Supermemory vs Zep & Graphiti

Supermemory vs Zep & Graphiti: side-by-side comparison of two dedicated memory layers systems — architecture, taxonomy, license, pricing, MCP/A2A support, and direct edges.

Supermemory · Zep & Graphiti

Recommend between these two →

Cost & capability

SupermemoryZep & Graphiti
Capability bandcompetentcompetent
Capability composite7268
Cost tierfreefree
$/Mtok input00
$/Mtok output00
Use casesLong Running Session, Memory Augmented Chat, Code Generation FocusedLong Running Session, Memory Augmented Chat, Multi Agent Coordination, Analytical Summarization

Where they differ (16)

Rows where both sides have data and the values disagree — the shortlist of dimensions that actually distinguish these two systems.

SupermemoryZep & Graphiti
Capability composite7268
Use casesLong Running Session, Memory Augmented Chat, Code Generation FocusedLong Running Session, Memory Augmented Chat, Multi Agent Coordination, Analytical Summarization
TypeMemory graph + extraction + RAGBi-temporal knowledge graph
Created2024-022024-08
Latest releaseno releasesv0.29.0 2026-04-27
LicenseMITApache-2.0
GitHub22.4k★ +113/mo TypeScript25.7k★ +137/mo Python
PricingOSSFree + paid
Funding$6M total Seed · 2025-10$3M total Seed (additional) · 2024-04
Backend storagecustomPostgres + Neo4j (Graphiti)
DeploymentBothManaged-only
API surfaceREST, SDK: Python, JS/TSREST, SDK: Python, JS/TS, Go
Multi-tenancyEnterprise: deploy in customer VPC for full tenant isolation; standard SaaS uses logical namespace per user/orgLogical namespace per user/session; AWS VPC self-hosted option for full data residency; HIPAA BAA on Enterprise plan
MCPnative (first-party) — claude-supermemory plugin + MCPnative (first-party) — Graphiti MCP server
Optimised formulti-channel capture (API, app, browser ext, MCP) + RAG over personal graphmemory operation tracing + drift / poisoning detection
Anti-fitno anti-fit explicitly statednot for use cases that don't run agent workloads in production

At a glance

SupermemoryZep & Graphiti
SectionDedicated memory layers Dedicated memory layers
TierT1 T1
TypeMemory graph + extraction + RAG Bi-temporal knowledge graph
Created2024-02 2024-08
Latest releaseno releases v0.29.0 2026-04-27
LicenseMIT Apache-2.0
GitHub22.4k★ +113/mo TypeScript 25.7k★ +137/mo Python
PricingOSS Free + paid
Funding$6M total Seed · 2025-10 $3M total Seed (additional) · 2024-04
Backend storagecustom Postgres + Neo4j (Graphiti)
DeploymentBoth Managed-only
API surfaceREST, SDK: Python, JS/TS REST, SDK: Python, JS/TS, Go
Embeddingmultiple supported multiple supported
Multi-tenancyEnterprise: deploy in customer VPC for full tenant isolation; standard SaaS uses logical namespace per user/org Logical namespace per user/session; AWS VPC self-hosted option for full data residency; HIPAA BAA on Enterprise plan
MCPnative (first-party) — claude-supermemory plugin + MCP native (first-party) — Graphiti MCP server
A2Anot documented publicly not documented publicly
OpenTelemetrynot documented publicly not documented publicly
Optimised formulti-channel capture (API, app, browser ext, MCP) + RAG over personal graph memory operation tracing + drift / poisoning detection
Anti-fitno anti-fit explicitly stated not for use cases that don't run agent workloads in production

Taxonomy

AxisSupermemoryZep & Graphiti
storagegraphgraph
retrievalsimilaritygraph-traversal
persistencelong-termlong-term
updateextractionappend-only
unitdocumentepisode
governanceopaqueauditable
conflictllm-arbitratebi-temporal

Pros & cons

Supermemory

Pros: Simple universal API surface — wraps memory in a single SDK call without forcing extraction/retrieval choices; YC-backed.

Cons: Architectural opacity is the cost of simplicity — limited control over structure or eviction; small team.

Zep & Graphiti

Pros: Bi-temporal graph captures event time + ingestion time, making contradiction tracking and chronological reasoning correct by construction.

Cons: KG storage is heavier than vector for the same data volume; smaller funding base than Mem0 ($2.3M vs $24M).

Rows last verified 2026-05-14 / 2026-05-14. Data is CC-BY-4.0 — see how to read this.