Quin · Intelligence
Xicoia · Private
This page is for invited clients. Enter the access code you were given.
Incorrect code — please try again.
Onboarding guide
This is how they get made. Not an app to log into — a guided conversation that turns what you know and how you think into clean, portable documents, which then drive her in text, in voice, and eventually on screen.
Worth being precise, because the word "platform" gets used loosely.
The Quintelligence Builder is a skill — a set of instructions you install into Claude. Once it's there, Claude knows how to interview you about Quin, one component at a time, and write up your answers as structured markdown files.
That's the whole mechanism. There's no server, no login, no subscription beyond Claude itself. The output is plain text files that you own and can read, edit and move anywhere.
We're deliberately starting here rather than building software. Existing platforms already do the hard parts — retrieval, voice, video, turn-taking. What they don't have is Quin: her positions, her voice, her limits. That's the part worth building carefully, and it's a writing job before it's an engineering one.
It doesn't host or run Quin. It produces the inputs; a platform does the talking.
It doesn't invent anything on your behalf. Where an answer is missing it writes [TO CONFIRM] and moves on. A visible gap is useful. A plausible guess is dangerous.
The distinction that saves the most time later is between who she is and what she thinks. Open each layer to see what sits in it.
What Quin knows and can reason over. Not documents dumped in a folder — your positions. For each topic: what you think, the argument behind it, the strongest objection and your answer to it, and the two or three sentences she'd actually say out loud.
Who she is, and how she comes across — pace, warmth, dryness, how she disagrees, what she'd never say. Held as seven adjustable settings plus the phrases captured in your own words. Her face and voice are already locked, so this layer references them rather than reopening them.
The same brain and personality every time, framed differently for each outing. Which slice of her knowledge is live, what she's there to do, what she politely declines here, and how much personality shows.
This is why a new context never means a new Quin. A board audience isn't a different character — it's a framing file with two settings shifted. Parallel versions drift; one core with per-context overrides doesn't.
Where she actually appears. Each surface has different constraints — a scripted film can be as considered as you like, because nothing is real-time. A live avatar trades depth against delay.
Read this as a map of what becomes possible over one well-built core, not a list of commitments.
If it describes how something is said — pace, humour, warmth — it's Personality. If it describes what is believed or known — a position, an argument, a fact — it's Brain.
A leadership philosophy is Brain. Delivering it drily is Personality.
A few minutes, once. After this it's simply there whenever you want to work on Quin.
One file. Open it in Claude and you'll see a card with a Save skill button — that installs it to your account. Quin's Phase 1 work is already inside it, so there's nothing else to set up.
Download the skillIn the sidebar, create a new Project. This is where the files you build up live, so every conversation picks up where the last one left off — and anyone you add to the Project works from the same version.
White papers, notes, bios, old decks — anything about Quin or about what you think. Drop them into the Project, or straight into a chat. Claude will tell you what's in them before it uses any of it, and you decide what counts.
Anything works — "let's work on Quin", "let's do her personality". It already knows what Phase 1 settled, so it opens by putting those decisions to you to confirm, then moves on to what's genuinely still open.
Easiest on desktop, where the finished files save straight out and Projects are simplest to manage. Mobile is fine for thinking aloud between sessions.
At the end of each session the updated files go into the Project, so it's always the current version that's picked up next time. Nothing from Phase 1 needs importing — it's already built in.
One question at a time. Your words on the left, the file taking shape on the right. Note the first move — what’s already settled gets carried over, not asked again. Step through it.
The conversation
personality.md — forming
Distinctive phrasing gets kept exactly as you said it. Tidying your words into something neater is precisely how a character stops sounding like anyone in particular.
So when Claude asks what you'd actually say out loud, that answer matters more than the description around it.
A folder of small, readable files. Select one to see what it holds and what you'll be asked.
Personality
Brain
Framing
Always on
Index
Two files covering the same ground in slightly different words is the single most reliable way to make her contradict herself. If new material overlaps an existing file, it extends that file rather than starting a rival.
One corpus, three surfaces. Each adds a single variable, so when something feels wrong you know what caused it.
Tests the brain · available first
The corpus compiles into a second skill that is Quin. Anyone on your team installs it and starts talking — no build, no accounts, no waiting. It's the cheapest possible way to find out whether her positions hold up and whether she sounds like herself.
Every copy is version-stamped, and feedback comes back to the files rather than being patched into the copy.
Tests the brain and the pacing · no build required
The same files configure a voice agent using Quin's locked voice. Her personality goes in as the standing instruction; her positions go in as knowledge. You get a link — talk to her from a phone, no account, no setup, and share it with whoever you want on your team.
This is also how we settle the intelligence-versus-speed question honestly. We build the same agent three times, changing only the underlying model, and you listen. A quicker model answers faster; a deeper one pauses longer. Hearing that difference beats any explanation of it.
Tests the whole experience · last, deliberately
The same corpus uploads to the avatar platform, which already accepts her voice and can be pointed at the reasoning engine of our choice. Nothing gets rebuilt — the brain moves across intact.
We've put this last on purpose. A genuinely live, unscripted fireside conversation with deep reasoning behind every answer is still the hard problem in this field: more thinking means a longer pause, and a long pause reads to an audience as a failure. Rather than talk around that, the voice step lets you hear exactly where the ceiling sits today and decide what's good enough.
We run the sessions roughly every other week, each one closing out a component or two. What helps most, per stage:
Before we start
Her age, how she describes what she is, and the exact words she uses if someone asks whether she's human. Short session — but everything else hangs on it.
Personality
How she disagrees. What she does when she doesn't know something. The one boundary you'd stop everything over. Contrasts help — who does she not sound like?
Positions
Bring the source material, but the valuable part is your stance on it — and the two or three sentences you'd say out loud. Two or three topics per session, no more.
Framing
A specific event beats a general case. Who's in the room, what they already know, and the ten or twenty questions you'd expect.
It happens here, in these sessions — slowly, with you in the room, and reviewed before anything is locked.
Which is the point: at conversation time Quin isn't working an answer out from first principles. She's delivering one you already thought through. That's how she can be both quick and considered.