BERNARD Brain Kit (free mini)
First clip: what this free kit does (a brand-new chat finds the saved note). Clips two and three: the rule hooks that come with the full kit.
What it does
Claude Code forgets everything between sessions. This kit gives it a memory that finds itself: your notes live
in a plain folder of Markdown files (a vault), a map of them is built for you, and on every prompt a small hook
puts the notes that match in front of Claude before it answers. You see a short BERNARD brain: on line when a
session starts and a BERNARD ROUTE packet under each prompt.
This is the free mini. The full kit adds: capture of every file Claude changes into a devlog and an in-progress
session note, rule hooks (small Python files that block Claude when a rule is broken, four ready to use), the
/bernard:save and /bernard:rule skills, and the guide.
Safety
No network calls, no keys, standard-library Python only. Scanned with AgentShield on 3 Nov 2026: 0 critical, 0 high, 0 medium findings (1 low: no session-end hook, by design in the free mini).
What you need
- A Mac (macOS 13 or later). Mac only for now: Linux and Windows are not supported.
- Claude Code 2.x, signed in.
- Python 3.8 or newer (a Mac with the command-line tools has 3.9). Nothing else
to install: the kit uses only Python's standard library.
Install in three steps
- Get it into a permanent folder (the plugin runs from there, so not
~/Downloads):
git clone https://github.com/albygira-cmyk/claude-brain-kit ~/bernard-brain-kit-free
- Open a terminal,
cd to that folder and run:
bash install.sh
- Open Claude Code and start a new session. Its first lines of context read
BERNARD brain: on.
What you will see
- Boot line - a short summary of the vault when the session starts.
- BERNARD ROUTE packet - on every prompt, up to five matching notes with their paths, or
NO_RELEVANT_MEMORY when nothing matches (Claude is told never to invent a memory).
Writing a note Claude can find
Put a Markdown file in ~/brain/topics/<domain>/<two-or-more-words>.md (or patterns/, references/,
projects/, sessions/). Under the title, write one plain sentence saying what the note is for, in the words
you would search with. The map rebuilds itself on the next prompt; python3 bin/brain.py map rebuilds it now.
Skills
/bernard:setup - creates the vault and turns memory on (install.sh already did this)./bernard:find - searches the vault by hand and answers with note paths.
Commands
python3 bin/brain.py status # where the vault is, how many notes, is the map fresh
python3 bin/brain.py map # rebuild MAP.md and the search index
python3 bin/brain.py check # names and first sentences the router cannot use
python3 bin/brain.py doctor # OK / FIX lines for every part of the install
Uninstall
bash uninstall.sh
Removes the plugin and its internal folder ~/.bernard. Your notes in ~/brain stay unless you add
--remove-vault --yes.
What is in the folder
.claude-plugin/ plugin manifest
hooks/ boot.py, route.py, brainlib.py
bin/brain.py command-line interface
vault-template/ starter notes and templates
skills/ /bernard:setup, /bernard:find
README.md (this file)
LICENSE.md MIT
install.sh / uninstall.sh
Tested and not tested
- Tested on macOS with Claude Code 2.1.x, Python 3.9 and 3.14.
- Not tested on Windows.
- Not tested on Linux.
Privacy
The plugin itself makes no network calls, stores no keys, and reads only your vault (~/brain by default) and its
own folder ~/.bernard. The notes it puts in front of Claude become part of your Claude Code conversation, so they go
to Anthropic like anything else you type there: keep passwords and keys out of your notes.
Licence
MIT (see LICENSE.md). Your notes are yours.
The full kit
The full kit (capture, four rule hooks, /bernard:save and /bernard:rule, the guide) is US$9, one time:
https://www.waar.com.hk/brain/
