Memory for coding agents

Your agent relearns
your project every
single morning.

It re-reads the same files, re-derives the decision you made last Tuesday and asks you a question you already answered. Scrollback is not memory. It dies at compaction and it dies again at session end. ns-brain is the memory that survives both. A new session opens with a short brief, like this one. Then, as you work, your agent is reminded only of what it learned that bears on the task in front of it. Months of memory and the context stays light.

One binary. SQLite underneath. No daemon, no account,
one version check a day and you can turn it off.

session start · before your first turn injected by a hook
════ ORCHARD BRAIN — SESSION BRIEF ════════════
  now      Mon 09:12 · last session 2d ago
  git      main · clean · 4 unpushed

## GUARDRAILS (binding, always shown)
  #48 [constraint] An invoice number is never
       reused, not even after a void
  #112 [constraint] No schema change ships
       without a migration that reverses it

## HOW THEY WANT TO WORK
  #63 Run the tests before every commit. That is
       standard behaviour, not something to ask about

## DUE NOW — 1 task
  [ ] #9 rotate the payment webhook secret
        before the next billing run

## THE LAST SESSION HERE LEFT NOTHING BEHIND
  2 commits, 0 memories. Whatever it worked out
  is gone. Store yours before you finish.
no prompt asked for this a hook did it
What you will notice

Install it on Friday.
Notice on Monday.

It arrives knowing

No more first-day questions

Your agent opens every session knowing what you decided, what broke last time and how you like to work. You stop answering the same question twice.

Corrections stick

Fix a fact once

Correct it and the old version stops coming back. Ask in last month's wording and you get this month's answer.

Every repo

Worked out once, found everywhere

Keep your projects in one shared store and what you work out in one repository can be found from any other, whichever agent you have open.

Lighter context

Fewer tokens, less re-reading

The brief is ranked and capped at about 1.5k tokens, with the binding rules first. Your agent asks the brain before it searches files, so it stops paying to re-derive what an earlier session already worked out.

Knows how you work

Your habits, kept

Say once how you like to work, what you want and what you never want. Every session starts with it, so you stop repeating yourself at the top of every chat.

New projects too

Day one follows your rules

Axioms bind every project you have. A repository you created this morning opens its first session already knowing your ways of working.

The more you use it,
the better it gets.

Free for one developer, on as many projects and machines as you want, for as long as you want it.

$ curl -fsSL https://brain.fightclub.pro/install.sh | sh
$ cd ~/projects/yours && ns-brain init