The graph database, rethought. Faster. · Smaller. · A fraction of the cost.
172 million edges from a single 3.6 GB file. 14/14 LDBC Interactive Complex queries at SF10 scale. 10× less RAM than the industry standard. Full openCypher, full Bolt protocol, $299/month flat. No per-GB scaling. No cluster required.
Head-to-Head
The benchmark Neo4j doesn’t want you to see.
We ran 105 Cypher queries. Same hardware. Same Bolt protocol. We gave Neo4j 2× the RAM budget.
Neo4j got 1 GB and 1 CPU. TETRA got 512 MB and 1 CPU. Both containerized. Both running the Recommendations dataset (28,863 nodes, 166,261 edges). Both speaking Bolt.
37 wins for TETRA. 19 for Neo4j. Neo4j OOM-crashed on every write.
Neo4j’s 19 wins are concentrated in multi-variable RETURN projections and OPTIONAL MATCH chains — known planner optimization targets for TETRA, not architectural limits. We publish them alongside our wins because you deserve to know.
Concurrency
Neo4j’s throughput collapses. TETRA’s stays flat.
Mixed workload. 1 to 64 concurrent Bolt clients. Neo4j got 1 GB of RAM. TETRA got 512 MB.
The JVM garbage collector fights the memory limit. Throughput degrades. Eventually the process dies. This is not a configuration problem — it’s the architecture.
TETRA has no JVM, no GC pauses, no heap pressure. When you eventually exhaust the CPU, performance degrades linearly. No cliffs. No crashes.
LDBC SF10
172M edges. Single file. 3.6 GB on disk.
While Neo4j was busy OOMing, we ran the hardest graph benchmark in the industry.
The LDBC Social Network Benchmark is the industry-standard test for graph databases. Scale Factor 10: 27.2 million nodes and 172.2 million edges.
The 14 Interactive Complex queries include variable-length path traversals, 7-hop chain queries through class hierarchies, correlated anti-joins, and bidirectional BFS across 68,000 persons.
Neo4j’s own wire protocol. Our engine. Single process. No cluster. No auxiliary infrastructure. Run it on a laptop, a VM, or alongside your app server.
Pricing
You’re paying Neo4j by the gigabyte. Forever.
Neo4j AuraDB Professional charges $65 per GB of RAM per month. Business Critical doubles that to $146/GB/month. Self-hosted Enterprise is “Contact Sales” — third-party data says $20,000 to $200,000+ per year.
Then come the add-ons. Bloom (visualization): $1,200–$2,500 per user per year. Graph Data Science: $10,000–$25,000+ per year.
TETRA is $299/month flat. Retina included. 30+ algorithms included. Full openCypher included. No per-GB scaling. No tier upgrades. No “Contact Sales.”
Monthly pricing
5-year cumulative cost
By year five, you’ve paid Neo4j $63,000 to crash on concurrent writes. The same workload on TETRA costs $17,940 — and it doesn’t crash.
Migration
Migration in an afternoon.
You’re already writing Cypher. TETRA implements 100% of the openCypher Technology Compatibility Kit — 1,611 of 1,611 scenarios. Full Bolt v4.4 protocol. Point your existing Neo4j driver at TETRA’s endpoint. Your application doesn’t know the difference.
Migration services: $10,864 one-time. Full schema conversion, CSV or JSON ingest, and validation against your existing Neo4j output. 10 GB of CSV data processed in under 7 minutes on 5 GB of RAM.
Or migrate yourself. The wire protocol is compatible.
Included
What $299 actually includes.
Everything Neo4j charges extra for. Included.
Retina — 3D Graph Viewer
WebGPU-driven 2D/3D graph explorer with server-side and client-side analytics. Run shortest path, verify migrations, visualize your data. Export to CSV or JSON.
Neo4j Bloom: $1,200–$2,500 per user per year (self-hosted).
30+ Graph Algorithms
Community detection, centrality, embeddings, pathfinding — built in.
Neo4j Graph Data Science: $10,000–$25,000+ per year.
Full openCypher
1,611/1,611 TCK scenarios. Bolt v4.4 protocol. Drop-in with existing Neo4j drivers.
Neo4j’s own standard. We pass 100% of it.
Post-Quantum Encryption
Files encrypted at rest with post-quantum resistant cryptography. The encrypted file IS the queryable database.
Neo4j: standard encryption only.
~38% Compression
Raw data compresses to ~38% of original, including all structures needed to query it. No separate indexes. Zero data resolution loss (formally proven).
Neo4j: 5–10× overhead from indexes, logs, caches, replication layers.
Co-location Ready
66 MB of RAM. 24 MB Alpine container. Runs alongside your app — no cluster, no network hop.
Neo4j: requires dedicated cluster, JVM, 710 MB+ for equivalent workload.
Architecture
One engine. Everything Neo4j needs a stack for.
A traditional graph database deployment is five systems, not one:
TETRA is one binary. One file. One process. The compressed, encrypted file is the query engine. Retina is built in. The algorithms are built in. Post-quantum encryption is built in.
You don’t manage a stack. You run a binary.
Transparency
19 queries. We publish them too.
Out of 105 Cypher queries, Neo4j beats TETRA on 19. They’re concentrated in multi-variable RETURN projections and OPTIONAL MATCH chains — known query planner optimization targets in our upcoming releases.
We publish them for the same reason we publish everything else: if you’re going to trust a database with your data, you deserve to see where it falls short. Not just where it wins.
See Every Query, Every Result →Partners
Building the future together.
Everyone else used this math to build language models.
We used it to build a database.