The memory that shows its work.
Any AI memory sounds confident. Virgil shows where the answer came from - what you said versus what a model inferred, dated, and corrected in the open.
One question. Two kinds of memory.
Where does my brother live now?
Dave lives in Austin. He recently started a new job there and has been spending more time in Denver lately.
No dates. No sources. No way to tell what you said from what it guessed.
Where does my brother live now?
Denver. You told me on 2026-06-16 (fact 3107). Austin stays in your record - superseded, not erased.
3 facts on Dave in your record
- [you said] "Dave moved to Denver"
- [you said] "Dave lives in Austin"
- [inferred] "The Denver move was for Dave’s new job"
The right card is a faithful illustration of what a live Virgil recall carries - labels, dates, fact ids, corrections kept.
-
Labeled
Every fact is [you said] or [inferred] - a closed vocabulary. Virgil never guesses which.
-
Dated
Every fact carries its date. Imports are a paste you vouch for - what you bring in lands dated and labeled [you said], and nothing enters your record unless you brought it.
-
Corrections kept
New facts don’t overwrite old ones. The old fact stays in the record, marked superseded by its successor.
Keep the receipts.
Connect Virgil once and every answer can show its work - on Claude, ChatGPT, Grok and whatever ships next.