brainkit
Every employee gets a personal knowledge vault their tools keep current — from the command line to any AI assistant they connect. The company gets a shared memory that grows from everyone's work — without anyone's private notes ever leaking to anyone else.
The one idea to hold onto
There is a single master vault for the company. A piece of software called the compiler produces a filtered personal copy for each person. Anything you're not allowed to see is never written into your copy at all — so no app, no AI, and no mistake downstream can ever show it to you.
Why it exists
The shared knowledge base nobody updates. Writing things down is extra work, so the wiki drifts out of date, people stop trusting it, and the company's real knowledge lives in inboxes and heads — and walks out the door with departures.
The tool that copies everything everywhere so it's never stale. Now your private notes — compensation conversations, candid feedback, half-formed ideas — are one search away from the whole company.
brainkit is the middle path. Knowledge flows from personal notes into the shared brain continuously — but only through a step a human approves, and private material physically stays private. The shared brain neither starves nor leaks.
The structure
Picture the company's knowledge as one brain. Every note in it sits in exactly one compartment, and the compartment decides who can read it. Nobody works from the whole brain — each person gets their own brain, built with only their compartments in it.
Both pictures are the same brain. The difference is what got written down when your copy was assembled.
Underneath the picture, that's one sealed master vault holding all company knowledge, organized into those same color-coded areas. The compiler makes each person their own copy containing only the areas they may read. Watch what happens to the orange and rose bars below — private areas belonging to other people simply don't exist in your copy.
The striped bar is the whole point: other people's private material isn't hidden, locked, or filtered out of your copy — it was never put there.
The single source of truth — one archive per company, holding every note as plain, readable files. Only the server ever touches it directly.
The security wall. It rebuilds every personal copy each time the master changes, and it "fails closed": if anything goes wrong, the worst case is you briefly see less than you should — never more.
When you edit your copy, the server checks every single change against what you're allowed to write before accepting it into the master. One out-of-bounds edit and the whole batch is declined.
The only way anything moves from a private area into a shared one. Your AI can draft a sanitized note and propose it; a person reviews and approves (or rejects) every single one.
Pointed at your vault alone. It files your meeting notes, tracks your action items, keeps your memory file a tidy overview — detail splits into its own note when a topic grows — distills the articles, PDFs, and screenshots you send it into shared reference notes, and drafts what's worth sharing.
Ask a question in plain language and get the right notes back — by keyword and by meaning. The index is built only from your own copy, so it inherits the same wall. Facts with dates attached — who's the current contact, and since when — can be looked up as they stood on any date, with the full history kept.
What lives inside
There are no per-document permission checkboxes to get wrong. Every note lives in exactly one area, and the area itself is the permission.
The operation
First: how does anything get in? Every brain has an Inbox, and capture is meant to feel like everyday habits — no new app to learn. There are five natural ways to send something to your brain.
Your brain has its own address. Forward a thread, or fire off a note to yourself, and it arrives in your Inbox. Because it's tied to your email, the system knows it's you — a message from an address it doesn't recognize is turned away, never filed onto the wrong person.
Text or DM your own brain like you'd message a colleague. Each person's chat assistant is locked to them alone, so a quick thought on the go lands in your Inbox and nowhere else.
Record a voice memo after a call or between meetings. It's transcribed and dropped into your Inbox — the fastest way to capture something before it slips away.
Save a transcript into your Inbox folder, or use your phone's "share" button to send a document straight to your brain. Or just hit Capture in your dashboard and type a thought — it names the exact Inbox it lands in before you save. If it's a file, it's ingestible — the original name is kept for reference.
Services that take notes for you can deliver them automatically: a meeting recorder like Fathom, or automations built in Zapier, Make, or Composio. Each connection gets its own verified delivery door — every incoming delivery must prove it came from the service you authorized (a cryptographic signature or a shared secret), and repeats are recognized and filed only once.
Every one of these runs through a single, built-in intake step the company controls. Whichever channel you use — including outside services delivering on their own — the note passes through one safe entry point that confirms who's allowed to write where, checks each automated delivery's credentials, and stamps every note with where it came from. Capture stays effortless without loosening the privacy wall. (A shared, company-level assistant can also take in org-wide material like all-hands recordings.)
