Research
Memory benchmarks measured and reproducible
How MemorySync performs on public memory benchmarks — measured under the standard protocol and reproducible against the public API.
92.1%
LoCoMo
94.8%
LongMemEval
LoCoMo
1,540 questions · 5 categories
92.1%
Overall
1189
Mean tokens
94.3%
93.6%
86.9%
91.8%
Single-hopMulti-hopTemporalOpen-domain
LoCoMo
LongMemEval
500 questions · 6 categories
94.8%
Overall
854
Mean tokens
96.4%
99.2%
89.8%
98.5%
91.6%
93.3%
Single-session (user)Single-session (assistant)Single-session (preference)Knowledge updateTemporal reasoningMulti-session
LongMemEval
From input to lasting memory
Useful information is understood, protected, and shaped into connected memory that stays relevant over time.
Memory creation
The right memory when it matters
Every query is understood in context, recalled across connected signals, and assembled into focused information ready to use.
Memory recall
Build AI systems that remember and improve over time
Deploy persistent memory through APIs tested across leading benchmarks.Give every agent the context it needs to improve.