SuperLocalMemory Logo — Local AI Memory Layer
SuperLocalMemory
ARCHITECTURAL COMPARISON

SuperLocalMemory
vs Zep

A technical comparison of two knowledge-graph approaches to AI agent memory: local-first with mathematical foundations vs cloud-hosted temporal knowledge graph.

Factual analysis — not marketing. Both systems solve real problems for different use cases.

At a Glance

SuperLocalMemory V3

  • —Local knowledge graph — on-device entity extraction
  • —74.8% LoCoMo (zero cloud) / 87.7% full power
  • —6-channel hybrid retrieval including query completion
  • —EU AI Act compliant by architecture (Mode A)
  • —AGPL v3 — fully open source
  • —No server infrastructure required

Zep

  • —Cloud knowledge graph — temporal entity relationships
  • —~85% LoCoMo (cloud LLM required)
  • —Graph-based temporal retrieval with LLM entity extraction
  • —Requires DPA for GDPR/EU AI Act compliance
  • —Community Edition (self-host) + Cloud (managed)
  • —Team collaboration + cross-device memory

Detailed Comparison

Dimension-by-dimension technical analysis of both systems.

Dimension SuperLocalMemory V3 Zep
Data Locality On-device (Mode A/B) or local storage + cloud synthesis (Mode C) Cloud-hosted (Zep servers) or self-hosted (Community)
Knowledge Graph Local entity graph — NLP extraction, no cloud calls (Mode A) Temporal knowledge graph — LLM-extracted entities and relationships
LoCoMo Score 74.8% (Mode A) / 87.7% (Mode C) ~85% (cloud required)
Installation npm install -g superlocalmemory — single command, no server Cloud signup or server deployment (Community Edition)
Offline Mode Full offline capability (Mode A/B) None — cloud connection required
EU AI Act (Mode A) Compliance by architecture — zero data transit Requires DPA and cloud compliance measures
Retrieval Channels 6-channel hybrid retrieval including query completion Temporal knowledge graph + semantic similarity
Team Collaboration Single-device by default Native team and organization-level memory
License AGPL v3 — open source Apache 2.0 (Community) / Commercial (Cloud)

Benchmark Context

Results on the LoCoMo benchmark (Long Conversation Memory). Field context shown.

System Score Cloud Required
SLM V3 Mode C 87.7% Yes (synthesis only)
Zep ~85% Yes
SLM V3 Mode A Retrieval 74.8% No
SLM V3 Mode A Raw (zero-LLM) 60.4% No

SLM V3 results from: arXiv:2603.14588. Zep score from published reports.

When Each Approach Fits

SuperLocalMemory fits when:

  • —Data sovereignty or EU AI Act compliance is required
  • —Offline or air-gapped operation is needed
  • —Individual developer workflow — personal assistant memory
  • —No cloud vendor relationship is acceptable
  • —Fully auditable, explainable retrieval is required

Zep fits when:

  • —Team-level shared memory with temporal relationships is needed
  • —Cross-device persistence is required natively
  • —Complex temporal knowledge graphs across users
  • —Cloud deployment fits existing organizational patterns
  • —Managed infrastructure reduces operational overhead

Frequently Asked Questions

What is the main architectural difference between SuperLocalMemory and Zep?

+
Zep is a cloud-hosted temporal knowledge graph memory service — conversation history, facts, and relationships are stored and processed on Zep's cloud infrastructure. SuperLocalMemory operates entirely on-device in Mode A, building a local knowledge graph without external connectivity. Both systems use graph-based approaches, but with fundamentally different data locality guarantees.

How do the benchmark scores compare on LoCoMo?

+
Zep scores approximately 85% on LoCoMo. SuperLocalMemory V3 scores 74.8% in Mode A (zero cloud) and 87.7% in Mode C (cloud LLM at every layer). SLM Mode C slightly outperforms Zep on LoCoMo in our evaluation. The key distinction: SLM Mode A achieves 74.8% without any cloud dependency.

Does Zep have a self-hosted option?

+
Zep offers a Community Edition for self-hosting, which requires infrastructure management. SuperLocalMemory is always local by design — no server to run, no infrastructure to maintain. Installation is a single npm or pip command.

How do the knowledge graph implementations differ?

+
Zep builds a temporal knowledge graph with entity and relationship extraction via cloud LLMs. SuperLocalMemory builds a local entity graph using on-device NLP (no cloud calls in Mode A/B), connected via the 6-channel retrieval pipeline including entity graph and query completion channels. The local approach trades team collaboration for data sovereignty.

Which is better for enterprise compliance?

+
SuperLocalMemory Mode A provides compliance-by-architecture — data never leaves the device, satisfying EU AI Act and GDPR data sovereignty requirements without DPA or additional infrastructure. Zep requires standard cloud compliance measures (DPA, data processing agreements with Zep) when processing personal data.

Explore the Research

SuperLocalMemory's mathematical techniques are open source and designed to benefit any memory architecture.

Research Papers Full Landscape Documentation