Slater: A Low-Memory GraphDB for Graphs That Don't Fit in RAM
Slater brings disk-native, low-memory graph storage with full neo4j Bolt compatibility, live writes, and built-in vector search.
Slater is a disk-native graph database that speaks standard Bolt, making it a drop-in replacement compatible with existing neo4j drivers. Instead of holding the entire graph resident in memory, it pages data from an on-disk image on demand — letting it serve graphs with hundreds of millions of nodes and billions of edges using only a few hundred MB of RAM, decoupling graph size from memory cost.
The architecture separates an immutable, offline-built core from a log-structured-merge (LSM) writable layer: writes accumulate in a write-ahead log and delta segments, then periodically fold back into a fresh core. This keeps read performance unaffected by writes, while business-key MERGE/SET/DELETE operations are group-committed and fsync-durable. A disk-native vector search (KNN) engine sits alongside the graph itself, enabling both graph traversal and embedding retrieval from a single system — useful for RAG and recommendation pipelines.
For engineers, the appeal lies in production-ready details: multi-tenant isolation, at-rest encryption, TLS-secured Bolt connections, a Rust codebase built with forbid(unsafe_code), and a minimal ~22MB Docker image. These combine to make Slater a scalable yet operationally lightweight alternative to memory-bound graph databases like neo4j or Memgraph.