Creator OS
Open beta, free while it runs 86 operations, one registry

Everything you post, on purpose.

One machine for the whole operation: your brand, your design system, your strategy, your corpus, the writing, the editing and the calendar. Your own AI drives all of it. Bought once, and yours for good.

The brand interview · 29 questions · 76 minutes
0 answered
Position

Loading the interview…

Brand Core

That is the whole core interview. In the product, these answers become the Brand Core every other layer reads. Anything left empty stays empty.

· your answers were kept in this tab and nowhere else.

This is the product's own interview, running in your browser. The product asks; the assistant you pick helps you answer. Your answers stay in this tab.

One machine, the whole arc

From a blank page to a post you can defend.

Eight stops, eight layers, 86 operations, one registry. What you decide at the top is read by every stop below it, so the voice, the look and the standards are the same all the way down to the export. Nothing here is a separate app you have to keep in sync.

BrandL1 · 20 ops

Who this is for, in their words.

A long interview, four micro engines, one Brand Core. It asks you to name your audience the way they would describe themselves, and to give their problems in their words rather than yours. Everything below reads this.

DesignL1 · 10 ops

One look, decided once.

Design system, formats, patterns, templates and canvas. Everything made further down reads from it, so nothing you publish is styled from scratch or from somebody else's taste.

CurationL2 · 7 ops

A corpus that is actually yours.

Scoped ingest, preflight, your own Council of voices, and Watchtower keeping it current without being asked. What comes in is chosen, not whatever a feed served you that week.

StrategyL3 · 10 ops

What this place rewards, and what worked here.

Platform rules, outlier analysis against your own normal, and what actually performed on your account. Enough to break a tie between candidates, never enough to overrule someone who knows the audience.

CreationL4 · 4 ops

Two doors, one generator.

Repurpose something you already have, or ideate on an ask. Both read the Brand Core before a word exists, and both hand the writing to your own agent.

ConsistencyL5 · 7 ops

Checked against you, not against average.

Voice scoring, the virality axes, and a register of named defects applied before generation and checked after. A draft that does not sound like you fails here, not after you post it.

EditingL6 · 10 ops

Footage in, file out.

Attach, transcribe, cut, caption, render. The editor is inside the product under the same licence, so a video never leaves halfway through for a tool you rent by the month.

PostingL7 · 9 ops

The week, and the record.

The calendar, the export and the publish record. It does not post for you. It records where you posted, which is the only reason measurement can ever reach the plan.

Counts are read off the live registry by site/tools/arc-counts.py. Nine more operations are the machinery every stop uses: resolve, cost, the assistant bridge and the licence.

AI native, end to end

Hand the whole thing to your own agent.

Creator OS is an MCP server. Point Claude Code, Codex, Cursor or anything else that speaks MCP at it, and all 86 operations are in reach from the terminal you already work in. Not a helper in a side panel. The whole product, by conversation, on the subscription you already pay for.

Plan next week from what actually worked.schedule.week · plan.derive · intel.what_is_working
Cut this footage and caption it.edit.attach_footage · edit.cut · edit.export_captions
Score this draft against my identity.voice.score · virality.score · constraint.review
Show me the outliers on my account.intel.outliers · analytics.by

Every line above is a registered operation, gated and traced exactly as it is in the panel. An operation that is not in the registry is not reachable from either surface.

Driven directly
Claude Code
sonnetopushaiku

The default. Sonnet is the right size for an assistant on rails.

Driven directly
Cursor Agent

Spawned by the product, and able to connect back through the same MCP config.

Driven directly
Codex

Same registry, same gates, same refusals.

Connects the same way
Gemini CLI, and anything that speaks MCP

Point it at run.py --mcp and it sees every operation the panel does. The three above are also spawned by the product itself; everything else connects inward.

No key of ours is stored, nothing is metered by us, and no token is resold. You pay your own AI bill, directly to whoever you already pay it to.

