In August 2026 we took one agency client's real folder, the same folder the team was uploading to AI chats, and counted. The folder held about 9,500 tokens of readable text, but one chat cost about 50,000 tokens to load it, most of that spent processing PDF pages the AI cannot read as text: a scanned agreement, a designed branding guide whose columns scramble on extraction, a metrics spreadsheet that turned out to be empty. Rebuilt and expanded into the Stak, that same client knowledge grew six times deeper, and a typical task reads about 7,000 tokens of it. Fewer tokens in, better context, every chat, every seat.
Stop re-explaining your clients to AI.
Your best account person knows every client cold.
Now your whole team’s AI does too.
Interactive walkthrough · No signup
Write the June email for Lakeside Marina. The promo is the new pontoon fleet.
Stak search›
Found their voice rules and approved proof points. One governed number: 212 five-star reviews.
Stak brand guide›
›
Here's the email, in their voice, with their numbers. I also found a stale claim in the old promo doc and flagged it for review.
june-email.md
Stak update request›approved by a human · merged
Reply…
This is the Stak working. Simulated conversation, real system.
The AI isn't the problem.
The re-brief is.
Your team already uses AI. But without the client's facts, every task starts from zero, in every seat, on every tool.
The re-brief
Every deliverable starts with re-explaining the client to the AI. Again. Every person, every tool, every time.
01
02
The generic draft
Without the client's facts, the AI guesses. The draft comes back sounding like nobody in particular.
The babysitting
A senior person rewrites the output line by line. That's your margin leaking into the rewrite step.
03
04
The slow ramp
A new hire takes months to learn each client. The knowledge lives in heads and old decks.
The re-brief tax, measured.
We counted it on a real client build: the folder the team was actually uploading to their AI, against the Stak we rebuilt and expanded from it.
~86% less tokens
spent per chat. Same client, same task.
Uploading the client folder
The same task from the Stak
~6x more knowledge
available to every seat, rebuilt and expanded into the Stak.
Readable client knowledge in the folder
Client knowledge in the Stak built from it
Measured on a real client build · August 2026
What's the story here?
We build each client a Stak.
Every seat reads from it.
Hand us the messy client folder. We refine it into one clean, governed source of truth.
Brand & voice
Audience & offers
Stats + sources
The Stak
Company overview
Service catalog
Messaging
Brand & voice
Audience & offers
Stats + sources
The Stak
Company overview
Service catalog
Messaging
Private
Read-only
Human-approved
Audited
TLS 1.3 encrypted
SOC 2 Type II hosting
The Process
How We Refine Your Stak
A Stak is only as good as the knowledge inside it. Four steps take it from the files you already have to a voice interview that fills in what only you know. No blank-page homework.
You hand off
01
Start With Your Files
Brand guides, decks, site copy, service docs: what you already have seeds the Stak. Nothing starts from scratch.
We research
02
Storystak Deep Dive
We research the brand end to end and structure it into the Stak's canonical files. Anything unverified is logged as a gap, never guessed.
We ask
03
Question Guide
The gaps become a short, prioritized question list: only what research can't answer, ordered by impact.
You (or your client) talk
04
Storystak Bot
Answer by talking. Dot, the voice interviewer, works through the question guide; answers come back as Stak updates you approve.
The Storystak Bot
The Last Step Is a Conversation
Not a form. Dot interviews you by voice, and the Stak gets sharper every time you talk to her.
Your Stak Asks.
You Just Talk.
Open your private link, enter your PIN, and talk for a few minutes. Dot works through your question guide, and every answer lands in your review queue as a pending Stak update.
- Don't have the answer? Your client does. Send them the same link; their answers land in the same review queue, and you still approve every change.
- Asked once, never again. A filed answer retires its question for good. Every session picks up where you left off.
- You approve every change. Dot never writes to a Stak. A person signs off on every update.
How It Works
Three moves. Then it just runs.
Once, at kickoff
01
We set up your agency
Your voice, your standards, your folder structure, mirrored to how your team already works. Training included.
