Notes & memory
Runs are independent: each one starts as a fresh pod with no memory of the last. What makes Brix feel like the same agent every time is its notes — the small, durable record it keeps for itself between runs.
What a note is
A note is a conclusion, not a transcript. After every run, Brix writes down what matters for next time:
- decisions and their reasons ("kept the old endpoint for one more release — two clients still use it");
- open threads ("PR #412 is waiting on review"; "the CI flake in the checkout tests is known, not caused by our change");
- standing preferences learned along the way ("keep changes small; the team reviews on weekdays").
The pod's full log is not a note — the log stays with the run (Wakes, runs & activity). Notes are the carry-forward, clipped to what's useful.
How notes are used
Every wake composes its task from three parts, fitted to a fixed budget:
- your instructions — the standing brief,
- the most recent notes, oldest first (up to 40),
- this wake's prompt.
That composition is why run 10 knows what run 9 did — and why "check back tomorrow on the PR you opened" works at all: tomorrow's run reads yesterday's note.
Where notes come from
Brix writes them itself during the settle step of each run. The store is also designed to take notes from you or from an agent during a run — the foundation for richer memory as the beta expands.
What notes are not
- Not your chat history — notes live with your Brix, apart from your conversations.
- Not a transcript — they're small and selective on purpose, because every note competes for space in the next run's brief.
- Not public — notes belong to your Brix record in your account.
Why this matters
Continuity is the whole point of an always-on agent: the Monday run should know what Friday left open without you re-explaining anything. Notes are that memory — everything else in Brix (wakes, runs, notifications) just moves work through it.