What it is

Three things that define it.

Everything above is a consequence of these. They are the reason it can be sold as a licence at all, and the reason it behaves the same on your machine in five years.

01

It never calls a model.

It hands your agent the Brand Core, the corpus, what the platform rewards and the named constraints, then stores what comes back. Your own agent writes. No key of ours, no metered spend, no tokens sold on by us.

02

It depends on nothing anyone else runs.

No shared service, no external brain, no server. A brand is one folder you own and can move to another machine. If we disappeared tomorrow, what you have keeps working.

03

You learn while you use it.

A score arrives with the axes it came from. A refusal says what was missing instead of failing quietly. The theories and the templates are shown where they apply rather than hidden behind the button, so a year in you are better at this, not just faster at a tool.

The panel

It reads this brand, so nothing here is written in general.

Real screens from the running product, on the demo brand that ships with it. Your theme, your screens.

The Identity screen: the Brand Core's sections, each filled in or left empty, with a readiness count.

Everything that keeps this brand consistent. Fill it in yourself, or ask your assistant to build it with you. Anything left empty stays empty.from the Identity screen

Measured 2026-08-20, ten generations per arm

A constraint prevents only what it names.

Why the constraint layer is built the way it is. Same model, same brief, three ways of asking it not to invent a number.

How generation was constrainedFabricated a figure
Free generation, filtered afterwards3 / 10
A full ban list that did not name the defect3 / 10
The same list, naming it0 / 10
Fabrication cannot be checked afterwards: a plausible number is indistinguishable from a real one once written. Prevention is not the better position there, it is the only one that exists.
What it will not do

Three refusals, on purpose.

It will not invent a cadence.

No platform carries a posting-frequency figure, because a cadence is a claim about a specific account, and the honest value before measurement is absent rather than a range that looks like knowledge.

It will not generate against an empty Brand Core.

It names the missing sections instead. A post generated from nothing could belong to anyone.

It will not publish.

It records where you published, which is a different thing, and the reason measurement can reach the plan at all.

A 70×7 tenet

Own your software, don't rent it.

A subscription rents access and takes it back. This one you keep. When a licence lapses you hold every build it ever covered, forever, and only stop receiving new ones. Nothing switches off. Nothing phones home. No clock is ever consulted.

Checked against the build, never the dayEntitlement compares a licence to the build's release date. Today's date is not an input, so nothing can expire.
Verified on your machineA key is a signed statement checked offline. No network call, no account, no check-in.
No subscription, no tokensNothing recurring and nothing metered by us. The AI bill is yours and you pay it to whoever you already pay.
One folder, yoursA brand is a folder you can move to another machine, and everything in it is yours outright.

Plainly, because it should be said plainly: the software itself stays ours to build and to license. Your right to run it does not lapse, is not revoked, and is not taken back, and everything you make with it is yours.

Open beta

Free while the beta is open.

Joining costs nothing and nothing is held back: there is no free tier and no paid tier during the beta, only the product. Put your address on the list and we will send you the build.

Free

Join the open beta

Tell us what you make, if you like. It helps decide what gets built next.

Your address goes on one list on our own server. No analytics, no third party, and nothing is sent to you until there is a build to send.

Optional, and not required to use the beta

Back it early and the price is locked.

These stay live so the payment path is exercised before it matters. Buying now permanently covers every build from the day you buy, and costs you nothing you would not pay later. Nothing in the beta is gated behind either one.

$100a year
One year of updates

Every build released in the next twelve months, yours permanently. Renews yearly. Renewing early extends from your date rather than from the day you pay, so it never costs you anything to renew ahead.

Buy a year
Never renew
$300once
Lifetime

Every build ever released. No renewal, no expiry, no check-in. One payment, and every future version is covered for good.

Own it outright

Your key arrives by email and on the page after checkout. Paste it under Settings, Plan & licence. It is verified on your machine.