DemoKit reads your codebase, plans the demo, drives the real app in a browser, films it in 4K with Cap's camera, then checks every step three ways — and writes no file unless the feature actually worked. CLI + agent skill. AGPL-3.0. npm i -g @dekai/demokit
Hey Product Hunt — I'm Dekai. I built DemoKit because my own demo videos kept coming out wrong: you record the take where it worked, quietly delete the four where it didn't, and six weeks later the video still shows a green checkmark long after the feature stopped producing one.
So DemoKit carries the claim with the recording. An agent reads your codebase and writes the plan, DemoKit drives the real app in headless Chromium, films it in 4K with the camera ported from Cap (the spring zoom, the real cursor), and then judges every step three times — state, source pixels, delivered frames. If they disagree, there's no file. Just a report naming the step.
Everything in the gallery is a real take on a real site, no fixtures: an Amazon errand that hits the protection-plan upsell and says no thanks (7/7), a 42s PR review on vercel/next.js (8/8), Google Maps on a WebGL canvas (4/4). The second video is the one it refused — that's the feature.
It's open source (AGPL-3.0, because the camera is Cap's) and one line to try: npm i -g @dekai/demokit. Where it loses, plainly: no narration or captions yet, macOS/Linux only, and it needs one `demokit login` for signed-in apps. Would love to see what you record with it.