BWhat changed
A long-running assistant may need a few relevant facts rather than an entire conversation history. The 2025 Mem0 paper describes extracting candidate memories, finding related stored items and deciding whether to add, update, delete or ignore each candidate. A graph variant also stores relationships. The researchers report conversational-memory results and query-time costs. The contribution is the specified incremental memory pipeline and its source-bound evaluation, rather than a certification of the current commercial product or ownership of the models it uses.
Scientific scope
A long-running assistant may need a few relevant facts rather than an entire conversation history. The 2025 Mem0 paper describes extracting candidate memories, finding related stored items and deciding whether to add, update, delete or ignore each candidate. A graph variant also stores relationships. The researchers report conversational-memory results and query-time costs. The contribution is the specified incremental memory pipeline and its source-bound evaluation, rather than a certification of the current commercial product or ownership of the models it uses. These are author/vendor-reported experiments. Full-context inference has a higher overall judge score in the reported table, while Mem0 uses less retrieved context and has lower measured latency. A headline relative gain compares against a particular manual memory proxy, not every current OpenAI system. Extraction and updating are not included in every cost figure. Adversarial questions are excluded from the main comparison, and a quoted competing result has a category-label discrepancy that prevents an unqualified cross-paper ranking.
Evidence
The linked original works support the scoped statement and contribution attribution. WRS has not independently replayed experiments or reconstructed a full mathematical proof. These are author/vendor-reported experiments. Full-context inference has a higher overall judge score in the reported table, while Mem0 uses less retrieved context and has lower measured latency. A headline relative gain compares against a particular manual memory proxy, not every current OpenAI system. Extraction and updating are not included in every cost figure. Adversarial questions are excluded from the main comparison, and a quoted competing result has a category-label discrepancy that prevents an unqualified cross-paper ranking.
Contributors
- The named authors in the research record · SCIENCE: Source-listed human researchers and engineers for the exact method or resource. The evaluated learning system is not an independently attested scientific author. Earlier methods, funding and data-collection roles remain separate. Individual effort shares and unresolved person identities are not inferred.
Provisional WRS assessment, edition 2026-10-23.9. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction