Overview

Zep

Rice vs Zep comparison.

Zep provides long-term memory for AI using temporal knowledge graphs.


Comparison

FeatureRiceZep
PhilosophyCognitive architecture mimicking human brain regionsKnowledge graph with structured data retrieval
Working MemoryDynamic decay and attention scoringSession buffer with summarization/windowing
Procedural MemoryServer-side WASM executionNone (relies on client-side tools)
Data StructureHDC with graph equivalentsTemporal knowledge graph with nodes and edges
RetrievalSeparate APIs for past (reminisce) vs current (drift)Hybrid semantic + graph + keyword search

Key Differences

Memory Model

Rice uses a four-component cognitive architecture (Working, Episodic, Procedural, Semantic). Zep uses a temporal knowledge graph that explicitly models nodes and edges with bi-temporal tracking.

Working Memory

Rice implements dynamic decay and attention scoring. Items naturally fade if not accessed, and drift() returns items sorted by relevance. Zep uses session buffers with summarization and windowing.

Execution

Rice includes Procedural Memory for server-side WASM skill execution. Zep relies on client-side tool execution via agent frameworks like LangChain.

Retrieval

Rice separates "remembering the past" (reminisce) from "orienting current context" (drift). Zep combines semantic vector search with graph traversal and keyword search.