Compare

Honest comparisons.

Choose the right tool, even when it isn't ours. Five AI memory layers, each steelmanned at its strongest, then measured against Papez under one frozen LoCoMo protocol. Every competitor fact is dated and sourced; where a system is genuinely better, this section says so.

Five comparisons

Papez vs. the field.

One graph you own, read everywhere over MCP, set against the alternatives, plainly. Pick the row closest to the decision you are actually making.

Papez vsMem0The category's most-adopted default for cheap, accumulated facts. Papez when a correction must be structure, a traversable supersede edge, not a read-time ranking guess.Papez vsZepThe most architecturally serious rival, bi-temporal validity at enterprise scale. Papez when the question is why the agent believes something, not just when a fact was valid.Papez vsSupermemoryStrongest at the front of the pipeline, wide ingestion, connectors, auto-built profiles. Papez is strongest at the back: how a memory links causally, ages, and can be explained.Papez vsLettaA full stateful-agent runtime with self-editing tiered memory. Papez is the memory underneath any runtime over MCP, engine-scored, auditable, and portable.Papez vsCogneeThe closest in spirit, graph-native with a real lifecycle stage. Cognee leads on ingestion breadth; Papez on causal recall you can audit, cite, and reproduce.
Every number in this section is measured under one frozen protocol. Papez certified at 85.55 ± 0.37 on LoCoMo, and no competitor number quoted unless it was published under a comparable configuration. The method, the ten runs, and the caveats are the LoCoMo citation surface.
Keep reading

Related.

Benchmark

LoCoMo results

The number and how to cite it, with the comparability caveats spelled out.

Concept

What is an AI memory layer?

Vector stores vs. temporal knowledge graphs vs. causal graphs, the honest tradeoffs.

Method

How we measure

The frozen protocol, the failure ledger, and the measured ceiling, the full dossier.

Research

The research behind Papez

The certified result, the multiplicative retention score, the preprint, and the open harness.

Product

The open-source engine

A causal-graph memory you own and can inspect, connected once, read everywhere.