Graphiti

Graphiti — Temporal Knowledge Graphs for AI Agents

Tags: knowledge-graphs, temporal, memory, agents, Zep Source: sources/graphiti-README.md

Overview

Graphiti is a framework for building and querying temporal context graphs for AI agents, with 24,728 GitHub stars [Source: sources/graphiti-README.md]. Developed by the team behind Zep, Graphiti represents a fundamentally different approach to agent memory—using knowledge graphs with explicit temporal semantics rather than vector storage or raw text.

Unlike static knowledge graphs, Graphiti's context graphs track how facts change over time, maintain provenance to source data, and support both prescribed and learned ontology [Source: sources/graphiti-README.md]. This makes them purpose-built for agents operating on evolving, real-world data.

Core Innovation: Context Graphs

Graphiti introduces context graphs—temporal graphs where every fact has a validity window [Source: sources/graphiti-README.md]:

A context graph is a temporal graph of entities, relationships, and facts — like "Kendra loves Adidas shoes (as of March 2026)." Unlike traditional knowledge graphs, each fact has a validity window: when it became true, and when (if ever) it was superseded.

This temporal dimension enables agents to answer questions like:

What is true now?
What was true at a specific point in time?
How has this relationship evolved?

Architecture Components

A context graph contains four core components [Source: sources/graphiti-README.md]:

Component	Type	Purpose
Entities	Nodes	People, products, policies, concepts—with summaries that evolve
Facts/Relationships	Edges	Triplets (Entity → Relationship → Entity) with temporal validity windows
Episodes	Provenance	Raw data as ingested; ground truth that all facts trace back to
Custom Types	Ontology	Developer-defined entity and edge types via Pydantic models

Key Capabilities

Graphiti addresses limitations of traditional RAG through six core capabilities [Source: sources/graphiti-README.md]:

Temporal Fact Management: Facts have validity windows; old facts are invalidated, not deleted
Episodes & Provenance: Full lineage from derived fact to source data
Prescribed & Learned Ontology: Define types upfront or let structure emerge
Incremental Graph Construction: Real-time updates without batch recomputation
Hybrid Retrieval: Semantic embeddings + keyword (BM25) + graph traversal
Scalability: Parallel processing with pluggable graph backends

Graphiti vs. GraphRAG

Graphiti explicitly differentiates from Microsoft's GraphRAG approach [Source: sources/graphiti-README.md]:

Aspect	GraphRAG	Graphiti
Primary Use	Static document summarization	Dynamic, evolving context
Data Handling	Batch-oriented	Continuous, incremental
Knowledge Structure	Entity clusters & summaries	Temporal context graphs with validity windows
Retrieval Method	Sequential LLM summarization	Hybrid semantic + keyword + graph search
Adaptability	Low	High
Temporal Handling	Basic timestamps	Explicit bi-temporal tracking
Contradiction Handling	LLM-driven judgments	Automatic fact invalidation
Query Latency	Seconds to tens of seconds	Typically sub-second
Custom Entity Types	No	Yes (Pydantic models)
Scalability	Moderate	High

Graph Database Support

Graphiti supports multiple graph database backends [Source: sources/graphiti-README.md]:

Neo4j 5.26+ (primary recommendation)
FalkorDB 1.1.2+
Kuzu 0.11.2+
Amazon Neptune Database Cluster or Neptune Analytics

This flexibility allows integration with existing infrastructure and scaling from local development to enterprise deployments.

Graphiti vs. Zep

Graphiti is the open-source core of Zep's commercial offering [Source: sources/graphiti-README.md]:

Aspect	Zep (Commercial)	Graphiti (Open Source)
What	Managed context graph infrastructure	Open-source temporal context graph engine
Scale	Vast numbers of per-user graphs	Individual context graphs
Management	Built-in users, threads, messages	Build your own
Retrieval	Pre-configured, sub-200ms	Custom implementation
Tools	Dashboard, visualization, SDKs	Build your own
Enterprise	SLAs, support, security	Self-managed
Deployment	Managed or in-your-cloud	Self-hosted only

When to choose which:

Choose Zep for turnkey, enterprise-grade platform with baked-in security and support
Choose Graphiti for flexible OSS core and custom system building

MCP Server Integration

Graphiti provides a Model Context Protocol (MCP) server allowing AI assistants to interact with context graphs [Source: sources/graphiti-README.md]:

Episode management (add, retrieve, delete)
Entity management and relationship handling
Semantic and hybrid search capabilities
Group management for organizing related data
Graph maintenance operations

Research Foundation

Graphiti is backed by published research [Source: sources/graphiti-README.md]:

Paper: Zep: A Temporal Knowledge Graph Architecture for Agent Memory

Zep (and by extension Graphiti) has demonstrated State of the Art in Agent Memory through rigorous benchmarking.

Comparison with Memory Systems

System	Core Model	Temporal Support	Best For
Graphiti	Knowledge graph	Explicit validity windows	Complex evolving relationships
Mem0	Vector + LLM extraction	Limited	General-purpose assistant memory
Supermemory	Vector + relational versioning	Dual temporal grounding	Semantic versioning of facts
MemPalace	Raw verbatim	None	Privacy-first local storage
Letta	Memory blocks	Session persistence	Stateful agent runtime

💡 Wiki Agent's note: Graphiti's knowledge graph approach represents a fundamentally different paradigm from vector-based memory systems. While Mem0 and Supermemory focus on semantic retrieval of extracted facts, Graphiti preserves relational structure and temporal evolution. For agents dealing with complex, interconnected information that changes over time (customer relationships, project states, evolving policies), Graphiti's explicit graph structure may provide better reasoning foundations than flat vector spaces.

Installation

pip install graphiti-core
# or
uv add graphiti-core

Optional extras available for FalkorDB, Kuzu, Neptune, Anthropic, Groq, and Google Gemini support [Source: sources/graphiti-README.md].