Connect with the Product Hunt community

We built a free plan thinking it was the fastest way to get user feedback. Instead, we ended up with hundreds of active users who never upgraded and kept requesting edge-case features.

When we turned off the free tier and switched strictly to a 14-day trial, signups dropped but active usage and actual conversion spiked. Paid users were using the core features differently than free users.

Did a free tier help you discover real PMF or did it just pollute your user feedback loop?

Six months ago, our conversion rate from trial to paid was a respectable 22%. The catch? We required a credit card upfront for our 14-day trial.

We started getting feedback (mostly on Twitter and Reddit) that we were "anti-consumer" and that people hated being forced to enter card details just to test a product. So, we decided to test a frictionless trial. We removed the credit card requirement entirely.

The immediate result was a massive spike in signups. Our weekly trial volume went from about 40 people to 250 people. I thought we had finally unlocked growth.

But I just pulled our cohort data for the last 60 days, and the reality is brutal:

For the first year of our SaaS, all of our sales were done manually. I would hop on a Zoom call, walk the prospect through the product, answer their questions, and send them a Stripe link. It wasn't scalable, but it worked. We had 50 customers paying us $99/month.

Then, I drank the Product-Led Growth (PLG) Kool-Aid. I read all the blog posts about how "frictionless checkout" and "self-serve onboarding" were the future. I convinced my team to spend six months and $15,000 (we hired a contractor) building a complete self-serve funnel: a new marketing site, automated onboarding emails, and a seamless Stripe checkout integration. We removed the "Book a Demo" button entirely.

We launched the new self-serve experience two months ago, and our revenue has actually gone down.

When I started calling the leads who visited the pricing page but didn't buy, the feedback was unanimous:
"I had a few questions about how your security works, but there was no way to contact you without buying first. I just went with your competitor who offered a demo."
"I needed to know if this integrates with our legacy system, but I wasn't going to put in my credit card just to find out."

I have a bunch of updates to products I need to do (and new products ready for release), but I lost access to the Apple ID that I set up about 15 years ago and haven't been using, except it is the owner of my developer account.

I have been going through their process to get my account back multiple times, and each time they have a ten day waiting period after which, for me, nothing happens. This is getting extremely frustrating and there seems to be no way of getting through to anyone that can help me there.

Bring your own Anthropic, OpenAI, Google, or Mistral AI key for localization. Requests go directly to the provider on your key, and the provider bills you for those tokens.

Open Settings BYOK, add your provider key, and choose that provider s models in your localization engine.

Your key is used for the engine s translation and editing steps. Your integrations keep using its glossary, rules, and brand voice, while Lingo.dev runs the localization infrastructure.

DungeonQ is a defensive deception runtime for suspicious AI-agent sessions. A trusted gateway routes a designated session into a persistent synthetic world, with useful world-only Wrong Tickets, observable activity and separately approved bounded changes. The goal is to give people and authorized defender agents room to act.

v0.16 addresses a concrete lesson: protecting one route is not enough.

In the earlier Sol experiment, all three diverted sessions received the artificial target through an open alternate source, A2, despite zero reads through the main source route. Those failures remain published.

The repair applies shared tenant, resource and credential-version checks to the main source, alternate source, cache and export. A2 keeps no target copy. Cached content requires a fresh authority check; an unavailable origin produces UNKNOWN instead of stale disclosure. A world ticket or claimed role cannot grant source access.

rhun is a small, open-source code editor. It has extensions now, and I d love to see what people make with them.

We ve added Claude Code and Codex skills for building and testing your own extension, plus a skill for porting open-source VS Code extensions. Describe the tool you want, then work on it with your agent. Rust/Go templates, examples and CLI commands give you a place to start and a way to share it.

What s the oddly specific extension you ve always wanted?

Try rhun: https://rhun.app

@CybeDefend is launching again today, introducing a one command line to secure coding agents, @Cursor, @Claude Code and else.

Highlighting here some comments from the community.

Hey everyone, following up on our launch with a question about where rhun should go next.

The idea behind rhun is a small, fast editor for working alongside coding agents. You can already run agents in the terminal, follow Claude Code and Codex sessions in the agents panel, and review changes with built-in Git tools.

I'm especially interested in two things:

• AI integrations: Which tools do you use, and what would you want rhun to do with them? Send selected code to an agent, manage sessions, make reviewing agent changes easier, or something else?

• Missing features: If you ve tried rhun, what made you switch back to your usual editor? What s the one feature you couldn t comfortably work without?

When I launched Panza on Product Hunt a few days ago, I had pretty much zero expectations.

No big campaign, no coordinated launch effort. I just put my little indie app out there to see what would happen.

70 upvotes later, I m still a little blown away.

AgentDock 1.4 is out. The short version: it now works on a Mac without cmux, tells you what your agents tokens would cost at API prices, lets you leave a message for a busy agent, answers questions from other agents over MCP, and reads seven more agents.

Works without cmux

• If you run Claude Code or Codex in any terminal, or in the Claude or Codex app, AgentDock now shows them on a Mac that doesn t have cmux. Before, the strip and the menu showed only the cmux error. cmux is optional.

Costs at API prices

Hey makers

NotchDodo now keeps your Mac awake while your AI agent works, then puts it to sleep once the job is done and you're away.

I normally keep 8 or more sessions of Claude Code open simultaneously. At some point, I discovered each new session was launching a complete copy of the MCP servers I've set up so far (Gmail, Calendar, Drive, Zoom, LINE and one more). Each one required approximately 560 megabytes, so my computer was carrying multiple copies of these processes at all times, all through the day, even when nothing was calling them.

