Every time I look at graph databases, I just cannot figure out what problem they're solving. Particularly in an LLM based world.
Don't get me wrong, graphs have interesting properties and there's something intriguing out these dynamic, open ended queries. But, what features/products/customer journeys are people building with a graph DB.
Every time I explore, I end up back at "yea, but a standard DB will do 90% of this as a 10% of the effort".
A standard DB ala Postgres will be a perfectly functional graph database unless you're doing very specialized network analysis queries, which is not what most of these "knowledge graph" databases are being used for. It's only querying and data modeling that's a bit fiddly (expressing the "graph" structure using SQL) and that's being improved by the new Property Graph Query (PGQ) in the latest SQL standards.
For starters, LLMs themselves are a graph database with probabilistic edge traversal.
Some apps want it to be deterministic.
I'm surprised this question comes up so often.
It's mainly from the vector embedding camp, who rightfully observe that vector + keyword search gets you to 70-80% on evals. What is all this hype about graphs for the last 20-30%?
You can judge for yourself what work has been done in the last 5 months. Many short videos here. New open source contributors who I didn't know before ramping up.
Does anyone have any experience with this DB? Or context about where it came from?
From the commit history it's obvious that this is an AI coded project. It was started a few months ago, 99% of commits are from 1 contributor, and that 1 contributor has some times committed 100,000 lines of code per week. (EDIT: 200,000 lines of code in the first week)
I'm not anti-LLM, but I've done enough AI coding to know that one person submitting 100,000 lines of code a week is not doing deep thought and review on the AI output. I also know from experience that letting AI code the majority of a complex project leads to something very fragile, overly complicated, and not well thought out. I've been burned enough times by investigating projects that turned out to be AI slop with polished landing pages. In some cases the claimed benchmarks were improperly run or just hallucinated by the AI.
So is anyone actually using this? Or is this someone's personal experiment in building a resume portfolio project by letting AI run against a problem for a few months?
That is a lot of code for what appears to be a vanilla graph database with a conventional architecture. The thing I would be cautious about is that graph database engines in particular are known for hiding many sharp edges without a lot of subtle and sophisticated design. It isn't obvious that the necessary level of attention to detail has been paid here.
I wasn't referring to the Pavlo bet but I would make the same one! Poor algorithm and architecture scalability is a serious bottleneck. I was part of a research program working on the fundamental computer science of high-scale graph databases ~15 years ago. Even back then we could show that the architectures you mention couldn't scale even in theory. Just about everyone has been re-hashing the same basic design for decades.
As I like to point out, for two decades DARPA has offered to pay many millions of dollars to anyone who can demonstrate a graph database that can handle a sparse trillion-edge graph. That data model easily fits on a single machine. No one has been able to claim the money.
Inexplicably, major advances in this area 15-20 years ago under the auspices of government programs never bled into the academic literature even though it materially improved the situation. (This case is the best example I've seen of obviously valuable advanced research that became lost for mundane reasons, which is pretty wild if you think about it.)
Almost all analytic graphs of general scope surpass 1T edges, see below. DARPA also has an unfilled objective for 1B edge real-time continuously updated operational graphs. These are smaller and the write throughput requirements are in line with non-graph analytical databases but graph databases struggle to meet that standard.
There are countless smaller graphs for narrow domains that may be <1B edges but many people have the ambition to stitch together these narrow graphs into a larger graph. When stitching graphs together, the number of edges is usually super-linear. A billion edges is kind of considered “Hello World” for system testing.
The Semantic Web companies in the 2000s had graphs that were 100B+ edges. They wanted to go much larger but hit hard scaling walls around that point. That scaling wall killed them.
Classic mapping data models are typically 10-100B edges. These could be much, much larger if they could process all the data available to them.
Of course, intelligence agencies had all kinds of graphs far beyond trillions of edges 20 years ago. People, places, things, events.
Any type of spatiotemporal entity graphs with large geographic scope are quadrillions of edges. It isn’t just a lot of inferred relationships between entities, the relationships evolve over time which also must be captured. These are probably the most commercially valuable type of graph. You could build hundreds of different graphs of this type with 1T+ edges in most regions, never mind doing it at scale. These are so large that we usually don’t store them. Subgraphs are generated on demand, which is computationally expensive.
These spatiotemporal entity graphs also have the largest write loads. Single sources generate tens of PB/day of new edges. There is a ton of industrial data that looks like this; it isn’t just people slinging structured data.
