Compare

recall beside every memory system it has been measured against

Nine comparisons, one page each: what the other system is, where the two were measured on the same benchmark, and what each costs in model calls, latency and privacy. Every figure names its source, and each page says who should pick which.

recall 0.5.1The charts on the home pageMethod and sources

The other systems’ figures are their own, from their papers, READMEs and sites, each with its own answer model and judge, or the SWE-ContextBench paper’s runs. recall’s are reproducible with recall eval. One difference holds on every page: the other systems run a model over every message they store, and recall runs none.

Comparison	What it is	Where both were measured	Theirs	recall
vs Mem0	Memory API, hosted or self-hosted	SWE-ContextBench, past fix in the top 3	39.4%	76.8%
vs Zep	Knowledge-graph memory, cloud	LongMemEval_S, answers correct	90.2% (site) · 71.2% (paper)	88.4%
vs Supermemory	Hosted memory API with agent plugins	SWE-ContextBench, past fix in the top 15	59.6%	97.0%
vs LangMem	Memory SDK for LangGraph agents	SWE-ContextBench, past fix in the top 10	73.3%	92.9%
vs Mastra	Observational memory in a TypeScript framework	LongMemEval_S, answers correct	94.9%	88.4%
vs Honcho	User-modelling memory, hosted or self-hosted	LongMemEval_S, answers correct	90.4%	88.4%
vs Hindsight	Open-source memory engine with a cloud	LongMemEval_S, answers correct	94.6% (board) · 91.4% (paper)	88.4%
vs EverMemOS	Memory runtime, now EverOS	LongMemEval_S, answers correct	83.0%	88.4%
vs OpenViking	Context database from ByteDance	SWE-ContextBench, past fix in the top 3	51.5%	76.8%