Graphiti is a framework for building and querying temporally-aware knowledge graphs, designed for AI agents operating in dynamic environments. It offers a real-time, incremental approach to knowledge graph construction, enabling AI agents to maintain context and adapt to changing information without full graph recomputation. The target audience includes AI developers and researchers building sophisticated agents that require robust memory and state management.

How It Works

Graphiti employs a bi-temporal data model to track both the occurrence and ingestion times of data, facilitating precise point-in-time queries. It utilizes an efficient hybrid retrieval system that combines semantic embeddings, keyword (BM25), and graph traversal methods. This approach allows for low-latency querying without relying on LLM summarization, and supports custom entity definitions via Pydantic models for flexible ontology creation.

Quick Start & Requirements

• Installation: pip install graphiti-core (or poetry add graphiti-core). Optional LLM provider extras available (e.g., graphiti-core[anthropic]).
• Prerequisites: Python 3.10+, Neo4j 5.26+, OpenAI API key (or Gemini/Anthropic/Groq with extras). Azure OpenAI support is also detailed.
• Setup: Requires setting up a Neo4j instance and configuring API keys.
• Documentation: Guides and API documentation.

Highlighted Details

• Real-time incremental updates for dynamic data.
• Bi-temporal data model for accurate historical queries.
• Hybrid retrieval (semantic, keyword, graph traversal) for low latency.
• Customizable entity definitions using Pydantic models.

Maintenance & Community

Graphiti is under active development. Community support is available via the Zep Discord server in the #Graphiti channel.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README text. Compatibility for commercial use or closed-source linking would require clarification on licensing.

Limitations & Caveats

The project is under active development, with ongoing work on expanding test coverage and enhancing retrieval capabilities. The README does not specify a license, which may impact commercial adoption.