Graphs are everywhere but we furiously avoid them because the scalability of operations over anything but severely constrained graphs is so poor. Selection bias.
NSA in particular heavily funded foundational theoretical and applied computer science research into scaling graph computing for decades. They had all kinds of boring graphs where trillions of edges was their Tuesday. The US military also uses large graph databases in fairly boring applications that probably didn’t require a graph database.
> many millions of dollars to anyone who can demonstrate a graph database that can handle a sparse trillion-edge graph.
I wonder why no one has claimed it. It's possible to compress large graphs to 1 byte per edge via Graph reordering techniques. So a trillion scale graph becomes 1TB, which can fit into high end machines.
Obviously it won't handle high write rates and mutations well. But with Apache Arrow based compression, it's certainly possible to handle read-only and read-mostly graphs.
Also the single machine constraint feels artificial. For any columnar database written in the last 5 years, implementing object store support is tablestakes.
Achieving adequate performance at 1T edges in one aspect requires severe tradeoffs in other aspects, making every implementation impractical at that scale. You touched on a couple of the key issues when I was working in this domain.
There is no single machine constraint, just the observation that we routinely run non-graph databases at similar scale on single machines without issue. It doesn't scale on in-memory supercomputers either, so the hardware details are unrelated to the problem:
- Graph database with good query performance typically has terrible write performance. It doesn't matter how fast queries are if it takes too long to get data into the system. At this scale there can be no secondary indexing structures into the graph; you need a graph cutting algorithm efficient for both scalable writes and join recursion. This was solved.
- Graph workloads break cache replacement algorithms for well-understood theory reasons. Avoiding disk just removes one layer of broken caching among many but doesn't address the abstract purpose for which a cache exists. This is why in-memory systems still scale poorly. We've known how to solve this in theory since at least the 1980s. The caveat is it is surprisingly difficult to fully reduce to practice in software, especially at scale, so no one really has. This is a work in progress.
- Most implementations use global synchronization barriers when parallelizing algorithms such as BFS, which greatly increases resource consumption while throttling hardware scalability and performance. My contribution to research was actually in this area: I discovered a way to efficiently use error correction algorithms to elide the barriers. I think there is room to make this even better but I don't think anyone has worked on it since.
The pathological cache replacement behavior is the real killer here. It is what is left even if you don't care about write performance or parallelization.
I haven't worked in this area for many years but I do keep tabs on new graph databases to see if someone is exploiting that prior R&D, even if developed independently.
Yes a graph database will happily lead you down a n^3 (or worse!) path when trying to query for a single relation if you are not wise about your indexes, etc.
That sounds like a ”graph” DB which implements edges as separate tables, like building a graph in a standard SQL RDB.
If you wish to avoid that particular caveat, look for a graph DB which materializes edges within vertices/nodes. The obvious caveat there is that the edges are not normalized, which may or may not be an issue for your particulat application.
Agreed, there's been a literal explosion in the last 3 months of new graph databases coded from scratch, clearly largely LLM assisted. I'm having to keep track of the industry quite a bit to decide what to add support for on https://gdotv.com and frankly these days it's getting tedious.
Six figures a week is a giant red flag. That kind of commit log usually means codegen slop or bulk reformatting, and even if some of it works I wouldn't trust the design, test coverage, or long-term maintenance story enough to put that DB anywhere near prod.
Sounds about right for someone who ships fast and iterates. 54 days for a v0 that probably needs refactoring isn't that crazy if the dev has a real DB background. We've all seen open source projects drag on for 3 years without shipping anything, that's not necessarily better
shipping fast matters a lot less than shipping something you actually understand. 200k lines in a week means nobody knows what's in there, including the author. that's not a codebase, it's a liability
jandrewrogers' point about cache replacement pathology in graph traversals is the kind of thing you only learn after burning real hardware cycles. I hit similar issues using Neo4j for knowledge graph queries — once traversal depth exceeded 4-5 hops, latency became unpredictable regardless of data size.
The embeddable angle is interesting though. For small-to-medium graphs (< 10M edges) that live in a single process, these scalability concerns matter less, and the developer experience of "just add a dependency" beats running a separate database server.
Serious question: are there any actually good and useful graph databases that people would trust in production at reasonable scale and are available as a vendor or as open source? eg. not Meta's TAO
- Does it need index free adjacency?
- Does it need to implement compressed sparse rows?
- Does it need to implement ACID?
