Onboarding — from a URL to a live radar.
Six screens, five extraction phases, about a minute. The one trap: the Re-extract button on the Confirm screen wipes your setup — it does not refresh it.
Onboarding takes one input — your website URL — and turns it into a working brand: colours, voice, competitors, and a brief queued for tomorrow morning. It takes about 30 to 45 seconds of reading, and everything it produces is a starting point you're expected to sharpen.
The six screens
Welcome
One screen, one button. Nothing is created yet.
Your brand name and website
Two fields. Type the URL and the brand name auto-fills from the hostname — edit it and we stop guessing. Hitting Start the scan creates the brand row, then hands you to the read.
The read
Five phases run while you watch. Roughly 30–45 seconds in the normal case. Each phase saves what it found as it lands, so nothing is lost if you leave.
Confirm what we read
One card mirroring back your industry, your palette, a three-adjective voice line and your competitors — each with a confidence chip. This is the cheapest place in the product to fix a wrong guess.
Your first brief
Pick what you want (a daily competitor read, or just the social posts) and where to send it — Telegram, Slack or Discord. You finish the connection itself on the Channels page afterwards.
You're live
We generate one real LinkedIn post from your new brand so you leave with something in your hand, not a confetti animation.
If no delivery channel ever gets connected, your brief doesn't generate at all — it isn't just undelivered. You'd have to press Generate by hand every day. Why →
The five extraction phases
The progress screen is not a loading animation with fake steps. Each line is a real job:
| Phase | What actually happens |
|---|---|
| 1 · Reading your homepage | We fetch your site and pull your About text, your industry, your social links, your target audience and — importantly — a first pass at your products & services. The industry we detect is classified against our taxonomy and saved as a proper industry, not free text. |
| 2 · Colours and logo | Fonts and favicon are read out of your HTML; the palette is counted off the pixels of your logo, with an AI pass only when the deterministic read is weak. If it can't lock a palette you'll be told so, and you pick the colours yourself. How this works → |
| 3 · Listening to your voice | We try your homepage plus the usual pages — /about, /about-us, /services, /work, /blog, /manifesto — and keep up to 6 as writing samples, skipping any page with under 400 characters of readable text. Then we analyse them into a voice fingerprint, so it's ready before you ever open the Voice tab. |
| 4 · Finding who you're up against | One AI call, using your hostname, industry and About as the fuel. It returns four to six real rivals, we probe each URL to check it resolves, and the top six land as suggestions. Not a search-results dump. |
| 5 · Tuning your radar | Sets up your brief settings and fires the first scrape against the rivals we just found, so your dashboard has something on it when you arrive rather than an empty state. |
Phases run with a generous ceiling — the countdown on screen is an estimate, not a deadline, and a slow site can push the whole read past a minute. If a phase comes back amber, it failed softly: the rest still ran, and you fix that one thing by hand later.
Reading the confidence chips
Every row on the Confirm screen carries a chip. There are two states, and the honest translation is:
- High confidence (green) — the deterministic signals agreed. Glance at it and move on.
- Best guess (orange) — we inferred it. Read it properly.
The competitor row is always labelled a best guess, no matter how many rivals we found. That's not false modesty — an AI's idea of your competitors and the businesses you actually lose deals to are different lists, and only you know which.
Re-extract wipes your progress
The Re-extract link on the Confirm screen is not a refresh — it is a reset. It throws away everything the read produced: your colours, your fonts, your favicon, the confidence scores, and your position in the brand setup chain. Then it runs the whole read again from scratch. Anything you'd already edited by hand is gone with it.
This is worth being precise about, because there is a second button called Re-extract that behaves completely differently:
| Onboarding · Confirm screen | Brand → Identity tab | |
|---|---|---|
| What it does | Wipes, then re-runs | Re-runs, then merges |
| Your edits | Lost | Kept — a field is only overwritten if the new run actually found something for it |
| Safe to press twice? | No | Yes |
So: if the Confirm screen shows one wrong colour, don't hit Re-extract. Click Edit details, fix the colour on the Identity tab, and carry on. Re-extract is for the case where the read landed on entirely the wrong site — a parked domain, a holding page, a redirect you didn't expect.
Skip and edit manually
There's a Skip and edit manually link on the progress screen. It doesn't cancel anything — the phases that already finished have already saved. It just stops you watching and drops you at the Confirm screen with whatever landed.
Use it when the read is clearly going nowhere: a site behind a login, a single-page app that serves nothing to a plain fetch, a brand-new domain with two sentences on it. You'll fill in the Brand tabs by hand, and everything works exactly the same from there.
What to fix first, afterwards
Onboarding gets you 70% of the way. The last 30% is where the quality lives:
- Your services. Homepages are marketing copy, and marketing copy is vague — so the services we pulled off yours probably are too. Rewrite them in the words a customer would use. Products & Services →
- Your competitors' links. A suggested rival arrives with whatever we could scrape off their homepage. Open each one and check they've got a LinkedIn and an Instagram at minimum, or they'll sit in your brief showing no activity forever.
- Your voice samples. Six pages of website copy is a start. Three things you actually wrote is better.