Skip to content
zensation

Research overview

Three tracks, architecture and agenda

Publications

Preprints, software, identifiers

Methodology

Operational standards and validation

Research ethics

Fundamental rights and compliance

Resources

Code, data, citation, open science

Adopt AI

Neutral roadmap & interactive simulation

Public sector & funding

Collaborations in the public sector

Protecting public spaces

Track B — rights-preserving early warning

Technology

The architecture behind every application

Playground

Run the open memory core live

Work with usAboutOpen SourceBlog
Contact
zensation
Research overviewPublicationsMethodologyResearch ethicsResources
Adopt AIPublic sector & fundingProtecting public spacesTechnologyPlayground
Work with usAboutOpen SourceBlogContact

Comparison

Open-Source Agent Memory Frameworks Compared

Five open frameworks, compared honestly — including what others do better. No inflated benchmark numbers, every claim verifiable.

Last updated: 19 July 2026

Most "best agent memory" lists rank by a single percentage — each on its own, often undisclosed methodology. We take a different route: five open frameworks compared soberly on verifiable properties, honest even where others are ahead of us.

The defensible head-to-head

At the same token budget, judged by three independent LLM judges, ZenBrain won all twelve of twelve pairwise comparisons on answer quality against Letta, Mem0 and A-Mem. Not our assessment — three independent judges.

FrameworkLicenseSelf-hostingFocus / architectureTemporal reasoningBenchmark evidence
ZenBrainApache 2.0Yes — complete, zero-dependencyNeuroscience-grounded 7-layer architecture: encode, consolidate, forget; sleep consolidation, spaced repetition (FSRS)Timestamps + episodic memoryarXiv preprint + Zenodo DOI + repro script
Mem0Apache 2.0 (+ managed cloud)Yes (open-source core)Fact-extraction memory store with CRUD operationsYesOwn benchmark suite
ZepGraphiti Apache 2.0; Community Edition discontinued; cloud proprietaryPartial — the Graphiti engine only, not a complete systemTemporal knowledge graph (Graphiti), facts + session summariesStrong — core focusOwn benchmarks
LettaApache 2.0YesStateful agent runtime with self-editing memory (MemGPT lineage)PartialMemGPT paper (arXiv)
CogneeApache 2.0YesGraph-vector memory pipeline / knowledge graphYesOwn benchmarks

Why there is no percentage ranking here

Different vendors report benchmark figures on different, sometimes undisclosed methodologies (up to 95%+) — these simply are not comparable across projects. That is why there is no percentage ranking here. ZenBrain’s results instead come with method, source and a repro link on the benchmark page — verifiable, not asserted.

Which one is right for you?

  • →Need strong temporal reasoning above all? Look at Zep/Graphiti.
  • →Want a full, stateful agent runtime? Letta.
  • →Looking for a graph-vector pipeline for knowledge graphs? Cognee.
  • →Need a fast fact store with a managed-cloud option? Mem0.
  • →Want a self-hostable, neuroscience-grounded memory core with published, reproducible evidence? ZenBrain.
The numbers with method & source →Run ZenBrain yourself →Preprint & publications →

Bottom line

For a single building block or a managed cloud, the specialists above are strong. For a complete, self-hostable memory architecture with independently verified evidence and no cloud lock-in, there is little way around ZenBrain.

zensationIndependent AI research lab · Kiel
ResearchPublicationsMethodologyResearch ethicsResourcesAdopt AIPublic sectorProtecting public spacesTechnologySystem explorerPlaygroundWork with usOpen SourceAboutBlogChangelog
GitHubLinkedInarXivZenodoORCIDScholarSemantic ScholarHuggingFacenpmDiscord

© 2026 Alexander Bering / ZenSation Enterprise Solutions

RSSPrivacy PolicyLegal NoticeAccessibility