Graphiti知识图谱库：如何为LLM应用构建动态时间感知图谱？

Hey Hacker News community, this is Paul, Preston, and Daniel from Zep. We're thrilled to introduce graphiti, an open-source library designed for constructing and querying dynamic, temporally aware knowledge graphs.

大家好，我们是 Zep 的 Paul、Preston 和 Daniel。我们非常高兴地向大家介绍 graphiti，这是一个用于构建和查询动态、具有时间感知指系统能够识别和处理与时间相关的信息，在知识图谱中表现为能够追踪实体关系随时间的变化。的知识图谱A structured knowledge base that represents entities and their relationships in a graph format.的开源库。

You can find the project here: https://git.new/graphiti. With graphiti, developers can model intricate, evolving relationships between entities across time. The library ingests both unstructured and structured data, and the resulting knowledge graph can be queried using a powerful fusion of temporal, full-text, semantic, and graph algorithm-based approaches.

项目地址：https://git.new/graphiti。借助 graphiti，开发者可以对实体之间随时间变化的复杂、演进的关系进行建模。该库可以处理非结构化和结构化数据，生成的知识图谱A structured knowledge base that represents entities and their relationships in a graph format.支持融合了时间、全文、语义和图算法等多种方法的强大查询。

Key Use Cases for LLM Applications

graphiti enables the creation of sophisticated LLM-powered applications. Here are two primary examples:

graphiti 支持创建复杂的由大语言模型驱动的应用程序。以下是两个主要示例：

Intelligent Assistants: Build assistants that learn from user interactions, seamlessly integrating personal knowledge with dynamic data from business systems like CRMs and billing platforms.
- 智能助手：构建能够从用户交互中学习的助手，将个人知识与企业系统（如客户关系管理和计费平台）中的动态数据无缝集成。
Autonomous Agents: Create agents that execute complex tasks independently, capable of reasoning with state changes from diverse, dynamic sources such as real-time traffic conditions or streaming voice transcriptions.
- 自主智能体：创建能够独立执行复杂任务的智能体，能够根据来自不同动态源（如实时交通状况或流式语音转录）的状态变化进行推理。

What Makes Graphiti Different?

graphiti is purpose-built for dynamic data and agentic applications, setting it apart from solutions like GraphRAG and other static graph libraries. Its core differentiators include:

graphiti 专为动态数据和智能体应用基于大型语言模型构建的自主或半自主系统，能够理解环境、做出决策并执行任务。而构建，这使其与 GraphRAG 等解决方案和其他静态图库区分开来。其核心差异化特性包括：

Smart Graph Updates

The library automatically evaluates new entities against the existing graph, intelligently revising both nodes and edges to reflect the latest context, ensuring the graph remains current and accurate.

该库会自动根据现有图谱评估新实体，智能地修订节点和边以反映最新上下文，确保图谱保持最新和准确。

Rich Edge Semantics

During graph construction, graphiti generates human-readable, semantic, and full-text searchable representations for edges. This enhances both the interpretability of the graph and the power of search queries.

在图谱构建过程中，graphiti 会为边生成人类可读、具有语义且支持全文搜索的表示。这既增强了图谱的可解释性，也提升了搜索查询的能力。

Temporal Awareness

It extracts and continuously updates time-based metadata for edges directly from input data. This foundational feature enables sophisticated reasoning over how relationships change and evolve.

它直接从输入数据中提取并持续更新基于时间的边元数据。这一基础特性支持对关系如何变化和演进进行复杂的推理。

Hybrid Search

graphiti offers a unified search interface that combines semantic search, BM25 (keyword-based) search, and graph-based search, with the ability to fuse results from these different modalities for comprehensive retrieval.

graphiti 提供了一个统一的搜索接口，融合了语义搜索、BM25（基于关键词）搜索和图搜索，并能够整合这些不同模式的搜索结果，实现全面的检索。

Performance & Schema Consistency

Fast: Delivers search results in under 100ms, with latency primarily dictated by the external embedding API call.
- 快速：在 100 毫秒内返回搜索结果，延迟主要取决于外部嵌入 API 的调用。
Schema Consistency: Maintains a coherent graph structure by reusing existing schema definitions, preventing the unnecessary proliferation of node and edge types and ensuring data integrity.
- 模式一致性在知识图谱构建和维护过程中，保持节点和边类型的定义和使用的一致性，防止模式混乱和数据冗余。：通过复用现有的模式定义来保持连贯的图谱结构，防止节点和边类型不必要的激增，确保数据完整性。

Our Motivation and Invitation

We originally built graphiti to power Zep Memory, our long-term memory layer for creating personalized and accurate LLM applications. However, we believe graphiti's potential extends far beyond memory systems. We have open-sourced it to support and foster innovation across a wider range of dynamic, agentic use cases.

我们最初构建 graphiti 是为了驱动 Zep Memory——我们用于创建个性化和准确的 LLM 应用程序的长期记忆层。然而，我们相信 graphiti 的潜力远不止于记忆系统。我们将其开源，是为了支持和促进更广泛的动态智能体用例的创新。

GitHub Repository: https://git.new/graphiti
- GitHub 仓库：https://git.new/graphiti
Documentation: https://help.getzep.com/graphiti
- 文档：https://help.getzep.com/graphiti

We would greatly appreciate your feedback, contributions, or simply hearing about the innovative projects you build with graphiti!

我们非常期待您的反馈、贡献，或者仅仅是听听您使用 graphiti 构建的创新项目！

– Paul, Preston, & Daniel

– Paul, Preston, & Daniel