Second brain: what Pinchy remembers

Pinchy’s memory is a plain folder of markdown files on your computer, called his second brain. He reads a short part of it at the start of every chat and looks the rest up when a question needs it. You can open, read and edit every file yourself.

What to ask for

“Remember that my sister’s birthday is 14 March.” He writes a note and adds it to his index.
“From now on, always answer in British English.” He writes it into his standing instructions.
“What did we decide about the Plantpal database?” He searches his notes and past chats.
“What do you know about me?” He reads it back from USER.md.
“Forget what I told you about my old job.” He finds the note and removes it (and says what he removed).
“What did the swarm build for the Acme report last week?” He looks it up in the note he keeps for that project.

What is in the folder

USER.md: who you are and what you like. Read at the start of every chat.
AGENTS.md: your standing instructions for him. Read at the start of every chat.
MEMORY.md: the index, one line per note. The notes themselves are in notes/ (by topic: people, projects, howto, decisions).
IMPULSIVE.md: the key points of your last few chats, so a new chat is not confused.
PLACES.md: folders and files he learned where to find.
journal/: one file per day with what was talked about and done.
notes/projects/, notes/agents/, notes/research/: what finished jobs found out, one note per project, per agent and per research question (see the next section).
CAPTURED.md: the list of what was saved from finished jobs, newest last.

Saving finished work

When a Work or Code job, a swarm, an agent run or a deep research finishes, Pinchy files its own summary in the brain: what was made, where the files are, what is still open.
Few notes that grow, not one note per job: everything done in one project folder goes into that project’s note with the newest entry first, an agent’s results go into that agent’s note, a research answer into a note named after the question.
The same job finishing again replaces its entry. A run that found nothing new is not saved.
These notes are not read at the start of a chat. He finds them when he looks something up, so they cost nothing until they are needed.
A line that looks like a password or a key is left out, and the entry says so.
Each entry is marked as saved from a finished job: it is information he can read, not instructions to follow.

How it works

A small model looks at the chat every few messages and notes what is worth keeping. When it saves something you see a small “Noted” message with Undo.
When you tell him to remember something, he saves it in the same turn.
Before answering a question that may have come up before, he searches his notes; when the notes have little on it, the search also brings back matching lines from past chats.
Claude Code keeps its own notes for a folder you also use with Claude Code. Those are not Pinchy’s memory and are never loaded into his chats. When his own notes have nothing on a question, his search also looks there and tells him whose notes they are.

Settings

Settings → Brain → Memory: switch memory on or off, pick the brain folder (any visible folder works, for example ~/Pinchy Brain), how often he updates his notes, the “Noted” message, recent chats, the daily recap, and how much memory is read at the start of a chat.
Settings → Brain → Memory → Saving finished work: one switch for all of it (“Save what finished jobs found out”), and one each for Work jobs, Code jobs, Swarms, Agent results and Research.
Settings → Brain → Memory → Open brain folder: opens the folder in your file manager.
Settings → Brain → Instructions: your standing instructions (the same text as AGENTS.md).
Settings → Safety & data → Backup & move: save the whole brain (with skills, tasks and settings) into one file.

Limits

Passwords and other secrets do not belong in notes. His own accounts are kept in a separate secrets folder inside the brain.
Only a capped part of the brain is read at the start of a chat (12,000 characters by default). Everything else is found by searching, so long notes cost nothing until they are needed.
An agent keeps its own memory in its own folder inside the brain; it does not write into the main Pinchy’s notes. What an agent REPORTS when a run ends is saved under notes/agents/ by Pinchy, not by the agent.
At most 40 finished jobs are saved per day, and a project’s note keeps its newest 30 entries.

All pages · Download Pinchy