Product

Give your AI agent deep, structured memory that grows smarter over time.

Core Operations

Three APIs that handle everything — store, search, and understand.

Memory Architecture

Four memory types inspired by cognitive psychology, each serving a distinct purpose.

📋

Fact

Discrete, objective, verifiable information extracted from conversations. Stored with vector embeddings and knowledge graph links.

🕐

Episode

Time-bound events and experiences with emotion metadata (valence, arousal, emotion labels) and temporal expressions.

🧠

Trait

Behavioral patterns discovered through multi-session reflection. Evolves through a 6-stage lifecycle from trend to core.

📄

Document

Coming Soon

Static reference materials uploaded for RAG-style retrieval. Specialized search for uploaded PDFs and files.

Trait Lifecycle

Traits progress through 6 stages as evidence accumulates. Confidence scores decay when contradicted and strengthen when reinforced.

trend
candidate
emerging
established
core
dissolved

Three-Layer Trait Classification

Traits are classified into three layers based on confidence: Behavior (observable patterns) → Preference (dispositional tendencies) → Core (stable personality traits).

Behavior

0.3 – 0.6

Preference

0.6 – 0.85

Core

> 0.85

Benchmark

Evaluated on LoCoMo, a public long-conversation memory benchmark testing single-hop, multi-hop, temporal, and open-domain reasoning.

81.7%

LoCoMo Overall Score

82.9%

Single-Hop

84.3%

Multi-Hop

81.1%

Open-Domain

76.6%

Temporal

neuromem
81.7%
Framework A
75.8%
Framework B
75.1%
Framework C
68.4%
Framework D
66.9%

Compared with mainstream AI memory frameworks on public benchmarks

See It in Action

A preview of neuromem's memory management dashboard — available immediately after sign-up.