You write
One sentence.
The length, the audience, and where the video should end. No selectors, no coordinates, no step list, no timings. The agent resolves every element live against the real screen.
Agent-driven demosThe whole production chain
Describe the video. Your agent records it.
One sentence in, a finished MP4 out. Fraime.it hands Claude Code, Codex, or Cursor the entire chain: staging the window, driving the app on screen, recording it, cutting the dead air, framing the shots, checking the result, and exporting.
Unlimited exports, no watermark, no signup. Works with your existing agent, and the bridge that drives it needs Fraime.it Pro.
What you type
“Record a 45 second walkthrough of our pricing page that ends on the enterprise tier.”
That is the entire input. The agent works out the rest against the live page.
What comes back
- An MP4 at the path you named, close to the length you asked for
- Chapters at each beat of the story
- Zooms framing whatever was clicked, and a smooth synthetic cursor
- No dead air, no automation banner, no narration
- An editable Fraime.it project, with the original intent saved alongside it
Division of labor
You bring the intent. The agent brings the script.
Most demo tools work the other way round: you write the steps, and the tool replays them.
The agent writes
The script.
It turns your sentence into ordered beats, each with an editorial label, a plain-language instruction, the visible state that proves it worked, and an estimated on-camera duration. Then it rehearses them against the real screen.
Fraime.it owns
The recording.
Window geometry, the capture clock, the timeline, undo history, and the export path. Everything lands in an ordinary project you can reopen, recut, or throw away.
The chain
Eight stages, from a sentence to a finished file.
Stage 1
Prompt
You describe the outcome. The agent asks a question only when the answer would change the story.
Stage 2
Script
It drafts four to eight beats, estimates the on-camera time for each, and checks the total against the length you asked for.
Stage 3
Rehearse
It walks the whole route once without recording, finding the cookie banners, lazy-loaded sections, and new-tab links that ruin a take.
Stage 4
Stage
The window is sized once and verified to the pixel. From here the geometry is frozen, and drift stops the run.
Stage 5
Record
Capture runs paused by default and resumes only around each visible action, so thinking time never reaches the timeline.
Stage 6
Cut
Pacing removes the remaining gaps using the exact recorded spans of each action, then trims the tail after the final beat.
Stage 7
Polish
Beats become chapters and zoom framing. Title cards, callouts, spotlights, and overlays go on through previewable edit plans.
Stage 8
Verify and ship
It renders composited frames to PNG and inspects them for framing and legibility, then exports to the path you named.
Browsers, terminals, IDEs
Anything you can put in a window.
A web flow, a terminal session, an IDE walkthrough, a coding agent recording its own work, or two windows in one story.
A web flow
“Record a 45 second walkthrough of our pricing page that ends on the enterprise tier.”
Fraime.it launches and owns an isolated Chrome window, so your real profile, tabs, cookies, and extensions stay out of the shot, with no automation banner.
A terminal session
“Record a terminal demo of installing the CLI and running the first command, about 30 seconds.”
Browser control is a separate opt-in download. Recording a terminal, an IDE, or a coding agent working through a task does not need it.
Two windows
“Record a two-window demo: run the build in the terminal, then swap to Chrome to show the deployed page.”
Swap between two windows or show them side by side. Each swap becomes its own editable clip, with clean cuts at the seams so you can add transitions later.
Why it does not look like automation
The agent thinks off camera.
Capture starts paused and stays paused. Each visible action resumes recording immediately before it happens and pauses again once it settles. The model’s reasoning, the snapshots it takes to decide what to click, and the latency between tool calls are never captured at all.
Every action is timestamped on the capture clock, so the cleanup afterwards works from exact recorded spans rather than guessing at video frames. What reaches the timeline is the demo.
It stops instead of guessing
A login screen, a CAPTCHA, a purchase, unexpected private data, or two elements that could both satisfy an instruction all end the run and come back to you.
An isolated browser
The agent drives a throwaway Chrome profile that Fraime.it launched, never your everyday browser, and never through a debugger extension.
You still watch it
Credential detection is an on-device tool you run deliberately, not part of this pipeline. Check the redactions and watch the export before publishing.
Describe your next demo instead of recording it.
Install the app, connect the agent you already use, and write the sentence.