UniDB: the universal all-in-one database
I enjoyed working with this one tremendously and it comes very close to KuzuDB (we all are still missing). Uni DB is a vector store for recall, graphDB for relationships, text index for search, rules engine for logic and more. Each with its own consistency model, its own ETL, its own failure mode.
I can’t summarize in a little post all the goodies but let me capture a few.
It’s structured Memory atop a typed property graph, OpenCypher with 36+ built-in graph algorithms (PageRank, Louvain, Dijkstra, betweenness).
It has associative recall via hybrid retrieval (fusing vector) ANN search (with 8 index algorithms, including RaBitQ/scalar/product quantization). BM25 full-text, reconciled through reciprocal-rank or weighted fusion.
It has its own variation of datalog and this is the one I find innovative. Locy is Uni’s logic layer and extends Cypher with recursive rules and stratified negation. It’s datalog integrated into Cypher, ain’t that cool.
It has an “ASSUME … THEN” type of construct whihc lets you mutate the graph inside a rollback boundary and query the counterfactual before committing.
Forks! Git-like branches of graphs you can query, sub-forks and commit to the main graph. Wished I had this in previous projects, so easy.
It has explainable decisions (EXPLAIN queries) giving you the proof trace for why something was derived; ABDUCE runs it backwards: what would need to change for a conclusion to no longer hold?
Underneath it all, it persists straight to object storage (S3), local disk or both. It terribly fast (yes, Rust based).
Still early days for the project, but totally love it for prototyping.
Uni: https://rustic-ai.github.io/uni-db Dragonscale: https://blog.dragonscale.ai/