The memory problem
Ask Claude, ChatGPT or Copilot to get you ready for tomorrow’s meeting and it goes looking — right then, in a few seconds, through whatever it can reach. KNET Brain adds the memory it has been missing: a day’s worth read properly and written down, every day, so the answer comes from the whole record instead of this morning’s hurried search.
See it in ninety seconds
1:33 · captioned · nothing to sign up for
Watch it come up short
one question · asked twice · shown side by side
›help me prepare for my meeting with Acme tomorrow
stopped at page 1 of 74 — the rest was never opened
summaries only — no transcript was read
most recent only — nothing before March
two opened, ten listed by filename
■ full — search stopped here
!Answered from 4 days of a two-year relationship.
›help me prepare for my meeting with Acme tomorrow
every page the brain holds on Acme — written up day by day, none skipped now
▪ room to spare — nothing had to be dropped
✓Answered from everything captured so far, not the last four days.
the reading already happened — a day at a time, since you started
What this is
KNET Brain speaks MCP, the open standard for connecting tools to AI clients — so it attaches to whichever one you already pay for. It has no connectors of its own, and that is deliberate: your client does the fetching, the reading and the thinking, using access it already has. This is where what it worked out is kept.
The client you already use
Your subscription, your interface, your habits. Unchanged.
KNET Brain
Added the way any other connector is. It stores and links — it has no connectors of its own, and never reads your mail itself.
The same client, with a memory
Nothing is replaced. Something is added — and it is the one thing that was missing.
What is actually happening
You ask them to prepare you for tomorrow’s meeting. They started five minutes ago. So they do the only thing they can: search your inbox for the client’s name, skim what comes back, and write you a summary.
It will be well written. It will sound assured. And it will be built out of whatever they happened to turn up in the time they had.
A brilliant researcher on their first morning — every single morning.
Three quiet failures
Three things happen in the background, and none of them look like an error.
Searching your inbox hands back a limited slice of what matched. For a busy mailbox that can be a couple of days’ worth. It has to stop somewhere — three months of email is more text than it can hold in its head at once.
Ask about a call and most note-takers return their own short recap rather than the transcript. The detail you actually remember saying is not in the room unless you insist on it.
If the transcript misheard your client’s name, that meeting does not exist as far as the search is concerned. And nothing tells you it was missed.
Nothing warns you. You get a confident, well-written answer built on a fraction of what you know — and it reads exactly like one built on all of it.
The objection everybody raises
It feels like it should be. Linking your inbox, your drive or a folder of documents to a project looks like handing the AI everything you know.
Every question sends it back through that doorway to start searching from scratch, with exactly the same limits as before. Nothing carries from Tuesday to Thursday. Nothing accumulates. Ask the same question next month and it does the same rushed search all over again.
A connection is a doorway,
not a memory.
A brain has already read everything, understood it, and written down what mattered — a little more each day, for as long as you have had it.
How it is written
Once a day, your client goes through what happened — email, calls, CRM, documents — and turns it into pages joined to what is already there. You file nothing, you tag nothing, and you do not decide the categories.
Renewal moved to Q3 after the March pricing conversation.
Procurement now routes through Dana Ruiz, not Tom — changed in June.
Sensitive to per-seat pricing; asked twice about a usage tier. See [[pricing-model-2025]].
Warehouse automation came up on the 14 May call — same theme as [[northwind]].written today
A day of email fits. Two years does not.
This is the whole trick, and it is worth being plain about it. The daily run reaches your material through the same connectors that came up short in the hero. Nothing here has better access than your client already has.
What changes is the size of the job. Yesterday is small enough to read all the way through — the transcript rather than the recap, the thread rather than the first twenty-five hits. Each run knows what previous runs already covered, so nothing is fetched or judged twice, and what it understood is written down instead of thrown away.
Do that every day and the impossible question — read two years by Thursday — never has to be asked. It was answered a day at a time, in advance.
The page is the unit, not the transcript. The entry that already exists is corrected rather than duplicated, every claim carries the source it came from so you can check it, and when two sources disagree the disagreement is recorded instead of one side quietly winning.
Side by side
Where the answer comes from
On its own — A quick search it runs the moment you ask
With a brain — Pages written from material that was read at leisure, days or months ago
How far back it reaches
On its own — The first handful of results — often days, not years
With a brain — As far back as you have been running it, without re-reading a thing
What it knows about a meeting
On its own — The note-taker’s short summary
With a brain — The transcript was read in full once — what mattered is on the page
What it remembers next week
On its own — Nothing. You start again
With a brain — All of it, plus whatever the runs added in between
Who keeps it current
On its own — You, by re-explaining yourself every time
With a brain — A scheduled run in your own client, once a day
The older idea
The idea is older than AI. People have kept second brains for years — in Obsidian, in Notion, in a shelf of notebooks. The principle is simple. Instead of one long diary, you keep one page per thing you care about: a page for a client, a page for a decision, a page for the problem that keeps coming back. Whenever one page mentions another, you join them.
