# memory

> Nothing forgets. Nothing forks. A memory is a short note an assistant writes down —
> a decision, a measured result, where a project stands. They all go to one database
> on machines we own, and any assistant we allow can read them and add to them. Three
> moves: read, so a tool gets the note that was actually written back rather than a
> guess reconstructed on the fly; write, so a fact learned in one app is recorded once
> and every other tool sees it next read; summarize, so long histories distil into a
> few lines instead of months of re-reading. Recall matches on meaning, not exact
> wording — ask about a deploy problem from last month and the right note comes back
> even if it never used those words. The model is swappable; the memory isn't. It
> lives outside the model, so changing which model we run doesn't move the history —
> memory kept inside a model makes every upgrade a small amnesia. And when the store
> can't be reached the tool says so out loud: the reply comes back marked stale rather
> than quietly answering from an old copy. memory isn't offered on its own yet — it is
> the substrate under the rest of the family, working every day.

## Elsewhere in the workshop

- [The whole family, mapped](https://strata2signal.com/family.html) · [the research index](https://research.strata2signal.com/llms.txt).
- [RuleSage](https://rulesage.strata2signal.com) · [RealKeep](https://realkeep.strata2signal.com) · [amble](https://amble.strata2signal.com) · [design](https://design.strata2signal.com/gallery.html) — what the substrate holds up.
- Say hello: [strata2signal.com/contact](https://strata2signal.com/contact/) or hello@strata2signal.com.

This site makes no third-party requests — no fonts, no analytics, no CDN.
Where your question goes: https://research.strata2signal.com/where-your-question-goes/