Overview
Graphiti is an open-source framework for building and querying temporally-aware knowledge graphs, specifically designed for AI agents operating in dynamic environments. It powers the core memory system in Zep's context engineering platform.
Key Features
- Temporal Awareness: Tracks when facts were true with explicit validity intervals (t_valid, t_invalid)
- Bi-Temporal Model: Stores both event time (when something happened) and ingestion time (when it was recorded)
- Dynamic Updates: Continuously integrates user interactions and enterprise data without batch recomputation
- Hybrid Search: Combines semantic search, keyword matching, and graph traversal
- Real-Time Processing: Incrementally processes incoming data with instant entity and relationship updates
- Low Latency: Precise queries without LLM summarization overhead
Architecture
Every graph edge (relationship) in Graphiti includes:
- Explicit validity intervals for temporal reasoning
- Relationship metadata and context
- Source attribution for traceability
The framework maintains communities of related entities that evolve over time.
Integration
Model Context Protocol (MCP)
Graphiti includes an MCP server, enabling integration with:
- Claude (Anthropic)
- Cursor IDE
- Other MCP-compatible clients
Supported Backends
- Neo4j (primary graph database)
- OpenAI (embeddings and LLM)
- Azure OpenAI
- Google Gemini
- Anthropic Claude
Performance
In the Deep Memory Retrieval (DMR) benchmark, Graphiti-powered systems achieve 94.8% accuracy versus 93.4% for MemGPT, demonstrating superior performance in memory retrieval tasks.
Use Cases
- AI agent memory systems
- Conversational AI with historical context
- Enterprise knowledge management
- Temporal reasoning applications
- Dynamic relationship tracking
- Multi-source data integration
Pricing
Free and open-source under the Apache 2.0 license.