Do that for a year and you no longer have notes. You have a map. And a map answers questions no single note can. What changed our mind on pricing. Everyone we know who is dealing with this. What we promised this client, and when.
a page for every client, decision and problem — and how they join up
So why does almost nobody have one?
Because keeping it up is a second job. The pattern is famous: you set it up over a weekend, feel brilliant for a fortnight, and then stop. Someone has to write every page, remember to join them up, and go back and correct the ones that have gone out of date. The upkeep grows faster than the payoff, and the whole thing goes quiet.
That is the part that has changed — and it is the only part that ever mattered.
What it feels like
Before a meeting
“Get me ready for Thursday with Acme.” Not the last twenty-five emails — every conversation you have had with them, what was promised, what changed, what is still open.
The name you half remember
“Who was it that mentioned warehouse automation?” One passing remark, on a call the run swept eighteen months ago. You had forgotten it. The page had not.
The link you would never have made
On a pipeline call you rattle off ten companies. Each one becomes a page. Six months later one of them turns up in an unrelated conversation — and the two are already joined.
Each of these depends on the material having been captured — the brain answers from its pages and says so plainly when it has none. Which is why the run matters more in the first month than in the tenth, and why it is worth pointing at the back catalogue once rather than only at yesterday.
Whose brain it is
This is the reason it is not simply a shared folder or a notes field in the CRM. Some of what is worth writing down is not for the whole company: your read on a negotiation, something said to you off the record, your own doubts about a deal. Everybody gets their own brain, and what goes in it stays in it.
Setup
Added like any other connector, in whichever client you already use — on a laptop or a phone. Nothing to install and nothing to run.
One standing instruction in your own client: each day, go through yesterday’s meetings and mail and write them up. It uses the connectors you already have, and it resumes rather than starting over.
Answers are drawn from the pages at once, and they name the ones they came from — so you can check the claim rather than trust the tone.
Free, either way
These are the standing instructions your client works to. Two run in conversation — one says check the brain before answering, the other says write to the brain when something happens. Two more run on a schedule, sweeping a mailbox and a transcript source; they name no provider, and find whichever mail server or transcript source you actually have connected. Knowing when to do each is the hard half — and the reason most knowledge bases go quiet after a fortnight. All four are free, and yours whether or not you ever have an account here.
Look in the brain before answering anything about your people, projects or decisions — and say plainly when there is nothing there, instead of falling back on a guess delivered in the same confident voice.
↓ knet-brain-consult.zipTurn a call, a thread or a decision made in a corridor into pages, linked as it goes. Update what is already there rather than piling up near-duplicates, and when two sources disagree, record the disagreement instead of quietly picking a side.
↓ knet-brain-capture-chat.zipSweep whichever mailbox you have connected, every few hours, and write up what mattered. It resumes rather than starting over, and a quiet window that produces nothing is a correct run.
↓ knet-brain-capture-email.zipSweep whichever transcript source you have connected, overnight. Late and re-processed recordings update one page instead of arriving as a second copy, so a missed night heals itself.
↓ knet-brain-capture-transcripts.zipRead your calendar before the day starts and brief the meetings your brain knows the most about — who you are seeing, what was agreed last time, what is still open — in your 7am email. A meeting it knows nothing about gets no brief, rather than a guess.
↓ knet-brain-prepare.zipEach night, read one older day of your mail and calls into the brain, going back from where set-up began, so it starts with your past and not just from today. It stops before it uses the writes your live jobs need, and you can turn it off or ask for longer.
↓ knet-brain-backfill.zipThe bundles are packaged as Claude skills, which is the format with the widest support today. The instructions inside them are plain English and work in any client that can be given standing instructions.
Installing one takes a person: a skill fires only once the client has registered it, so an agent that downloads a bundle for you has done nothing yet. In Claude that is Settings → Capabilities, where skills are listed — upload the zip as it is, or have an owner add it once for a whole team. In Claude Code, unzip it into ~/.claude/skills/. Everywhere else, SKILL.md goes wherever that client keeps standing instructions.
No lock-in
One link returns the whole brain as ordinary files — one page each, readable in any text editor. Open the folder in Obsidian and every link still works. Nothing in there needs us in order to be read, which is the only version of “no lock-in” worth anything.
Start now
Install the plugin, connect it, and ask your assistant to set up your brain. Signing in creates the account — there is no invitation to wait for.
A paid plan on your assistant is needed for the recurring capture, because free plans cannot schedule work.