Slater是一款专为读密集型图设计的低内存图数据库[1]。该系统通过磁盘分页技术，能够在数百MB RAM的约束下服务数亿节点和数十亿边的图数据[1]。现有版本v0.24.1已能在数百MB内存中处理包含9000万节点和15亿边的Wikidata图数据[1]。
Slater采用离线构建与在线服务分离的架构[1]。其设计支持标准Bolt协议，与任何Neo4j驱动直接兼容[1]。系统集成了磁盘原生向量搜索功能，支持余弦、L2和点积等距离度量方式，向量数阈值为50,000[1]。
在存储灵活性方面，Slater支持文件系统、Amazon S3和Google Cloud Storage三种后端存储[1]。系统还提供可选的本地SSD二级缓存层，用于加速对象存储的访问性能[1]。
关于数据写入，Slater集成了可选的LSM可写层，支持MERGE、SET和DELETE操作以及业务键写入，同时保持读路径性能不受影响[1]。安全方面采用基于argon2id的ACL认证机制，并支持至rest加密（使用XChaCha20-Poly1305）[1]。
该项目采用Apache License 2.0开源许可[1]。
Slater is a graph database engineered to minimize memory consumption while handling large-scale graph data [1]. The system can serve graphs containing 90 million nodes and 1.5 billion edges within just hundreds of megabytes of RAM, as demonstrated by its support for the Wikidata graph [1].
The database implements disk-based paging technology to achieve this efficiency [1]. It adheres to the standard Bolt protocol, enabling compatibility with any Neo4j driver without modification [1].
Slater employs a separated architecture consisting of an offline build phase (slater-build) and an online serving phase (slater) [1]. For write operations, the system offers an optional LSM-based writable layer that supports MERGE, SET, and DELETE operations along with business key writes, all while maintaining read path performance [1]. The current version is v0.24.1 [1].
The platform integrates native disk-based vector search capabilities, supporting cosine, L2, and dot product distance metrics, with support for up to 50,000 vectors [1]. Storage flexibility is provided through three backend options: filesystem, Amazon S3, and Google Cloud Storage [1]. An optional local SSD secondary cache layer can accelerate access to object storage [1].
Security features include ACL authentication based on argon2id and optional at-rest encryption using XChaCha20-Poly1305 [1]. The project is distributed under the Apache License 2.0 [1].