Feature Comparison
A transparent, feature-by-feature comparison against the leading graph databases. We show what we have, what they have, and where gaps exist.
| Feature | xrayGraphDB | Neo4j | Memgraph | KuzuDB | GalaxyDB |
|---|---|---|---|---|---|
| Cypher support | Full + Neo4j compat | Native | Partial | Partial | Partial |
| GFQL dataframe query | Yes | — | — | — | — |
| Per-tenant encryption | Yes | — | — | — | — |
| Vectorized execution | Yes | — | — | TBD | — |
| Native vector embeddings | Yes (EMBED) | Plugin | — | — | — |
| Built-in code intelligence | 9 procedures | — | — | — | — |
| Columnar wire protocol | xrayProtocol | — | — | — | — |
| Bolt compatibility | v5 | v5 | v1 | — | — |
| Plan cache | Yes (425x) | Yes | — | — | — |
| Lock-free adjacency | Yes | — | — | — | — |
| HA clustering | Yes | Yes | Yes | — | — |
| RBAC + audit | Yes | Yes | Partial | — | — |
| Neo4j syntax compat | CREATE INDEX, SHOW PROCS | Native | Partial | — | — |
| Free tier | Full engine | Limited | Source | Free | Free |
Performance Numbers
Internal benchmarks on commodity hardware. Phase 2 external benchmarks are in progress.
(plan cache hit)
(pipelined)
(vs Bolt row-based)
(bulk create)
Phase 2 measured benchmarks coming — these are from our internal testing on a single 8-core machine with 32GB RAM.
Where the Speed Comes From
DataChunk Pipeline
1,024 tuples per batch. Column-oriented processing maximizes CPU cache utilization and enables SIMD vectorization.
xrayProtocol
Columnar wire format with LZ4 compression. Results stream column-by-column instead of row-by-row. 24x faster than Bolt.
Plan Cache
AST fingerprinting with 425x speedup. Parameterized queries hit cache immediately. Auto-invalidation on schema changes.
PMR Arenas
Per-query memory allocation with zero GC pauses. Deterministic cleanup. No fragmentation, no leaks, no stop-the-world.