perfstaq · the content machine

Dump the footage.
Reels ship themselves.

Drop your raw talking-head takes into perfstaq. It finds every cut and shows you why. You say yes. Finished vertical reels come out the other side.

Get early access →Watch the machine run
perfstaq — machine
local
DUMPCUTREEL
MOV

take_03.mov

12:41 · 4K · one ramble

transcribing · per-word timestamps

So,um,thethingaboutlaunchweekis—okaywait,letmetryagain.Thethingaboutlaunchweekisthatnobodycaresyet.

Local-first — no cloud, no uploads

Non-destructive — the source is sacred

Human-gated — nothing ships without you

Early access · rolling invitesBuilt in India · Ships everywhere
SILENCE — CUT“UM” — CUT“SO BASICALLY” — CUTTAKE 1 — CUTTAKE 3 — KEPTDEAD AIR — CUTYOUR TANGENT — KEPT (IT WAS GOOD)“WAIT, LET ME REDO THAT” — CUTTHE POINT — KEPTSILENCE — CUT“UM” — CUT“SO BASICALLY” — CUTTAKE 1 — CUTTAKE 3 — KEPTDEAD AIR — CUTYOUR TANGENT — KEPT (IT WAS GOOD)“WAIT, LET ME REDO THAT” — CUTTHE POINT — KEPT

How footage becomes a stack

01 / DUMP

Empty your camera roll into it.

Drag in the raw takes — long rambles, false starts, all of it. The machine transcribes every word with its timestamp, on your machine, while you go record the next one.

whisper-grade transcription · per-word timestamps · runs local

drop the mess here

MOVtake_01.mov
8:12
MOVtake_02.mov
3:47
MOVrant_about_pricing.mov
12:41

every word, timestamped · nothing uploaded

02 / CUT

It proposes. You dispose.

Silence, fillers, repeated takes, the sentence you fumbled three times — each cut arrives with a reason, timestamps, and a confidence score. Unsure? It keeps the footage. You hold the scissors.

every cut explained · conservative by default · one-tap approve

proposed cuts · you decide

human gate: on
SILENCE

1.9s of dead air

confidence 96%

cutkeep
FILLER

“um, so, basically”

confidence 91%

cutkeep
BAD TAKE

take 1 of 3 — fumbled

confidence 88%

cutkeep
PAUSE

might be dramatic?

confidence 34%

cutkeep

34% sure? it keeps. footage is not the enemy.

03 / REEL

Born vertical.

Approved cuts become tight 9:16 reels with word-synced captions burned in and audio loudness-normalized. The kind of output you'd pay an editor for, minus the editor and the week of waiting.

word-synced captions · 9:16 reframe · loudness normalized

nobodycaresyet.

04 / SHIP

The machine queues. You pull the trigger.

Reels stack into storylines, slot into a schedule, and wait at the publish gate. Instagram first, more rails coming. Nothing — nothing — goes out without your explicit yes.

storyline pools · scheduled posting · human publish gate

publish rail · instagram

gate armed

reel_04.mp4

Tue · 9:00 AM

queued

reel_05.mp4

Thu · 9:00 AM

queued

reel_06.mp4

Sat · 11:00 AM

draft
Waiting at the gateonly you can press this

What the machine believes

01

The source is sacred.

Your original footage is never modified. Every output is derived, every decision reversible.

02

Every cut has a reason.

Silence, filler, repetition, bad take — with timestamps and confidence. No black-box removals.

03

When unsure, keep.

The machine is conservative by design. Footage is not the enemy; losing a good moment is.

04

Never cut mid-word.

Every boundary snaps to a word edge. Your speech is never clipped, ever.

05

You are the editor-in-chief.

The machine proposes; you approve. Your face and your voice ship only on your yes.

06

Local-first, always.

Transcription, detection, rendering — all on your machine. No cloud, no uploads, no leaks.

Fair questions

Ask the hard ones.

Anything a machine that touches your face and voice should answer plainly.

No. Transcription, cut detection, and rendering all run locally. The only network calls are the ones you approve at the publish gate — and those go straight to the platform you're posting to.

Early access

Get in before the machine goes loud.

We onboard in small batches — the people drowning in raw footage first. Tell us what you record and we'll slot you in.

I'm a…

Rolling invites. No newsletters, no spam.