MoM: Memory of Memory
Bowen Qin and Yao Lu at the National University of Singapore propose Memory of Memory (MoM): agent memory that commits the current value on arrival while keeping the displaced value and its provenance.
Ask this paper
The missing combination. Archival memories store everything and reconcile at query time, so stale values come back; CRUD memories overwrite on write, so a wrong update cannot be undone. MoM commits on arrival and retains what it displaces.
P-Mem. A typed provenance graph whose active frontier exposes one current value per key; each new observation is typed as support, supersede, contest, reject, revoke or resolve.
Validity over accuracy. On LongMemEval P-Mem matches the strongest retrieval memory in accuracy with about 4x fewer read tokens, and graph-guided pruning cuts the stale-answer rate on knowledge updates from 19.4% to 10.9%.
Revision chains. On ALFWorld-Revision chains P-Mem stays at 100% while query-time reading falls to 25%.
Recovering errors. Because displaced values are kept, P-Mem recovers committed errors that a CRUD memory cannot (100% vs 0%).
Abstract
For a long-horizon LLM agent, the memory question is not what was once recorded but what \emph{currently holds}. Most designs answer it only indirectly: every interaction is stored, and the present is reconstructed at query time by retrieving and reconciling records, so stale values re-enter and the same conflicts are re-litigated. Committing the current value at write time avoids this, but existing write-time (CRUD) memories overwrite, so a wrong update is unrecoverable and prior state is lost. We take the missing combination---\emph{commit on arrival while retaining what is displaced}---and formalize it as \textsc{Memory of Memory} (MoM): memory tracks not only content but the provenance, status, and history of its own entries. We instantiate MoM as \textsc{Provenant Memory} (P-Mem), a typed provenance graph whose \emph{active frontier} exposes one current value per resolved key while displaced values are retained as provenance; typed operations decide whether a new observation supports, supersedes, contests, rejects, revokes, or resolves an existing value. P-Mem's decisive gain is validity rather than accuracy: its turn-level read matches the strongest retrieval memory in accuracy at $\sim$4$\times$ fewer read tokens---a retrieval-granularity effect---while graph-guided turn pruning cuts the knowledge-update stale-answer rate (19.4\%$\rightarrow$10.9\%); on revision chains it stays at 100\% where query-time reading collapses to 25\%, and, because displaced values are retained rather than overwritten, it recovers committed errors a CRUD memory cannot (100\% vs.\ 0\%).