Intent vs. Reality: Bridging the Structural Gap Between Configuration State and Physical Topology
Every network engineer has hit this at least once. You look at an official network inventory document or a central tracking ledger, and it tells you one thing. Then you log into a physical core switch CLI, execute a manual diagnostic command, and find something entirely different. The static documentation states that a specific port is completely isolated, but the physical, live reality shows it is actively broadcasting traffic across a multi-tenant corridor.
That gap between operational intent and physical reality is a common source of unexpected downtime, routing loops, and hidden vulnerabilities. Legacy network management tools struggle with it because they treat two different kinds of data as one. They attempt to force the abstract, relational rules of network design and the fluid, interconnected physical topology of physical hardware cables into a single software storage model.
Why a Single Database Cannot Support a Network Digital Twin
The data describing what a network should do is fundamentally different from the mathematical data describing how packets actually traverse physical circuits. Forcing one database engine to handle both workloads tends to give slow multi-hop queries, fragile table designs, and no practical way to detect configuration drift before an operator reports an outage.
Managing infrastructure rules demands a database built for transactional safety. Conversely, mapping a live topology demands a system optimized for rapid path tracing and multi-hop relationship indexing.
OmniTwin solves this structural challenge by engineering a dual-engine data tier. Instead of relying on a single, compromised storage architecture, our platform uses two specialized database engines that are kept in sync. Intent and live reality are stored separately.
The OmniTwin Dual-Engine Architecture
To maintain a computational twin that can validate network changes before they ship, OmniTwin pairs a relational engine with a native graph database.
| Database Engine | Architecture Role inside OmniTwin | Key Data Types & Operations |
|---|---|---|
| PostgreSQL 17 | The Source of Intent | Tracks tenant spaces, multi-tenant boundaries, approved IP prefixes, and administrative access control policies. |
| Graph Database | The Source of Reality | Maps live interface states, routing neighbors, VLAN propagation boundaries, and physical hardware links as explicit nodes and edges. |
1. The Intent Tier: PostgreSQL 17 & Semantic Vectors
PostgreSQL is the anchor for transactional rules. Within this layer, OmniTwin stores explicit system policies, multi-tenant virtual routing and forwarding (VRF) schemas, and authorized IPAM boundaries.
We also use pgvector to store semantic vector embeddings. These embeddings translate high-level engineering requirements and written compliance specifications into structured mathematical data objects. The goal is for the Go-based control plane to interpret the intent behind a configuration, so that it knows what the network is authorized to look like at any point in time. The vector-based interpretation is in development; the relational intent tier is what the platform relies on today.
2. The Reality Tier: Topology Graphs
Instead of forcing deeply nested networking pathways into rigid database rows and columns, the graph tier treats every physical switch, router, cloud gateway, and local database container as an independent node.
The physical and logical connections between these devices are mapped as explicit relationship edges. This graph matrix tracks real-time link states and telemetry anomalies. When our automated agents execute high-throughput path traversal algorithms to simulate how a packet moves across the fabric, the graph engine computes the path across the multi-vendor topology.
Eliminating State Drift via Continuous Reconciliation
The core of the OmniTwin control plane is the reconciliation loop running between these two distinct environments. Our background daemons constantly cross-reference the relational intent state in PostgreSQL with the physical topology reality held in the graph tier.
How the Sync Engine Prevents Downtime:
If an unauthorized configuration change occurs on a remote switch, or if a physical path drift violates a security embedding, OmniTwin catches the delta at the software layer. By comparing runtime topology against the validated mathematical model, the engine identifies the structural drift, so engineers can be alerted before the mismatch causes an outage. Automated drift detection is in development.
By decoupling intent from reality at the database level, OmniTwin gives infrastructure teams one view of what the network is meant to be and one of what it is, with the differences between them made explicit.