Launched Loamly today - AI brand monitoring + traffic detection (open source)
Hey IH 👋
I'm Marco. Just launched Loamly after about a year of building and wanted to share it here.
The problem I kept noticing
I use ChatGPT, Claude, and Perplexity for maybe half my research now. Need a tool recommendation? Ask AI. Comparing options? AI.
I started wondering: what do brands actually see on their end? When AI recommends you (or doesn't), can you even know?
Turns out: no. This is completely invisible to most businesses.
What Loamly does
Two things:
-
Brand monitoring — See what AI platforms actually say when someone asks about your brand, your competitors, or your industry. Track how these answers change over time.
-
Traffic detection — Know when AI sends visitors to your site. This traffic exists but most analytics can't identify it.
The traffic detection piece is fully open source: github.com/loamly/loamly
Where I'm at
- $120 MRR from a couple of test users including the company I work for
- Free tier: 5K credits/month
- Also launched today on PH and HN
Would genuinely love feedback from this community. What's unclear? What would make this more useful for your product?
Validating an idea to help professionals reply safely to difficult work messages
This is solving a real blind spot. The "AI visibility" problem is going to become a bigger deal as more purchase decisions flow through LLM recommendations.
A few thoughts:
The attribution challenge - How are you handling the detection when AI traffic doesn't have clear referrer data? I imagine a lot of it shows up as direct traffic. Are you fingerprinting based on user behavior patterns or something else?
Competitive monitoring angle - The brand monitoring piece could be really valuable for tracking how AI models position you vs competitors. Do you show side-by-side comparisons or sentiment trends?
Open source strategy - Smart move making the traffic detection open source. That's the kind of thing that could become an industry standard if adoption spreads. What made you decide to open source that part specifically vs keeping it proprietary?
The $120 MRR on launch day is a solid signal. Congrats on shipping - this feels like good timing given how fast AI search is growing.
Thank you very much for your thoughts @demogod_ai really appreciate it.
On detection without referrer data:
I've tried and decided to use a few different approaches depending on the AI platform:
ChatGPT (OpenAI's browsing agent) sends cryptographically signed requests using RFC 9421 HTTP Message Signatures. We verify these against OpenAI's public keys. This gives us 100% accuracy for that traffic!
For Claude, Perplexity, Gemini - we use behavioral detection: navigation timing patterns (paste vs click), how the browser interacts with the page, request header analysis. It's pattern matching based on actual AI platform behavior.
The traffic detection code is fully open source so you can see exactly how it works: github.com/loamly/loamly
On competitive monitoring:
Yes, we show side-by-side. You can track how AI answers change over time for your brand vs competitors. The Intelligence dashboard lets you see which prompts mention you, which mention competitors, and how that shifts. It's basically "what does AI actually say when someone asks about your category?"
On open sourcing the detection:
Honestly, trust. This is a new category and people are rightfully skeptical about what's being tracked. By open sourcing the detection, developers can audit the code and verify there's no dark pattern. It also means the community can improve the detection methods. I personally have used probably 20+ tools this year to build Loamly and noticed robustness and loved most of the products that were open source (supabase, firecrawl.dev, tons of UI components etc...)
The brand monitoring side stays proprietary because it needs infrastructure (API calls to AI platforms, scheduled jobs, data storage). But the core detection logic being open felt like the right call for adoption.
Thanks for the kind words on the $120 MRR - small but and to be honest it was already before the "official" public launch.
The RFC 9421 signed requests from OpenAI is clever - didn't realize they were doing cryptographic verification at that level. That's a much stronger signal than behavioral heuristics alone.
The behavioral detection for Claude/Perplexity/Gemini makes sense as a fallback. Curious if you've seen false positive rates vary much across different site architectures (SPAs vs traditional server-rendered pages, etc.)?
The trust angle on open sourcing makes sense. In the analytics space especially, there's so much baggage from tracking scripts that being auditable is a real differentiator. Similar playbook to what made Plausible/Fathom resonate.
One thing that could be interesting for the roadmap: correlation between AI mention sentiment and actual traffic conversion. If you're tracking both "what AI says about you" and "when AI sends traffic," there's probably signal in connecting those - like seeing if certain phrasings in AI responses correlate with higher intent visitors.
Good luck with the PH/HN launches - triple-launching is ambitious but smart for coverage.
On SPA vs server-rendered: The navigation timing detection happens on initial page load only - how the visitor first arrived. Doesn't really matter if it's an SPA or traditional site for that signal. The RFC 9421 verification is just header validation, architecture-agnostic.
Haven't seen meaningful differences in false positives between site types yet, but still early on volume :)
On sentiment → conversion:
Yeah, that's exactly where I want to go. Right now the two pieces are separate - you can see what AI says, and you can see who visited. Connecting them is the obvious next step. Not built yet but it's top of the list.
Thanks for thinking through this with me.