ZenBrain: A Neuroscience-Inspired 7-Layer Memory Architecture for Autonomous AI Systems

Head-to-head: On LongMemEval-500, ZenBrain wins all nine answer-quality comparisons against Letta, Mem0 and A-Mem (3 competitors x 3 LLM judges) under Bonferroni-corrected significance, and reaches 91.3% of a full-context oracle's binary-judge accuracy at 1/106 of the per-query token cost.

Built by: Alexander Bering — Division Manager in a mid-sized company and Principal Investigator. What is researched here has to survive a Monday morning, not just a review.

⭐ If this is useful, a star on GitHub helps other agent developers find it.

Overview

ZenBrain is a 7-layer neuroscience-inspired memory architecture for autonomous AI systems. It bridges the gap between biological memory principles and practical AI system design, evaluated on long-context recall, memory stability, and knowledge retrieval tasks across ten experiments and a 15-algorithm ablation study (results below).

Paper: ZenBrain v8 (Zenodo) | DOI: 10.5281/zenodo.19353663 (concept DOI — always resolves to the latest version)

Status: arXiv preprint (April 2026). Independent replications, counter-results, and reviewer feedback welcome — contact: research@zensation.ai

Key Results (9 Experiments)

Experiment	Metric	Result
Exp 1 — LoCoMo Retrieval	F1 score	+20.7% vs. Flat Store
Exp 2 — Layer Ablation	Storage efficiency	+47.4% vs. single-layer
Exp 3 — Retention Curves	Retention@30d	89.9% (vs. 0% pure Ebbinghaus)
Exp 4 — Sleep Consolidation	Memory stability	+37.0% vs. no-sleep baseline
Exp 5 — Hebbian Retrieval	Precision@5	0.955
Exp 6 — Bayesian Confidence	Confidence AUC	0.797 (+49.5% vs. 0.533)
Exp 8 — Moderate Ablation	15-algorithm suite	Cooperative redundancy (no single removal matters)
Exp 9 — Challenging Ablation	15-algorithm suite	7 of 15 algorithms individually significant
Exp 10 — Stress Ablation	15-algorithm suite	9 of 15 algorithms individually critical

The full 15-algorithm ablation reveals a cooperative survival network with a measurable gradient across three difficulty levels. Sleep consolidation acts as a 1.92x multiplier. 95 reproducible tests across 4 experiment suites, all with Mulberry32 seeded PRNG (10 seeds).

Architecture: 7 Memory Layers

Layer 1: Working Memory       — Active task focus (capacity-limited, 7±2 items)
Layer 2: Episodic Memory      — Concrete experiences with temporal context
Layer 3: Semantic Memory      — Abstracted facts and relationships
Layer 4: Procedural Memory    — Skills and how-to knowledge
Layer 5: Short-Term Memory    — Session context buffer
Layer 6: Long-Term Memory     — Persistent cross-session knowledge
Layer 7: Core Memory          — Pinned identity and values (Letta-pattern)

Key algorithms:

• Hebbian Learning (co-activation strengthening, decay, normalization)
• FSRS Spaced Repetition (optimal review scheduling)
• Bayesian Confidence Propagation (uncertainty quantification)
• Sleep Consolidation (Stickgold & Walker 2013 — memory replay simulation)
• Ebbinghaus Decay (forgetting curve modeling)
• Contextual Retrieval (Anthropic method: reduced retrieval failure rate by 67% with reranking, 49% without)

Installation

# Core algorithms (zero dependencies)
npm install @zensation/algorithms

# Memory layer orchestration
npm install @zensation/core

# PostgreSQL + pgvector adapter
npm install @zensation/adapter-postgres

# SQLite adapter (zero-config)
npm install @zensation/adapter-sqlite

Quick Start

import { MemoryCoordinator } from '@zensation/core';
import { PostgresAdapter } from '@zensation/adapter-postgres';

const memory = new MemoryCoordinator({
  adapter: new PostgresAdapter({ connectionString: process.env.DATABASE_URL }),
});

// Store a memory across all relevant layers
await memory.store({
  content: 'ZenBrain uses Hebbian learning for knowledge graph strengthening',
  type: 'semantic',
  importance: 0.9,
});

// Recall with confidence scores
const results = await memory.recall('Hebbian learning', { topK: 5 });
// Returns facts with 95% confidence intervals

Links

• Paper (v8, 9 Aug 2026): https://zenodo.org/records/21858218
• PDF: https://zenodo.org/records/21858218/files/zenbrain-v8.pdf
• GitHub: https://github.com/zensation-ai/zenbrain
• Website: https://zensation.ai/technologie
• Live demo (HF Space): https://huggingface.co/spaces/zensation-ai/zenbrain-playground — runs the real open library in your browser
• npm packages (7): @zensation/algorithms · @zensation/core · @zensation/adapter-postgres · @zensation/adapter-sqlite · @zensation/mcp · @zensation/ai-sdk · @zensation/cli
• Reproduction packages: Mechanism ablation, Tables 7-9 (Apache-2.0) · Measurement package: judged outputs, flag manifests, analysis scripts (CC BY 4.0)
• ORCID: https://orcid.org/0009-0001-1793-012X

Citation

@misc{bering2026zenbrain,
  title   = {ZenBrain: A Neuroscience-Inspired 7-Layer Memory Architecture for Autonomous AI Systems},
  author  = {Bering, Alexander},
  year    = {2026},
  doi     = {10.5281/zenodo.19353663},
  url     = {https://doi.org/10.5281/zenodo.19353663},
  note    = {Zenodo Preprint}
}

@article{bering2026zenbrainarxiv,
  title         = {ZenBrain: A Neuroscience-Inspired 7-Layer Memory Architecture for Autonomous AI Systems},
  author        = {Bering, Alexander},
  year          = {2026},
  eprint        = {2604.23878},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2604.23878}
}

License

Code (the @zensation/* npm packages and the GitHub repository) — Apache-2.0, see LICENSE.

Paper and preprint records (zenbrain-v8.pdf, the arXiv preprint, the Zenodo preprint records) — CC BY 4.0. Reproduction packages carry their own licence, stated on each record (CC BY 4.0 or Apache-2.0).