Switch from AGE to falkorDB or ladybugDB

[antonymarion]

Feature Title

choice of the DB for the graph

Describe the feature

Hello Raphael and thanks for this interesting repo!

The choice has been made of AGE, I believe to be fully PostgreSQL native and compliant.

But why don't you use Falkor or Ladybug that both are much more performant for cypher queries with dense graph (number of node, degree) and deep graph traversal (n hops); since AGE use pure SQL behind the hood with join.. so clearly not suited in case we want to traverse deep graphs.

By the way ladybug is embedded using wasm and transformer.js, so it is even better if edgequake want to be offline for instance and browser native.

you can look at what https://github.com/abhigyanpatwari/GitNexus did for instance.

cheers,
Antony

Motivation

No response

Alternatives considered

No response

Additional context

No response

[raphaelmansuy]

Thanks. Very good suggestion, I will add this on the roadmap.

[lvca]

Dropping in with one more candidate for the roadmap, with the appropriate conflict disclosure: I maintain ArcadeDB, so treat this as a data point rather than an objective comparison.

antonymarion's critique of AGE is accurate — Cypher over SQL joins collapses on deep traversal. The three alternatives have different shapes worth naming:

• LadybugDB — unique because it embeds in WASM with transformer.js, so it is the only option if edgequake ever wants to run entirely in the browser. Production maturity is still early.
• FalkorDB — fast, drop-in for Redis users, but BSL-licensed. Relicensing concerns bite anyone wanting to ship a pure Apache/MIT stack.
• ArcadeDB — Apache 2.0, single jar, multi-model (graph + document + vector + time series). Native Cypher (97.8% TCK), plus Gremlin and SQL against the same data. Multi-hop MATCH path = (a)-[*1..n]-(b) is a first-class path operator, not compiled to joins.

The Rust side talks to ArcadeDB over the Bolt protocol with any Neo4j-compatible driver (the neo4rs crate works out of the box), so the integration cost is a thin adapter, not a new driver.

Rough rule of thumb I use when sketching these decisions:

• browser-offline mandatory → LadybugDB
• Redis-native shop, license non-issue → FalkorDB
• standalone process, open license, multi-model room to grow → ArcadeDB

Happy to write a quick edgequake adapter PR against whichever candidate you pick, if it helps the roadmap call. Also happy to host a comparison benchmark on a representative edgequake workload if @antonymarion has a shape in mind.

[antonymarion]

Yep @lvca in my case I have some browser-offline features in mind so Ladybug is my best candidate.

For the workload I am not sure of what you are expecting from me if you can precise :)

[worldwidehigh]

Hey everyone,

If low RAM usage in our Docker container is a key constraint (e.g., to keep hosting costs down or run on resource-constrained servers), the database rankings shift dramatically.

Here is how the candidates compare regarding memory footprint:

• LadybugDB (🥇 Best for low RAM): Since it is disk-backed (on-disk columnar) and runs in-process (written in C++), it only loads what is necessary into memory. It has no JVM overhead and no separate server process, keeping the baseline RAM footprint extremely low (just a few megabytes).
• ArcadeDB (🥈 Medium RAM): It stores data on disk, which is good. However, because it is JVM-based (Java), it has a higher baseline memory footprint. You typically need to allocate at least 512MB–1GB of RAM just for the Java Virtual Machine to run stably.
• FalkorDB (🥉 High RAM): As a pure in-memory database, the entire graph must fit into RAM. As the dataset grows, memory usage grows linearly, making it unsuitable for low-RAM environments.

Updated Ranking for Low-RAM Docker Setup:

1. LadybugDB (Highly efficient, C++ native, disk-backed, very low baseline RAM)
2. ArcadeDB (Disk-backed but burdened by JVM memory overhead)
3. FalkorDB (Demanding due to its in-memory architecture)

If we need to be strictly mindful of memory consumption, LadybugDB is definitely the most resource-friendly choice for our container.

[lvca]

About performance, don't get scared about the JVM! We measured ArcadeDB vs FalkorDB vs LadybugDB with the standard LDBC benchmark (for graph databases) and these are the results: https://arcadedb.com/blog/graph-olap-engine-the-fastest-graph-analytics-with-zero-compromises/ by turning on the OLAP.