VibeFrames
Open Studio
alpha · chat-first video editor

Compose videos
through conversation.

You describe what you want. The agent reasons, calls tools, and builds a HyperFrames composition — clip by clip, track by track — while you watch in real time.

See the engineering
~/vibeframes/composition.html
HTML Player

How it works

Three steps from idea to video.

Step 01
Describe

Tell the agent what you want in plain language.

describe a change…
Step 02
Compose

Typed tools mutate the HyperFrames composition tree.

add-cliptrack-bg · 0–3s218ms
add-cliptrack-title · 3–8s218ms
update-cliptrack-bg · 1.2s offset218ms
Step 03
Preview

Watch your video render in real time via SSE streaming.

preview · vibeframes0:02 / 0:30
Launch.
from VibeFrames
live
bgtxtout
The engineering behind it

Agentic Harness, end‑to‑end.

Not a wrapper around streamText. A real Mastra Harness: typed state, mode switching, a tool catalog, on‑demand skills, durable LibSQL persistence, and a custom SSE event bus.

harness · director mode
run · 0R0XKJ2A
make me a 30s launch teaser with a title at the start and an outro CTA at the end
add-cliptrack-bg · 0–3s
<h1 class="text-6xl ...">Launch.</h1>
218ms
add-cliptrack-outro · 27–30s
<a href="/start" class="...">Start now →</a>
218ms
VF

Done — added a hero title from 0–3s on track-bg and an outro CTA from 27–30s on a new track-outro. Preview is live on the right.

composition.delta · clipCount: 2 · trackCount: 2
describe a change…