How I built memory that stops repeating itself: dedup at 0.85, conflicts flagged, nothing deleted
My desktop companion remembers things about me. That is the whole point of it — until it remembers the same thing twice, and then remembers the opposite of it, and then confidently asserts whichever one it read last. That is the failure mode I fixed in Ankita v2.7.5. The problem was not that the memory store lost facts. It was that it never questioned them. Every new fact went in as a new row, so…
Ankita v2.7.5, the desktop companion, now remembers things more accurately by preventing it from remembering the same fact twice. The issue was that the memory store never questioned the facts, storing each new fact as a separate row and leading to duplicate entries. The solution implemented is a deduplication system with two layers of comparison.
The lexical fallback ensures memory quality is not dependent on a provider being configured. When a new fact is near-identical to an existing fact, it updates that fact instead of creating a duplicate. If the facts disagree, they are kept as conflicting entries and the companion is instructed to ask which is current rather than asserting either.
The new rule is that a new fact is compared against what is already stored before being written, with a similarity threshold of 0.85 for merging. The system also flags conflicts and does not delete or rewrite facts automatically, ensuring the memory remains a personal store for accurate recollections.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
