HydraDB vs Neo4j
Neo4j makes sense when your team designs the graph and its ontology, then loads data into it. HydraDB is worth a look when your agents work across a large body of unstructured data that keeps changing, because the graph and its ontology form from that data and evolve as it does.
Based on public docs and pricing as of Nov 2026. Something out of date? Suggest a correction →
Neo4j is the most popular graph database by a wide margin. It created Cypher that shaped GQL. Its tooling is the most mature in the market, having been in production since 2010 and it’s the go-to for fraud, identity and recommendation graphs, where the schema is settled and the data can be modelled and loaded ahead of the questions you'll ask.
The friction shows up when the entities aren't clear, the relationships keep moving, and the data arrives as messages and documents you can't map ahead of time.
The caveats worth knowing
It expects a schema, and someone who knows Cypher. You decide the nodes and relationships before loading anything. Then you map every source onto that schema and write Cypher against it. Neo4j's text-to-Cypher tools help with queries, but they work off the schema you designed.
You pay memory prices for data you rarely touch. Neo4j's cloud bills per GB of memory, with storage bundled in. An agent's context mostly grows and mostly sits cold, but you size the instance for all of it.
Agent memory is write-heavy. Every message, ticket and update is a write. Reads scale out with replicas, but writes go through a single leader, and sharding is a separate subscription.
What changes with HydraDB?
No schema or Cypher required up front. Connect Slack, Gmail, Drive, Jira or any of 100+ sources, and the graph is created as data arrives, structured to your ontology or one proposed for you. It updates incrementally and restructures when new kinds of facts appear. Agents query in natural language, and openCypher works if your team prefers it.
Connect 3 apps, export 1 table, keep 1 export fresh, then ask.
Design the model your queries rely on, write 3 imports, map 4 sources onto your schema, set up extraction for 3 sources, keep 3 imports running, run 1 load, then write Cypher.
Only tables? Neo4j is the safe pick: your schema is settled, the data arrives structured, and its tooling has been in production since 2010.
Storage stops driving the bill. The graph lives on object storage, with hot data cached in memory and NVMe. Months of cold Slack history cost $0.50 per GB a month on the managed cloud (as opposed to about $65 per GB).
Agents can find what changed, when, and why. Every fact is versioned like code in Git, with when, why and where it came from. So "what was the refund policy before last week, and who changed it?" has an answer.
What was the refund window on 15 Sep?
HydraDB: 14 days, since 12 Mar. First written policy, from policy.pdf.
Neo4j: returns today's value, 30 days. An update replaces the value. To keep history, model versions yourself. Neo4j versioning guide ↗
Technical comparison
| Neo4j | HydraDB | |
|---|---|---|
| Architecture | Native graph store on disk, JVM-based, with an in-memory page cache | Written in Rust; the graph lives on object storage, with memory and NVMe as caches |
| Scaling | Read replicas; writes through a single leader; sharding via Infinigraph, a separate subscription | Compute and storage scale separately |
| Building the graph | You design the schema and write the imports; LLM Graph Builder (Labs) for documents | 100+ app connectors and file uploads, structured to your ontology or a proposed one, then updated incrementally |
| Protocols and drivers | Bolt, HTTP, official drivers for major languages | Bolt 5.x (works with Neo4j drivers), HTTPS JSON API, Python and TypeScript SDKs |
| Retrieval | Cypher, plus vector and full-text indexes | Vector search, BM25 keyword search and graph traversal combined in one query, with metadata filters and recency weighting |
| Graph algorithms | Graph Data Science library (centrality, community detection, similarity and more) | Native path procedures (shortest and bounded paths); no general algorithm library |
| When a fact changes | New value replaces the old | Every version kept, Git-style |
| Managed pricing | $0.09 per GB of memory per hour (Professional); Enterprise on request | Managed from $25/month: $0.70 per 1M tokens ingested, $0.50 per GB-month storage; open source free (pricing) |
HydraDB speaks Bolt 5.x and works with Neo4j drivers. Pick your language or paste your own code.
What changes
Pasted code is matched against patterns (driver URIs, auth calls, gds.*, apoc.*), not parsed. It stays in your browser.