Architecture & Philosophy

How Onyx Works

Onyx is designed around a simple principle: Data work should be expressive, declarative, related, and have minimal constraints.

To make this possible, Onyx introduces an AI-controlled database management system that separates data storage, relationship definition, and data workflow orchestration.

1. A Layered Database Architecture

Traditional databases tightly couple storage, schema, relationships, queries, and analytics. Onyx intentionally separates these concerns.

The Data Plane

Stores data, executes queries, resolves relationships, and manages indexing.

The Relationship Layer

Defines how entities connect using schema-level resolver logic.

The AI Control Plane

Plans, generates, and orchestrates data workflows with humans in the loop.

2. Modeling Data as Connected Graphs

Onyx models data as entities organized into collections and graphs. Unlike traditional SQL databases, relationships are not limited to static foreign keys or join tables.

Unlike many graph databases, relationships are not required to be permanently stored as edges. Instead, relationships are treated as first-class schema constructs that can evolve as data and requirements change.

3. Resolver-Defined Relationships

Relationships are defined by logic, not storage layout.

In Onyx, relationships are defined using resolver queries declared at the schema level. A resolver specifies:

  • how two entities or collections relate
  • the logic used to establish that relationship
  • the shape of the relationship when traversed

Resolvers can connect entities within a single graph, across multiple graphs, or across collections or domains.

4. Resolution Inside the Database

Resolver execution happens inside the database system. Onyx does not federate graphs over the network.

Graph attachment occurs at the model level, using indexed internal structures rather than network-bound joins. Relationship expansion is incremental and locality-aware.

This approach avoids N+1 query patterns while preserving predictable performance on modern SSD-backed systems.

5. Efficient Traversal and Indexing

To maintain performance with flexible relationships, Onyx uses:

  • indexed reference assignment
  • multidimensional hash-based lookup
  • skip-list–backed traversal structures

Relationship resolution is computed once and reused across traversals, enabling efficient multi-hop graph queries.

The AI Workflow Engine

6. From Queries to Data Workflows

Queries are the wrong abstraction for most data work. Onyx replaces this with workflow-based data interaction. A data workflow represents a sequence of steps that express intent, retrieve and transform connected data, and produce meaningful outputs.

7. The AI Control Plane

Onyx introduces an AI-driven control plane that manages data workflows. Using a flash-model AI, the control plane interprets user intent, proposes workflow steps, generates queries, and explains each step.

8. Workflow Trees

Workflows in Onyx are structured as trees. This allows workflows to branch into alternative paths, explore multiple hypotheses, recombine results, and evolve over time.

9. Persisted Scripts and Reuse

Approved workflows can be persisted as scripts. These scripts capture intent and logic, rerun as data changes, serve as living documentation, and can be shared across teams.

10. First-Class Outputs

Workflow execution produces first-class outputs, not just raw rows. Outputs can include statistical summaries, derived datasets, and visualization-ready results.

11. Dashboards Built from Workflows

Dashboards are composed from workflow outputs. This means dashboards inherit workflow logic, changes are explainable, and visualizations remain stable.

12. Human-in-the-Loop

At every stage—modeling, relationship definition, planning, and execution—humans remain in control. The AI acts as an accelerator, not a black box.

What This Enables

Flexible without chaosIntelligent without guessworkReusable by default

“Onyx is not a query language, a BI tool, or an AI assistant bolted onto a database. It is an AI-controlled database management system designed for connected data and repeatable data workflows.”