Ultrademo is a capture-and-render pipeline for demo videos of any website or SaaS application, driven by a Claude Code skill (an AGENTS.md brings Codex, Cursor, and other coding agents along). Point it at a URL or your codebase, review the script it drafts, and get a finished MP4 with high quality narration. It all runs on your machine: your credentials and app data never leave your environment.
Apache-2.0 · runs locally · bring your own ElevenLabs key or use free offline voices
A launch video costs $2,000 to $15,000 with an agency, or an entire weekend with a video editing software. And it becomes stale the moment your UI changes.
Showcase
The hero video above and every video below came straight out of the pipeline: scouted, scripted, captured, narrated, and rendered by the skill. No timeline, no post-production, no retouching.
Narration: ElevenLabs · captions burned in · 1080p MP4, with vertical and GIF cuts from the same capture.
How it works
01
"Make a 2-minute walkthrough of our onboarding flow." Point it at a URL or your codebase, hand it test credentials, add context if you want.
02
Claude scouts your app and drafts a scene-by-scene outline: what's on screen, what the voiceover says. You edit it inline before anything gets recorded. Nothing ships without your sign-off.
03
A browser agent walks through your live app: real clicks, real typing, real screens. TTS then voices every line. Remotion cuts it together with zooms, cursor motion, and captions.
04
Your UI changed next month? Same project, one command, fresh video. Unchanged narration is cached, so a visual refresh costs nothing extra.
The storyboard is the single source of truth. Edit one narration line and re-render in minutes. Re-capture only the affected scenes when your UI changes. Reprompt, not reshoot.
The skill · available now
The full pipeline is on GitHub under Apache-2.0. Not a trial, not a crippled tier: the same code we use for all our own demos.
# get it git clone https://github.com/new-xp/ultrademo ultrademo-workspace cd ultrademo-workspace && npm install npx playwright install chromium npm run doctor # checks node, ffmpeg, voices # optional, for premium narration echo 'ELEVENLABS_API_KEY=sk_...' > .env # then, in Claude Code: › make a 2-minute demo of our onboarding flow
Open the folder in Claude Code and ask for a demo video. The bundled skill scouts your app, drafts the script for your sign-off, writes the capture flow, records, narrates, and renders.
The repo ships an AGENTS.md that hands Codex, OpenCode, Cursor, Gemini CLI, and other coding agents the same playbook. Built and tested with Claude Code; any capable agent can follow it.
Copy the flow template, describe your scenes, then npm run capture · tts · render. Every video is a self-contained project folder you can re-run any time.
Local-first
Demo tools that host your capture also host your credentials and your customers' data. Ultrademo doesn't. The browser runs on your laptop, the frames render on your laptop, and in-page redaction blurs sensitive content before a single frame is saved. Unredacted pixels never exist anywhere but your screen.
The skill also runs a secrets sweep on the final frames (URL tokens, keys in forms) before anything is presented back to you.
Blur by pattern or selector. Emails, account numbers, a whole customer-name column. It survives scrolling and navigation, and never modifies your app or its data.
Re-runs
Every video is a stored project in your machine: the flow, the script, the voice, the settings. When you ship an update, just re-run the skill. Cosmetic refresh? One command, done. A selector broke? The skill re-scouts just that surface and patches the flow. Feature changed? Edit the narration and only the edited lines get re-synthesized. And if you work in the repo, the skill reads your diff since the last capture to know which scenes need repair before it re-runs anything.
One command whenever you ship. Renders archive automatically, so every re-run leaves a before and after pair.
A deploy triggers a fresh capture. Your demo updates itself on the same pipeline that ships your code.
We watch for changes, repair broken selectors, verify the new video is faithful, and hand you the diff. You just approve.
Coming soon · join the waitlist
The skill is for builders. The app is for everyone who'd rather click than type, plus the things that only work in a hosted environment.
Every video as an editable filmstrip. Reword a narration line, reorder scenes, re-voice, re-capture one shot, re-render. No timeline, no keyframes, no editor skills.
Paste a URL, add a one-paragraph brief, approve the script. Watch the video assemble itself scene by scene, live.
No API keys. Metered studio-grade narration out of the box, with voice cloning so the founder can narrate without recording a take.
Every render gets a branded, embeddable page. Send a link instead of a 60 MB file.
Connect GitHub or Linear. When a feature ships, Ultrademo re-captures, re-renders, and tells you exactly which scenes changed.
Started local? Open the same project in the app. One format end to end, nothing to migrate.
Triggered by deploy #214 · main · the invite modal moved to the top bar. The flow selector was repaired automatically and the new take matches your approved script. 9 scenes unchanged, narration untouched (nothing re-billed).
Web App · Early Access
Early access rolls out in order, and early users lock in exclusive launch discounts and founding-member perks. Skill users get bumped up the list: include your GitHub handle and we'll find your star.
We promise to never spam you. You'll just get the notification when it's your turn.
FAQ
Yes. Apache-2.0, full pipeline, no watermark, no time limit. Our Web App will be a paid product (coming soon) with visual editing, included voices, share pages, and automatic re-generation. Local and DIY through the skill stays free forever.
For the best narration, yes, bring your own key (when re-generating scenes, unchanged lines are cached, so re-renders don't re-bill). If you don't want to use ElevenLabs, Piper TTS gives you free offline narration, and if you're a macOS user, you can fall back to macOS's built-in voice.
The pipeline itself costs nothing. Your only spend is narration: ElevenLabs usage on your own key, with per-line caching keeping re-runs close to zero, or nothing at all with Piper or the macOS voice.
If a browser can reach it, yes: production apps, staging, localhost - works on all. You provide the URL and test credentials; a one-time login command handles signed-in sessions. Web apps only for now, no native apps.
The skill scouts read-only, never saves settings on camera, warns you before actions that cost money, and redacts sensitive data in-page before frames are saved. And it all runs on your machine, so nothing sensitive is uploaded anywhere.
In-page redaction blurs it live during capture, by pattern (emails, account numbers) or by selector (a whole column). It applies to screenshots and video clips alike, and survives scrolling and navigation.
No. It's the recommended way to run the skill and what we build and test with, but the repo ships an AGENTS.md that gives Codex, OpenCode, Cursor, Gemini CLI, and other coding agents the same playbook. The pipeline itself is plain Node and works fully standalone too. The web app (waitlist above) will need none of this.
No. A URL, test credentials, and a one-line brief are enough; that's how we produce videos of third-party apps. But if your code is available, the skill puts it to work: stable selectors straight from source, feature flags and empty states accounted for, demo data seeded against the real schema, and on re-runs it reads your diff to know which scenes need re-capturing.
It deserves a demo video. Try the Ultrademo skill for free.
Early access to the web app comes with exclusive launch discounts and founding-member perks. We promise to never spam you. You'll just get the notification when it's your turn.
We promise to never spam you. You'll just get the notification when it's your turn.
