Live webinar Nov 5: save your seatLive webinar Nov 5: Building Golden Records & Entity Resolution with Fluree. Save your seat
The engine under Fluree AI
The verifiable knowledge graph database.
FlureeDB is a source-available knowledge graph database you can run yourself. Every transaction signed. Every state queryable. Every policy enforced at the triple, inside the engine.
Most databases store records. Fluree stores knowledge: the relationships between entities, the history of every change, and the policies that govern who may see what. It is built on W3C standards all the way down, because your data will outlive every tool you use to query it.
As AI moves from answering questions to taking actions, the hard problem is no longer whether a model can find data. It is whether anyone can reconstruct and defend what it did with it. That needs a ledger where provenance, time and policy are properties of the data itself, not of the application wrapped around it.
We spent eight years building it, and it is the engine under everything else we make. Fluree AI is the managed product on top, with golden records from your tables, graph facts from your documents, one governed vocabulary, and agents that reason over the result.
Ten things most graph databases can’t do.
Side by side with the others- 01ImmutabilityEvery transaction signed and appended. An update is a new commit, never an overwrite.
- 02SecurityABAC, RBAC and relationship policies, evaluated per triple at query time.
- 03ProvenanceJWS-signed transactions. W3C Verifiable Credentials and DIDs, native.
- 04Integrated searchBM25 with Block-Max WAND and HNSW vectors inside the engine. Search joins like any pattern.
- 05ReasoningRDFS, OWL 2 RL and user-defined Datalog rules, materialized on demand.
- 06InteroperabilityFull SPARQL 1.1 with zero W3C test-suite failures, plus openCypher and a GraphQL endpoint on every ledger. JSON-LD, Turtle, TriG, N-Quads.
- 07Time travelQuery by transaction, ISO timestamp or commit hash. Same engine, same indexes.
- 08Git-like branchingFork, rebase, merge and diff for data. Rollback in one command.
- 09FederationIceberg, Parquet, relational via R2RML and remote SPARQL as one graph. No copy.
- 10Embeddable engineBinary, HTTP server, or a Rust library in your process. One engine, three surfaces.
#1 on SPARQLoscope DBLP.
105 queries over 561 million triples, seven engines on matched hardware. Fluree completes every query with a 17.5 ms geometric mean. QLever takes 11.5× longer, Virtuoso 17.1×. At 8.19 billion triples, on Wikidata, QLever’s latency is still 10.4× Fluree’s. Bulk import runs above 2 million triples per second on a single node.
Full report, raw timings and reproduction scripts| Engine | Geometric mean · log scale | Time | Passed |
|---|---|---|---|
| Fluree#1 | 17.5 ms | 105/105 | |
| QLever | 202 ms | 105/105 | |
| Virtuoso | 300 ms | 103/105 | |
| MillenniumDB | 1.66 s | 103/105 | |
| Jena | 67.7 s | 34/105 | |
| Oxigraph | 87.0 s | 39/105 | |
| Blazegraph | 333 s | 3/105 |
Geometric mean latency over the 105 DBLP-core queries, log scale, lower is better. Failed queries count as twice the 180-second timeout. Fluree v4.2.0 measured September 2026; other engines June 2026, same instance class and dataset.
Install. Query. Ship.
No JVM, no Python environment, no ZooKeeper. Pick a package manager and you are querying a knowledge graph in under a minute, in SPARQL, openCypher, GraphQL or JSON-LD, on the same engine. Moving off Neo4j? Most of your Cypher queries carry over. See Fluree vs. Neo4j, or six Neo4j alternatives compared.
Read the quickstart# install the single binary brew install fluree/tap/fluree # create a ledger fluree create movies # insert a few triples fluree insert 'ex:alien a schema:Movie ; schema:name "Alien" .' # query fluree query 'SELECT ?title WHERE { ?m a schema:Movie ; schema:name ?title }' → "Alien" # the same question, as of any moment fluree query --at 2026-09-08T09:00Z '…'
On-prem, air-gapped
Docker images or a single binary. Full data sovereignty.
Your cloud
AWS, Azure or GCP. S3 + DynamoDB for HA production.
Fluree managed
Single-tenant SaaS, region selection, failover across zones.
FlureeDB is source-available under the Business Source License 1.1: free in production, including inside your own product; a commercial license is needed only to offer it as a hosted database service. Each release converts to Apache 2.0 on its fourth anniversary.
Put it on your data.
Query a knowledge graph in under a minute with the single binary, or bring a real dataset and we will stand up a governed graph with you.
