Technology
LatticeDB released as single-file embedded graph database
Written in Zig, the open-source database combines graph traversal, vector similarity search, and full-text search in a single file.
The short version
- LatticeDB has been introduced as an embedded, single-file property-graph database designed for single-machine, relationship-heavy workloads.
- The database integrates graph traversal, HNSW vector similarity search, and BM25 full-text indexing into a unified query layer and query language.
- Targeted at local application development, agent memory, and retrieval-augmented generation pipelines, it uses an embedded single-writer model.
Key facts
- LatticeDB is an embedded single-file property-graph database written in Zig with zero external dependencies.[Hacker News]
- The engine unifies graph traversal, HNSW vector search, and BM25 full-text indexing under a single query layer.[Hacker News]
- It operates on a single-writer model with write-ahead log (WAL) durability and single-process ownership.[Hacker News]
- LatticeDB provides native C API bindings alongside support for Python, TypeScript, Go, and a CLI.[Hacker News]
- Developer benchmark testing on an Apple M1 recorded single-threaded node lookups at 0.13 microseconds and vector searches at 0.83 milliseconds across 1 million vectors.[Hacker News]
What remains uncertain
- The reported benchmark figures (0.13 μs node lookups and 0.83 ms vector searches) are based on single-threaded developer tests on an Apple M1 and have not been independently verified.[Hacker News]