Quickstart
Create an account, type a topic, write a short brief, press start. A run takes a few minutes depending on how much conversation there is to read.
- Sign up and confirm your email.
- Write your brief: voice, audience, and anything you never want written.
- Type one topic in plain language. Not a keyword: a topic.
- Press start and watch. Nothing else is required from you.
- Export the finished piece as HTML or Markdown.
Brief & voice
Every run takes a brief: who the piece is for, the voice you want, and the product context the writer needs. You give it per run, so you can point two runs at the same topic and get two different pieces out.
Keep the brief somewhere you can paste it from. Reusing the same one is what makes the second piece sound like the first.
What to put in it
- Audience. Who is reading, and what they already know. This does more work than anything else in the brief.
- Voice. Two or three real sentences you would publish, not adjectives. "We write like a senior engineer explaining to a peer" beats "professional yet approachable".
- Never. Claims, comparisons and phrasings that are off limits. Include them in the brief and check the draft against them before you publish.
Starting a run
A run takes exactly one topic. Give it the thing you would say out loud to a colleague, "why teams abandon their design system", rather than a keyword string. The run frequently comes back having sharpened the topic into the question underneath it; that is working as intended, and it tells you when it moves.
Optional inputs: target length, an angle to prefer, and keywords you already care about. All three are hints, not instructions. If the evidence contradicts a hint, the run follows the evidence and notes the conflict.
Watching the work
The run view streams each of the six moves as it happens: threads being read, clusters forming, the convergence call, the keyword ranking, the outline, then the draft written in front of you. Runs execute server side, so the work continues whether or not you are watching.
Each move can be expanded, which is how you check rather than trust: each thread is listed with the excerpt it was read for, the cluster shows what fell into it, and the convergence figure shows what it was measured over.
The evidence
Everything the run read stays attached to it: every source with a link back to the original, the excerpts it was read for, the cluster, the score, the ranked keywords and the reasoning behind each decision.
Steering the output
The cheapest place to intervene is the outline, which lands before any prose is written. Cut a section, reorder them, or reject the angle and take a different headline, the draft is then written against your version.
Exports
Every finished run produces the same set of artifacts:
| Artifact | What it is for |
|---|---|
| HTML | Paste straight into a CMS. Clean semantic markup, no wrapper divs, no inline styles. |
| Markdown | For static sites and docs repos. Headings, lists and links, no front matter. |
| SEO meta | Title, description and slug, generated from the finished piece. |
| JSON-LD | Article structured data, ready to drop in a script tag. |
Credits & billing
Runs are metered in credits. Plans grant a monthly balance: Starter $12 for 87,200 credits (about nine finished pieces), Growth $29 for 210,000 (about twenty-three), Scale $99 for 712,000 (about seventy-nine). Top-ups are available at 7,200 credits per dollar with a $10 minimum and require an active subscription.
Credits are held when a run starts and only spent when it finishes with something publishable. Failed and stalled runs release the hold in full, and stopping a run yourself charges only the work already done. Unused credits roll over at renewal and do not expire, though you need an active subscription to start a run. Upgrading grants the full credit difference between plans immediately rather than a prorated fraction.
Billing is handled by Paddle as merchant of record, in USD, with local pricing where available.
Common questions
How long does a run take?
A few minutes. Topics with very large communities behind them take longer to read.
What if there is not much demand?
The convergence figure comes back low and the threads behind it show why. You still get the piece and the reading; what the number buys you is the evidence to decide whether it is worth publishing, which is cheaper to learn here than three weeks afterwards.
Does it publish for me?
No. It hands you artifacts and you publish. An unreviewed draft going live is nobody's idea of a good day.
Who owns the output?
You do. See the terms.