Your AI is fat.Feeding it is expensive.
Every conversation drags months of dead weight along, and when the memory lapses, the whole pile gets re-read at full price. sweap.it weighs your AI's memory, puts it on a diet, and keeps it off. Local-only. Reversible. Zero dependencies.
Weigh it. Diet it. Keep it off.
The program
three lifts, run daily
Step on the scale
sweap audits the agent's persistent store and shows exactly what each conversation costs: rows, megabytes, duplicated thinking blobs. You cannot diet what you do not weigh.
Cut, don't burn
Duplicate thinking blobs get quarantined, not deleted. Every diet is reversible: restore the quarantine and the store comes back byte for byte.
Watch the regrowth
A watcher samples the store every minute and flags when the weight creeps back. It flags, it does not prevent. The scale stays honest.
Know the bill
A lapse price turns megabytes into money at live model list prices: what a memory lapse costs when the whole pile gets re-read. Present tense, always.
Measured, not promised
one store · real rows · daily weigh-ins
finding On the tested store schema, every thinking blob was stored twice, across all models tested: 9,456 rows, 26 MB. weigh-in 2026-10-17 244.1 MB → 206.5 MB dieted 37.57 MB bytes lost: 0 (quarantine = full undo) watcher sampling every minute · regrowth gets flagged
Price board
what a lapse costs at list prices