Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-08-13 · source-backed
Adaption Labs inverts the usual framing: retrieval can only search memories that exist, so extraction is the leverage point. They write two forms, narrative memories preserving causal context, atomic memories preserving verbatim facts, and never delete, appending supersession markers instead. Reported LongMemEval 90.6% vs Mem0 OSS 71.6% and full-history 60.6%, with 97% fewer tokens. Vendor benchmarks, no independent replication, weight accordingly. The "we expected to need knowledge graphs and didn't" note is the honest part.
Each link below shares sources, entities, or timing with this story.
Adaption Labs benchmarked against mem0 / Shared entity: Agent / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Adaption Labs benchmarked against mem0); both cover Agent; overlapping topics (agent, benchmark, context, memory).
Shared entity: Agent / Shared topic / Earlier coverage / Tension
Both cover Agent; overlapping topics (agent, benchmark, memory); earlier Agent coverage from 2026-06-05.
Both cover Agent; overlapping topics (agent, context, memory); earlier Agent coverage from 2026-03-05.
Adaption Labs benchmarked against mem0 / Shared topic
Linked by a graph relationship (Adaption Labs benchmarked against mem0); overlapping topics (agent, context, fact, memory).
Shared entity: Vendor / Shared topic / Earlier coverage / Tension
Both cover Vendor; overlapping topics (accordingly, benchmark); earlier Vendor coverage from 2026-08-02.
Shared entity: Agent / Shared topic / Earlier coverage
Both cover Agent; overlapping topics (agent, context, memory); earlier Agent coverage from 2026-08-02.
Both cover Agent; overlapping topics (agent, context, memory); earlier Agent coverage from 2026-07-19.
Shared entity: Vendor / Shared topic / Earlier coverage
Both cover Vendor; overlapping topics (accordingly, agent, benchmark); earlier Vendor coverage from 2026-07-18.