From there, everything is automatic. Here's the full journey of a single piece of information — say, the transcript of this morning's client call — from your Inbox to the company's shared memory.
Via any of the routes above — emailed, messaged, spoken, dropped in, typed into the dashboard's Capture box, or delivered automatically by a connected tool. That's the entire effort required from you.
Your assistant reads it, writes a summary, pulls out the decisions that were made and the action items (with owners and deadlines), and files each piece where it belongs. Anything it can't confidently place, it flags for you instead of guessing.
On a schedule, your changes flow back to the master vault — after the server verifies that every changed file is one you're allowed to write. Then everyone's personal copy is rebuilt fresh.
The call surfaced a pricing decision the whole company should know. Your AI drafts a sanitized version — only what's meant to be shared, with the source noted — and places it in your outgoing tray. It can propose; it can never publish. (You can propose one yourself, too, straight from a note in the dashboard.)
A designated owner reviews the pending queue — in the CLI or right in the admin dashboard's Promotions tab — and approves or rejects each draft after reading it. Approved notes land in the shared brain stamped with who wrote it, who approved it, and where it came from. Rejections keep a written reason — which teaches everyone what this company considers shareable.
On the next rebuild, the note appears in every person's copy. Six months from now, anyone can ask "why did we price it that way?" and get the decision — with its reasoning — in seconds.
The window in
A brain you can't look inside is a brain you won't trust. So both sides of the system come with a dashboard — a live, local page that updates itself the moment the brain changes, served only to your own machine and reaching nothing but its own origin. (A single self-contained HTML snapshot is still one flag away when you want something to email or archive.)
Open your vault in any browser and watch it stay current: how many notes you have and where they live, what's waiting in your inbox, your open action items, and an interactive map of your knowledge where every note is a dot connected to what it references — click one to see everything that links in and out.
Search by keyword and meaning, then read the note itself — rendered, with its backlinks beside it, in light or dark. And when a thought strikes, Capture writes it straight into your Inbox. Because it's built only from your copy, it can only ever show — and add to — what you're allowed to see.
Whoever runs the system gets the whole picture at a glance: each person's vault — its size, how fresh its search index is, whether edits are waiting to sync back — the system's own health checks, and a table of exactly who may read and write each area.
It's also where the human gate lives: the Promotions tab shows each pending share with its destination, a warning naming who will be able to read it, and its full text — with Approve and Reject right there, and a badge that flags new arrivals from any tab. You approve under your own name, picked from a list scoped to whoever may approve that particular item — an admin for anything, plus a team's lead for a note headed into their team's space — so the record can't carry a typo, a blank, or a name that wasn't entitled to say yes. What stays read-only is the permission configuration itself: it's edited in the audited files where every change leaves a record.
The dashboards observe by default and write only through the same guarded doors as everything else — capture can reach nothing but your own Inbox, approval still passes the human gate, and no view can rebuild an index or change a permission.
The trust story
Most AI tools handle privacy with instructions: the system tells the AI not to reveal things and hopes it listens. brainkit doesn't ask. It's built so that the sensitive material isn't there to reveal.
A useful analogy
Think of a briefing binder prepared for each executive before a board meeting. Nobody hands everyone the master file and asks them not to peek at each other's sections — each person's binder is assembled to contain only their pages. brainkit does that, automatically, for every person, on every update.
Your copy physically contains only what you may read. There's no lock to pick and no filter to fool — what you can't see was never written to your copy. Even links pointing at notes outside your view are replaced with a stub, so the contents can't travel through them.
If the rebuild ever hits an error, your previous copy stays in place untouched. Every failure path leads to seeing less, temporarily — there is no error that shows you more.
Your assistant has no ability to write into shared areas. The only door from private to shared is the promotion queue, and a human holds the key to it — every time, with a named approver on record.
The privacy claim is tested adversarially: automated tests generate thousands of random company shapes — including planted "honeypot" content designed to escape — and verify that no one's private material ever survives into anyone else's copy.
The payoff