@KuzuDB has straightforward integrations with Polars, NetworkX, Pandas, PyArrow and...Torch Geometric! Meaning that you can export data towards #GraphMachineLearning with one method call. If only they had a bolt driver for JavaScript (hint). https://kuzudb.com
Thanks for trying out Kùzu! The tag is Kùzu Inc., so am just including that here :) Out of curiosity, the use case for a JS bolt driver would be for high-quality frontend visualization, correct? Being an embedded database that runs in process, there is no need for a protocol like bolt in Kùzu's case, as there's no server and this no network transport of data when returning a query response. Do you see a scenario where an embedded architecture like Kùzu's can incorporate a binary output format to an external client for performance reasons? If so, it's likely going to look different from bolt.
Associate Professor at University of Waterloo, Co-founder & CEO of Kùzu Inc.
8moMoving the conversation from X: Glad to hear you like Kuzu! For embeddability: Many use cases of graphs are analytical, which requires doing some batch and offline querying & analytics, e.g., finding nodes with the largest degrees or ranking nodes based on some graph algorithms. Having an embeddable dbms simplifies such pipelines as you simply import them into your scripts without setting up any servers/clients. Embeddable DBMSs also have the advantage that you can build API servers around them to support client-server applications by building API servers around them. It would be much harder to take a server-client dbms and make it embeddable. So we hope to get the easy of use benefit of embeddable dbms's for analytics-oriented applications but also support client-server applications with some API server support we give. For bolt: We need to think about how to plug Kuzu into these software directly. We could consider supporting a Bolt protocol but if these software issue Neo4j's Cypher, it would not directly work with Kuzu as there are still differences in the query languages. Overall though think we need an API server of some kind to support applications that need a server-client architecture. That should be the first step.