- Does translating Cypher to SQL count as a graph database?
That's a difficult question and I would like to avoid giving a direct answer (because I co-lead a nonprofit benchmarking graph databases) but even knowing what you need for a graph database can be a tricky decision. See my FOSDEM 2025 talk, where I tried to make sense of the field:
plenty of those - I've had to work with dozens of different graph databases integrating them on https://gdotv.com, save for maybe 1-2 exceptions in the list of supported databases on our website, they're all production ready and either backed by a vendor or open-source (or sometimes both, e.g. Apache AGE for Azure PostgreSQL).
There are some technologies that have been around for a long time but really flying under the radar, despite being used a lot in enterprise (e.g. JanusGraph).
When you actually need to run graph algorithms against your relational data, you export the subset of that data into something like Grafeo (embedded mode is a big plus here) and run your analysis.
That importing is expensive and prevents you from handling billion scale graphs.
It's possible to run cypher against duckdb (soon postgres as well via duckdb's postgres extension) without having to import anything. That's a game changer when everything is in the same process.
Speaking of embeddable, we just announced cypher syntax for gfql, so the first OSS CPU/GPU cypher query engine you can use on dataframes
Typically used with scaleout DBs like databricks & splunk for analytical apps: security/fraud/event/social data analysis pipelines, ML+AI embedding & enrichment pipelines, etc. We originally built it for the compute-tier gap here to help Graphistry users making embeddable interactive GPU graph viz apps and dashboards and not wanting to add an external graph DB phase into their interactive analytics flows.
We took a multilayer approach to the GPU & vectorization acceleration, including a more parallelism-friendly core algorithm. This makes fancy features pay-as-you-go vs dragging everything down as in most columnar engines that are appearing. Our vectorized core conforms to over half of TCK already, and we are working to add trickier bits on different layers now that flow is established.
The core GFQL engine has been in production for a year or two now with a lot of analyst teams around the world (NATO, banks, US gov, ...) because it is part of Graphistry. The open-source cypher support is us starting to make it easy for others to directly use as well, including LLMs :)
It's not clear that graph-bench in "Tested with the LDBC Social Network Benchmark via graph-bench" is a benchmark that you made. It seems more robust and reliable than "we built a db and a benchmark tool, and our benchmark tool says we're the best". Just a thing to be careful about. You should just state that it's your tool and you welcome feedback to help make it so that other projects being compared are compared in their best light. Something like that might help, I don't know though it's a hard problem.
There seem to be a lot of these, how does it compare to Helix DB for example? Also, why would you ever want to query a database with GraphQL, for which it was explicitly not made for that purpose?
I just spent an hour with Grafeo, trying to also get the associated library grafeo_langchain working with a local Ollama model. Mixed results. I really like the Python Kuzu graph database, still use it even though the developers no longer support it.
I added a super cheap and bad embedding database in a project that allows the agent to call a tool for searching all the content it's built, it seems to work pretty well! This way the agent doesn't need to call a bunch of list tools (which I was worried would introduce lost of data to the context), and can find things based on fuzzy search.
Because Rust is an excellent language that pushes you into the "pit of success", and consequently software written in Rust tends to be fast, robust and easy to deploy.
There's no big mystery. No conspiracy or organised evangelism. Rust is just really good.
Worth noting that “robust” and “correct” are orthogonal. Graph databases (well, any database) seem like an area where correctness particularly matters, and I doubt Rust gives any meaningful advantage there.
I wonder if people are using (or intend to use) vibe-coded projects like the one linked.
I mean - I understand, some people have fun looking at new tech no matter the source, but my question is is there a person who would be designated to pick a GraphQL in language and would ignore all the LLM flags and put it in production.
Don't get me wrong, graphs have interesting properties and there's something intriguing out these dynamic, open ended queries. But, what features/products/customer journeys are people building with a graph DB.
Every time I explore, I end up back at "yea, but a standard DB will do 90% of this as a 10% of the effort".
I know pglite, and while it's great someone made that, it's definitely not the same
Some apps want it to be deterministic.
I'm surprised this question comes up so often.
It's mainly from the vector embedding camp, who rightfully observe that vector + keyword search gets you to 70-80% on evals. What is all this hype about graphs for the last 20-30%?
Writing it in Rust gets visibility because of the popularity of the language on HN.
Here's why we are not doing it for LadybugDB.
Would love to explore a more gradual/incremental path.
