Mem0's delete_all() silently kept most memories intact

Mem0 fixed a delete_all() bug on August 1 that left most stored memories in place while telling callers the deletion succeeded, per the project's v2.0.15 release notes.

That's a compliance problem: anyone who assured a user their agent memory was deleted can't trust that answer anymore. Vector stores typically cap a single listing page around 100 entries, so larger accounts kept the overflow after a supposed full wipe.

The function listed the vector store once instead of paging through it, so it only ever saw the first page. The call still returned success regardless.

The same release also caps Supabase's search and list top_k at the vecs limit of 1000, instead of erroring like it used to. It also sets size on Elasticsearch KNN queries, which had quietly capped results at 10 hits no matter the requested top_k. All three share one shape: the return value looked fine while the underlying store did less than asked.

Mem0's notes don't say how many accounts crossed that page-size threshold before the fix landed. My bet: that number stays undisclosed, because admitting it invites the exact compliance exposure this bug created. Watch whether mem0 or any product built on it ever publishes an affected-account count.