Turn the team's Google Drive into a searchable, linked knowledge base that
Claude can read in full, scoped automatically to what each person is allowed to see.
Everything the team knows is scattered across hundreds of documents. Asking an
AI assistant to search Drive gets you one keyword query at a time and a few snippets. Nobody
has read all of it, and nobody ever will.
Read once, reuse forever
The tool reads every document once and writes short, linked notes. That reading is the
expensive part, and it happens a single time for the whole team.
Everything cites its source
Every claim points back to the document it came from. Nothing is invented, and
disagreements between documents are flagged rather than smoothed over.
Nobody sees more than they should
When a teammate installs it, it checks their own Google permissions and keeps only what
they can already open.
How the knowledge gets built
Four steps. The first three happen once, on the team lead's machine.
01
Copy the Drive
Every document becomes a plain text file on the computer. Read-only:
nothing in Drive is ever changed.
02
Understand what is in every document
Hundreds of readers work in parallel, one document each. A reader's job is
to work out what the document is actually about: what it argues, what it decides, which
programme or partner or number it concerns. That understanding comes out as a few short
notes, one idea each, every one citing the document it came from. Meeting invitations and
logistics are skipped.
03
Connect documents by what they are about
One librarian files every note in turn. Notes about the same idea, from
documents that never mention each other, are merged into a single entry that cites all of
them. A topic touched by twenty documents becomes one synthesised page rather than twenty
fragments, and documents end up linked through the subjects they share. Where two documents
disagree, both claims are kept and the conflict is flagged.
04
Everyone gets their own copy
Each teammate installs it and signs in. The tool checks what they can open
in Drive and keeps only that. It then updates itself every night.
Start it with one paste
For the person setting this up for a team. Open Claude Code in a terminal, paste
this, and it will do the rest: fetch the tool, install what it needs, and open your browser to
sign in to Google. If someone has sent you a bundle instead,
skip to the two-minute version.
Set up Hivemind Team Vault for me on this machine.
1. Clone https://github.com/jamesbedford-beep/hivemind_super_badass_repo into my Documents folder.
2. Make sure Python 3.10+ and rclone are installed (install rclone with winget if it is missing).
3. Start the Team Vault app: run teamvault/server.py, pointing the vault at Documents/TeamVault.
4. When my browser opens, walk me through connecting my Google account (read-only) and picking which drives my team's context comes from.
Explain each step in plain language as you go, and stop and ask me before anything that costs money or touches Drive.
Paste it into Claude Code. It clones the tool and checks what your machine needs.
Sign in to Google. A browser tab opens. Access is read-only, and the token stays on your computer.
Pick your team's drives and let it scan. It suggests the documents worth including, with a reason for each.
Review and build. You see the cost before anything is spent, and anything personal is held back automatically.
Share one bundle. Everyone installs it and gets their own correctly-scoped copy.
Nothing leaves your computer. The Drive copy, the notes and the
sign-in token all stay local. The tool never uploads your documents anywhere.
Personal material is held back by default. Performance reviews,
compensation, hiring and candidate files and donor profiles are excluded even when the person
installing could open them in Drive. Access is not the only test.
Been sent a bundle? Two minutes.
You do not need any of the above. Someone on your team has already done the
expensive part; you just need your own copy of it.
Unzip TeamVault-Setup.zip anywhere on your machine, and read
START-HERE.md if you want the longer version.
Run python install.py from that folder. You need Python 3.10+
and rclone; the file tells you how to get them.
Sign in to your own Google account when the browser opens. It is read-only,
so nothing here can change anything in Drive.
Open the wiki folder in Obsidian, or point an AI assistant at
it and start asking questions.
I have been sent a Hivemind Team Vault bundle as a zip file. Please set it up for me:
1. Unzip it into my Documents folder and read START-HERE.md.
2. Check I have Python 3.10+ and rclone, and install rclone with winget if it is missing.
3. Run install.py from that folder and walk me through signing in to my own Google account.
4. Tell me how many notes I ended up with, and open the folder when it is done.
Explain what is happening in plain language, and stop if it reports any unresolved links.
You will not receive everything in the bundle. On install it
checks, document by document, what you can already open in Drive, and keeps only the notes built
entirely from those. Links to anything held back become plain text, so you never hit a dead end.
Re-run it any time: it refreshes your copy and removes anything you no longer have access to.
Questions people ask
Do I need to be technical?
No. One paste into Claude Code, one Google sign-in, and a few clicks in a browser.
Will it show me things I should not see?
No. Your copy is built from your own Drive permissions, checked file by file.
Does it go stale?
It syncs each night and adds anything new you can see, so it stays current on its own.
What does it cost?
The reading is the only real cost, and the team lead pays it once. Installing is free.