Blog · July 30, 2026 · 1 min read
One message in, four memories out
What extraction actually does to a paragraph, and why we split aggressively instead of storing whole messages.
People often ask why Esfis does not simply store your messages and search them later. It does store them, as documents. But what it recalls are memories, and memories are smaller.
The paragraph
Had a great call with Alex. He's enjoying the new PM role at Stripe, though the payments work is intense. He moved to Seattle for the job, Capitol Hill. Wants dinner next time I'm in town.
The memories
- Alex works at Stripe as a PM. since August 2026
- Alex finds the payments work intense. as of September 4
- Alex lives in Capitol Hill, Seattle. supersedes: Alex lives in Austin
- Alex wants dinner when I am next in Seattle. open intent
Four atomic facts, each with enough context to stand on its own, each linked to the entity Alex and to a date. The third one found an older fact it contradicts and marked it superseded. The fourth is tagged as an intent, so it can be surfaced when a trip to Seattle shows up in the calendar and quietly closed afterwards.
Why split
- Recall precision. A question about where Alex lives should not drag in his opinion on payments work.
- Updates. You can supersede one fact without touching the other three.
- Budgets. Eight small facts fit where one long message does not.
Why keep the document too
Because extraction is lossy and sometimes wrong. The original message is one click away from every memory, and if a fact looks off you can read what was actually said. We would rather show our work than ask for faith.