Your AI agents keep grabbing your screen. Make them stay backstage.
Anyone running browser automation — Claude, CDP seats, Playwright, agent stacks —
knows the moment: a window you didn't ask for jumps in front of you, your dictation dies
mid-sentence, your typing lands in the wrong app. Backstage ends it.
macOS 13+ · notarized · runs entirely on your machine · no account, no permissions requested
Why this exists
We built Backstage because it happened to us. Our own agent's browser stole the founder's
screen every fifteen minutes, mid-dictation, until the words he was speaking were being typed
into the agent's window instead of his own. We traced the exact line of code that does it —
a standard call every browser-automation stack makes — and realized every AI-agent user has
this problem. Backstage is the fix: your agents keep working, in the background, where they
belong. Your screen stays yours.
What it does
Bounces uninvited grabs and puts you back where you were — measured in
single-digit milliseconds on the shipped build.
Knows your browser from your agent's browser, per running process —
your own Chrome is never touched, even though it's the same app.
Your clicks always win. When you open a guarded app, it opens.
Only programmatic grabs get bounced.
Dictation shield. Engage it and nothing may steal focus at all while you speak.
Agents can ask permission. A one-line CLI grants a timed on-screen pass —
for the login your agent genuinely needs you to type. It expires by itself.
Storm-proof by design. A misbehaving agent trips a breaker instead of
turning the guard into its own problem. Every decision is logged, locally.
Runs entirely on your machine. No account, no network calls, nothing leaves your Mac.
What that's worth
Your minutes back. Every steal costs the sentence you were dictating, the
copy-paste recovery dance, and the thirty seconds of finding your place — dozens of times a
day if your agents run on timers.
Your words land where you aimed them. No more dictation silently typed
into an agent's browser.
Your agents get more useful, not less. With focus theft off the table, you
can run more automation with confidence instead of dreading the next pop-up.
Trust you can check. A published checksum, a signed release manifest, and
a claims table generated from tests that actually ran.
Proven, not promised
Backstage v0.1.1 — what this build is proven to do (generated 2026-08-31 06:13Z)
Backstage is sold as the current version it is at: the first, smallest high-impact build.
Every claim below is generated from a falsifier that ran end-to-end on real machines; the
scorecard rows are the source of truth. Nothing on this page is hand-written marketing.
Proven, end-to-end
S1+S6 CDP raise storm bounced then breaker — 6 bounces max 4ms, BREAKER_OPEN at 7th — guard can never become the load (mode: Backstage.app v0.1.0, live seat, HS guard stopped; 2026-08-31_0231Z)
Backstage is built by Vox Ordo — we run AI agents
all day on our own machines; this tool exists because the problem drove us crazy first.
Support: hello@voxordo.ai