Graphiti
Temporal knowledge graph framework for AI agents. Built by the Zep team, Graphiti ingests episodic data (conversations, facts, events) and builds a knowledge graph that preserves temporal relationships — when facts were learned, how they changed over time, and which supersede older facts. Designed for long-term agent memory where fact evolution and temporal reasoning matter.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
Open source (MIT) for full auditability. Data stored in your Neo4j instance — full data sovereignty. LLM API keys required for extraction. Self-hosted keeps all data under your control.
⚡ Reliability
Best When
You're building agents that need to reason about how information changes over time — fact supersession, temporal context, and evolving entity states.
Avoid When
Your agent needs simple session memory or static knowledge retrieval — Zep API or standard RAG are simpler and better supported.
Use Cases
- • Build agent memory systems that track how user preferences, facts, and context change over time with temporal graph structure
- • Enable agents to reason about fact temporality — 'what was the user's address last month vs now?' — with versioned knowledge graph nodes
- • Ingest agent conversation history and extract evolving entity relationships for persistent multi-session agent context
- • Create knowledge graphs from structured data streams where entities and relationships change frequently over time
- • Build research agents that track how information about topics evolves across ingested sources over time
Not For
- • Simple conversation history storage — Zep's higher-level memory API is simpler for just storing and retrieving conversation context
- • Static knowledge bases without temporal dynamics — standard vector RAG or Neo4j knowledge graphs are simpler for non-temporal use cases
- • Teams without Neo4j or graph database infrastructure — Graphiti requires Neo4j as the backing store
Interface
Authentication
Graphiti is a Python library — no auth required for the framework itself. Neo4j authentication (username/password or certificates) manages data store access. LLM API keys (OpenAI, etc.) required for entity extraction.
Pricing
Graphiti itself is MIT-licensed and free. Running costs are: Neo4j hosting (Community is free, Aura from $65/month), LLM API calls for entity extraction. Maintained by Zep team.
Agent Metadata
Known Gotchas
- ⚠ Graphiti is a Python library, not a service — no REST API; requires embedding in Python agent code
- ⚠ Requires Neo4j with APOC plugin installed — Neo4j setup is non-trivial, particularly for managed cloud databases
- ⚠ Entity extraction makes LLM calls per ingested episode — extraction costs can be significant for high-volume ingestion
- ⚠ Temporal graph queries require familiarity with Graphiti's query API — not standard Cypher/SQL; learning curve applies
- ⚠ Early-stage library with active development — API may change between minor versions; pin version carefully
- ⚠ Entity deduplication quality depends on LLM quality — similar entities may or may not be merged depending on LLM classification accuracy
Alternatives
Full Evaluation Report
Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for Graphiti.
AI-powered analysis · PDF + markdown · Delivered within 30 minutes
Package Brief
Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.
Delivered within 10 minutes
Score Monitoring
Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.
Continuous monitoring
Scores are editorial opinions as of 2026-03-07.