Patterns

Norn is a building block. These are the shapes we expect it to fit first. Each one uses only the calls in these docs.

For a content tool, a social media manager or a CMS: show each customer a short list of ideas for their brand, ready to use or skip.

1. Create a brand for each customer when they sign up, from the website they gave you.
2. Ask for three to five ideas when they open the planner, or once each morning.
3. Show title, hook and why_now, and keep sources one click away so people can check where an idea came from.
4. Send every decision back: used, changed or skipped. Ask for a reason now and then, not on every skip. Native does the same, so reasons keep coming without nagging.

For a website builder, a booking system or a CRM: the moment a business signs up, you know its website. Turn that into its first week of posts.

const brand = await norn('POST', '/brands', {
  website: signup.website,
  country: signup.country,
})
await waitUntilReady(brand.id)

const batch = await norn('POST', '/ideas', { brand: brand.id, count: 5 })
const done = await waitUntilDone(batch.id)
for (const idea of done.data) {
  await drafts.create({ customerId: signup.id, title: idea.title, prompt: idea.prompt })
}

norn is a small helper around fetch that adds your access token, as in the Quickstart. waitUntilReady asks for the brand every ten seconds until its status is ready, and waitUntilDone asks for the batch every 15 seconds until it is done. Both throw when the status is failed, so a failure ends the job instead of waiting forever. Run this in a background job: a brand and a batch together take minutes, not milliseconds.

For agencies and teams that approve everything before it goes out: let Norn propose and let people decide. Ask for more ideas than you need, show the top few to whoever approves, and send every decision back. Approvals tell Norn what good looks like for this client. Reasons tell it why.

If you build an agent that writes and posts on its own, have it call Norn first and write from the prompt. The agent keeps doing what language models are good at, the writing, and Norn decides what is worth writing about.

When predictions arrive, the loop can also run the other way: your team or your models write, and Norn ranks the drafts before anything is published.