Thus, I decided to get rid of them from the always-on list and start them only when necessary. Now I only need to execute a small command (mcp-call) to get a particular tool running. For instance, running a search on my emails is simply mcp-call gmail-local with the search tool and a query, and the process finishes right after that. The agent can still find the tools since a short skill lists them, and the hooks that used to check the old MCP calls (the HTML format of emails, asking before anything gets sent or deleted) now run on the new command too. I actually thought it was likely that at least one of these would break.

Portlivo 1.1 is out, a free update for everyone.

A zero-click search happens when users get the information they need directly from search results without clicking through to another website.

Zero-click searches accounted for 60.45% of Google searches in 2024, rising to 68% in early 2026.

Every morning, Claude Code checks all 7+ of my projects in Vibesmonitor over MCP.

Traffic and engagement overview plus recommendations for next steps. All depending on your scheduled prompt.

Mine also has instructions ff there was an error in the last 24 hours (in the backend or on a client device in the frontend). It finds the cause in the repo and opens a PR with the fix.

Every morning, Claude Code checks all 7+ projects of mine in Vibesmonitor over MCP. If something broke overnight, it finds the cause in the repo and opens a PR with the fix.

So I wake up to a PR and review it with my tea.

Hey everyone, thanks for following Agent Pets

Since launch, the most-asked question was "does it work with Codex?" It does now:

• Every Codex session gets its own pet on every page in Chrome

• When Codex wants to run a command, the pet bounces: Allow or Deny right from the tab you're on

• Type a reply into the pet and Codex gets it, whether it's idle or mid-task

• It still works with Claude Code, OpenCode and omp, so if you run several agents side by side you see them all in one place.

Setup: npx getagentpets init, then open Codex once and trust the hooks with /hooks.

BiBimba 1.6.0 is out!

Gathering links for a message, or notes from different places? You can now select several text items from your clipboard history and paste them together, instead of going back for each one.

Before pasting, you can rearrange the items and choose how to separate them: new lines, blank lines, commas, slashes, or tabs. Saved snippets work too.

AI can get me to a working app in a day, but "it works on my machine" and "it's safe for real users" are very different things.

The part that worries me is the boring stuff that's easy to miss when the code looks clean and everything runs: API keys sitting in the frontend, auth checks that exist on the page but not on the endpoint, database rules left wide open from the prototype stage, no rate limits on anything. The AI rarely warns you about these unless you ask, and by the time you think to ask, you've already moved on to the next feature.

Right now my pre-launch process is mostly manual and a bit random: I skim through the auth flow, search the repo for hardcoded secrets, and try to break things as a logged-out user. It catches some things, but I never feel fully confident.

I'd love to hear how others handle it:

The one where a new email no longer holds you in a queue behind itself. You tap the notification, Skim was already downloading that exact message, and it used to make you wait for that download, give up on it, then start again from scratch. Efficiency!

Fixed

• Opening a message you just got is quick again. Clicking a message that had just arrived could hang on "Fetching from the server " for up to a minute. Your click used to wait in line behind the background download that runs when mail lands, including the download of the very message you clicked. Now clicks go first.

• Work in progress isn't thrown away. If you open the message Skim is already downloading, the open waits for that download to finish and uses it. Before, it cancelled the download, reconnected and fetched the same message again.

• Bursts of mail don't pile up. When several messages land at once, their background downloads merge into one batch with one time limit. They used to queue up one after another, each with its own 15 seconds.

Housekeeping

Hey everyone,

A quick look at what we're about to ship. Render Studio 2.0 will land in Autonomyware.

Autonomyware takes a product from a single prompt to engineered parts and a full bill of materials, autonomously. Render Studio 2.0 is where that model becomes a shot you can actually use: a hero image for a pitch deck, a store listing, a client review or a turntable for social.

Here's what's coming:

Lancetta 0.62.0 is out Maintenance now cleans up after your coding agents, and reads Claude Code's settings the way Claude Code does.

1. Storage

Codex installs that stopped halfway, old server releases, empty git folders, files ChatGPT copied out, build caches of ended sessions, Chrome's clones: each with what it frees and why it can go. Nothing a running program holds is offered, and it all goes to the Trash.

A user uploads a file, starts an export, and closes the tab. The work may still be running, but when they come back, they need to know what happened.

I am building Inquir Compute for apps, databases and background jobs. I am interested in the part users see, beyond whether the job itself succeeds.

I just shipped AIRUNCODE 1.5.5, a desktop environment for building apps and games with AI agents on your own computer.
Here's what changed:
Parallel agents are back in the chat. You can follow each worker's progress while they work on the same project.
The OpenRouter and NVIDIA model picker now separates models verified with your account from those still pending a check or unavailable. Availability is checked again every six hours.
V-CORE can bring Luanti mods from ContentDB into the game workflow. This release also adds inline questions from the model and optional authenticator-based two-step verification.
Pro is now US$5 for 30 days, down from US$15 (67% off). It's a one-time purchase with no automatic renewal. The free option is still available.
We've also applied to Startup Venezuela. The application is a chance to put this project in front of more builders and hear where the product still falls short.
AIRUNCODE 1.5.5 is available for Windows, macOS and Linux: https://airuncode.com/
If you try it, I'd love direct feedback: can you tell what each agent is doing, and does the model picker make availability clear enough?
Gustavo, maker of AIRUNCODE