Also focusing on just one query language: strongly typed cypher.
https://github.com/LadybugDB/ladybug/discussions/141
https://vldb.org/cidrdb/2023/kuzu-graph-database-management-...
You can judge for yourself what work has been done in the last 5 months. Many short videos here. New open source contributors who I didn't know before ramping up.
https://youtube.com/@ladybugdb
From the commit history it's obvious that this is an AI coded project. It was started a few months ago, 99% of commits are from 1 contributor, and that 1 contributor has some times committed 100,000 lines of code per week. (EDIT: 200,000 lines of code in the first week)
I'm not anti-LLM, but I've done enough AI coding to know that one person submitting 100,000 lines of code a week is not doing deep thought and review on the AI output. I also know from experience that letting AI code the majority of a complex project leads to something very fragile, overly complicated, and not well thought out. I've been burned enough times by investigating projects that turned out to be AI slop with polished landing pages. In some cases the claimed benchmarks were improperly run or just hallucinated by the AI.
So is anyone actually using this? Or is this someone's personal experiment in building a resume portfolio project by letting AI run against a problem for a few months?
https://news.ycombinator.com/item?id=29737326
Kuzu folks took some of these discussions and implemented them. SIP, ASP joins, factorized joins and WCOJ.
Internally it's structured very similar to DuckDB, except for the differences noted above.
DuckDB 1.5 implemented sideways information passing (SIP). And LadybugDB is bringing in support for DuckDB node tables.
So the idea that graph databases have shaky internals stems primarily from pre 2021 incumbents.
4 more years to go to 2030!
As I like to point out, for two decades DARPA has offered to pay many millions of dollars to anyone who can demonstrate a graph database that can handle a sparse trillion-edge graph. That data model easily fits on a single machine. No one has been able to claim the money.
Inexplicably, major advances in this area 15-20 years ago under the auspices of government programs never bled into the academic literature even though it materially improved the situation. (This case is the best example I've seen of obviously valuable advanced research that became lost for mundane reasons, which is pretty wild if you think about it.)
There are countless smaller graphs for narrow domains that may be <1B edges but many people have the ambition to stitch together these narrow graphs into a larger graph. When stitching graphs together, the number of edges is usually super-linear. A billion edges is kind of considered “Hello World” for system testing.
The Semantic Web companies in the 2000s had graphs that were 100B+ edges. They wanted to go much larger but hit hard scaling walls around that point. That scaling wall killed them.
Classic mapping data models are typically 10-100B edges. These could be much, much larger if they could process all the data available to them.
Of course, intelligence agencies had all kinds of graphs far beyond trillions of edges 20 years ago. People, places, things, events.
Any type of spatiotemporal entity graphs with large geographic scope are quadrillions of edges. It isn’t just a lot of inferred relationships between entities, the relationships evolve over time which also must be captured. These are probably the most commercially valuable type of graph. You could build hundreds of different graphs of this type with 1T+ edges in most regions, never mind doing it at scale. These are so large that we usually don’t store them. Subgraphs are generated on demand, which is computationally expensive.
These spatiotemporal entity graphs also have the largest write loads. Single sources generate tens of PB/day of new edges. There is a ton of industrial data that looks like this; it isn’t just people slinging structured data.
Graphs are everywhere but we furiously avoid them because the scalability of operations over anything but severely constrained graphs is so poor. Selection bias.
NSA in particular heavily funded foundational theoretical and applied computer science research into scaling graph computing for decades. They had all kinds of boring graphs where trillions of edges was their Tuesday. The US military also uses large graph databases in fairly boring applications that probably didn’t require a graph database.
I wonder why no one has claimed it. It's possible to compress large graphs to 1 byte per edge via Graph reordering techniques. So a trillion scale graph becomes 1TB, which can fit into high end machines.
Obviously it won't handle high write rates and mutations well. But with Apache Arrow based compression, it's certainly possible to handle read-only and read-mostly graphs.
Also the single machine constraint feels artificial. For any columnar database written in the last 5 years, implementing object store support is tablestakes.
There is no single machine constraint, just the observation that we routinely run non-graph databases at similar scale on single machines without issue. It doesn't scale on in-memory supercomputers either, so the hardware details are unrelated to the problem:
- Graph database with good query performance typically has terrible write performance. It doesn't matter how fast queries are if it takes too long to get data into the system. At this scale there can be no secondary indexing structures into the graph; you need a graph cutting algorithm efficient for both scalable writes and join recursion. This was solved.
