One engine, one audit trail: graph-native storage with bitemporal reasoning, causal traceability, hybrid retrieval, and federated query, deployed inside your perimeter.
Graph-native. Sovereign. Auditable. The foundation your AI stack should have started with.
Columnar in-memory storage engineered for traversal, with measured 9µs server-side writes.
Reconstruct what was known at any point in time: no snapshots, no restore jobs.
CAUSED_BY edges make explanation a traversal, not a log parse.
HNSW vector + BM25 full-text + k-hop traversal in one query surface.
One Cypher surface over PostgreSQL, MongoDB, MySQL, Neo4j, and S3 with predicate pushdown, and no migration.
Every change observable; drives downstream audit and reactivity.
When BSA/AML examiners ask why, the answer is a query: replayable, inspectable, and identical every time it runs.
// A regulator asks: why was this transaction flagged? MATCH (t:Transaction {id: $flagged_id}) MATCH path = (t)-[:CAUSED_BY*1..5]->(reason) RETURN path, reason.source, reason.recorded_at // → the exact evidence chain, replayable at examination time
Benchmarked on official LDBC FinBench SF1 data with identical row counts, not synthetic best cases.
Every acknowledged write survives kill -9. High availability replicates the WAL cross-host
over HTTP with replica lag metrics and one-command promotion. All 7 failover checks pass,
including durability of the promoted node.
Cypher surface: 79/79 audited features passing in both storage modes; conformance measured continuously against the official openCypher TCK.
The classic graph workloads: the ones relational joins and vector lookups struggle with.
Multi-hop money flows, shared identifiers, ring detection, all in real time. Patterns that stay invisible in tables surface as paths.
Entities and relationships as the substrate for search, reasoning, and GenAI grounding. One model of the business, queryable by people and machines.
Entity resolution across CRM, ERP, and support systems into one golden record. Relationship context does the disambiguating.
Co-purchase and behaviour graphs drive relevance. Every suggestion comes with an explainable “because” path.
Supplier → part → product dependencies as a live graph. Single-source risk and ripple analysis become one query.
Topology-aware root cause: what depends on the failing thing. Blast radius before you reboot, not after.
Who can reach what, through which grants: access paths as queries. Entitlement creep shows up as a traversal, not an audit season.
Retrieval that returns paths and citations, not just similar text. Answers your AI can defend, not just produce.
Source → transformation → consumer graphs across your estate. “What breaks if this changes?” answered before the change ships.
A vector store can find a similar paragraph. It cannot tell you why, and “why” is the question the examiner asks.
| Vector RAG | ATHERGraph | |
|---|---|---|
| Retrieval basis | Statistical similarity of embeddings: “this text feels related” | ✓ Explicit relationships: this entity is connected to that one, typed and directed |
| Explaining an answer | Re-prompt the model and hope the story is consistent | ✓ Click prediction → walk CAUSED_BY edges to source |
| Point-in-time truth | Embeddings have no clock, so yesterday's truth contaminates today's answers | ✓ Bitemporal layer: reconstruct exactly what was known, when |
| Multi-hop reasoning | Chunk stuffing; hops beyond one document degrade fast | ✓ Native k-hop traversal: account → device → counterparty → source, in one query |
| Audit artifact | A transcript of what the model said | ✓ A replayable graph path a regulator can inspect |
| Data residency | Typically a managed cloud vector service | ✓ Sovereign: on-prem or air-gapped, and data never leaves your perimeter |
Scoped to your deployment. Talk to us and we'll give you a number, not a maze.
A 30-minute walkthrough on real data: the flag, the CAUSED_BY chain, and the examination-ready artifact at the end of it.