You hit your limit.The subscription caps out mid-task.
Your device goes off.Or you move to another one — a laptop, a phone, a box that isn't yours.
Context gets too expensive.The history grows until carrying on costs more than starting over.
Every one of them forces a fresh session, and every fresh session forgets you.
Handoff notes don't fix it: a note carries information, not how you work.
Any AI that speaks MCP — pick yours
Loading…
No key in any of these, and nothing to sign up for first. Yours not listed? If it
speaks MCP it already works — that's the last chip.
Before you put your life in it
Anyone holding one of your keys can read and write your memory — that is what a key
is for. And the operator of this service has database access, so it is technically
possible for him to read your files. He does not, but choose what you store knowing
that he could. Nothing here is end-to-end encrypted.
So: a good place for how a project stands and how you like to be worked with. Not a place
for passwords or card numbers. Your memory is never used to train anything, sold, or
shared. Run by Godspeed Digital Ventures, Accra, Ghana —
the full privacy and terms say the rest, plainly.
The handoff, as it actually happens
"Can we continue?"
How it works
Run the commandIt holds no key. Your browser opens once so you can sign in.
Ask it anything about youIt reads your memory before it answers, on any machine you land on.
It writes as you goDecisions, where a thing stands, how you want to be worked with — saved without being told.
Notes are the easy half
What a new session actually misses is not facts about your project — it's how you want
to be worked with. That lives in one file, rules.md, and every AI that
connects reads it before it answers you. Write it once; it lands on the next machine,
the next model, the next tool.
1. One step at a time
2. Real answer first
3. Mark what you verified and what you're guessing
4. Build the command for my shell
5. Don't shrink my ideas
Only you can write it. A connected AI can read your rules and can never
change or delete them — that's enforced on the server, not asked for politely.
You edit them in your dashboard.
When the AI runs in the cloud
Some AIs don't run on your computer: web apps, and coding agents that work on a
server. They can't see anything you installed, so the command above does nothing
for them. Add your memory in that AI's own settings instead, with this address:
Then start a new chat or task. Some of them only pick up a new
connection when a session starts.
ChatGPT — Settings → Security and login → turn on
Developer mode, then add an app with the address. Paid plans, on the website.
Claude (the app, and Claude Code on the web) — Settings
→ Connectors → Add custom connector.
Cursor cloud agents — cursor.com/agents → the MCP menu.
Devin — Customize → MCPs → Add custom MCP. Choose
Personal: Organization gives your whole team one shared
memory — yours.
Gemini — gemini.google.com → Settings → Connected
Apps → Add a custom app. For now only in the US, 18 and over, on a personal
account. Set it up on the website; it then works on your phone too.
GitHub Copilot's coding agent — it can't sign in through a
browser, so it needs a key. Someone who runs the repo adds the
address and the key in the repo's Copilot settings.
Grok — grok.com/connectors → New Connector → Custom.
Le Chat (Mistral) — Connectors → Add Connector →
Custom MCP Connector. Works on the free plan.
Not yet: Codex's cloud tasks and Google's Jules don't let you add a
memory of your own. Codex on your own computer does — use the command above.
Not listed? If its settings have a Connectors or MCP screen, paste
the address there.
Rather have a key?
You don't need one to start. Run the command above and your browser signs you in,
making the account if you haven't got one. This is for a script, a bot or an agent
that cannot open a browser — it gets a key of its own that you can revoke on
its own.
Do not run the setup command again. It will only answer
already exists in user config. The connection is still there —
it is your sign-in that ended, not the setup.
Sign in from inside the AI instead:
Claude Code — type /mcp, choose
knowsme (or portable-memory, if you set it up before October 2026), and sign in.
An AI app (Claude, ChatGPT and the like) — open its
connector settings and reconnect knowsme (called Portable Memory if you added it before October 2026).
Another CLI or editor — look for its MCP or connectors
screen. Restarting it will usually prompt you.
You will see this whenever you end a session yourself, or when a sign-in on a
computer that is not yours reaches its 8-hour limit. It is normal.
Leaving a computer that is not yours
Setup writes knowsme into that computer's own user account, so it appears
in every folder and stays after your session ends. To leave it clean:
End the session. Open your memory from any device and end it
under Signed-in AIs. Your key stops working straight away.
Then remove the connection, so its owner is not asked to sign
in for a memory that is not theirs:
claude mcp remove knowsme --scope user
Set it up before October 2026? Then the connection is called portable-memory:
claude mcp remove portable-memory --scope user
Only remove it if that person does not use knowsme themselves. One computer
holds one connection, shared by everyone who uses it — removing it removes
theirs too. If they do use it, stop after ending your session; their own sign-in is
untouched.
Your token
Save this now. It is stored hashed — this is the only time it can be shown, and it is the
only thing that unlocks your memory.