Research & Benchmarks

    We built MemoryStack to solve a problem that kept us up at night: AI agents that forget. To prove it works, we tested it against the toughest memory benchmarks in the industry—and the results speak for themselves. Here's what we discovered when we put our system head-to-head with existing solutions.

    Benchmark Results

    LongMemEval-S Benchmark:
    MemoryStack
    vs
    Zep
    vs
    Full Context
    020406080100100%72%85%Single-SessionUser91.1%80.4%94.6%Single-SessionAssistant89.5%55%64%Multi-Session93.3%56.7%20%Preference90.2%45%60%TemporalReasoning97.4%83.3%78.2%KnowledgeUpdate

    Research Paper