I'm building PulseBoard an uptime monitor with an AI assistant (Vigil) that explains why your site broke, not just that it broke.

Here's a real example from a live site recently:

The Incident:

• A 404 error appeared on the site at 04:10:01 UTC

• Average response time: 570ms (consistent with baseline of 583ms)

• Uptime dropped to 93% (from baseline 95.8%)

Hi Hunters, my name is Alfred and i'm launching on ProductHunt 17th of april.

I know that it's a bit close to the deadline but will anyone be the hunter for my project? Project link is: https://rainz.net and the link to the launch is: https://www.producthunt.com/prod...

Please any hunter answer as we need help.

Over the past few weeks, I've been focused on one question:

Monitoring tells you something broke.
But how do you figure out why?

That's where PulseBoard is heading.

What's new

Hey everyone, I'm building TheSockle, an AI tool for founders who feel their brand messaging is inconsistent before they've even nailed the strategy.

You share context about your brand, answer 18 focused questions, and a workflow of specialized AI agents turns it into a 45-slide Strategic Brand Platform within 24h: positioning, messaging, visual direction, personas and a go-to-market plan.

Hi PH, M lodie here, PM at GetMint
Quick one from me, because these threads are always the most interesting part of the launch.

We're building GetMint, which shows brands how they show up when someone asks ChatGPT or Perplexity about their category, who gets recommended instead of them, and what to actually fix. The tricky part is that AI answers aren't stable, so the real challenge is turning that noise into something a marketer can trust and act on.

Page here if you're curious: GetMint Product Hunt

Now tell me about you: what are you building, who's it for, and what's the hardest part you're solving right now?

Hey Product Hunt solo builder here.

Every multi-agent thing I built ended up with one "manager" object that knew every agent and every step, and an agent harness that was a giant while-loop owning the model call, the tools, the memory, and the approvals. Add a fourth agent and both became the thing that breaks.

So I built Cosmonapse, a distributed agent-to-agent protocol where agents are peers on a shared bus. The model is a nervous system, and the names tell you where logic goes: Neuron thinks (your agent, a pure function), Axon sends, Dendrite reacts, Synapse connects (the bus), Engram remembers.

To build harnesses tool calls, memory, human approval, and retries are typed signals and hooks, not lines in a loop. You grow a harness by adding nodes. An orchestrator is optional, and it's the same primitive as a worker. LLM-agnostic too: each neuron runs whatever model it wants.

claude code/codex can make the deck. the weird part starts after i send it to someone else.

the note usually comes back as a screenshot + "slide 5 here", then i have to work out which version they saw and paste it back into the agent.

i've been testing that handoff with PreApp. one public single-file deck kept all 12 slides, button/keyboard nav, and exact text selection. a second deck used child iframes: playback worked, but precise selection needed slide anchors. that's the kind of edge case i want to find, not another polished demo.

looking for 3 people who already have a non-sensitive HTML deck or report from Claude Code/Codex. i'll help get the first link out, ask one real viewer for one note, pull it back into the agent, and follow it through v2.