Comparison

Feature Comparison

Capability	Mem0	Zep	Letta	Cognee	Mnemo Core	Mnemo Cloud
Vector search	Yes	Yes	Yes	Yes	Yes	Yes
BM25 keyword search	No	Yes	No	Yes	Yes	Yes
Knowledge graph	Yes	Yes	No	Yes	Yes	Yes
Forgetting model	No	Basic	Agent-managed	No	Weibull	Weibull
Memory tiers	No	No	No	No	Yes	Yes
Cross-encoder rerank	Yes	Yes	No	No	Yes	Yes
Contradiction detection	Graph layer	Temporal versioning	No	Partial	LLM 3-layer	LLM 3-layer
Triple-path fusion	No	No	No	No	Yes	Yes
Scope isolation	Basic	No	No	No	Yes	Yes
Self-hosted	Yes	CE deprecated	Docker	Yes	Yes	Yes
$0 local (Ollama)	Partial	No	Partial	No	Yes	Yes
TypeScript SDK	Yes	Yes	Yes	No	Yes	Yes
Python SDK	Yes	Yes	Yes	Yes	Yes	Yes

Note: Features and pricing may change. Check each project's official site for current information. Last verified: April 2026.

Architecture Differences

Each framework takes a different approach:

• Mem0 — Cloud-first SaaS with open-source option. Simple API, large community. Graph memory with conflict detection.
• Zep — Built around Graphiti temporal knowledge graph. BM25 hybrid search. Community Edition deprecated, cloud-first.
• Letta — Full agent platform where agents manage their own memory (virtual memory paradigm). Apache 2.0.
• Cognee — Python-first knowledge engine with auto-generated ontologies and pipeline architecture.
• Mnemo — Cognitive science-first. Weibull decay, triple-path retrieval, lightweight and embeddable. MIT licensed.

When to Choose Mnemo

• You want intelligent forgetting, not infinite storage
• You need a lightweight, embeddable memory layer (not a full agent platform)
• You want to run locally for free with Ollama
• You care about retrieval quality (triple-path + rerank + decay)
• You're building in TypeScript/Node.js

Benchmark

See LOCOMO Benchmark for retrieval accuracy comparison tested under identical conditions.