Answers are grounded in this paper and cite their sources.
Keep every word. Load only the meaning. A local memory layer for people who work with an AI all day, every day. The full conversation log is never summarized away. A meaning layer with **line-level addresses** sits on top of it, so a new session pulls in only the few sections it needs — and can still walk back to the exact lines they came from. A dated public record of the design. The implementation is not published. More: https://subtractlab.com
No takes yet. Share an insight, caveat, or question.
Koji Okuda (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: