Knowledge Graph Memory by Anthropics (Graphiti)

Source: Graphiti: Knowledge Graph Memory for a Post-RAG Agentic World

Overview

Knowledge Graph Memory, also known as Graphiti, is an MCP server designed to provide real-time, dynamic, and temporally-aware contextual memory for agentic AI systems. Built on Neo4j, it enables developers to track relationships between data entities, maintain continuity across sessions, and manage evolving data in complex projects.

Features

• Real-Time, Incremental Updates: Continuously ingests and integrates new data episodes (events/messages) without requiring batch recomputation of the entire knowledge graph.
• Temporally-Aware Knowledge Graph: Tracks both when events occur and when they are ingested, maintaining explicit validity intervals for each relationship (t_valid, t_invalid).
• Conflict Resolution with Temporal Metadata: Uses temporal information to update or invalidate outdated knowledge, preserving historical accuracy and enabling historical queries.
• Handles Multiple Data Types: Supports ingestion of chat histories, structured JSON, and unstructured text, allowing unified or multiple graphs within a single setup.
• Fast Query Retrieval: Achieves low-latency (P95 ~300ms) retrieval using hybrid search (semantic embeddings, BM25 keyword search, direct graph traversal) without requiring LLM calls at query time.
• Scalable Indexing: Utilizes vector and BM25 indexes for near-constant time access to nodes and edges, regardless of graph size.
• Custom Ontology and Entity Types: Automatically generates ontologies and supports the definition of custom, domain-specific entities using Pydantic models.
• Historical Query Capability: Enables users to reconstruct the state of knowledge at any point in time or analyze data evolution.
• Unified Context for Agents: Provides a continuously updated, unified memory layer for agentic applications, improving context management and continuity.

Use Cases

• Autonomous AI agents requiring dynamic, up-to-date memory
• Applications where data changes frequently
• Scenarios needing accurate historical state reconstruction
• Complex workflows and multi-session interactions

Pricing

No pricing information is provided in the available content.

Tags

• knowledge-graph
• memory
• context-management
• developer-tools

Category

ai-integration-mcp-servers

Note: For technical details and implementation, see the original blog post and related resources.