@zensation/core

7-layer memory system for AI agents. Pluggable storage, embeddings, and LLM providers. Built on @zensation/algorithms.

latest

Source

npm

Version: 0.3.3

Version published: 5 days ago

Maintainers: 1

Created: 6 months ago

Source

@zensation/core

7-layer memory system for AI agents. Pluggable storage, embeddings, and LLM providers.

What's Inside

The orchestration layer that turns @zensation/algorithms into a complete memory system:

MemoryCoordinator — orchestrates all 7 layers: auto-routing store(), cross-layer recall(), consolidate(), decay(), FSRS review queue
7 Memory Layers — Working, Short-Term, Episodic, Semantic, Procedural, Core, Cross-Context
Sleep Consolidation — memory replay simulation (Stickgold & Walker, 2013)
Pluggable Adapters — bring your own storage (@zensation/adapter-postgres, @zensation/adapter-sqlite)

Quick Start

npm install @zensation/core

import { MemoryCoordinator } from '@zensation/core';

const memory = new MemoryCoordinator({ storage: adapter, embedding: embedder });

// Auto-routes to the right layer (semantic, episodic, procedural, or core)
await memory.store('User prefers TypeScript', { type: 'auto' });

// Cross-layer search with ranked, deduplicated results
const results = await memory.recall('programming preferences');

// Consolidate: promote episodic -> semantic, apply decay
await memory.consolidate();

// FSRS review queue across all layers
const dueItems = await memory.getReviewQueue();

Tree-Shakeable Imports

// Full coordinator
import { MemoryCoordinator } from '@zensation/core';

// Individual layers
import { WorkingMemory } from '@zensation/core/layers/working';
import { EpisodicMemory } from '@zensation/core/layers/episodic';
import { SemanticMemory } from '@zensation/core/layers/semantic';
import { ProceduralMemory } from '@zensation/core/layers/procedural';
import { CoreMemory } from '@zensation/core/layers/core';
import { CrossContextMemory } from '@zensation/core/layers/cross-context';

// Adapter interfaces
import type { StorageAdapter, EmbeddingProvider } from '@zensation/core/interfaces';

Research

These algorithms are documented in an open-access technical disclosure: ZenBrain: A Neuroscience-Inspired 7-Layer Memory Architecture (Zenodo). See also: HuggingFace Model Card.

Part of ZenBrain

This package is part of the ZenBrain monorepo — the neuroscience-inspired memory system for AI agents.

Package	Description
`@zensation/algorithms`	Pure algorithms (FSRS, Hebbian, emotional, Bayesian)
@zensation/core	Memory layers + coordinator (this package)
`@zensation/adapter-postgres`	PostgreSQL + pgvector storage
`@zensation/adapter-sqlite`	SQLite + sqlite-vec storage

License

Apache 2.0 — see LICENSE.

About ZenBrain

ZenBrain is a seven-layer, neuroscience-derived memory architecture for LLM agents, built as zero-dependency TypeScript and published under Apache-2.0. On LongMemEval-500 three of nine head-to-head answer-quality comparisons hold against Letta, Mem0 and A-Mem — all three against A-Mem, the remaining six are ties, none lost (three competitors x three LLM judges, Bonferroni-corrected, version-matched) — reaching 91.3% of a full-context oracle's binary-judge accuracy at 1/109.6 of the per-query token cost.