featurecast

Your AI agent films your product videos itself.

Open Source, 2026

37.5 frames per second that carry a new picture

Code on GitHub

The script the agent wrote

export default async function transkriptSprung(
  page: RecordPage,
  demo: Demo,
): Promise<void> {
  await demo.hold(1000)
  await ruhigKlicken(demo, ABSPIELEN)
  await demo.hold(2000 - NACH_KLICK_MS)
  await inDieMitte(page, demo, SATZ, { tempo: FILM_SCROLL_TEMPO })
  await ruhigKlicken(demo, SATZ)
  await warteAuf(page, SATZ_AKTIV, 3000)
  await demo.hold(NACHLAUF_MS - NACH_KLICK_MS)
}

One take, two formats: desktop and phone.

Every new feature needs a demo video, for the website, for LinkedIn, in portrait for phones. So you record your screen, cut it, and after the next update none of it matches. The phone version ends up as a cropped desktop video with a mouse pointer on it.

What I built

Finished product video from a single take

I built featurecast as a tool for AI agents. You hand it to the agent, it writes the script, opens your web product in a browser and records it on its own. Pointer, zoom and camera are added afterwards, from a log of the recording. So a different look costs seconds of rendering and no second take.