- Graph workloads break cache replacement algorithms for well-understood theory reasons. Avoiding disk just removes one layer of broken caching among many but doesn't address the abstract purpose for which a cache exists. This is why in-memory systems still scale poorly. We've known how to solve this in theory since at least the 1980s. The caveat is it is surprisingly difficult to fully reduce to practice in software, especially at scale, so no one really has. This is a work in progress.
- Most implementations use global synchronization barriers when parallelizing algorithms such as BFS, which greatly increases resource consumption while throttling hardware scalability and performance. My contribution to research was actually in this area: I discovered a way to efficiently use error correction algorithms to elide the barriers. I think there is room to make this even better but I don't think anyone has worked on it since.
The pathological cache replacement behavior is the real killer here. It is what is left even if you don't care about write performance or parallelization.
I haven't worked in this area for many years but I do keep tabs on new graph databases to see if someone is exploiting that prior R&D, even if developed independently.
> There are some additional optimizations that are specific to graphs that a relational DBMS needs to incorporate: [...]
This is essentially what Kuzu implemented and DuckDB tried to implement (DuckPGQ), without touching relational storage.
The jury is out on which one is a better approach.
If you wish to avoid that particular caveat, look for a graph DB which materializes edges within vertices/nodes. The obvious caveat there is that the edges are not normalized, which may or may not be an issue for your particulat application.
Trying to make it optional.
Try
explain match (a)-[b]->(c) return a.rowid, b.rowid, c.rowid;
> We've all seen open source projects drag on for 3 years without shipping anything, that's not necessarily better
There are more options than “never ship anything” and “use AI to slip 200,000 lines of code into a codebase”
The embeddable angle is interesting though. For small-to-medium graphs (< 10M edges) that live in a single process, these scalability concerns matter less, and the developer experience of "just add a dependency" beats running a separate database server.
Author of ArcadeDB critiques many nominally open source licenses here:
https://www.linkedin.com/posts/garulli_why-arcadedb-will-nev...
What is a graph database is also relevant:
https://archive.fosdem.org/2025/schedule/event/fosdem-2025-5...
Full history here: https://www.linkedin.com/pulse/brief-history-graphs-facebook...
When you actually need to run graph algorithms against your relational data, you export the subset of that data into something like Grafeo (embedded mode is a big plus here) and run your analysis.
It's possible to run cypher against duckdb (soon postgres as well via duckdb's postgres extension) without having to import anything. That's a game changer when everything is in the same process.
Typically used with scaleout DBs like databricks & splunk for analytical apps: security/fraud/event/social data analysis pipelines, ML+AI embedding & enrichment pipelines, etc. We originally built it for the compute-tier gap here to help Graphistry users making embeddable interactive GPU graph viz apps and dashboards and not wanting to add an external graph DB phase into their interactive analytics flows.
Single GPU can do 1B+ edges/s, no need for a DB install, and can work straight on your dataframes / apache arrow / parquet: https://pygraphistry.readthedocs.io/en/latest/gfql/benchmark...
We took a multilayer approach to the GPU & vectorization acceleration, including a more parallelism-friendly core algorithm. This makes fancy features pay-as-you-go vs dragging everything down as in most columnar engines that are appearing. Our vectorized core conforms to over half of TCK already, and we are working to add trickier bits on different layers now that flow is established.
The core GFQL engine has been in production for a year or two now with a lot of analyst teams around the world (NATO, banks, US gov, ...) because it is part of Graphistry. The open-source cypher support is us starting to make it easy for others to directly use as well, including LLMs :)
Claude helped a lot but it's all reviewed and curated by me.
(Which may not all be true, but perhaps moreso than your average project)
There's no big mystery. No conspiracy or organised evangelism. Rust is just really good.
I mean - I understand, some people have fun looking at new tech no matter the source, but my question is is there a person who would be designated to pick a GraphQL in language and would ignore all the LLM flags and put it in production.
https://github.com/agnesoft/agdb
Ah, yeah, a different query language.
* it is possible to write high quality software using GenAI
* not using GenAI could mean project won't be competitive in current landscape
From examine this codebase it doesn’t appear to be written carefully with AI.
It looks like code that was promoted into existence as fast as possible.
why? this is false in my opinion, iterating fast is not a good indicator of quality nor competitiveness
Because the latter is really dumb. I don't mind a software written in C, although I personally wouldn't want to write it anymore.