Cognee vs Zep & Graphiti
Cognee vs Zep & Graphiti: side-by-side comparison of two dedicated memory layers systems — architecture, taxonomy, license, pricing, MCP/A2A support, and direct edges.
Cost & capability
| Cognee | Zep & Graphiti | |
|---|---|---|
| Capability band | competent | competent |
| Capability composite | 65 | 68 |
| Cost tier | free | free |
| $/Mtok input | 0 | 0 |
| $/Mtok output | 0 | 0 |
| Use cases | Long Running Session, Memory Augmented Chat, Analytical Summarization, Multi Agent Coordination | Long Running Session, Memory Augmented Chat, Multi Agent Coordination, Analytical Summarization |
Where they differ (14)
Rows where both sides have data and the values disagree — the shortlist of dimensions that actually distinguish these two systems.
| Cognee | Zep & Graphiti | |
|---|---|---|
| Capability composite | 65 | 68 |
| Use cases | Long Running Session, Memory Augmented Chat, Analytical Summarization, Multi Agent Coordination | Long Running Session, Memory Augmented Chat, Multi Agent Coordination, Analytical Summarization |
| Type | Knowledge graph + ECL pipeline | Bi-temporal knowledge graph |
| Created | 2023-08 | 2024-08 |
| Latest release | v1.0.8 2026-05-06 | v0.29.0 2026-04-27 |
| GitHub | 17.1k★ +152/mo Python | 25.7k★ +137/mo Python |
| Funding | $10M total Seed · 2026-02 | $3M total Seed (additional) · 2024-04 |
| Backend storage | hybrid (vector + graph) | Postgres + Neo4j (Graphiti) |
| Deployment | Both | Managed-only |
| API surface | REST, SDK: Python | REST, SDK: Python, JS/TS, Go |
| Multi-tenancy | Logical namespace per project/dataset; self-hosted OSS deployment | Logical namespace per user/session; AWS VPC self-hosted option for full data residency; HIPAA BAA on Enterprise plan |
| MCP | native (first-party) — cognee-mcp | native (first-party) — Graphiti MCP server |
| Optimised for | typed knowledge graph extraction (ECL pipeline) over RAG | memory operation tracing + drift / poisoning detection |
| Anti-fit | no anti-fit explicitly stated | not for use cases that don't run agent workloads in production |
At a glance
| Cognee | Zep & Graphiti | |
|---|---|---|
| Section | Dedicated memory layers | Dedicated memory layers |
| Tier | T1 | T1 |
| Type | Knowledge graph + ECL pipeline | Bi-temporal knowledge graph |
| Created | 2023-08 | 2024-08 |
| Latest release | v1.0.8 2026-05-06 | v0.29.0 2026-04-27 |
| License | Apache-2.0 | Apache-2.0 |
| GitHub | 17.1k★ +152/mo Python | 25.7k★ +137/mo Python |
| Pricing | Free + paid | Free + paid |
| Funding | $10M total Seed · 2026-02 | $3M total Seed (additional) · 2024-04 |
| Backend storage | hybrid (vector + graph) | Postgres + Neo4j (Graphiti) |
| Deployment | Both | Managed-only |
| API surface | REST, SDK: Python | REST, SDK: Python, JS/TS, Go |
| Embedding | multiple supported | multiple supported |
| Multi-tenancy | Logical namespace per project/dataset; self-hosted OSS deployment | Logical namespace per user/session; AWS VPC self-hosted option for full data residency; HIPAA BAA on Enterprise plan |
| MCP | native (first-party) — cognee-mcp | native (first-party) — Graphiti MCP server |
| A2A | not documented publicly | not documented publicly |
| OpenTelemetry | not documented publicly | not documented publicly |
| Optimised for | typed knowledge graph extraction (ECL pipeline) over RAG | memory operation tracing + drift / poisoning detection |
| Anti-fit | no anti-fit explicitly stated | not for use cases that don't run agent workloads in production |
Taxonomy
| Axis | Cognee | Zep & Graphiti |
|---|---|---|
| storage | graph | graph |
| retrieval | graph-traversal | graph-traversal |
| persistence | long-term | long-term |
| update | extraction | append-only |
| unit | fact | episode |
| governance | inspectable | auditable |
| conflict | llm-arbitrate | bi-temporal |
Pros & cons
Cognee
Pros: Pipeline-as-code ECL (extract/cognify/load) makes the memory build path inspectable and replayable; fully OSS.
Cons: Smaller community than Mem0/Letta; more end-to-end engineering required to deploy.
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).