page 4 of this version of this document, confidence 0.8, reviewed by nobody yet, valid from March .

A landing page on static hosting almost never goes down. The API, the app and the database behind it do, and the homepage stays green the whole time.