///104 Ads OS

The ads team

Turns your monthly briefing into ad concepts, copy, images and video, checked before you see them.

In buildBeing built now. Join the list to hear when it is ready.

How the team works

Collect your ad data and see what works
Read the monthly briefing and write hooks in every language
Build a varied set of concepts
You approve the concepts
Make images, statics and video, with an automatic quality check
You approve the assets
Go live after your go, then feed the results back

Who is on the team

ads-leadLead
Runs the ad work from briefing to approved ads and hands each step to the right agent.
ads-plan-orchestrator, ads-briefing-parser
Read your monthly briefing and build a set of concepts, then stop for your review.
ads-performance-analyst
Reads your ad data and says what works now.
ads-concept-strategist, ads-diversity-critic
Turn copy into concepts and remove duplicates so the set stays varied.
ads-generate-orchestrator, ads-render-router
Make the assets for each approved concept and pick the right model.
ads-image-prompt-writer, ads-script-writer, ads-video-prompt-writer
Write the image prompts, video scripts and video prompts.
ads-qc-agent
Checks each new asset on product, brand rules and copy. Pass or fail.
ads-recipe-adapter
Adapts a winning layout to a different number of products.
ads-deliver-orchestrator
Puts approved ads live on Meta, only after your explicit go.
ads-creative-feature-extractor, ads-feedback-matcher, ads-learnings-analyst
Link live results back to each concept and turn them into lessons.

Every task it can do 28

ad-copyWrites hooks and on-image copy for all your market languages in one go.
hook-craftChecks that ad hooks are strong, varied and add new information.
transcreationKeeps copy native in each language instead of translated word for word.
concept-matrixBuilds a set of ad concepts, each with copy, a brief and a hypothesis.
ads-performance-analysisTurns performance data into insights for the next concepts.
ad-teardownBreaks down one ad into an analysis and a reusable prompt template.
template-teardownTurns a reference ad into a layout template and checks it against the original.
image-prompt-builderWrites ready-to-use image prompts per image model from an approved concept.
image-bankGenerates image variations for a layout and stores them for later runs.
static-imageGenerates the base image for a static ad from your product photos.
static-canvas-overlayAdds headline, proof, button and logo on top of the base image.
static-html-overlayAdds text and branding as HTML, for product shots and many languages.
static-canva-overlayAdds text and branding through Canva instead of HTML.
static-ad-htmlBuilds static ads as HTML templates with your brand built in.
brand-overlayPlaces logo, brand colours, headline and call to action on a generated image.
figma-renderTurns a finished static ad into an editable Figma frame.
ads-video-prompt-builderWrites shot-by-shot video prompts from a creative brief.
ads-ugc-genWrites the prompt for a clip of a person speaking a line to camera.
video-genGenerates the video from the prompt with the chosen video model.
vision-qcChecks a generated image or video on product, brand rules and copy.
meta-publishPuts an approved asset live on the right Meta ad account.
ads-dataCollects Meta Ads and Google Ads numbers into one daily table.
creative-winnaarsSorts your ads into winners, middle and laggards on the goal that counts.
creative-vermoeidheidFinds tired ads whose click rate drops while people see them more often.
budget-pacingChecks if this month's ad budget is on track.
utm-checkChecks that every running ad has clean UTM tags.
ads-weekrapportWrites the weekly ads report: winners, tired ads, pacing, actions.
ads-lessenTurns lessons that keep coming back into proposals for the brain.

Also included: intake, brain-voorstel, humanizer, anti-ai-slop, visual, animatie, commercial-animatie, carousel, still-animatie.

What it does for you

The ads team starts from data. It reads how your ads perform and what works now. Then it takes your monthly briefing and writes hooks and on-image copy in every language you sell in, written natively, not translated. From that it builds a varied set of concepts, each with copy, a brief and a hypothesis you can test.

You approve, change or kill each concept. For each approved concept the team makes images, static ads and video, and every asset gets an automatic check on product, brand rules and copy. You approve the assets. Only after your go does an ad go live on Meta.

After launch, the results are linked back to the concept they came from. The weekly report shows winners, tired ads, budget pacing and missing UTM tags, so the next month starts from what you learned.

What you need

Claude Code with a paid Claude plan.
The company brain (BX-001), recommended. Without it, /intake fills a brand folder from your website.
Python 3.11 or newer. Node 18 or newer and Chrome or Chromium for static ads.
A Meta Ads account. Google Ads, optional.
Supabase for the ad data, and Notion for the briefing and the two review boards.
Keys for an image and video model service, such as fal.ai. Figma and Canva, optional.

The team runs end to end on test data out of the box. Several live connections are still being built.

What it never does

It never publishes on its own. Publishing is locked in the settings and blocked by a hook, and an ad only goes live after you approve the concept, approve the asset and give the go.