Your agent ships the video too.
Paste one line where your agent works on your app. It records the app, cuts the video, and hands you a link.
Read https://vos.so/agent.md, do what it says, and give me the link.
> Read https://vos.so/agent.md, do what it says, and give me the link.
- Installed @vosjs/cli and ran vos setup
- Found the app: pnpm dev → localhost:3000
- Scripted the walk: pick a preset, tune a colour, copy the CSS
- Recorded a 29 s take, zooms planned from the clicks
- Pushed it to vos.so
- Here is your link: vos.so/vos/995a0b5d
tweakcn: tune a preset, copy the CSS
vos.so/vos/995a0b5d
Data, not pixels
Every decision lives in a document your agent can read, patch and hand back.
"zoom": [ - { "at": 2.1, "level": 2.2 }+ { "at": 2.1, "level": 1.8 } ]
Edits are patches. A value changes; nothing re-records.
Rules in one file. vos setup writes them for you.
A person takes over. The studio opens on the same take.
On every pull request
The Action records the product in CI and comments with the video.
name: release media
on:
pull_request:
push:
tags: ['v*']
permissions:
contents: read
pull-requests: write
jobs:
media:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-node@v4
with: { node-version: 22 }
- uses: vosjs/action@v1
with:
key: ${{ secrets.VOS_API_KEY }}Questions
Can an AI agent make videos with vosso?
Yes. The vos CLI runs the whole pipeline headless: vos record drives a page from an actions.json script and records it, the take’s doc.json holds every editable decision as plain JSON, and vos render produces the final video. Logs go to stderr, results to stdout, and --json streams NDJSON events, so no TTY is needed.
How does an agent edit a vosso video?
By patching JSON, not regenerating code. A take’s doc.json contains zoom spans, trims, speed ramps, and styling as data; the rendering program is constant. An agent changes a number and re-renders: the browser flow never re-runs, and manual edits survive re-planning.
Is the output deterministic?
Yes, by construction: the engine owns the master clock and its own tween sampler, so a given take and doc.json render the same frames every time, on the agent’s machine, in CI, or on a server. That makes video diffs reviewable the way code diffs are.
Can my agent put a video it already made on vos.so?
Yes. Paste the host line to it: it rewrites a Remotion or HyperFrames composition, an HTML page or a script so the words and colours are data, pushes it, and hands you a link that plays and keeps every version. Without a credential the push is claimable, and you claim it with a free account.
Is there machine-readable documentation?
vos.so/llms.txt carries the CLI commands, config schemas, and pipeline description in plain text. The @vosjs/cli README documents the take format and exit codes. Both are written to be pasted into an agent’s context.
Something else? hello@vos.so
Ship the video with the release.
Free. Export up to 4K with no watermark, ever, and record takes up to 30 minutes.