dotdotduck

Turn your existing site into an AI-native site

Open Source

Developer Tools

Artificial Intelligence

GitHub

Visit WebsiteSee on Product HuntTwitter ⧉Github ⧉

Hey Product Hunt, I'm Kuei, co-founder of dotdotduck — and we're relaunching. There's a new demo video at the top of the page. If you've got a minute, watch that first; it shows the whole thing end to end. When we first posted, one line did all the work in the comments: clicks tell you what happened; intent signals tell you what the user wanted to happen. Analytics stops at the click. We're trying to capture the gap between what someone intended and what they actually did — every accept, reject, hesitation, and dead-end — as a structured signal. The rest is deliberately boring. Instead of bolting a chatbot to the corner, we leave your interface exactly as it is and let AI work inside it: a command palette, an on-page agent that operates the real page, inline AI on any selection, voice, and long-press. Our take on trust: it doesn't come from replacing your UI with AI; it comes from letting AI work inside an interface people already understand. So every agent action asks for approval first — it can do most of the work, but the last step stays with you — and every rejection becomes another signal showing where it misread you. To be straight about where we are: we're observing behavior, not claiming proof. We treat these signals as research inputs, not validation. The hard part was never collecting them — it's knowing which ones actually mean something before they're allowed to steer a roadmap. That last part is where I'd genuinely love your thinking: 🔍 Signal vs. noise: UI-level intent capture can produce a flood of data. Where's the line between a signal that's truly actionable and noise dressed up as insight? 🤝 Making value legible: Behavioral capture only feels fair when the payoff is visible. How do you connect "I was observed" to "this got better" for the user, not just the team? 🕹️ Recovery and trust: When an agent operates the real page, where should the human-in-the-loop line sit — approve before, interrupt during, or undo after? Does asking for approval kill the magic, or build it? 📐 The shift itself: If AI becomes part of the interface, do buttons and dashboards start to matter less — or more? Want to feel it rather than read about it? Open the demo, hit Ctrl/⌘+K, and run /introduce. Then select some text for Inline AI, or hold Space to talk to the agent. (If it ever feels underwhelming, it's almost always because /introduce got skipped — that flow builds the history the dashboard reads.) We're proudly open source (AGPL-3.0). Fork it, break it, or open an issue on GitHub — we read every single one. Thanks for checking out dotdotduck!