Per client
02
We build each client a Stak
Hand us the messy folder. We refine it into one clean reference kit, every stat tied to a source.
Day one
03
Your team plugs in
Each person points their own AI at it. No new software, no prompts to memorize. The AI stops guessing.
Why I Built This
AI without human expertise is just a super calculator.
I spent 16 years building marketing systems before AI existed. When it arrived, my first thought was that my career had a shelf-life. Then I remembered Photoshop, when everyone said it was the end of real design. The people who adapted didn't lose their careers. They unlocked a new level of them.
I jumped in head first and found the same thing. AI without human expertise is just a fast calculator. But apply 16 years of knowing how great work actually gets made. And the ceiling disappears.
That's what we encode into your systems.
As the Principal Architect, my job isn't to sell you a generic AI license. It's to look at your existing data stack and design the secure, human-governed roads that actually make your AI investments functional.
Founder & Principal Architect
FAQ
The short version, answered.
Questions
Answer
What is Storystak?
Storystak builds each of a marketing agency's clients a governed Client Stak: one source of truth for brand, voice, audience, offers, and sourced stats that every team member's AI tools read from. It ships white-label under your agency's own name.
What is a Client Stak?
A Client Stak is a single governed reference kit for one client: brand guide, voice rules, audience notes, approved offers, and statistics with every number tied to a source. Whatever AI your team already uses points at it, so drafts come back on-brand and factually sourced.
What is the Storystak Bot?
A voice interviewer named Dot, who fills the gaps research can't. You open a private link, enter a PIN, and talk through a short question guide; every answer comes back as a pending Stak update a person approves. Dot never writes to a Stak directly, and an answered question is never asked again. Try the live demo.
Which AI tools does it work with?
The AI your team already uses, over MCP, an open standard. Claude is the one we test end to end; ChatGPT connects too, read only. There is no new software to install and no prompts to memorize; each person points their own assistant at the Stak.
Do I need a team ChatGPT or Claude account?
No, but it can help. Your team can point individual ChatGPT or Claude accounts at the Stak just as easily as a team plan. Access is your call, person by person: full-time employees, freelancers, or a contractor brought in for one project. When the project ends, so does the access.
Does everyone have to switch to Claude?
No. The Stak connects to ChatGPT too. What you should know: ChatGPT limits connectors to read only, so nobody can file a correction from inside the chat; it needs a paid plan; and setup runs through developer settings instead of three clicks. Claude is the team-shaped platform and the better prompt writer, which makes every other tool better: Claude plus the Stak writes the prompt, and you take it to ChatGPT, Nano Banana, Higgsfield or Gemini for the image or the video. That is why we recommend Claude, and the reason is capability, not preference. Keep every tool you like.
How is this different from a prompt library?
A prompt library still relies on each person re-explaining the client every time. The Stak is a governed source the AI reads directly, with every stat tied to a source and stale claims flagged, so output stays consistent across people, tools, and time.
Why not just upload my own context files?
It works, just not well. Files get re-uploaded chat by chat, nothing ties a number to a source, and old decks linger long after the facts change. Each person also grabs different files, so every seat gets a different picture of the same client. The Stak replaces that pile with one governed source every tool reads the same way.
Can I do this myself?
In part, yes. A solo marketer with clean files and discipline can get real value uploading their own context. The gap shows up with a team: everyone keeps their own version, the versions drift, and nobody notices the stat that went stale. Storystak adds governance at team scale: one source every seat reads, every number tied to a source, stale claims flagged, and every change approved by a person. That keeps the tenth seat's drafts as consistent as the first's.
How long does setup take?
Your agency is set up once at kickoff: voice, standards, and folder structure, with training included. From there each client Stak is built per client from your existing folder, and your team plugs in on day one.
How do we get started, and what does it cost?
Start with a free 30-minute call. Bring your most complicated client and we'll map exactly what their Stak would carry, whether you work with us or not. Book Now.
Bring one client.
We'll map their Stak.
Free 30-minute call. Pick your most complicated client and we'll show you exactly what their Stak would carry, whether you work with us or not.
Not ready to talk?