Draft:Graphiti (framework)

Graphiti is an open-source Python framework developed by Zep for building and querying temporally-aware knowledge graphs. It is designed to serve as a dynamic memory layer for artificial intelligence (AI) agents and applications using large language models (LLMs). Unlike traditional retrieval-augmented generation (RAG) systems that rely on static document retrieval, Graphiti enables agents to track how facts, entities, and relationships evolve over time.

History

Graphiti was developed by the team at Zep to address the "stale data" problem in LLM-based applications. It was first released as an open-source project on GitHub in late 2024, with its formal public launch and research debut in early 2025. The project emerged from a need for a memory system that could resolve contradictions, such as a user changing their preferences or a business project moving through different statuses, without losing historical context.

In January 2025, the developers published a research paper titled Zep: A Temporal Knowledge Graph Architecture for Agent Memory[1] , which detailed the framework's performance against existing state-of-the-art memory systems like MemGPT.

Architecture

Graphiti utilizes a three-tiered graph architecture to manage information:

  1. Episode Subgraph: Acts as a non-lossy audit trail. It stores raw data (e.g., chat logs, documents, or JSON) as discrete "episodes" with precise timestamps.
  2. Semantic Entity Subgraph: Extracted from episodes, this layer contains structured entities (people, concepts, objects) and their relationships. Relationships in this layer use **bi-temporal metadata**, tracking both when an event occurred and when it was recorded.
  3. Community Subgraph: This high-level layer groups related entities into clusters, facilitating thematic reasoning and broader contextual understanding.

Key features

Temporal intelligence

Graphiti employs a bi-temporal data model that assigns "valid at" and "invalid at" timestamps to facts. This allows the system to perform point-in-time queries, enabling an AI agent to answer questions like "What was the project status last Tuesday?" or "What did the user believe before they received the update?"

Hybrid retrieval

The framework achieves low-latency retrieval (typically sub-300ms) by combining multiple search methods:

  • Semantic search: Using vector embeddings to find similar concepts.
  • Keyword search: Utilizing BM25 for precise term matching.
  • Graph traversal: Navigating relationships between entities to find connected context.

Model Context Protocol (MCP)

Graphiti includes a dedicated server for the Model Context Protocol (MCP), an open standard for AI tools. This allows AI clients such as Anthropic's Claude Desktop or the Cursor IDE to use Graphiti as a persistent, shared memory backend across different sessions.

Performance

According to the Deep Memory Retrieval (DMR) benchmark, Graphiti-powered systems have demonstrated superior accuracy compared to baseline implementations. In evaluations such as LongMemEval, which tests long-term context maintenance, the framework achieved accuracy improvements of up to 18.5% while reducing response latency by 90% through the elimination of LLM-based summarization at query time.

Support and integrations

Graphiti supports several major graph database backends:

It also integrates with major LLM and embedding providers, including OpenAI, Anthropic, Google Gemini, and Azure OpenAI Service.

See also

  1. ^ https://arxiv.org/abs/2501.13956

Content Disclaimer

Informasi ini disarikan dari Wikipedia dan disajikan kembali untuk tujuan edukasi. Konten tersedia di bawah lisensi CC BY-SA 3.0. Kami tidak bertanggung jawab atas ketidakakuratan data yang bersumber dari kontribusi publik tersebut.

  1. The information displayed on this website is sourced in part or in whole from Wikipedia and has been adapted for the purpose of restating it. We strive to provide accurate and relevant information, however:
  2. There is no guarantee of absolute accuracy. Wikipedia is an open, collaborative project that can be edited by anyone, so information is subject to change.
  3. It is not intended to constitute professional advice. The content displayed is for informational and educational purposes only. For important decisions (e.g., medical, legal, or financial), please consult a professional.
  4. Content copyright. Wikipedia is licensed under the Creative Commons Attribution-ShareAlike License (CC BY-SA). This means that content may be reused with appropriate attribution and shared under a similar license.
  5. Responsible use. Any risk arising from the use of information from this website is entirely the responsibility of the user.