Agents need memory, not just access.
The design keeps data next to the agent runtime rather than behind API calls to stale exports. The twin is meant to be the context window. Agents are in development.
The people building OmniTwin spent their careers working in network infrastructure, and kept running into the same unsolved problem from the inside. OmniTwin is in private beta, now open, and this site says plainly which parts of it are wired and which are not.
Network infrastructure management hasn't kept pace with the complexity it serves. Teams still reconcile spreadsheets against CLI output. Drift between intended state and actual state goes undetected until something breaks.
OmniTwin was built to close that gap - a network twin that models topology, with drift detection and AI agents that operate on a shared model of the network both in development.
Kortesa Labs was founded by product, engineering and technical marketing executives out of Tier 1 network providers, the carriers whose backbones other networks are built on top of. Between them they have spent their careers inside operations at that scale, and inside the infrastructure Fortune 500 companies depend on to stay up.
OmniTwin did not come out of a market study. It came out of watching network engineers, DevOps teams and systems engineers do the same work by hand, year after year. Reconciling an IPAM record against what DHCP actually leased. Tracing what a subnet change would break, in their head, before making it. Rebuilding a picture of the estate from spreadsheets and vendor exports that went stale the day they were produced.
The pattern was always the same. The data existed. It was just in five systems, none of which agreed with each other, and none of which could answer a question about any of the others. Building something resilient on that foundation meant doing the reconciliation yourself, every time, and being right every time.
The design keeps data next to the agent runtime rather than behind API calls to stale exports. The twin is meant to be the context window. Agents are in development.
Intent (Postgres) and reality (a graph database) are structurally different. Forcing both into one schema hides drift. We keep them separate on purpose.
An HTTP webhook and gRPC are available today. An agent protocol is planned. No proprietary SDK required.
The goal is one platform from a single binary on a laptop to air-gapped on-premises, with the same data model throughout. Air-gapped deployment is on the roadmap.
The hard problems are graph reconciliation and drift detection, not dashboards. The interface is a command bar, not a control panel.
OmniTwin is built on open foundations: Postgres, open-source graph databases, the Go and Rust ecosystems. These projects represent decades of collective engineering work, and we take that seriously.
As the platform matures, we plan to contribute tooling back - particularly around graph reconciliation, CIDR arithmetic libraries, and agent protocol definitions. The infrastructure community gave us the building blocks. We intend to return useful primitives.
Postgres 17 · Graph DB · Rust math · Go control plane · gRPC
We're onboarding the first teams now. If your network is where AI will operate, we'd like to talk.
Join the beta