Clip pipeline for agents

Your agent can
do this alone.It'll cost 21× more.

Point us at a channel. We find the moments worth cutting, cut them, caption them, and hand back finished vertical clips with the plan for which account posts what, when. One call instead of a four-hour agent loop.

$39/ month  ·  100 clips  ·  REST + MCP  ·  no per-render billing
Start a run
Shelves of tape archives lit by a single shaft of light
The problem

Your best thirty seconds are buried in hour four.

Nobody is going to watch two thousand videos to find them. That is the job, and it is the part that costs money.

Same job, two ways

Where the credits actually go

Token counts measured across our own first build of this pipeline.

Agent, unassisted

1.4M tokens
Inventory channel, pick sources94K
Work out where timings come from71K
Score and rank candidates168K
Choose and verify every cut point243K
Solve 3 source layouts152K
Build a caption renderer206K
Chase drift, re-render twice331K
Write the posting plan96K
1,361Ktokens

Agent, through us

66K tokens
Describe the job, one POST2K
Read back ranked candidates31K
Approve or swap a few picks9K
Collect clips and posting plan24K
0
0
0
0
66Ktokens
21×

The expensive part was never the rendering. It was the figuring out — which moments are worth cutting, where a sentence actually starts, why the captions drift a quarter second by the end. We already paid for that once.

Measured on a real 2,247-video channel. Your number moves with how much of the candidate list you read back before approving.

The output

Not one clip. A formation.

Twenty hooks, two cuts each, five voices, spaced across weeks — so a batch never reads as one operator working a schedule.

What lands in your bucket

Forty files and a plan

Vertical clip: hard truth
Hard truth0:28  ·  style A
Vertical clip: receipts
Receipts0:23  ·  style B
Vertical clip: called it
Called it0:37  ·  style C
Vertical clip: origin story
Origin0:31  ·  style D
Vertical clip: contrarian
Contrarian0:25  ·  style E
Vertical clip: rules
Rules0:35  ·  style A
Seven stages, in order

What runs when you call it

STAGE 1

Inventory

Pull the whole back catalogue and rank it by what actually performed.

STAGE 2

Timings

Word-level timings straight from the captions. No transcription pass, no bill for one.

STAGE 3

Mine

Every window scored against seven archetypes — receipts, hard truth, origin, called it, contrarian, rules, spicy.

STAGE 4

Cut

In and out points snapped to real pauses, so nothing opens or lands mid-word.

STAGE 5

Reframe

Face crop, recovered portrait, or blurred pillar — picked per clip from what the source actually is.

STAGE 6

Caption

Word-by-word burned captions in five distinct styles.

STAGE 7

Route

Each clip gets an account, a voice, copy, hashtags and a slot.

Integration

The whole thing

# one call, then read the manifest
POST api.repostarmy.com/v1/runs

{
  "channel":  "youtube.com/@yourchannel",
  "hooks":    20,
  "variants": 2,
  "accounts": 5,
  "format":   ["shorts", "x"]
}
What comes back
40finished 1080×1920 MP4s, loudness normalised
1posting plan — account, copy, hashtags, slot
20ranked alternates you didn't take
source timestamps, so every cut is checkable

Same surface over MCP, so an agent drives it with no wrapper.

One real channel, one run

From the build

2,247videos inventoried
192segments scored
20hooks kept
40clips delivered
17.9minutes of footage
The one rule

Verified owners only

You connect the channel you're clipping and we check that you own it. That's the whole gate, and it isn't negotiable — a tool that lets anyone cut up anyone is a copyright problem wearing a product's clothes.

It's also why this works. Clips from a catalogue you own, posted by accounts you run, are marketing. Everything else is someone else's lawsuit.

Point it at your catalogue.

$39 / month  ·  100 clips  ·  cancel whenever  ·  first run free

Start a run