Headless CLI & Agent Skill

Let your AI agent film the feature it just shipped.

Coding agents can write features, run tests, and open pull requests — but they usually stop short of showing the UI in motion. Ocelo ships a complete headless CLI inside the app bundle so scripts and agents can capture a window, apply eased zooms and captions, and export a finished MP4 or GIF in one pass.

Download for macOS See pricing

macOS 15 (Sequoia)+ · Apple silicon · 5.5 MB download · Every feature. Clean 720p exports up to 60 seconds.

Deterministic JSON in, MP4 out

Pass `--record --window "Safari" --zooms @zooms.json --captions @captions.json --json`. Ocelo streams progress on stderr (`· recording`) and prints a machine-readable JSON summary on stdout when the export lands.

Choreographed storyboards (`shoot.sh`)

Coordinate browser scrolls, cursor paths, AppleScript actions, camera zooms, and timed captions from a single JSON storyboard so camera movement and UI actions never drift apart.

No Accessibility keystroke tap in the app

Ocelo never asks for Accessibility permission to record your screen. Synthetic clicks for automated shoots live in a separate local helper (`ocelo-drive`) that never runs during normal customer recordings.

Headless workflow

Install the Ocelo agent skill in one line, or call the binary directly:

curl -fsSL https://ocelo.app/install-skill.sh | sh

/Applications/Ocelo.app/Contents/MacOS/Ocelo --record \
  --window "Safari" \
  --seconds 10 \
  --zooms '[{"at":2.0,"u":0.5,"v":0.42,"hold":3.0,"scale":1.9}]' \
  --captions '[{"at":0.6,"hold":2.8,"text":"Shipped in 10 seconds"}]' \
  --out demo.mp4 --json

Read the complete machine-readable reference at /skills.md or /llms.txt.

How it compares

Designed for macOS, not a cloud dashboard.

What matters	Ocelo	Typical cloud / Electron recorder
Agent & CLI control	Built-in `--list-windows --json` and `--record --json` CLI	GUI-only picker requiring manual mouse clicks
Automated camera work	Declarative `--zooms`, `--captions`, and `--pointer` JSON specs	Static Playwright/FFmpeg capture with no zoom or framing
Synchronization	Explicit `· recording` stderr readiness signal	Fragile `sleep` guesses that clip the first seconds
Agent skill manual	Published at ocelo.app/skills.md & llms.txt	None

Pricing

Pay once, or pay a little each year.

Both give you the same app with nothing held back. Tax is handled by Polar, our merchant of record, and shown before you pay.

Best value

$79once

Yours to keep. No renewal.

Buy it outright

Every feature — there is no upper tier
Up to 3 Macs
Captions generated on your Mac, no per-minute charge
Full-resolution exports with no duration limit
Works offline
Every update to version 1

$24a year

Cancel whenever you like.

Subscribe yearly

Every feature — there is no upper tier
Up to 3 Macs
Captions generated on your Mac, no per-minute charge
Full-resolution exports with no duration limit
Works offline
Updates for as long as you subscribe

Free, and it doesn't expire

Every feature, on your real work, for as long as you like. Export clean 720p videos up to 60 seconds — record as long as you want and cut it down. Nothing else is held back: no feature is greyed out and there is nothing to sign up for.

Download and try it

Questions

Straight answers.

How does an agent know where to zoom if nobody clicked the mouse?

When a script drives the browser, there are no human clicks for auto-zoom to cluster. You pass explicit normalized coordinates (`u`, `v` from 0 to 1) and hold times via `--zooms`, and Ocelo computes the same critically damped spring camera path and motion blur as a manual take.

Can it run on a headless Linux CI container?

No. Ocelo uses Apple’s ScreenCaptureKit and Metal-accelerated Core Image on macOS 15+, which require a logged-in macOS GUI session on Apple silicon.

Where is the full agent specification?

Agents can fetch https://ocelo.app/skills.md directly or use the one-line installer (`curl -fsSL https://ocelo.app/install-skill.sh | sh`) to drop the skill into `.agents/skills/ocelo-site-demo/SKILL.md`.

More workflows

No mandatory subscription